1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.33 (SLES 10),
5 2.6.18-128.1.1.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: 1.41.5.sun2
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
18 Descriptoin: don't return error if have particaly created objects for file.
19 Details : lov_update_create_set uses set->set_success as index for created objects,
20 so if some requests will be failed, they will have hole at end of
21 array and we can use qos_shrink_lsm for allocate correct lsm.
23 Severity : enhancement
25 Description: Update OFED support to 1.4.1
29 Description: Reduce small size read RPC
30 Details : Set read-ahead limite for every file and only do read-ahead when
31 available read-ahead pages are bigger than 1M to avoid small size
34 Severity : enhancement
36 Description: move AT tunable parameters for more consistent usage
37 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
39 Severity : enhancement
41 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
42 Details : allow skip disconnected ost for send statfs request and hide error
48 Description: fix racy locking of mballoc block bitmaps causing BUG
49 Details : The locking of the mballoc buddy bitmap and the in-memory
50 block bitmap was using two different spin locks in some
51 cases. This made it possible to incorrectly access the
52 mballoc bitmap while another process was modifying it,
53 causing a sanity assertion to fail. While no on-disk corruption
54 was reported, there was some risk of this happening.
57 Frequency : rare, on llog test 6
59 Descriptoin: don't allow connect to already connected import
60 Details : allowing connect to already connected import is hide connecting problem.
65 Descriptoin: allow kill process which wait statahead result
66 Details : for some reasons 'ls' can stick in waiting result from statahead,
67 in this case need way for kill this process.
69 Severity : enhancement
71 Description: Add state history info file, enhance import info file
72 Details : Track import connection state changes in a new osc/mdc proc file;
73 add overview-type data to the osc/mdc import proc file.
75 Severity : enhancement
77 Description: MDS create should not wait for statfs RPC while holding DLM lock.
79 Severity : enhancement
81 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
86 Descriptoin: OST grant too much space to client even there are not enough space.
87 Details : Client will shrink its grant cache to OST if there are no write
88 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
89 this grant cache if there are already not enough avaible space
90 (left_space < total_clients * 32M).
93 Frequency : start MDS on uncleanly shutdowned MDS device
95 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
96 Details : stay in waiting mds<>ost recovery finished produce random bugs
97 due race between two ll_sync thread for one lov target. send
98 ACTIVATE event only if connect realy finished and import have
102 Frequency : rare, connect and disconnect target at same time
104 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
105 Details : don't call obd_disconnect under lov_lock. this long time
106 operation and can block ptlrpcd which answer to connect request.
111 Descriptoin: don't lose wakeup for imp_recovery_waitq
112 Details : recover_import_no_retry or invalidate_import and import_close can
113 both sleep on imp_recovery_waitq, but we was send only one wakeup
117 Frequency : always with long access acl
119 Descriptoin: mds can't pack reply with long acl.
120 Details : mds don't control size of acl but they limited by reint/getattr
123 Severity : enhancement
125 Description: Update to SLES10 kernel-2.6.16.60-0.33.
127 Severity : enhancement
129 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
132 Frequency : start MDS on uncleanly shutdowned MDS device
134 Descriptoin: aborting recovery hang on MDS
135 Details : don't throttle destroy RPCs for the MDT.
138 Frequency : on remount
140 Description: external journal device not working after the remount
141 Details : clear dev_rdonly flag for external journal devices in
147 Description: shutdown vs evict race
148 Details : client_disconnect_export vs connect request race.
149 if client will evicted at this time - we start invalidate
150 thread without referece to import and import can be freed
154 Frequency : rare, need acl's on inode.
156 Description: client can't handle ost additional correctly
157 Details : if ost was added after client connected to mds client can have
158 hit lnet_try_match_md ... to big messages to wide striped files.
159 in this case need teach client to handle config events about add
160 lov target and update client max ea size at that event.
162 Severity : enhancement
164 Description: Changelogs
165 Details : Changelogs are a lightweight mechanism to track filesystem
166 metadata and namespace changes. The changelog is recorded
167 permanently on the MDTs, and is periodically "consumed" / purged
168 when records are no longer needed.
170 Severity : enhancement
172 Description: compact fld format with extents
173 Details : Store range of seq rather than every seq in FLD. Seq
174 controller update FLD rather than clients. In Case of CMD, mdt0
175 has FLD, all other metadata server act as non persistent proxy
176 for FLD queries and cache fld entries in fld cache.
181 Description: don't skip ost target if they assigned to file
182 Details : Drop slow OSCs if we can, but not for requested start idx.
183 This means "if OSC is slow and it is not the requested
184 start OST, then it can be skipped, otherwise skip it only
185 if it is inactive/recovering/out-of-space.
189 Description: more cleanup in mds_lov
190 Details : not send LOV EA under replay, we can't know about they size at this
191 time. Don't allow client connect to mds before any ost connected,
192 for avoid problems with LOV EA size and returning EIO to client.
194 Severity : enhancement
196 Description: Interoperability at server side (Disk interoperability)
198 Severity : enhancement
200 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
202 Severity : enhancement
204 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
206 Severity : enhancement
208 Description: New client IO stack (CLIO).
210 Severity : enhancement
212 Description: Commit on sharing. Eliminate inter-client dependencies between
213 uncommitted transactions by doing transaction commits.
214 Thereby clients may recovery independently.
217 Frequency : Create a symlink file with a very long name
219 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
220 Details : If there is no extra space in the request for early cancels,
221 ldlm_req_handles_avail() returns 0 instead of a negative value.
223 Severity : enhancement
225 Description: Add /proc entry for import status
226 Details : The mdc, osc, and mgc import directories now have
227 an import directory that contains useful import data for debugging
230 Severity : enhancement
232 Description: Re-disable certain /proc logging
233 Details : Enable and disable client's offset_stats, extents_stats and
234 extents_stats_per_process stats logging on the fly.
237 Frequency : Only on FC kernels 2.6.22+
239 Description: oops in statahead
240 Details : Do not drop reference count for the dentry from VFS when lookup,
241 VFS will do that by itself.
243 Severity : enhancement
245 Description: Generic /proc file permissions
246 Details : Set /Proc file permissions in a more generic way to enable non-
247 root users operate on some /proc files.
251 Description: Hitting mdc_commit_close() ASSERTION
252 Details : Properly handle request reference release in
253 ll_release_openhandle().
257 Description: quota recovery deadlock during mds failover
258 Details : This patch includes att18982, att18236, att18237 in bz14840.
260 1. fix osts hang when mds does failover with quotaon
261 2. prevent watchdog storm when osts threads wait for the
266 Frequency : only patchless client
267 Description: add workaround for race between add/remove dentry from hash
269 Severity : enhancement
271 Description: Allow OST glimpses to return PW locks
275 Description: LBUG when llog conf file is full
276 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
280 Description: Prevent import from entering FULL state when server in recovery
284 Description: service mount cannot take device name with ":"
285 Details : Only when device name contains ":/" will mount treat it as
291 Description: replace ptlrpcd with the statahead thread to interpret the async
292 statahead RPC callback
296 Frequency : on recovery
297 Description: I/O failures after umount during fail back
298 Details : if client reconnected to restarted server we need join to recovery
299 instead of find server handler is changed and process self
300 eviction with cancel all locks.
302 Severity : enhancement
304 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
306 Severity : enhancement
308 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
310 Severity : enhancement
312 Description: Upcall on Lustre log has been dumped
313 Details : Allow for a user mode script to be called once a Lustre log has
314 been dumped. It passes the filename of the dumped log to the
315 script, the location of the script can be specified via
316 /proc/sys/lnet/debug_log_upcall.
321 Description: avoid idr_remove called for id which is not allocated.
322 Details : Move assigment s_dev for clustered nfs to end of initialization,
323 to avoid problem with error handling.
328 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
329 Details : When connection is reused this not moved from CONN_UNUSED_HASH
330 into CONN_USED_HASH and this prodice warning when put connection
331 again in unused hash.
334 Severity : enhancement
336 Description: File striping can now be set to use an arbitrary pool of OSTs.
338 Severity : enhancement
340 Description: Export bytes_read/bytes_write count on OSC/OST.
344 Description: Early reply size mismatch, MGC loses connection
345 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
346 the connect flags are properly negotiated.
351 Description: quotas are not honored with O_DIRECT
352 Details : all writes with the flag O_DIRECT will use grants which leads to
353 this problem. Now using OBD_BRW_SYNC to guard this.
357 Description: add quota statistics
358 Details : 1. sort out quota proc entries and proc code.
359 2. add quota statistics
361 Severity : enhancement
363 Description: enable quota support for HEAD.
367 Description: Properly propagate oinfo flags from lov to osc for statfs
368 Details : restore missing copy oi_flags to lov requests.
370 Severity : enhancement
372 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
373 : l_getgroups(8), lst(8), routerstat(8)
374 Details : included man pages for llobdstat(8), llstat(8),
375 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
377 Severity : enhancement
379 Description: configure's --enable-quota should check the
380 : kernel .config for CONFIG_QUOTA
381 Details : configure is terminated if --enable-quota is passed but
382 : no quota support is in kernel
386 Description: Remove portals compatibility
387 Details : Remove portals compatibility, not interoperable with releases
392 Description: Resolve device initialization race
393 Details : Prevent proc handler from accessing devices added to the
394 obd_devs array but yet be intialized.
396 Severity : enhancement
398 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
400 Severity : enhancement
402 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
407 Description: Using wrong pointer in osc_brw_prep_request
408 Details : Access to array[-1] can produce panic if kernel compiled with
409 CONFIG_PAGE_ALLOC enabled
413 Description: Client runs out of low memory
414 Details : Consider only lowmem when counting initial number of llap pages
418 Description: *optional* service tags registration
419 Details : if the "service tags" package is installed on a Lustre node
420 When the filesystem is mounted, a local-node service tag will
421 be created. See http://inventory.sun.com/ for more information
422 about the Service Tags asset management system.
426 Description: Kernel BUG tries to release flock
427 Details : Lustre does not destroy flock lock before last reference goes
428 away. So always drop flock locks when client is evicted and
429 perform unlock regardless of successfulness of speaking to MDS.
433 Description: add refcount protection for osc callbacks, avoid panic on shutdown
437 Description: sanity test 65a fails if stripecount of -1 is set
438 Details : handle -1 striping on filesystem in ll_dirstripe_verify
443 Description: ASSERTION(CheckWriteback(page,cmd)) failed
444 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
447 Severity : enhancement
449 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
453 Description: do not process already freed flock
454 Details : flock can possibly be freed by another thread before it reaches
455 to ldlm_flock_completion_ast.
459 Description: LBUG during stress test
460 Details : Need properly lock accesses the flock deadlock detection list.
464 Description: oops in page fault handler
465 Details : kernel page fault handler can return two special 'pages' in error
466 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
470 Description: timeout with invalidate import.
471 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
472 should be handled by ptlrpcd. This produce long age waiting and
473 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
475 Severity : enhancement
477 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
481 Description: enable MGS and MDT services start separately
482 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
483 a co-located MGS without starting the MGS, which is a complement
484 to 'nosvc' mount option.
488 Frequency : after recovery
489 Description: precreate to many object's after del orphan.
490 Details : del orphan st in oscc last_id == next_id and this triger growing
491 count of precreated objects. Set flag LOW to skip increase count
492 of precreated objects.
496 Frequency : rare, on clear nid stats
497 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
498 Details : when clean nid stats sometimes try destroy live entry,
499 and this produce panic in free.
503 Description: Stack overflow during MDS log replay
504 ease stack pressure by using a thread dealing llog_process.
508 Description: wait until IO finished before start new when do lock cancel.
509 Details : VM protocol want old IO finished before start new, in this case
510 need wait until PG_writeback is cleared until check dirty flag and
511 call writepages in lock cancel callback.
513 Severity : enhancement
515 Description: using special macro for print time and cleanup in includes.
519 Description: mds_mfd_close() ASSERTION(rc == 0)
520 Details : In mds_mfd_close(), we need protect inode's writecount change
521 within its orphan write semaphore to prevent possible races.
525 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
529 Frequency : rare, on shutdown ost
530 Description: don't hit live lock with umount ost.
531 Details : shrink_dcache_parent can be in long loop with destroy dentries,
532 use shrink_dcache_sb instead.
536 Description: don't panic with use echo client
537 Details : echo client pass NULL as client nid pointer and this produce null
542 Description: fix build on ppc32
543 Details : compile code with -m64 flag produce wrong object file for ppc32.
547 Description: add message levels for liblustreapi
551 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
552 Details : Move the 'good_osts' check before the 'total_bavail' check. This
553 will result in an -EAGAIN and in the exit call path we call
554 alloc_rr() which will with increasing aggressiveness attempt to
555 aquire precreated objects on the minimum number of required OSCs.
559 Description: Use old size assignment to avoid deadlock
560 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
561 the scheduling while holding a spinlock. We do not need locking
562 for size in ll_update_inode() because size is only updated from
563 the MDS for directories or files without objects, so there is no
564 other place to do the update, and concurrent access to such inodes
565 are protected by the inode lock.
569 Description: resolve "_IOWR redefined" build error on SLES10
573 Description: dump the memory debugging after all modules are unloaded to
574 suppress false negative in conf_sanity test 39
576 Severity : enhancement
578 Description: build kernel-ib packages for OFED 1.3 in our release cycle
583 Description: fix SLES kernel versioning
584 Details : the kernel version for our SLES 10 kernel did not include a "-"
585 before the "smp" at the end. while this was not a problem in
586 general, it did mean that software trying to use the kernel
587 version to try to detect a vendor specific kernel would fail.
588 this was most evident by the OFED build scripts.
592 Description: Don't update lov_desc members until making sure they are valid
593 Details : When updating lov_desc members via proc fs, need fix their
594 validities before doing the real update.
598 Description: don't put request into delay list while invalidate in flight.
599 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
600 import in flight. this produce timeout for invalidate and sometimes
601 can cause stale data.
605 Frequency : on ppc only
606 Description: not convert ost objects for directory because it's not exist.
607 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
611 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
613 Severity : enhancement
615 Description: Update to support 2.6.22.14 vanilla kernel.
619 Frequency : rare, on recovery
620 Description: read procfs can produce deadlock in some situation
621 Details : Holding lprocfs lock which send rpc can produce block for destroy
622 obd objects and this also block reconnect with -EALREADY. This
623 isn't fix all lprocfs bugs - but make it rare.
625 Severity : enhancement
627 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
630 Frequency : frequent on X2 node
632 Description: mdc_set_open_replay_data LBUG
633 Details : Set replay data for requests that are eligible for replay.
637 Description: lustre_mgs: operation 101 on unconnected MGS
638 Details : When MGC is disconnected from MGS long enough, MGS will evict the
639 MGC, and late on MGC cannot successfully connect to MGS and a lot
640 of the error messages complaining that MGS is not connected.
644 Frequency : on network error
645 Description: panic with double free request if network error
646 Details : mdc_finish_enqueue is finish request if any network error ocuring,
647 but it's true only for synchronus enqueue, for async enqueue
648 (via ptlrpcd) this incorrect and ptlrpcd want finish request
651 Severity : enhancement
653 Description: client-side metadata stat-ahead during readdir(directory readahead)
654 Details : perform client-side metadata stat-ahead when the client detects
655 readdir and sequential stat of dir entries therein
658 Frequency : on start mds
660 Description: Implement get_info(last_id) in obdfilter.
663 Frequency : occasional
665 Description: Correctly check stale fid, not start epoch if ost not support SOM
666 Details : open with flag O_CREATE need set old fid in op_fid3 because
667 op_fid2 was overwritten with new generated fid, but mds can answer
668 with one of these two fids and both is not stale. Setattr
669 incorrectly started an epoch and assume will be called
670 done_writing, but without SOM done_writing ever being called.
673 Frequency : rare, depends on device drivers and load
675 Description: MDS or OSS nodes crash due to stack overflow
676 Details : Code changes in 1.8.0 increased the stack usage of some functions.
677 In some cases, in conjunction with device drivers that use a lot
678 of stack the MDS (or possibly OSS) service threads could overflow
679 the stack. One change which was identified to consume additional
680 stack has been reworked to avoid the extra stack usage.
683 Frequency : occasional
685 Description: Do not fail import if osc_interpret_create gets -EAGAIN
686 Details : If osc_interpret_create got -EAGAIN it immediately exits and
687 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
688 oscc_has_objects and see OSC has no objests and call
689 oscc_internal_create to resend create request.
691 Severity : enhancement
693 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
695 Severity : enhancement
697 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
700 Frequency : very rare
702 Description: Fix replay if there is an un-replied request and open
703 Details : In some cases, older replay request will revert the
704 mcd->mcd_last_xid on MDS which is used to record the client's
707 Severity : enhancement
709 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
711 Severity : enhancement
713 Description: Add rhel5 support to HEAD.
715 Serverity : enhancement
717 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
722 Description: Don't allow skipping OSTs if index has been specified.
723 Details : Don't allow skipping OSTs if index has been specified, make
724 locking in internal create lots better.
728 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
729 Details : access to bitfield in structure is always rounded to long
730 and this produce problem with not atomic change any bit.
734 Description: Lustre make rpms failed.
735 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
736 builds Lustre rpms from tarball.
739 Frequency : rare on shutdown ost
741 Description: If llog cancel was not send before clean_exports phase, this can
742 produce deadlock in llog code.
743 Details : If llog thread has last reference to obd and call class_import_put
744 this produce deadlock because llog_cleanup_commit_master wait when
745 last llog_commit_thread exited, but this never success because was
746 called from llog_commit_thread.
750 Description: allow userland application know is lost one of stripes.
751 Details : fill lvb_blocks with error code on ost and return it to
752 application if error flag found.
756 Description: NULL lov_tgts causing MDS oops
757 Details : more safe checks for NULL lov_tgts for avoid oops.
759 Severity : enhancement
761 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
765 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
766 Details : If an OST is down the MDS will hang indefinitely in
767 obd_statfs_rqset() waiting for the statfs data. While for
768 MDS QOS usage of statfs, it should not stuck in waiting.
770 Severity : enhancement
772 Description: remote_acl support
773 Details : Support ACL-based permission check for remote user.
774 Support setfacl/getfacl for remote user with the utils
775 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
776 the system "{s,g}etfacl" utils.
778 Severity : enhancement
780 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
782 Severity : enhancement
784 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
788 Description: make mgs_setparam() handle fsname containing dash
789 Details : fsname containing a dash does not work with lctl conf_param
791 Severity : enhancement
793 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
795 Severity : enhancement
797 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
801 Frequency : rare, at shutdown
802 Description: access already free / zero obd_namespace.
803 Details : if client_disconnect_export was called without force flag set,
804 and exist connect request in flight, this can produce access to
805 NULL pointer (or already free pointer) when connect_interpret
806 store ocd flags in obd_namespace.
810 Frequency : only at startup
811 Description: not alloc memory with spinlock held.
812 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
813 if any spinlock held.
815 Severity : enhancement
817 Description: make lustre randomly fail allocating memory
818 Details : Make lustre randomly failed allocating memory for testing purpose.
820 Severity : enhancement
822 Description: lost problems with lov objid file
823 Details : Fixes some scability and access to not inited memory problems
824 in work with lov objdid file.
829 Description: lfs find does not continue on file error
830 Details : Continue other files processing when a file/dir is absent.
834 Description: Inconsistent usage of lustre_pack_reply()
835 Details : Standardize the usage of lustre_pack_reply() such that it
836 always generate a CERROR on failure.
839 Frequency : occasional
841 Description: MDS hang or stay in waiting lock
842 Details : If client receive lock with CBPENDING flag ldlm need send lock
843 cancel as separate rpc, to avoid situation when cancel request
844 can't processed due all i/o threads stay in wait lock.
848 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
852 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
856 Description: Update to SLES9 kernel-2.6.5-7.287.3.
860 Description: Update to RHEL4 latest kernel.
862 Severity : enhancement
864 Description: Build SLES10 patchless client fails
865 Details : The configure was broken by run ./configure with
866 --with-linux-obj=.... argument for patchless client. When the
867 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
868 can't search header adequately. Use absolute path such as
869 -I($LINUX)/include instead.
873 Description: interrupt oig_wait produce painc on resend.
874 Details : brw_redo_request can be used for resend requests from ptlrpcd and
875 private set, and this produce situation when rq_ptlrpcd_data not
876 copyed to new allocated request and triggered LBUG on assert
877 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
878 ptlrpcd set if request is changed and can be safety changed to use
883 Description: LASSERT_{REQ,REP}SWAB macros are buggy
884 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
885 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
886 these macros and replace them with their difinitions instead.
890 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
891 Details : Update which_patch & target file for SLES10 latest kernel.
894 Description: Remove client patches from SLES 10 kernel.
895 Details : This causes SLES 10 clients to behave as patchless clients
896 even on a Lustre-patched (server) kernel.
898 Severity : enhancement
900 Description: self-adjustable client's lru lists
901 Details : use adaptive algorithm for managing client cached locks lru
902 lists according to current server load, other client's work
903 pattern, memory activities, etc. Both, server and client
904 side namespaces provide number of proc tunables for controlling
907 Severity : enhancement
909 Description: light-weight GSS support
910 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
911 and reduce performance overhead.
913 Severity : enhancement
915 Description: Linux keyring support
916 Details : Support using service of Linux keyring for Lustre GSS internal
917 context refresh/cache mechanism.
921 Description: Fix errors in lfs documentation
922 Details : Fixes man pages
926 Description: don't use GFP_* in generic Lustre code.
927 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
931 Description: obdclass is limited by single OBD_ALLOC(idarray)
932 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
936 Description: warnings with build patchless client with vanila 2.6.19 and up
937 Details : change old ctl_table style and replace ctl_table/ctl_table_header
938 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
942 Frequency : Only for SLES
943 Description: sanity_quota fail test_1
944 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
945 file, which makes TSTID[2] inunique.
949 Frequency : Only for SLES9
950 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
951 Details : Update target/ChangeLog/which_patch .
955 Frequency : Only for test_10 in sanity-quota.sh
956 Description: a bug in quota test code
957 Details : checking a test flag in a wrong place
961 Frequency : Only for test_13 in sanity-quota.sh
962 Description: a bug in quota test script
963 Details : shouldn't check overfull allocated quota
967 Description: jbd statistics
968 Details : Port older jbd statistics patch for sles10
970 Severity : enhancement
972 Description: Add printing inode info into message about error in writepage.
977 Description: reply_lock_interpret crash due to race with it and lock cancel.
978 Details : Do not replay locks that are being cancelled. Do not reference
979 locks by their address during replay, just by their handle.
983 Frequency : When flocks are used.
984 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
985 Details : Flock locks might destroy just granted lock if it could be merged
986 with another existing flock, this is done in completion handler,
987 so teach ldlm_cli_enquque_fini that this is a valid case for
993 Description: Oops in read and write path when failing to allocate lock.
994 Details : Check if lock allocation failed and return error back.
998 Description: lstripe command fails for valid OST index
999 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1000 instead of lov->desc.ld_active_tgt_count.
1004 Description: sanity.sh failed test 103
1005 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1010 Description: df doesn't work properly if diskfs blocksize != 4K
1011 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1015 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1016 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1018 Severity : enhancement
1020 Description: Adaptive timeouts
1021 Details : RPC timeouts adapt to changing server load and network
1022 conditions to reduce resend attempts and improve recovery time.
1026 Description: llapi_file_create() does not allow some changes
1027 Details : add llapi_file_open() that allows specifying the mode and
1028 open flags, and also returns an open file handle.
1032 Description: merge and cleanup kernel patches.
1033 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1037 Description: Add journal checksum support.(Kernel part)
1038 Details : The journal checksum feature adds two new flags i.e
1039 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1040 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1041 indicates that the commit block contains the checksum for
1042 the blocks described by the descriptor blocks. Now commit
1043 record can be sent to disk without waiting for descriptor
1044 blocks to be written to disk. This behavior is controlled
1045 using JBD2_FEATURE_ASYNC_COMMIT flag.
1049 Description: OSS needs mutliple precreate threads
1050 Details : Add ability to start more than one create thread per OSS.
1054 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1055 Details : execute lfs setstripe on client
1059 Description: mds_obd_create error creating tmp object
1060 Details : When the user sets quota on root, llog will be affected and can't
1061 create files and write files.
1064 Frequency : Always on ia64 patchless client, and possibly others.
1066 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1067 Details : This allows the patchless client to be loaded on architectures
1068 without this export.
1072 Description: RedHat Update kernel for RHEL5
1073 Details : Add the kernel config file for RHEL5.
1077 Description: RedHat Update kernel for RHEL5
1078 Details : Modify the kernel config file more closer RHEL5.
1082 Description: Build failure against Centos5 (RHEL5)
1083 Details : Define PAGE_SIZE when it isn't present.
1087 Description: test_7 on sanity-quota.sh can't be used on separate machines
1088 Details : add facet to handle it
1092 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1093 Details : with NFS, the anon dentry's parent was set to itself in
1094 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1095 corresponding dentry other than use rec->ur_name.
1097 Severity : enhancement
1099 Description: lfs setstripe enhancement
1100 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1104 Description: enable data checksumming by default
1105 Details : enable checksum by default, allow --disable-checksum
1106 configure option and "-o nochecksum" mount option. Checksums
1107 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1111 Description: lustre support for RHEL5
1112 Details : Add support for RHEL5.
1116 Description: Client eviction due to failover config
1117 Details : after a connection loss, the lustre client should attempt to
1118 reconnect to the last active server first before trying the
1119 other potential connections.
1123 Description: when mds and osts use different quota unit(32bit and 64bit),
1124 quota will be released repeatly.
1125 Details : void sending multiple quota reqs to mds, which will keep the
1126 status between the reqs.
1130 Description: osts not allocated evenly to files
1131 Details : change the condition to increase offset_idx
1134 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1136 Description: Data corruption for OSTs that are formatted larger than 2TB
1138 Details : When generating the bio request for lustre file writes the
1139 sector number would overflow a temporary variable before being
1140 used for the IO. The data reads correctly from Lustre (which
1141 will overflow in a similar manner) but other file data or
1142 filesystem metadata may be corrupted in some cases.
1146 Description: Tune the kernel for good SCSI performance.
1147 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1148 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1153 Description: rewrite ext2-derived code in obdclass/uuid.c
1154 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1155 from scratch preserving functionality.
1159 Description: "lfs find -obd UUID" prints directories
1160 Details : "lfs find -obd UUID" will return all directory names instead
1161 of just file names. It is incorrect because the directories
1162 do not reside on the OSTs.
1166 Description: use i_size_read and i_size_write in 2.6 port
1167 Details : replace inode->i_size access with i_size_read/write()
1170 Frequency : when removing large files
1172 Description: scheduling issue during removal of large Lustre files
1173 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1174 It causes scheduling issues when removing large files (17TB in the
1178 Frequency : only with liblustre clients on XT3
1180 Description: evictions taking too long
1181 Details : allow llrd to evict clients directly on OSTs
1184 Frequency : only on ppc
1186 Description: /proc/fs/lustre/devices broken on ppc
1187 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1188 We should make sure the type of 'index' is loff_t and then cast
1189 explicitly as needed below. Do not assign an explicitly cast
1193 Frequency : only for rhel5
1195 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1196 Details : Modify the target file & which_kernel.
1198 Severity : enhancement
1200 Description: omit set fsid for export NFS
1201 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1202 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1206 Frequency : Always, for kernels after 2.6.16
1207 Description: Fix warning idr_remove called for id=.. which is not allocated.
1208 Details : Last kernels save old s_dev before kill super and not allow
1209 to restore from callback - restore it before call kill_anon_super.
1213 Description: buffer overruns could theoretically occur
1214 Details : llapi_semantic_traverse() modifies the "path" argument by
1215 appending values to the end of the origin string, and a
1216 overrun may occur. Adding buffer overrun check in liblustreapi.
1220 Description: Fix error on 'ls .' at the top of the Lustre mount.
1221 Details : Don't revalidate dentry if it is a root dentry.
1225 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1226 Details : Modify vm-tunables-rhel4.patch.
1230 Description: Kernel config for 2.6.18-vanilla.
1231 Details : Modify targets/2.6-vanilla.target.in.
1232 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1233 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1234 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1235 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1239 Description: improve handling recoverable errors
1240 Details : if request processig with error which can be recoverable on server
1241 request should be resend, otherwise page released from cache and
1246 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1247 Details : Modify target file & which_patch.
1248 A flaw was found in the IA32 system call emulation provided
1249 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1250 value could be stored in the %RAX register, which could trigger an
1251 out-of-bounds system call table access. An untrusted local user
1252 could exploit this flaw to run code in the kernel
1253 (ie a root privilege escalation). (CVE-2007-4573).
1257 Description: O_DIRECT bypasses client statistics.
1258 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1259 read_ahead_stats were updated. copied the stats section from
1260 osc_send_oap_rpc() into async_internal().
1264 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1268 Description: change order of libsysio includes
1269 Details : '#include sysio.h' should always come before '#include xtio.h'
1273 Description: handle "serious error: objid * already exists" more gracefully
1274 Details : If LAST_ID value on disk is smaller than the objects existing in
1275 the O/0/d* directories, it indicates disk corruption and causes an
1276 LBUG(). If the object is 0-length, then we should use the existing
1277 object. This will help to avoid a full fsck in most cases.
1279 Severity : enhancement
1281 Description: adapt the lustre_config script to support the upgrade case
1282 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1283 server targets to 1.6 in parallel.
1287 Description: To avoid grant space > avaible space when the disk is almost
1288 full. Without this patch you might see the error "grant XXXX >
1289 available" or some LBUG about grant, when the disk is almost
1291 Details : In filter_check_grant, for non_grant cache write, we should
1292 check the left space by if (*left > ungranted + bytes), instead
1293 of (*left > ungranted), because only we are sure the left space
1294 is enough for another "bytes", then the ungrant space should be
1295 increase. In client, we should update cl_avail_grant only there
1296 is OBD_MD_FLGRANT in the reply.
1300 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1301 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1302 described in CVE-2007-4573. This problem could allow untrusted
1303 local users to gain root access.
1306 Frequency : when using O_DIRECT and quotas
1308 Description: Incorrect file ownership on O_DIRECT output files
1309 Details : block usage reported by 'lfs quota' does not take into account
1310 files that have been written with O_DIRECT.
1315 Description: touch file failed when fs is not full
1316 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1317 otherwise we can get ENOSP while fs is not full.
1321 Description: parallel lock callbacks
1322 Details : Instead of sending blocking and completion callbacks as separated
1323 requests, adding them to a set and sending in parallel.
1326 Frequency : only for Cray XT3
1327 Bugzilla : 12829/13455
1328 Description: Changing primary group doesn't change the group lustre assigns to
1330 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1331 overridden with the primary group provided by the group upcall,
1332 whereas the supplied fsgid can be trusted if it is in the list of
1333 supplementary groups returned by the group upcall.
1335 Severity : enhancement
1337 Description: Allow masking D_WARNING, D_ERROR messages from console
1338 Details : Console messages can now be disabled via lnet.printk.
1342 Description: User code with malformed file open parameter crashes client node
1343 Details : Before packing join_file req, all the related reference should be
1344 checked carefully in case some malformed flags cause fake
1345 join_file req on client.
1349 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1350 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1351 mode is equal to the granted mode. However, it should also check
1352 that LDLM_FL_CP_REQD is not set.
1356 Description: Heavy nfs access might result in deadlocks
1357 Details : After ELC code landed, it is now improper to enqueue any mds
1358 locks under och_sem, because enqueue might want to decide to
1359 cancel open locks for same inode we are holding och_sem for.
1363 Description: Client eviction while running blogbench
1364 Details : A lot of unlink operations with concurrent I/O can lead to a
1365 deadlock causing evictions. To address the problem, the number of
1366 oustanding OST_DESTROY requests is now throttled to
1367 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1368 callbacks are processed in priority.
1372 Description: enable ACLs on MDS by default
1373 Details : ACLs must be enabled on MDS by default.
1376 Frequency : PPC/PPC64 only
1378 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1379 Details : fix duplicated definitions between asm-ppc64/types.h and
1380 lustre_types.h on PPC.
1383 Frequency : PPC/PPC64 only
1385 Description: asm-ppc/segment.h does not exist
1386 Details : fix compile issue on PPC.
1390 Description: better handle error messages in extents code
1393 Frequency : RHEL4 only
1395 Description: mkfs is very slow on IA64/RHEL4
1396 Details : A performance regression has been discovered in the MPT Fusion
1397 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1398 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1399 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1401 Severity : enhancement
1403 Description: SNMP support enhancement
1404 Details : Adding total number of sampled request for an MDS node in snmp
1407 Severity : enhancement
1409 Description: Optimize ldlm waiting list processing for PR extent locks
1410 Details : When processing waiting list for read extent lock and meeting read
1411 lock that is same or wider to it that is not contended, skip
1412 processing rest of the list and immediatelly return current
1413 status of conflictness, since we are guaranteed there are no
1414 conflicting locks in the rest of the list.
1418 Description: Time out and refuse to reconnect
1419 Details : When the failover node is the primary node, it is possible
1420 to have two identical connections in imp_conn_list. We must
1421 compare not conn's pointers but NIDs, otherwise we can defeat
1422 connection throttling.
1426 Description: port llog fixes from b1_6 into HEAD
1427 Details : Port llog reference couting and some llog cleanups from b1_6
1428 (bug 10800) into HEAD, for protect from panic and access to already
1429 free llog structures.
1433 Description: Detect stride IO mode in read-ahead
1434 Details : When a client does stride read, read-ahead should detect that and
1435 read-ahead pages according to the detected stride pattern.
1439 Description: data checksumming impacts single node performance
1440 Details : add support for several checksum algorithm. Currently, only CRC32
1441 and Adler-32 are supported. The checksum type can be changed on
1442 the fly via /proc/fs/lustre/osc/*/checksum_type.
1446 Description: use adler32 for page checksums
1447 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1452 Description: build for x2 fails
1453 Details : fix compile issue on Cray systems.
1457 Description: Properly match for duplicate locks
1458 Details : Due to different lock order from skiplists code, we need to
1459 traverse entire list for now
1462 Frequency : only on PPC/SLES10
1464 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1465 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1466 be defined. Add a hack in mkfs_lustre.c to work around this compile
1471 Description: LASSERT on MDS when client holding flock lock dies
1472 Details : ldlm pool logic depends on number of granted locks equal to
1473 number of released locks which is not true for flock locks, so
1474 just exclude such locks from consideration.
1478 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1479 Details : Use fsfilt_sync() for both the whole filesystem sync and
1480 individual file sync to eliminate dangerous inode locking
1481 with I_LOCK that can lead to a deadlock.
1485 Description: performance in 1.6.3
1486 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1487 sure that 1MB requests are not fragmented by the block layer.
1489 Severity : enhancement
1491 Description: organize the server-side client stats on per-nid basis
1492 Details : Change the structure of stats under obdfilter and mds to
1500 The "uuid"s file would list the uuids of _active_ exports.
1501 And the clear entry is to clear all stats and stale nids.
1503 Severity : enhancement
1505 Description: eliminate client locks in face of contention
1506 Details : file contention detection and lockless i/o implementation
1507 for contended files.
1511 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1514 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1516 Description: MDS may lose file striping (and hence file data) in some cases
1517 Details : If there are additional extended attributes stored on the MDS,
1518 in particular ACLs, SELinux, or user attributes (if user_xattr
1519 is specified for the client mount options) then there is a risk
1520 of attribute loss. Additionally, the Lustre file striping
1521 needs to be larger than default (e.g. striped over all OSTs),
1522 and an additional attribute must be stored initially in the
1523 inode and then increase in size enough to be moved to the
1524 external attribute block (e.g. ACL growing in size) for file
1529 Description: skiplist implementation simplification
1530 Details : skiplists are used to group compatible locks on granted list
1531 that was implemented as tracking first and last lock of each
1532 lock group the patch changes that to using doubly linked lists
1536 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1537 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1538 because LASSERT on both the data supplied by a client, and the
1539 data on disk is dangerous and incorrect.
1541 Severity : enhancement
1543 Description: Slow truncate/writes to huge files at high offsets.
1544 Details : Directly associate cached pages to lock that protect those pages,
1545 this allows us to quickly find what pages to write and remove
1546 once lock callback is received.
1550 Description: more ldlm soft lockups
1551 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1552 starve other threads from the resource lock for a long time in
1553 case of long waiting queue, so change the debug level from
1554 D_OTHER to the less frequently used D_INFO.
1556 Severity : enhancement
1558 Description: add -gid, -group, -uid, -user options to lfs find
1562 Description: Hung threads in invalidate_inode_pages2_range
1563 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1564 once one is completed. As a result, some RPCs are stuck in the
1565 queue and are never sent.
1569 Description: filter threads hungs on waiting journal commit
1570 Details : Cleanup filter group llog code, then only filter group llog will
1571 be only created in the MDS/OST syncing process.
1575 Description: Procfs and llog threads access destoryed import sometimes.
1576 Details : Sync the import destoryed process with procfs and llog threads by
1577 the import refcount and semaphore.
1579 Severity : enhancement
1581 Description: openlock cache of b1_6 port to HEAD
1586 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1587 Details : If insertion of an extent fails, then discard the inode
1588 preallocation and free data blocks else it can lead to duplicate
1593 Description: don't always update ctime in ext3_xattr_set_handle()
1594 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1595 In some cases the ctime should not be updated, for example for
1596 2.0->1.8 compatibility it is necessary to delete an xattr and it
1597 should not update the ctime.
1601 Bugzilla : 15713/16362
1602 Description: Assertion in iopen_connect_dentry in 1.6.3
1603 Details : looking up an inode via iopen with the wrong generation number can
1604 populate the dcache with a disconneced dentry while the inode
1605 number is in the process of being reallocated. This causes an
1606 assertion failure in iopen since the inode's dentry list contains
1607 both a connected and disconnected dentry.
1611 Description: assertion failure in ldlm_handle2lock()
1612 Details : fix a race between class_handle_unhash() and class_handle2object()
1613 introduced in lustre 1.6.5 by bug 13622.
1618 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1619 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1624 Description: oops in sd_iostats_seq_show()
1625 Details : unloading/reloading the scsi low level driver triggers a kernel
1626 bug when trying to access the sd iostat file.
1631 Description: Kernel panics during QLogic driver reload
1632 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1633 patch, causing memory corruption.
1638 Description: journal_dev option does not work in b1_6
1639 Details : pass mount option during pre-mount.
1641 Severity : enhancement
1643 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1644 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1645 extent information of a file. It can be used to map logical blocks
1646 in a file to physical blocks in the block device.
1650 Description: LDLM soft lockups - improvement
1651 Details : It is be possible to send the lock handle along with each read
1652 or write request because the client is already doing a lock match
1653 itself so there isn't any reason the OST should have to re-do that
1657 Frequency : only on Cray X2
1659 Description: X2 build failures
1660 Details : fix build failures on Cray X2.
1664 Description: xid & resent requests
1665 Details : Initialize RPC XID from clock at startup (randomly if clock is
1668 Severity : enhancement
1670 Description: Add lustre_start utility to start or stop multiple Lustre servers
1675 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1676 Details : In case of memory pressure, list_del() can be called twice on
1677 req->rq_history_list, causing a kernel oops.
1681 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1682 Details : fix stack overflow in the distributed lock manager by defering
1683 export eviction after a failed AST to the elt thread instead of
1684 handling it in the dlm interpret routine.
1688 Description: Convert some comments to new format.
1689 Details : Update documenting comments to match doxygen conventions.
1693 Description: Grammar fixes.
1694 Details : A couple of trivial spelling fixes.
1698 Description: OSD_COUNTERS-mandatory
1699 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1700 to simplify the code.
1704 Description: simplify cmm_device freeing logic.
1705 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1709 Description: Add lockdep support to dt_object_operations locking interface.
1710 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1711 parameter indicating lock ordering. Update mdd code to use new
1716 Description: Introduce failloc constants for lockless IO tests.
1717 Details : Add two new failloc constants to test lockless IO. Only one of
1718 them in implemented---another is checked in yet to be landed
1723 Description: Add lockdep support for inode mutex.
1724 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1725 in the situations where Lustre has to lock more than one inode
1730 Description: Add optional invariants checking support.
1731 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1732 configure switch. This macro is to be used for consistency and
1733 sanity checks that are too expensive to be left in `production'
1738 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1739 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1744 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1745 Details : Use spin_lock_nested() in (the only) situation where more than
1746 one ldlm_lock is locked simultaneously. Also, fix possible
1747 dead-lock in ldlm_lock_change_resource() by enforcing particular
1752 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1753 Details : Instead of specifying each ldlm_lock call-back through separate
1754 parameter, wrap them into struct ldlm_callback_suite.
1758 Description: Kill join_lru obd method and its callers.
1759 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1760 in memory---a requirement implemented through ->o_join_lru() obd
1761 method in HEAD. Remove this method and its users.
1765 Description: Add asynchronous ldlm ENQUEUE completion handler.
1766 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1767 case that stock ldlm_completion_ast() cannot handle as it waits
1768 until lock is granted. Introduce new ldlm_completion_ast_async()
1769 for this. Also comment ldlm_completion_ast().
1773 Description: ldlm_error <-> errno conversion.
1774 Details : Add functions to map (rather arbitrary) between LDLM error codes
1775 and standard errno values. CLIO needs this to prevent LDLM specific
1776 constants from escaping ldlm and osc.
1780 Description: Kill unused ldlm_handle2lock_ns() function.
1781 Details : Kill unused ldlm_handle2lock_ns() function.
1785 Description: Add lu_ref support to ldlm_lock
1786 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1787 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1788 and ldlm_resource. LDLM interface has to be changed, because code that
1789 releases a reference on a lock, has to "know" what reference this is.
1790 In the most frequent case
1792 lock = ldlm_handle2lock(handle);
1794 LDLM_LOCK_PUT(lock);
1796 no changes are required. When any other reference (received _not_ from
1797 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1798 instead of LDLM_LOCK_PUT().
1800 Arguably, changes are pervasive, and interface requires some discipline
1801 for proper use. On the other hand, it was very instrumental in finding
1802 a few leaked lock references.
1806 Description: Add ldlm_lock_addref_try().
1807 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1808 attempts to addref a lock that might be being canceled
1813 Description: Add ldlm_weigh_callback().
1814 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1815 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1816 hard-coded `number of pages' logic.
1820 Description: Add lockdep annotations to llog code.
1821 Details : Use appropriately tagged _nested() locking calls in the places
1822 where llog takes more than one ->lgh_lock lock.
1826 Description: Add loi_kms_set().
1827 Details : Wrap kms updates into a helper function.
1831 Description: Constify instances of struct lsm_operations.
1832 Details : Constify instances of struct lsm_operations.
1836 Description: lu_conf support.
1837 Details : On a server, a file system object is uniquely identified
1838 by a fid, which is sufficient to locate and load all object
1839 state (inode). On a client, on the other hand, more data are
1840 necessary instantiate an object. Change lu_object_find() and
1841 friends to take additional `lu_conf' argument describing object.
1842 Typically this includes layout information.
1846 Description: lu_context fixes.
1847 Details : Introduce new lu_context functions that are needed on the client
1848 side, where some system threads (ptlrpcd) are shared by multiple
1849 modules, and so cannot be stopped during module shutdown.
1853 Description: Add start and stop methods to lu_device_type_operations.
1854 Details : Introduce two new methods in lu_device_type_operations, that are
1855 invoked when first instance of a given type is created and last one
1856 is destroyed respectively. This is need by CLIO.
1860 Description: Add lu_ref support to struct lu_device.
1861 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1862 track leaked references.
1866 Description: Introduce lu_kmem_descr.
1867 Details : lu_kmem_descr and its companion interface allow to create
1868 and destroy a number of kmem caches at once.
1872 Description: Fix lu_object finalization race.
1873 Details : Fix a race between lu_object_find() finding an object and its
1874 concurrent finalization. This race is (most likely) not possible
1875 on the server, but might happen on the client.
1879 Description: Introduce lu_ref interface.
1880 Details : lu_ref is a debugging module allowing to track references to
1881 a given object. It is quite cpu expensive, and has to be
1882 explicitly enabled with --enable-lu_ref. See usage description
1887 Description: Factor lu_site procfs stats into a separate function.
1888 Details : Separate lu_site stats printing code into a separate function
1889 that can be reused on a client.
1893 Description: Constify instances of struct {lu,dt,md}_device_operations.
1894 Details : Constify instances of struct {lu,dt,md}_device_operations.
1898 Description: Introduce struct md_site and move meta-data specific parts of
1899 struct lu_site here.
1900 Details : Move md-specific fields out of struct lu_site into special struct
1901 md_site, so that lu_site can be used on a client.
1905 Description: Kill mdd_lov_destroy().
1906 Details : Remove unused mdd code.
1910 Description: Add st_block checking to multistat.c.
1911 Details : Add st_block checking to multistat.c.
1915 Description: Add lu_ref support to struct obd_device.
1916 Details : Add lu_ref tracking to obd_device.
1920 Description: Kill obd_set_fail_loc().
1921 Details : Remove unused code.
1925 Description: Add special type for ptlrpc_request interpret functions.
1926 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1927 there. Actual usage will be in CLIO.
1931 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1932 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1933 doesn't work with lockdep.
1937 Description: Add rwv.c test program.
1938 Details : New testing program exercising readv(2) and writev(2) (Qian).
1942 Description: Add sendfile.c test program.
1943 Details : New testing program exercising sendfile(2) (Jay).
1947 Description: Ratelimit a message that can be very frequent.
1948 Details : Ratelimit a memory allocation failure message that can
1953 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1954 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1958 Description: Liblustre build fixes.
1959 Details : Liblustre build fixes.
1963 Description: libcfs: add cfs_{need,cond}_resched() interface.
1964 Details : libcfs: add cfs_{need,cond}_resched() definition and
1965 implementations for Linux, NT, and liblustre.
1967 Severity : enhancement
1969 Description: More exported tunables for mballoc
1970 Details : Add support for tunable preallocation window and new tunables for
1971 large/small requests
1975 Description: Detect corruption of block bitmap and checking for preallocations
1976 Details : Checks validity of on-disk block bitmap. Also it does better
1977 checking of number of applied preallocations. When corruption is
1978 found, it turns filesystem readonly to prevent further corruptions.
1982 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1983 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1984 window to 0. If it does hit cache miss, and read-pattern is still
1985 stride-io mode, does not reset the stride window, but also does not
1986 increase the stride window length in this case.
1990 Frequency : only for big-endian servers
1991 Description: Check if system is big-endian while mounting fs with extents feature
1992 Details : Mounting a filesystem with extents feature will fail on big-endian
1993 systems since ext3-based ldiskfs is not supported on big-endian
1994 systems. This can be over-riden with "bigendian_extents" mount option.
1996 Severity : enhancement
1998 Description: The root squash functionality
1999 Details : A security feature, which is to prevent users from being able
2000 to mount lustre on their desktop, run as root, and delete
2001 all of the files in the filesystem. The goal is accomplished by
2002 remapping user id (UID) and group id (GID) of the root user to
2003 a UID and GID specified by the system administartor via Lustre
2004 configuration management server (MGS). The functionality also
2005 allows to specify sets of clients for which the remapping does
2010 Description: Excessive recovery window
2011 Details : With AT enabled, the recovery window can be excessively long (6000+
2012 seconds). To address this problem, we no longer use
2013 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2014 timeout no longer depends on the service time, it is set to
2015 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2016 time via pb_service_time.
2020 Description: Watchdog triggered on MDS failover
2021 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2022 the OSTs know that the MDS "UUID" can be reused for the same export
2023 from a different NID, so we do not need to wait for the export to be
2027 Frequency : rare, only if using MMP with Linux RAID
2029 Description: MMP doesn't work with Linux RAID
2030 Details : While using HA for Lustre servers with Linux RAID, it is possible
2031 that MMP will not detect multiple mounts. To make this work we
2032 need to unplug the device queue in RAID when the MMP block is being
2033 written. Also while reading the MMP block, we should read it from
2034 disk and not the cached one.
2036 Severity : enhancement
2038 Description: open file using fid
2039 Details : A file can be opened using just its fid, like
2040 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2043 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2046 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2047 Details : add proper locking for creation of jbd_4k slab cache
2051 Description: MMP check in ext3_remount() fails without displaying any error
2052 Details : When multiple mount protection fails during remount, proper error
2055 --------------------------------------------------------------------------------
2057 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2059 * Support for kernels:
2060 2.6.5-7.283 (SLES 9),
2061 2.6.9-55.EL (RHEL 4),
2062 2.6.16.46-0.14 (SLES 10),
2063 2.6.18.8 vanilla (kernel.org)
2064 * Client support for unpatched kernels:
2065 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2066 2.6.16 - 2.6.22 vanilla (kernel.org)
2067 * Due to recently discovered recovery problems, we do not recommend
2068 using patchless RHEL 4 clients with this or any earlier release.
2069 * Recommended e2fsprogs version: 1.39.cfs8
2070 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2071 * Starting with this release, the ldiskfs backing filesystem required
2072 by Lustre is now in its own package, lustre-ldiskfs. This package
2073 should be installed. It is versioned separately from Lustre and
2074 may be released separately in future.
2078 Description: block reactivating mgc import until all deactivates complete
2079 Details : Fix race when failing back MDT/MGS to itself (testing)
2082 Frequency : at statup only
2084 Description: mds_lov_synchronize race leads to various problems
2085 Details : simultaneous MDT->OST connections at startup can cause the
2086 sync to abort, leaving the OSC in a bad state.
2088 Severity : enhancement
2090 Description: add optional extra BUILD_VERSION info
2091 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2092 to override the lustre version.
2095 Frequency : 2.6.18 servers only
2097 Description: ll_kern_mount() doesn't release the module reference
2098 Details : The ldiskfs module reference count never drops down to 0
2099 because ll_kern_mount() doesn't release the module reference.
2104 Description: server LBUG when using old ost_num_threads parameter
2105 Details : Accept the old ost_num_threads parameter but warn that it
2106 is deprecated, and fix an off-by-one error that caused an LBUG.
2111 Description: Transient SCSI error results in persistent IO issue
2112 Details : iobuf->dr_error is not reinitialized to 0 between two
2116 Frequency : sometimes when underlying device returns I/O errors
2118 Description: OSTs not going read-only during write failures
2119 Details : OSTs are not remounted read-only when the journal commit threads
2120 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2121 instead of the ext3 wrappers.
2125 Description: poor connect scaling with increasing client count
2126 Details : Don't run filter_grant_sanity_check for more than 100 exports
2127 to improve scaling for large numbers of clients.
2130 Frequency : SLES10 only
2132 Description: sanity-quota.sh quotacheck failed: rc = -22
2133 Details : Quotas cannot be enabled on SLES10.
2136 Frequency : liblustre clients only
2138 Description: getdirentries does not give error when run on compute nodes
2139 Details : getdirentries does not fail when the size specified as an argument
2140 is too small to contain at least one entry
2142 Severity : enhancement
2144 Description: Add LNET router traceability for debug purposes
2145 Details : If a checksum failure occurs with a router as part of the
2146 IO path, the NID of the last router that forwarded the bulk data
2147 is printed so it can be identified.
2152 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2153 Details : Due to a race condition, liblustre clients were occasionally
2154 evicted incorrectly.
2156 Severity : enhancement
2158 Description: lfs setstripe use optional parameters instead of postional
2161 Severity : enhancement
2163 Description: Nanosecond timestamp support for ldiskfs
2164 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2165 resolution timestamps. There is not yet support for this at
2166 the Lustre filesystem level.
2169 Frequency : during server recovery
2171 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2172 Details : request with rq_no_resend flag not awake l_wait_event if they get
2176 Frequency : nfs export on patchless client
2178 Description: connectathon hang when test nfs export over patchless client
2179 Details : Disconnected dentry cannot be found with lookup, so we do not need
2180 to unhash it or make it invalid
2183 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2184 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2185 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2190 Description: open req refcounting wrong on reconnect
2191 Details : If reconnect happened between getting open reply from server and
2192 call to mdc_set_replay_data in ll_file_open, we will schedule
2193 replay for unreferenced request that we are about to free.
2194 Subsequent close will crash in variety of ways.
2195 Check that request is still eligible for replay in
2196 mdc_set_replay_data().
2201 Description: disable writes to filesystem when reading health_check file
2202 Details : the default for reading the health_check proc file has changed
2203 to NOT do a journal transaction and write to disk, because this
2204 can cause reads of the /proc file to hang and block HA state
2205 checking on a healthy but otherwise heavily loaded system. It
2206 is possible to return to the previous behaviour during configure
2207 with --enable-health-write.
2209 Severity : enhancement
2211 Description: 64-bit inode version
2212 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2213 to the inode. This will be required for version-based recovery.
2218 Description: MDS fails to start if a duplicate client export is detected
2219 Details : in some rare cases it was possible for a client to connect to
2220 an MDS multiple times. Upon recovery the MDS would detect this
2221 and fail during startup. Handle this more gracefully.
2223 Severity : enhancement
2225 Description: Add -o localflock option to simulate old noflock
2227 Details : This will achieve local-only flock/fcntl locks
2233 Description: log_commit_thread vs filter_destroy race leads to crash
2234 Details : Take import reference before releasing llog record semaphore
2239 Description: Wrong request locking in request set processing
2240 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2241 imp_delayed_list, in this place should be used imp_lock.
2244 Frequency : when reconnection
2246 Description: Grant Leak when osc reconnect to OST
2247 Details : When osc reconnect ost, OST(filter) should check whether it
2248 should grant more space to client by comparing fed_grant and
2249 cl_avail_grant, and return the granted space to client instead
2250 of "new granted" space, because client will call osc_init_grant
2251 to update the client grant space info.
2254 Frequency : when client reconnect to OST
2256 Description: Grant Leak when osc do resend and replay bulk write
2257 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2258 bulk write request, because the grant info while be sync between
2259 OSC and OST when reconnect, and we should ignore the grant info
2260 these of resend/replay write req.
2265 Description: Grant space more than avaiable left space sometimes.
2266 Details : When then OST is about to be full, if two bulk writing from
2267 different clients came to OST. Accord the avaliable space of the
2268 OST, the first req should be permitted, and the second one
2269 should be denied by ENOSPC. But if the seconde arrived before
2270 the first one is commited. The OST might wrongly permit second
2271 writing, which will cause grant space > avaiable space.
2274 Frequency : when client is evicted
2276 Description: Grant might be wrongly erased when osc is evicted by OST
2277 Details : when the import is evicted by server, it will fork another
2278 thread ptlrpc_invalidate_import_thread to invalidate the
2279 import, where the grant will be set to 0. While the original
2280 thread will update the grant it got when connecting. So if
2281 the former happened latter, the grant will be wrongly errased
2282 because of this race.
2287 Description: Checking Stale with correct fid
2288 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2289 to check whether it is stale, because sometimes, we want the
2290 enqueue happened anyway, and op_data.fid2 will not be initialized.
2292 Severity : enhancement
2294 Description: update patchless client
2295 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2298 Frequency : only with 2.4 kernel
2300 Description: random memory corruption
2301 Details : size of struct ll_inode_info is to big for union inode.u and this
2302 can be cause of random memory corruption.
2307 Description: Memory leak in recovery
2308 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2309 It should also check obd_fail before fsfilt_start, otherwise if
2310 fsfilt_start return -EROFS,(failover mds during mds recovery).
2311 then the req will return with repmsg->transno = 0 and rc = EROFS.
2312 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2313 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2314 be freed no matter whether fsfilt_commit success or not.
2317 Frequency : only with huge count clients
2319 Description: Prevents from taking the superblock lock in llap_from_page for
2321 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2322 for llap_from_page call in ll_removepage prevents from taking the
2323 superblock lock for a soon died page.
2328 Description: Not check open intent error before release open handle
2329 Details : in some rare cases, the open intent error is not checked before
2330 release open handle, which may cause
2331 ASSERTION(open_req->rq_transno != 0), because it tries to release
2332 the failed open handle.
2337 Description: Set cat log bitmap only after create log success.
2338 Details : in some rare cases, the cat log bitmap is set too early. and it
2339 should be set only after create log success.
2343 Description: Accessing a block bevice can re-enable I/O when Lustre is
2344 tearing down a device.
2345 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2350 Description: service threads may hog cpus when there are a lot of requests
2351 Details : Insert cond_resched to give other threads a chance to use some CPU
2356 Description: the cat log was not initialized in recovery
2357 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2358 unlink log on mds will not be initialized until mds post
2359 recovery. And also in mds post recovery, the unlink log will
2360 initialization will be done asynchronausly, so there will be race
2361 between add unlink log and unlink log initialization.
2365 Description: brw_stats were being printed incorrectly
2366 Details : brw_stats were being printed as log2 but all of them were not
2367 recorded as log2. Also remove some code duplication arising from
2368 filter_tally_{read,write}.
2372 Frequency : rare, only in recovery.
2373 Description: ASSERTION(req->rq_type != LI_POISON) failed
2374 Details : imp_lock should be held while iterating over imp_sending_list for
2375 prevent destroy request after get timeout in ptlrpc_queue_wait.
2379 Description: replay-single.sh test 52 fails
2380 Details : A lock's skiplist need to be cleanup when it being unlinked
2381 from its resource list.
2385 Description: Short directio read returns full requested size rather than
2387 Details : Direct I/O operations should return actual amount of bytes
2388 transferred rather than requested size.
2390 Severity : enhancement
2392 Description: metadata RPC reduction (e.g. for rm performance)
2393 Details : decrease the amount of synchronous RPC between clients and servers
2394 by canceling conflicing lock before the operation on the client
2395 and packing thier handles into the main operation RPC to server.
2397 Severity : enhancement
2399 Description: Async OSC create to avoid the blocking unnecessarily.
2400 Details : If a OST has no remain object, system will block on the creating
2401 when need to create a new object on this OST. Now, ways use
2402 pre-created objects when available, instead of blocking on an
2403 empty osc while others are not empty. If we must block, we block
2404 for the shortest possible period of time.
2408 Description: improve handling recoverable errors
2409 Details : if request processig with error which can be recoverable on server
2410 request should be resend, otherwise page released from cache and
2413 Severity : enhancement
2415 Description: refine locking for avoid write wrong info into lov_objid file
2416 Details : fix possible races with add new target and write/update data in
2419 --------------------------------------------------------------------------------
2421 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2426 Frequency : on some architectures
2428 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2429 Details : Uninitialized flags sometimes cause configuration commands to
2433 Frequency : patchless clients only
2435 Description: missing __iget() symbol export
2436 Details : The __iget() symbol export is missing. To avoid the need for
2437 this on patchless clients the deathrow inode reaper is turned
2438 off, and we depend on the VM to clean up old inodes. This
2439 dependency was during via the fix for bug 12181.
2441 --------------------------------------------------------------------------------
2443 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2445 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2446 INTEROPERATE with older versions automatically. In many cases a
2447 special upgrade step is needed. Please read the
2448 user documentation before upgrading any part of a live system.
2449 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2450 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2451 * WARNING: Lustre configuration and startup changes are required with
2452 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2456 Severity : enhancement
2458 Description: MountConf
2459 Details : Lustre configuration is now managed via mkfs and mount
2460 commands instead of lmc and lconf. New obd types (MGS, MGC)
2461 are added for dynamic configuration management. See
2462 https://mail.clusterfs.com/wikis/lustre/MountConf for
2465 Severity : enhancement
2467 Description: dynamic OST addition
2468 Details : OSTs can now be added to a live filesystem
2470 Severity : enhancement
2472 Description: startup order invariance
2473 Details : MDTs and OSTs can be started in any order. Clients only
2474 require the MDT to complete startup.
2476 Severity : enhancement
2478 Description: parallel, asynchronous orphan cleanup
2479 Details : orphan cleanup is now performed in separate threads for each
2480 OST, allowing parallel non-blocking operation.
2482 Severity : enhancement
2484 Description: optimized stripe assignment
2485 Details : stripe assignments are now made based on ost space available,
2486 ost previous usage, and OSS previous usage, in order to try
2487 to optimize storage space and networking resources.
2489 Severity : enhancement
2491 Description: Permanently set tunables
2492 Details : All writable /proc/fs/lustre tunables can now be permanently
2493 set on a per-server basis, at mkfs time or on a live system.
2495 Severity : enhancement
2497 Description: Lustre message v2
2498 Details : Add lustre message format v2.
2500 Severity : enhancement
2502 Description: client OST exclusion list
2503 Details : Clients can be started with a list of OSTs that should be
2504 declared "inactive" for known non-responsive OSTs.
2508 Description: SPEC SFS validation failure on NFS v2 over lustre.
2509 Details : Changes the blocksize for regular files to be 2x RPC size,
2510 and not depend on stripe size.
2512 Severity : enhancement
2514 Description: Multiple MD RPCs in flight.
2515 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2516 To avoid overly-overloading MDS, introduce a limit on number of
2517 MDS RPCs in flight for a single client and add /proc controls
2518 to adjust this limit.
2520 Severity : enhancement
2522 Description: client read/write statistics
2523 Details : Add client read/write call usage stats for performance
2524 analysis of user processes.
2525 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2526 file access. extents_stats shows chunk size distribution.
2527 extents_stats_per_process show chunk size distribution per
2530 Severity : enhancement
2532 Description: mds statistics
2533 Details : Add detailed mds operations statistics in
2534 /proc/fs/lustre/mds/*/stats.
2538 Description: Failure of copying files with lustre special EAs.
2539 Details : Client side always return success for setxattr call for lustre
2540 special xattr (currently only "trusted.lov").
2545 Description: Refcount LNET uuids
2546 Details : The global LNET uuid list grew linearly with every startup;
2547 refcount repeated list entries instead of always adding to
2550 Severity : enhancement
2552 Description: Dynamic service threads
2553 Details : Within a small range, start extra service threads
2554 automatically when the request queue builds up.
2557 Frequency : mixed-endian client/server environments
2559 Description: mixed-endian crashes
2560 Details : The new msg_v2 system had some failures in mixed-endian
2563 Severity : enhancement
2565 Description: Easy OST removal
2566 Details : OSTs can be permanently deactivated with e.g. 'lctl
2567 conf_param lustre-OST0001.osc.active=0'
2569 Severity : enhancement
2571 Description: MGS proc entries
2572 Details : Added basic proc entries for the MGS showing what filesystems
2575 Severity : enhancement
2577 Description: provide MGS failover
2578 Details : Added config lock reacquisition after MGS server failover.
2580 Severity : enhancement
2582 Description: add Linux 2.4 support
2583 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2587 Description: a large application tries to do I/O to the same resource and dies
2588 in the middle of it.
2589 Details : Check the req->rq_arrival time after the call to
2590 ost_brw_lock_get(), but before we do anything about
2591 processing it & sending the BULK transfer request. This
2592 should help move old stale pending locks off the queue as
2593 quickly as obd_timeout.
2596 Frequency : when an incorrect nid is specified during startup
2598 Description: ptlrpc connect to non-existant node causes kernel crash
2599 Details : LNET can't be re-entered from an event callback, which
2600 happened when we expire a message after the export has been
2601 cleaned up. Instead, hand the zombie cleanup off to another
2604 Severity : enhancement
2606 Description: plain/inodebits lock performance improvement
2607 Details : Grouping plain/inodebits in granted list by their request modes
2608 and bits policy, thus improving the performance of search through
2612 Frequency : only if OST filesystem is corrupted
2614 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2615 Details : for a short time RPCs with bulk IO are in the replay list,
2616 but replay of bulk IOs is unimplemented. If the OST filesystem
2617 is corrupted due to disk cache incoherency and then replay is
2618 started it is possible to trip an assertion. Avoid putting
2619 committed RPCs into the replay list at all to avoid this issue.
2622 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2624 Description: System hang on startup
2625 Details : This bug allowed the liblustre (e.g. catamount) client to
2626 return to the app before handling all startup RPCs. This
2627 could leave the node unresponsive to lustre network traffic
2628 and manifested as a server ptllnd timeout.
2630 Severity : enhancement
2632 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2633 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2634 boolean to control whether to print peer debug info when a
2635 client's RPC times out.
2638 Frequency : only for kernels with patches from Lustre below 1.4.3
2640 Description: Remove old rdonly API
2641 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2644 Frequency : only for devices with external journals
2646 Description: Set external device read-only also
2647 Details : During a commanded failover stop, we set the disk device
2648 read-only while the server shuts down. We now also set any
2649 external journal device read-only at the same time.
2652 Frequency : when upgrading from 1.4 while trying to change parameters
2654 Description: The wrong (new) MDC name was used when setting parameters for
2655 upgraded MDT's. Also allows changing of OSC (and MDC)
2656 parameters if --writeconf is specified at tunefs upgrade time.
2659 Frequency : when setting specific ost indicies
2661 Description: QOS code breaks on skipped indicies
2662 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2663 created with --index need not be sequential.
2665 Severity : enhancement
2667 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2668 Details : The uninit_groups feature works in conjunction with the kernel
2669 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2670 up the pass1 processing of e2fsck. This is a read-only feature
2671 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2672 mount filesystems that have had this feature enabled.
2674 Severity : enhancement
2676 Description: Improve multi-block allocation algorithm to avoid fragmentation
2677 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2678 allocation locality and avoid filesystem fragmentation.
2680 ------------------------------------------------------------------------------
2682 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2684 * Support for kernels:
2685 2.4.21-47.0.1.EL (RHEL 3)
2686 2.6.5-7.283 (SLES 9)
2687 2.6.9-42.0.10.EL (RHEL 4)
2688 2.6.12.6 vanilla (kernel.org)
2689 2.6.16.27-0.9 (SLES 10)
2690 * Recommended e2fsprogs version: 1.39.cfs5
2692 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2696 Frequency : occasional, depends on client load and configuration
2697 Bugzilla : 12181, 12203
2698 Description: data loss for recently-modified files
2700 Details : In some cases it is possible that recently written or created
2701 files may not be written to disk in a timely manner (this should
2702 normally be within 30s unless client IO load is very high).
2703 The problem appears as zero-length files or files that are a
2704 multiple of 1MB in size after a client crash or client eviction
2705 that are missing data at the end of the file.
2707 This problem is more likely to be hit on clients where files are
2708 repeatedly created and unlinked in the same directory, clients
2709 have a large amount of RAM, have many CPUs, the filesystem has
2710 many OSTs, the clients are rebooted frequently, and/or the files
2711 are not accessed by other nodes after being written.
2713 The presence of the problem can be detected by looking at
2714 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2715 smaller than the second (nr_unused) then dirty files will not
2716 be flushed automatically to disk. "sync; sleep 10" should be
2717 run several times on the node before unmounting it to update
2718 Lustre (this is also safe to run on nodes without this problem).
2720 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2721 that can cause this same problem, so customers using that kernel
2722 also need to update the kernel in addition to Lustre. In order
2723 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2725 It is normal that files written just before a client crash (less
2726 than 30s) may not yet have been flushed to disk, even for local
2730 Frequency : frequent on thin XT3 nodes
2732 Description: UUID collision on thin XT3 Linux nodes
2733 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2734 randomly, since we relied on an insufficiently-seeded PRNG.
2739 Description: OSS hangs after "All ost request buffers busy"
2740 Details : A deadlock between quota and journal operations caused OSS
2741 hangs after printing "All ost request buffers busy."
2744 Frequency : always on liblustre builds
2746 Description: Cleanup compiler warnings on liblustre
2749 Frequency : always on liblustre builds on XT3
2751 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2755 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2756 > 32000 subdirectories
2757 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2758 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2759 subdirectory count crosses 32000. This will aid e2fsck to
2760 correctly handle more than 32000 subdirectories.
2763 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2765 Description: System hang on startup
2766 Details : This bug allowed the liblustre (e.g. catamount) client to
2767 return to the app before handling all startup RPCs. This
2768 could leave the node unresponsive to lustre network traffic
2769 and manifested as a server ptllnd timeout.
2771 Severity : enhancement
2773 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2774 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2775 boolean to control whether to print peer debug info when a
2776 client's RPC times out.
2781 Description: make O_SYNC working on 2.6 kernels
2782 Details : 2.6 kernels use different method for mark pages for write,
2783 so need add a code to lustre for O_SYNC work.
2788 Description: Failure to close file and release space on NFS
2789 Details : Put inode details into lock acquired in ll_intent_file_open.
2790 Use mdc_intent_lock in ll_intent_open to properly
2791 detect all kind of errors unhandled by mdc_enqueue.
2796 Description: proc file read during shutdown sometimes raced obd removal,
2798 Details : Add lock to prevent obd access after proc file removal.
2801 Frequency : Only for files larger than 4GB on 32-bit clients.
2803 Description: improperly doing page alignment of locks
2804 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2805 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2810 Description: RPCs being resent when they shouldn't be
2811 Details : Some RPCs that should not be resent are being resent. This
2812 can cause inconsistencies in the RPC state machine. Do not
2813 resend such requests.
2816 Frequency : rare, only with NFS export
2818 Description: Crash on NFS re-export node
2819 Details : under very unusual load conditions an assertion is hit in
2820 ll_intent_file_open()
2823 Frequency : only if OST filesystem is corrupted
2825 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2826 Details : for a short time RPCs with bulk IO are in the replay list,
2827 but replay of bulk IOs is unimplemented. If the OST filesystem
2828 is corrupted due to disk cache incoherency and then replay is
2829 started it is possible to trip an assertion. Avoid putting
2830 committed RPCs into the replay list at all to avoid this issue.
2835 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2836 Details : Large single O_DIRECT read and write calls can fail to allocate
2837 a sufficiently large buffer to process the request. In case of
2838 allocation failure the allocation is retried with a smaller
2839 buffer and broken into smaller requests.
2841 Severity : enhancement
2843 Description: Add -o localflock option to simulate old noflock behaviour.
2844 Details : This will achieve local-only flock/fcntl locks coherentness.
2849 Description: versioning check is incomplete
2850 Details : Checking the version difference of client vs. server, report
2851 error if the gap is too big.
2856 Description: add support PG_writeback bit
2857 Details : add support for PG_writeback bit for Lustre, for more carefull
2858 work with page cache in 2.6 kernel. This also fix some deadlocks
2859 and remove hack for work O_SYNC with 2.6 kernel.
2861 Severity : enhancement
2863 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2864 Details : The uninit_groups feature works in conjunction with the kernel
2865 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2866 up the pass1 processing of e2fsck. This is a read-only feature
2867 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2868 mount filesystems that have had this feature enabled.
2870 Severity : enhancement
2872 Description: Improve multi-block allocation algorithm to avoid fragmentation
2873 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2874 allocation locality and avoid filesystem fragmentation.
2876 ------------------------------------------------------------------------------
2878 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2880 * Support for kernels:
2881 2.6.9-42.0.3.EL (RHEL 4)
2882 2.6.5-7.276 (SLES 9)
2883 2.4.21-47.0.1.EL (RHEL 3)
2884 2.6.12.6 vanilla (kernel.org)
2885 2.6.16.21-0.8 (SLES10)
2886 * Recommended e2fsprogs version: 1.39.cfs2-0
2888 * The backwards-compatible /proc/sys/portals symlink has been removed
2889 in this release. Before upgrading, please ensure that you change
2890 any configuration scripts or /etc/sysctl.conf files that access
2891 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2892 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2893 in advance of the upgrade on any system running Lustre 1.4.6 or
2894 newer, since /proc/sys/lnet was added in that version.
2895 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2899 Frequency : only when quota is used
2901 Description: avoid scanning export list for quota master
2902 Details : Change the algorithms to avoid scanning export list in order
2903 to improve the efficiency.
2906 Frequency : MDS failover only, very rarely
2908 Description: "went back in time" messages on mds failover
2909 Details : The greatest transno may be lost when the current operation
2910 finishes with an error (transno==0) and the client's last_rcvd
2911 record is over-written. Save the greatest transno in the
2912 mds_last_transno for this case.
2915 Frequency : always for specific kernels and striping counts
2917 Description: client may get "Matching packet too big" without ACL support
2918 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2919 when trying to access files in certain configurations. The
2920 clients should in fact be denied when mounting because they do
2921 not understand ACLs.
2924 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2926 Description: many clients connecting with IO in progress causes connect timeouts
2927 Details : Avoid synchronous journal commits to avoid delays caused by many
2928 clients connecting/disconnecting when bulk IO is in progress.
2929 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2930 OST_IO_PORTAL to avoid delays behind potentially many pending
2934 Frequency : occasionally with multiple writers to a single file
2936 Description: shared writes to file may result in wrong size reported by stat()
2937 Details : Allow growing of kms when extent lock is cancelled
2940 Frequency : always with random mmap IO to multi-striped file
2942 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2943 Details : Only if the hole is at the end of OST object so that kms is too
2944 small. Fix is to increase kms accordingly in ll_nopage.
2947 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2949 Description: writes to a missing object would leak memory on the OST
2950 Details : If there is an inconsistency between the MDS and OST filesystems,
2951 such that the MDS references an object that doesn't exist, writes
2952 to that object will leak memory due to incorrect cleanup in the
2953 error handling path, eventually running out of memory on the OST.
2958 Description: Creating too long symlink causes lustre errors
2959 Details : Check symlink and name lengths before sending requests to MDS.
2962 Frequency : only if flock is enabled (not on by default)
2964 Description: posix locks not released on fd closure on 2.6.9+
2965 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2966 kernels, this caused such locks not to be released on fd close and
2967 then assertions on fs unmount about still used locks.
2970 Frequency : MDS failover only, very rarely
2972 Description: clients may get ASSERTION(granted_lock != NULL)
2973 Details : When request was taking a long time, and a client was resending
2974 a getattr by name lock request. The were multiple lock requests
2975 with the same client lock handle and
2976 mds_getattr_name->fixup_handle_for_resent_request found one of the
2977 lock handles but later failed with ASSERTION(granted_lock != NULL).
2982 Description: handle->h_buffer_credits > 0, assertion failure
2983 Details : h_buffer_credits is zero after truncate, causing assertion
2984 failure. This patch extends the transaction or creates a new
2988 Frequency : NFS re-export or patchless client
2989 Bugzilla : 11179, 10796
2990 Description: Crash on NFS re-export node (__d_move)
2991 Details : We do not want to hash the dentry if we don't have a lock.
2992 But if this dentry is later used in d_move, we'd hit uninitialised
2993 list head d_hash, so we just do this to init d_hash field but
2994 leave dentry unhashed.
2997 Frequency : NFS re-export or patchless client
2999 Description: NFS exports has problem with symbolic link
3000 Details : lustre client didn't properly install dentry when re-exported
3001 to NFS or running patchless client.
3004 Frequency : NFS re-export or patchless client
3006 Description: Various nfs/patchless fixes.
3007 Details : fixes reuse disconected alias for lookup process - this fixes
3008 warning "find_exported_dentry: npd != pd",
3009 fix permission error with open files at nfs.
3010 fix apply umask when do revalidate.
3013 Frequency : occasional
3015 Description: Crash on NFS re-export node
3016 Details : calling clear_page() on the wrong pointer triggered oops in
3017 generic_mapping_read().
3020 Frequency : rarely, using O_DIRECT IO
3022 Description: unaligned directio crashes client with LASSERT
3023 Details : check for unaligned buffers before trying any requests.
3026 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3028 Description: stale data returned from RAID cache
3029 Details : If only a small amount of IO is done to the RAID device before
3030 reading it again it is possible to get stale data from the RAID
3031 cache instead of reading it from disk.
3034 Frequency : always for sles10 kernel
3036 Description: sles10 support
3037 Details : ll_follow_link: compile fixes and using of nd_set_link
3038 under newer kernels.
3041 Frequency : depends on arch, kernel and compiler version, always on sles10
3044 Description: recursive or deep enough symlinks cause stack overflow
3045 Details : getting rid of large stack-allocated variable in
3049 Frequency : depends on hardware
3051 Description: lustre write performance loss in the SLES10 kernel
3052 Details : the performance loss is caused by using of write barriers in the
3053 ext3 code. The SLES10 kernel turns barrier support on by
3054 default. The fix is to undo that change for ldiskfs.
3056 ------------------------------------------------------------------------------
3058 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3060 * Support for kernels:
3061 2.6.9-42.0.3EL (RHEL 4)
3062 2.6.5-7.276 (SLES 9)
3063 2.4.21-47.0.1.EL (RHEL 3)
3064 2.6.12.6 vanilla (kernel.org)
3068 Frequency : quota enabled and large files being deleted
3070 Description: releasing more than 4GB of quota at once hangs OST
3071 Details : If a user deletes more than 4GB of files on a single OST it
3072 will cause the OST to spin in an infinite loop. Release
3073 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3078 Description: statfs data retrieved from /proc may be stale or zero
3079 Details : When reading per-device statfs data from /proc, in the
3080 {kbytes,files}_{total,free,avail} files, it may appear
3081 as zero or be out of date.
3084 Frequency : systems with MD RAID1 external journal devices
3086 Description: lconf's call to blkid is confused by RAID1 journal devices
3087 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3088 of returning all block devices that match the journal UUID.
3091 Frequency : always, for aggregate stripe size over 4GB
3093 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3094 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3095 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3098 Frequency : always if "lfs find" used on a local file/directory
3100 Description: "lfs find" segfaults if used on a local file/directory
3101 Details : The case where a directory component was not specified wasn't
3102 handled correctly. Handle this properly.
3105 Frequency : always on ppc64
3107 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3108 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3109 perform endianness conversion.
3112 Frequency : rarely (truncate to non-zero file size after write under load)
3113 Bugzilla : 10730, 10687
3114 Description: Files padded with zeros to next 4K multiple
3115 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3116 it is possible that files that are truncated to a non-zero size
3117 immediately after being written are filled with zero bytes beyond
3118 the truncated size. No file data is lost.
3120 Severity : enhancement
3122 Description: Allow recovery/failover for liblustre clients.
3123 Details : liblustre clients were unaware of failover configurations until
3126 Severity : enhancement
3128 Description: user file locks should fail when not mounting with flock option
3129 Details : Set up an error-returning stub in ll_file_operations.lock field
3130 to prevent incorrect behaviour when client is mounted without
3131 flock option. Also, set up properly f_op->flock field for
3135 Frequency : always on ia64
3137 Description: "lfs df" loops on printing out MDS statfs information
3138 Details : The obd_ioctl_data was not initialized and in some systems
3139 this caused a failure during the ioctl that did not return
3140 an error. Initialize the struct and return an error on failure.
3143 Frequency : SLES 9 only
3145 Description: Error of copying files with lustre special EAs as root
3146 Details : Client side always return success for setxattr call for lustre
3147 special xattr (currently only "trusted.lov").
3150 Frequency : rarely on clusters with both ia64+i386 clients
3152 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3153 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3154 clients are doing concurrent writes to the same file it is
3155 possible that the ia64 clients may LASSERT because the OST
3156 extent locks are not PAGE_SIZE aligned. Ensure that grown
3157 locks are always aligned on the request boundary.
3160 Frequency : specific use, occasional
3162 Description: Overwriting in use executable truncates on-disk binary image
3163 Details : If one node attempts to overwrite an executable in use by
3164 another node, we now correctly return ETXTBSY instead of
3165 truncating the file.
3167 Severity : enhancement
3169 Description: Async OSC create to avoid the blocking unnecessarily.
3170 Details : If a OST has no remain object, system will block on the creating
3171 when need to create a new object on this OST. Now, ways use
3172 pre-created objects when available, instead of blocking on an
3173 empty osc while others are not empty. If we must block, we block
3174 for the shortest possible period of time.
3179 Description: chmod on Lustre root is propagated to other clients
3180 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3181 invalid by the follow_mount time.
3184 Frequency : liblustre clients only
3186 Description: Race in 'instant cancel' lock handling could lead to such locks
3187 never to be granted in case of SMP MDS
3188 Details : Do not destroy not yet granted but cbpending locks in
3192 Frequency : replay/resend of open
3194 Description: non null lock assetion failure in mds_intent_policy
3195 Details : Trying to replay/resend lockless open requests resulted in
3196 mds_open() returning 0 with no lock. Now it sets a flag if
3197 a lock is going to be returned.
3199 Severity : enhancement
3201 Description: Checksum enhancements
3202 Details : New checksum enhancements allow for resending RPCs that failed
3205 Severity : enhancement
3207 Description: Tunables on number of dirty pages in cacche
3208 Details : Allow to set limit on number of dirty pages cached.
3213 Description: client crash on unmount - lock still has references
3214 Details : In some error handling cases it was possible to leak a lock
3215 reference on a client while accessing a file. This was not
3216 harmful to the client during operation, but would cause the
3217 client to crash when the filesystem is unmounted.
3220 Frequency : specific case, rare
3222 Description: ETXTBSY on mds though file not in use by client
3223 Details : ETXTBSY is no longer incorrectly returned when attempting to
3224 chmod or chown a directory that the user previously tried to
3225 execute or a currently-executing binary.
3228 Frequency : extremely rare except on liblustre-based clients
3230 Description: Lustre space not freed when files are deleted
3231 Details : Clean up open-unlinked files after client eviction. Previously
3232 the unlink was skipped and the files remained as orphans.
3237 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3238 Details : In some cases it was possible to send a blocking callback to
3239 a client doing a glimpse, even though that client didn't get
3240 a lock granted. When the glimpse lock is cancelled on the OST
3241 the freed lock is left on the waiting list and corrupted the list.
3244 Frequency : all core dumps
3246 Description: Broke core dumps to lustre
3247 Details : Negative dentry may be unhashed if parent does not have UPDATE
3248 lock, but some callers, e.g. do_coredump, expect dentry to be
3249 hashed after successful create, hash it in ll_create_it.
3251 ------------------------------------------------------------------------------
3253 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3255 * Support for kernels:
3256 2.6.9-42.0.2.EL (RHEL 4)
3257 2.6.5-7.276 (SLES 9)
3258 2.4.21-40.EL (RHEL 3)
3259 2.6.12.6 vanilla (kernel.org)
3263 Frequency : always on RHEL 3
3265 Description: Number of open files grows over time
3266 Details : The number of open files grows over time, whether or not
3267 Lustre is started. This was due to a filp leak introduced
3268 by one of our kernel patches.
3270 ------------------------------------------------------------------------------
3272 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3274 * Support for kernels:
3275 2.6.9-42.EL (RHEL 4)
3276 2.6.5-7.276 (SLES 9)
3277 2.4.21-40.EL (RHEL 3)
3278 2.6.12.6 vanilla (kernel.org)
3283 Bugzilla : 5719, 9635, 9792, 9684
3284 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3285 Details : If a server is under heavy load and cannot reply to new
3286 connection requests before the client resends the (re)connect,
3287 the connection handling code can behave badly if two service
3288 threads are concurrently handing separate (re)connections from
3289 the same client. Add better locking to the connection handling
3290 code, and ensure that only a single connection will be processed
3291 for a given client UUID, even if the lock is dropped.
3293 Severity : enhancement
3295 Description: add TCP zero-copy support to kernel
3296 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3297 sends if the hardware supports scatter-gather and checksumming.
3298 This allows socklnd to do client-write and server-read more
3299 efficiently and reduce CPU utilization from skbuf copying.
3302 Frequency : only if NFS exporting from client
3304 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3305 Details : If setting attributes on a file created under NFS that had
3306 never been opened it would be possible to oops the client
3307 if the file had no objects.
3310 Frequency : always for liblustre
3312 Description: liblustre client does MDS+OSTs setattr RPC for each write
3313 Details : When doing a write from a liblustre client, the client
3314 incorrectly issued an RPC to the MDS and each OST the file was
3315 striped over in order to update the timestamps. When writing
3316 with small chunks and many clients this could overwhelm the MDS
3317 with RPCs. In all cases it would slow down the write because
3318 these RPCs are unnecessary.
3320 Severity : enhancement
3322 Description: allow number of MDS service threads to be changed at module load
3323 Details : It is now possible to change the number of MDS service threads
3324 running. Adding "options mds mds_num_threads={N}" to the MDS's
3325 /etc/modprobe.conf will set the number of threads for the next
3326 time Lustre is restarted (assuming the "mds" module is also
3327 reloaded at that time). The default number of threads will
3328 stay the same, 32 for most systems.
3333 Description: OST crash if filesystem is unformatted or corrupt
3334 Details : If an OST is started on a device that has never been formatted
3335 or if the filesystem is corrupt and cannot even mount then the
3336 error handling cleanup routines would dereference a NULL pointer.
3341 Description: NULL pointer deref in llap_from_page.
3342 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3343 incorrect) mapping if the page was truncated/reclaimed while it was
3344 searched for. Check for this condition and skip such pages when
3345 doing readahead. Introduce extra check to llap_from_page() to
3346 verify page->mapping->host is non-NULL (so page is not anonymous).
3349 Frequency : Sometimes when using sys_sendfile
3351 Description: "page not covered by a lock" warnings from ll_readpage
3352 Details : sendfile called ll_readpage without right page locks present.
3353 Now we introduced ll_file_sendfile that does necessary locking
3354 around call to generic_file_sendfile() much like we do in
3358 Frequency : with certain MDS communication failures at client mount time
3360 Description: NULL pointer deref after failed client mount
3361 Details : a client connection request may delayed by the network layer
3362 and not be sent until after the PTLRPC layer has timed out the
3363 request. If the client fails the mount immediately it will try
3364 to clean up before the network times out the request. Add a
3365 reference from the request import to the obd device and delay
3366 the cleanup until the network drops the request.
3369 Frequency : occasionally during client (re)connect
3371 Description: assertion failure during client (re)connect
3372 Details : processing a client connection request may be delayed by the
3373 client or server longer than the client connect timeout. This
3374 causes the client to resend the connection request. If the
3375 original connection request is replied in this interval, the
3376 client may trip an assertion failure in ptlrpc_connect_interpret()
3377 which thought it would be the only running connect process.
3380 Frequency : only with obd_echo servers and clients that are rebooted
3382 Description: kernel BUG accessing uninitialized data structure
3383 Details : When running an obd_echo server it did not start the ping_evictor
3384 thread, and when a client was evicted an uninitialized data
3385 structure was accessed. Start the ping_evictor in the RPC
3386 service startup instead of the OBD startup.
3388 Severity : enhancement
3389 Bugzilla : 10193 (patchless)
3390 Description: Remove dependency on various unexported kernel interfaces.
3391 Details : No longer need reparent_to_init, exit_mm, exit_files,
3392 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3395 Frequency : rare (only users of deprecated and unsupported LDAP config)
3397 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3398 Details : LDAP apparently contains 'lustreName' attributes instead of
3399 'name'. A simple remapping of the name is sufficient.
3402 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3404 Description: waiting_locks_callback trips kernel BUG if client is evicted
3405 Details : Running with the dump_on_timeout debug flag turned on makes
3406 it possible that the waiting_locks_callback() can try to dump
3407 the Lustre kernel debug logs from an interrupt handler. Defer
3408 this log dumping to the expired_lock_main() thread.
3410 Severity : enhancement
3412 Description: Support NFS exporting on 2.6 kernels.
3413 Details : Implement non-rawops metadata methods for NFS server to use without
3414 changing NFS server code.
3417 Frequency : very rare (synthetic metadata workload only)
3419 Description: two racing renames might cause an MDS thread to deadlock
3420 Details : Running the "racer" program may cause one MDS thread to rename
3421 a file from being the source of a rename to being the target of
3422 a rename at exactly the same time that another thread is doing
3423 so, and the second thread has already enqueued these locks after
3424 doing a lookup of the target and is trying to relock them in
3425 order. Ensure that we don't try to re-lock the same resource.
3428 Frequency : only very large systems with liblustre clients
3430 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3431 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3432 all OSTs in parallel. This allows the eviction of stale liblustre
3433 clients to proceed much faster than if they were done in series,
3434 and also offers similar improvements for other set_info RPCs.
3439 Description: excessive CPU usage during initial read phase on client
3440 Details : During the initial read phase on a client, it would agressively
3441 retry readahead on the file, consuming too much CPU and impacting
3442 performance (since 1.4.5.8). Improve the readahead algorithm
3443 to avoid this, and also improve some other common cases (read
3444 of small files in particular, where "small" is files smaller than
3445 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3450 Description: MDS crash when receiving packet with unknown intent.
3451 Details : Do not LBUG in unknown intent case, just return -EFAULT
3453 Severity : enhancement
3454 Bugzilla : 9293, 9385
3455 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3456 Details : Do not serialize getattr (non-intent version) and statfs.
3459 Frequency : occasional, when OST network is overloaded/intermittent
3461 Description: client evicted by OST after bulk IO timeout
3462 Details : If a client sends a bulk IO request (read or write) the OST
3463 may evict the client if it is unresposive to its data GET/PUT
3464 request. This is incorrect if the network is overloaded (takes
3465 too long to transfer the RPC data) or dropped the OST GET/PUT
3466 request. There is no need to evict the client at all, since
3467 the pinger and/or lock callbacks will handle this, and the
3468 client can restart the bulk request.
3471 Frequency : Always when mmapping file with no objects
3473 Description: client crashes when mmapping file with no objects
3474 Details : Check that we actually have objects in a file before doing any
3475 operations on objects in ll_vm_open, ll_vm_close and
3481 Description: Request leak when working with deleted CWD
3482 Details : Introduce advanced request refcount tracking for requests
3483 referenced from lustre intent.
3485 Severity : Enhancement
3487 Description: Cache open file handles on client.
3488 Details : MDS now will return special lock along with openhandle, if
3489 requested and client is allowed to hold openhandle, even if unused,
3490 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3491 closing files for every read/write openration.
3493 Severity : Enhancement
3495 Description: Cache open negative dentries on client when possible.
3496 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3497 negative dentries on lock revocation.
3502 Description: Remounting a client read-only wasn't possible with a zconf mount
3503 Details : It wasn't possible to remount a client read-only with llmount.
3505 Severity : enhancement
3506 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3507 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3508 MPICH is included to add Lustre-specific ADIO interfaces.
3509 This is based closely on the UFS ADIO layer and only differs
3510 in file creation, in order to allow the OST striping to be set.
3511 This is user-contributed code and not supported by CFS.
3516 Description: extended inode attributes (immutable, append-only) work improperly
3517 when 2.4 and 2.6 kernels are used on client/server or vice versa
3518 Details : Introduce kernel-independent values for these flags.
3520 Severity : enhancement
3523 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3524 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3525 granularity. Now, Fractional MB granularity is be supported,
3526 this is very useful for low memory system.
3528 Severity : enhancement
3530 Description: Getattr by fid
3531 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3532 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3533 have locks and avoids getattr rpc requests for it on every stat.
3536 Frequency : Always, for filesystems larger than 2TB
3538 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3539 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3540 filesystems larger than 2TB to be mounted with mballoc enabled.
3543 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3545 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3546 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3547 filesystems larger than 2TB is mounted then the kernel will
3548 silently corrupt the start of the filesystem. CONFIG_LBD is
3549 enabled for all CFS-supported kernels, but the possibility of
3550 this happening with a modified kernel config exists.
3552 Severity : enhancement
3554 Description: add client O_DIRECT support for 2.6 kernels
3555 Details : It is now possible to do O_DIRECT reads and writes to files
3556 in the Lustre client mountpoint on 2.6 kernel clients.
3558 Severity : enhancement
3560 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3561 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3562 parallel, not waiting for response from every OST before sending
3563 a rpc to the next OST.
3568 Description: setattr vs write race when updating file timestamps
3569 Details : Client processes that update a file timestamp into the past
3570 right after writing to the file (e.g. tar) it is possible that
3571 the updated file modification time can be reset to the current
3572 time due to a race between processing the setattr and write RPC.
3574 Severity : enhancement
3576 Description: Bring 'lfs find' closer in line with regular Linux find.
3577 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3578 -print0 options and obtains all the needed info through the lustre
3581 Severity : enhancement
3583 Description: support up to 1024 configured devices on one node
3584 Details : change obd_dev array from statically allocated to dynamically
3585 allocated structs as they are first used to reduce memory usage
3590 Description: Flush dirty partially truncated pages during truncate
3591 Details : Immediatelly flush partially truncated pages in filter_setattr,
3592 this way we completely avoid having any pages in page cache on OST
3593 and can retire ugly workarounds during writes to flush such pages.
3598 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3599 Details : protect inode from truncation within vfs_unlink() context
3600 just take a reference before calling vfs_unlink() and release it
3601 when parent's i_sem is free.
3606 Description: last_id value checked outside lock on OST caused LASSERT failure
3607 Details : If there were multiple MDS->OST object precreate requests in
3608 flight, it was possible that the OST's last object id was checked
3609 outside a lock and incorrectly tripped an assertion. Move checks
3610 inside locks, and discard old precreate requests.
3613 Frequency : always, if extents are used on OSTs
3615 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3616 Details : OSTs using the extents format would not zero the high 16 bits of
3617 the index physical block number. This is not a problem for any
3618 OST filesystems smaller than 16TB, and no kernels support ext3
3619 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3620 new/modified files) and can be fixed for existing filesystems
3621 with e2fsprogs-1.39-cfs1.
3626 Description: import connection selection may be incorrect if timer wraps
3627 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3628 import connections to be ignored if the 32-bit jiffies counter
3629 wraps. Use a 64-bit jiffies counter.
3632 Frequency : very large clusters immediately after boot
3634 Description: LNET request buffers exhausted under heavy short-term load
3635 Details : If a large number of client requests are generated on a service
3636 that has previously never seen so many requests it is possible
3637 that the request buffer growth cannot keep up with the spike in
3638 demand. Instead of dropping incoming requests, they are held in
3639 the LND until the RPC service can accept more requests.
3642 Frequency : Sometimes during replay
3644 Description: Assertion failure in ll_local_open after replay.
3645 Details : If replay happened on an open request reply before we were able
3646 to set replay handler, reply will become not swabbed tripping the
3647 assertion in ll_local_open. Now we set the handler right after
3648 recognising of open request
3651 Frequency : very rare
3653 Description: kernel reports "badness in vsnprintf"
3654 Details : Reading from the "recovery_status" /proc file in small chunks
3655 may cause a negative length in lprocfs_obd_rd_recovery_status()
3656 call to vsnprintf() (which is otherwise harmless). Exit early
3657 if there is no more space in the output buffer.
3659 Severity : enhancement
3661 Description: clear OBD RPC statistics by writing to them
3662 Details : It is now possible to clear the OBD RPC statistics by writing
3663 to the "stats" file.
3668 Description: Client mtime is not the same on different clients after utimes
3669 Details : In some cases, the client was using the utimes() syscall on
3670 a file cached on another node. The clients now validate the
3671 ctime from the MDS + OSTs to determine which one is right.
3676 Description: Inability to activate failout mode
3677 Details : lconf script incorrectly assumed that in pythong string's numeric
3678 value is used in comparisons.
3681 Frequency : always with multiple stripes per file
3683 Description: Inefficient object allocation for mutli-stripe files
3684 Details : When selecting which OSTs to stripe files over, for files with
3685 a stripe count that divides evenly into the number of OSTs,
3686 the MDS is always picking the same starting OST for each file.
3687 Return the OST selection heuristic to the original design.
3692 Description: mount failures may take full timeout to return an error
3693 Details : Under some heavy load conditions it is possible that a
3694 failed mount can wait for the full obd_timeout interval,
3695 possibly several minutes, before reporting an error.
3696 Instead return an error as soon as the status is known.
3698 Frequency : quota enabled and large files being deleted
3700 Description: releasing more than 4GB of quota at once hangs OST
3701 Details : If a user deletes more than 4GB of files on a single OST it
3702 will cause the OST to spin in an infinite loop. Release
3703 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3708 Description: statfs data retrieved from /proc may be stale or zero
3709 Details : When reading per-device statfs data from /proc, in the
3710 {kbytes,files}_{total,free,avail} files, it may appear
3711 as zero or be out of date.
3714 Frequency : systems with MD RAID1 external journal devices
3716 Description: lconf's call to blkid is confused by RAID1 journal devices
3717 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3718 of returning all block devices that match the journal UUID.
3721 Frequency : always, for aggregate stripe size over 4GB
3723 Description: assertion fails when trying to use 4GB stripe size
3724 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3725 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3728 Frequency : always on ppc64
3730 Description: the first write on an ext3 filesystem with mballoc got stuck
3731 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3732 perform endianness conversion.
3734 ------------------------------------------------------------------------------
3736 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3738 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3739 INTEROPERATE with older versions automatically. Please read the
3740 user documentation before upgrading any part of a live system.
3741 * WARNING: Lustre networking configuration changes are required with
3742 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3745 * Support for kernels:
3746 2.6.9-22.0.2.EL (RHEL 4)
3747 2.6.5-7.244 (SLES 9)
3748 2.6.12.6 vanilla (kernel.org)
3751 Severity : enhancement
3752 Bugzilla : 7981/8208
3753 Description: Introduced Lustre Networking (LNET)
3754 Details : LNET is new networking infrastructure for Lustre, it includes
3755 a reorganized network configuration mode (see the user
3756 documentation for full details) as well as support for routing
3757 between different network fabrics. Lustre Networking Devices
3758 (LNDs) for the supported network fabrics have also been
3759 created for this new infrastructure.
3761 Severity : enhancement
3762 Description: Introduced Access control lists
3763 Details : clients can set ACLs on files and directories in order to have
3764 more fine-grained permissions than the standard Unix UGO+RWX.
3765 The MDS must be started with the "-o acl" mount option.
3767 Severity : enhancement
3768 Description: Introduced filesystem quotas
3769 Details : Administrators may now establish per-user quotas on the
3772 Severity : enhancement
3774 Description: Configuration change for the XT3
3775 The PTLLND is now used to run Lustre over Portals on the XT3
3776 The configure option(s) --with-cray-portals are no longer used.
3777 Rather --with-portals=<path-to-portals-includes> is used to
3778 enable building on the XT3. In addition to enable XT3 specific
3779 features the option --enable-cray-xt3 must be used.
3784 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3785 Details : A race between allocating a new llcd and re-getting the llcd_lock
3786 allowed another thread to grab newly-allocated llcd.
3788 Severity : enhancement
3790 Description: 2.6 OST async journal commit and locking fix to improve performance
3791 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3792 2.6 kernels are now async as they already were in 2.4 kernels so
3793 that they can commit concurrently with the network bulk transfer.
3794 For block-allocated files the filter allocation semaphore is held
3795 to avoid filesystem fragmentation during allocation. BKL lock
3796 removed for 2.6 xattr operations where it is no longer needed.
3801 Description: lconf incorrectly determined whether two IP networks could talk
3802 Details : In some more complicated routing and multiple-network
3803 configurations, lconf will avoid trying to make a network
3804 connection to a disjoint part of the IP space. It was doing the
3805 math incorrectly for one set of cases.
3810 Description: Fix for potential infinite loop processing records in an llog.
3811 Details : If an llog record is corrupted/zeroed, it is possible to loop
3812 forever in llog_process(). Validate the llog record length
3813 and skip the remainder of the block on error.
3816 Frequency : occasional (liblustre only)
3818 Description: liblustre could not open files whose last component is a symlink
3819 Details : sysio_path_walk() would incorrectly pass the open intent to
3820 intermediate path components.
3823 Frequency : rare (liblustre only with non-standard tuning)
3824 Bugzilla : 7201 (7350)
3825 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3826 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3827 after it has been released, only do this for referenced locks
3829 Severity : enhancement
3831 Description: specifying an (invalid) directory default stripe_size of -1
3832 would reset the directory default striping
3833 Details : stripe_size -1 was used internally to signal directory stripe
3834 removal, now use "all default" to signal dir stripe removal
3835 as a directory striping of "all default" is not useful
3838 Frequency : common for large clusters running liblustre clients
3840 Description: doing an ls when liblustre clients are running is slow
3841 Details : sending a glimpse AST to a liblustre client waits for every AST
3842 to time out, as liblustre clients will not respond. Since they
3843 cannot cache data we refresh the OST lock LVB from disk instead.
3845 Severity : enhancement
3847 Description: doing an ls at the same time as file IO can be slow
3848 Details : enqueue and other "small" requests can be blocked behind many
3849 large IO requests. Create a new OST IO portal for non-IO
3850 requests so they can be processed faster.
3853 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3855 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3856 created file with mode 000
3857 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3858 allow an ftruncate() as a non-root user to a file with mode 000.
3859 Lustre now respects this flag to disable mode checks when
3860 truncating a file owned by the user
3863 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3866 Description: Revoking locks from clients that went dead or catatonic might take
3868 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3869 cancellation of such locks instant on servers without waiting for
3870 any reply from clients. Clients drops these locks when cancel
3871 notification from server is received without replying.
3874 Frequency : liblustre-only, when liblustre client dies or becomes busy
3876 Description: Doing ls on Linux clients can take a long time with active
3878 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3879 granting such locks to it in the first place if possible. Locks
3880 are taken by proxy on the OST during the read or write and
3881 dropped immediately afterward. Add connect flags handling, do
3882 not grant locks to liblustre clients for glimpse ASTs.
3884 Severity : enhancement
3886 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3887 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3888 sequential reads and behaves badly with applications doing random
3889 reads. Improve it by reading ahead at least read region, and
3890 avoiding excessive large RPC for small reads.
3892 Severity : enhancement
3894 Description: Creating more than 1000 files for a single job may cause a load
3895 imbalance on the OSTs if there are also a large number of OSTs.
3896 Details : qos_prep_create() uses an OST index reseed value that is an
3897 even multiple of the number of available OSTs so that if the
3898 reseed happens in the middle of the object allocation it will
3899 still utilize the OSTs as uniformly as possible.
3904 Description: OST or MDS may oops in ping_evictor_main()
3905 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3906 but doesn't restart at beginning of obd_exports_timed list
3909 Severity : enhancement
3911 Description: improve by-nid export eviction on the MDS and OST
3912 Details : allow multiple exports with the same NID to be evicted at one
3913 time without re-searching the exports list.
3916 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3918 Description: MDS may oops in groups_free()
3919 Details : in rare race conditions a newly allocated group_info struct is
3920 freed again, and this can be NULL. The 2.4 compatibility code
3921 for groups_free() checked for a NULL pointer, but 2.6 did not.
3924 Frequency : common for liblustre clients doing little filesystem IO
3925 Bugzilla : 9352, 7313
3926 Description: server may evict liblustre clients accessing contended locks
3927 Details : if a client is granted a lock or receives a completion AST
3928 with a blocking AST already set it would not reply to the AST
3929 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3932 Frequency : lfs setstripe, only systems with more than 160 OSTs
3934 Description: unable to set striping with a starting offset beyond OST 160
3935 Details : llapi_create_file() incorrectly limited the starting stripe
3936 index to the maximum single-file stripe count.
3939 Frequency : LDAP users only
3941 Description: lconf did not handle in-kernel recovery with LDAP properly
3942 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3944 Severity : enhancement
3946 Description: bind OST threads to NUMA nodes to improve performance
3947 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3948 node and do their allocations there to localize memory access
3950 Severity : enhancement
3952 Description: llmount can determine client NID directly from Myrinet (GM)
3953 Details : the client NID code from gmnalnid was moved directly into
3954 llmount, removing the need to use this or specifying the
3955 client NID explicitly when mounting GM clients with zeroconf
3958 Frequency : if client is started with down MDS
3960 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3961 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3962 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3963 ptlrpc_import_delayed_req() because we didn't check if the
3964 request was interrupted, and we also didn't break out of the
3965 event loop if there was no timeout
3970 Description: data loss during non-page-aligned writes to a single file from
3971 both multiple nodes and multiple threads on one node at same time
3972 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3973 inconsistency led to false short-reads, that were cached and later
3974 used by ->prepare_write() to fill in partially written page,
3975 leading to data loss.
3978 Frequency : always, if lconf --abort_recovery used
3980 Description: lconf --abort_recovery fails with 'Operation not supported'
3981 Details : lconf was attempting to abort recovery on the MDT device and not
3984 Severity : enhancement
3986 Description: remove cleanup logs
3987 Details : replace lconf-generated cleanup logs with lustre internal
3988 cleanup routines. Eliminates the need for client-cleanup and
3991 Severity : enhancement
3993 Description: add support for EAs (user and system) on lustre filesystems
3994 Details : it is now possible to store extended attributes in the Lustre
3995 client filesystem, and with the user_xattr mount option it
3996 is possible to allow users to store EAs on their files also
3998 Severity : enhancement
4000 Description: Add possibility (config option) to show minimal available OST free
4002 Details : When compiled with --enable-mindf configure option, statfs(2)
4003 (and so, df) will return least minimal free space available from
4004 all OSTs as amount of free space on FS, instead of summary of
4005 free spaces of all OSTs.
4007 Severity : enhancement
4009 Description: do not expand extent locks acquired on OST-side
4010 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4011 by server: they are not cached anyway.
4014 Frequency : when mmap is used/binaries executed from Lustre
4016 Description: Unmmap pages before throwing them away from read cache.
4017 Details : llap_shrink cache now attempts to unmap pages before discarding
4018 them (if unmapping failed - do not discard). SLES9 kernel has
4019 extra checks that trigger if this unmapping is not done first.
4024 Description: lconf didn't resolve symlinks before checking to see whether a
4025 given mountpoint was already in use
4028 Frequency : when migrating failover services
4029 Bugzilla : 6395, 9514
4030 Description: When migrating a subset of services from a node (e.g. failback
4031 from a failover service node) the remaining services would
4032 time out and evict clients.
4033 Details : lconf --force (implied by --failover) sets the global obd_timeout
4034 to 5 seconds in order to quickly disconnect, but this caused
4035 other RPCs to time out too quickly. Do not change the global
4036 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4038 Severity : enhancement
4039 Frequency : if MDS is started with down OST
4040 Bugzilla : 9439,5706
4041 Description: Allow startup/shutdown of an MDS without depending on the
4042 availability of the OSTs.
4043 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4044 Add appropriate locking and lov-osc refcounts for safe
4045 cleaning. Add osc abort_inflight calls in case the
4046 synchronize never started.
4049 Frequency : occasional (Cray XT3 only)
4051 Description: root not authorized to access files in CRAY_PORTALS environment
4052 Details : The client process capabilities were not honoured on the MDS in
4053 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4054 been accessed by an authorized user then root was able to access
4055 the file on the local client also. The root user capabilities
4056 are now allowed on the MDS, as this environment has secure UID.
4059 Frequency : occasional
4061 Description: ldiskfs "too long searching" message happens too often
4062 Details : A debugging message (otherwise harmless) prints too often on
4063 the OST console. This has been reduced to only happen when
4064 there are fragmentation problems on the filesystem.
4069 Description: Division by zero in statfs when all OSCs are inactive
4070 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4071 lov_statfs divides by value returned by lov_get_stripecnt().
4075 Bugzilla : 9489, 3273
4076 Description: First write from each client to each OST was only 4kB in size,
4077 to initialize client writeback cache, which caused sub-optimal
4078 RPCs and poor layout on disk for the first writen file.
4079 Details : Clients now request an initial cache grant at (re)connect time
4080 and so that they can start streaming writes to the cache right
4081 away and always do full-sized RPCs if there is enough data.
4082 If the OST is rebooted the client also re-establishes its grant
4083 so that client cached writes will be honoured under the grant.
4088 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4089 Details : Now I/O RPCs go to different portal number and (presumably) fast
4090 lock requests (and glimses) and other RPCs get their own service
4091 threads pool that should be able to service those RPCs
4094 Severity : enhancement
4096 Description: Ability to exchange lustre version between client and servers and
4097 issue warnings at client side if client is too old. Also for
4098 liblustre clients there is ability to refuse connection of too old
4100 Details : New 'version' field is added to connect data structure that is
4101 filled with version info. That info is later checked by server and
4105 Frequency : rare, liblustre only.
4106 Bugzilla : 9296, 9581
4107 Description: Two simultaneous writes from liblustre at offset within same page
4108 might proceed at the same time overwriting eachother with stale
4110 Details : I/O lock withing llu_file_prwv was released too early, before data
4111 actually was hitting the wire. Extended lock-holding time until
4112 server acknowledges receiving data.
4115 Frequency : extremely rare. Never observed in practice.
4117 Description: avoid generating lustre_handle cookie of 0.
4118 Details : class_handle_hash() generates handle cookies by incrementing
4119 global counter, and can hit 0 occasionaly (this is unlikely, but
4120 not impossible, because initial value of cookie counter is
4121 selected randonly). Value of 0 is used as a sentinel meaning
4122 "unassigned handle" --- avoid it. Also coalesce two critical
4123 sections in this function into one.
4125 Severity : enhancement
4127 Description: allow liblustre clients to delegate truncate locking to OST
4128 Details : To avoid overhead of locking, liblustre client instructs OST to
4129 take extent lock in ost_punch() on client's behalf. New connection
4130 flag is added to handle backward compatibility.
4132 Severity : enhancement
4133 Bugzilla : 4928, 7341, 9758
4134 Description: allow number of OST service threads to be specified
4135 Details : a module parameter allows the number of OST service threads
4136 to be specified via "options ost ost_num_threads={N}" in the
4137 OSS's /etc/modules.conf or /etc/modprobe.conf.
4141 Bugzilla : 6146, 9635, 9895
4142 Description: servers crash with bad pointer in target_handle_connect()
4143 Details : In rare cases when a client is reconnecting it was possible that
4144 the connection request was the last reference for that export.
4145 We would temporarily drop the export reference and get a new
4146 one, but this may have been the last reference and the export
4147 was just destroyed. Get new reference before dropping old one.
4149 Severity : enhancement
4150 Frequency : if client is started with failover MDS
4152 Description: Allow multiple MDS hostnames in the mount command
4153 Details : Try to read the configuration from all specified MDS
4154 hostnames during a client mount in case the "primary"
4157 Severity : enhancement
4159 Description: Stop sending data to evicted clients as soon as possible.
4160 Details : Check if the client we are about to send or are sending data to
4161 was evicted already. (Check is done every second of waiting,
4162 for which l_wait_event interface was extended to allow checking
4163 of exit condition at specified intervals).
4166 Frequency : rare, normally only when NFS exporting is done from client
4168 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4170 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4173 Frequency : always, if srand() is called before liblustre initialization
4175 Description: Liblustre uses system PRNG disturbing its usage by user application
4176 Details : Introduce internal to lustre fast and high-quality PRNG for
4177 lustre usage and make liblustre and some other places in generic
4178 lustre code to use it.
4180 Severity : enhancement
4181 Bugzilla : 9477, 9557, 9870
4182 Description: Verify that the MDS configuration logs are updated when xml is
4183 Details : Check if the .xml configuration logs are newer than the config
4184 logs stored on the MDS and report an error if this is the case.
4185 Request --write-conf, or allow starting with --old_conf.
4187 Severity : enhancement
4189 Description: Handle symlinks in the path when checking if Lustre is mounted.
4190 Details : Resolve intermediate symlinks when checking if a client has
4191 mounted a filesystem to avoid duplicate client mounts.
4196 Description: lconf can hit an error exception but still return success.
4197 Details : The lconf command catches the Command error exception at the top
4198 level script context and will exit with the associated exit
4199 status, but doesn't ensure that this exit status is non-zero.
4204 Description: failure of ptlrpc thread startup can cause oops
4205 Details : Starting a ptlrpc service thread can fail if there are a large
4206 number of threads or the server memory is very fragmented.
4207 Handle this without oopsing.
4210 Frequency : always, only if liblustre and non-default acceptor port was used
4212 Description: liblustre cannot connect to servers with non-default acceptor port
4213 Details : tcpnal_set_default_params() was not called and was therefore
4214 ignoring the environment varaible TCPNAL_PORT, as well as other
4215 TCPNAL_ environment variables
4220 Description: two objects could be created on the same OST for a single file
4221 Details : If an OST is down, in some cases it was possible to create two
4222 objects on a single OST for a single file. No problems other
4223 than potential performance impact and spurious error messages.
4227 Bugzilla : 5681, 9562
4228 Description: Client may oops in ll_unhash_aliases
4229 Details : Client dcache may become inconsistent in race condition.
4230 In some cases "getcwd" can fail if the current directory is
4236 Description: Inode refcounting problems in NFS export code
4237 Details : link_raw functions used to call d_instantiate without obtaining
4238 extra inode reference first.
4242 Bugzilla : 9942, 9903
4243 Description: Referencing freed requests leading to crash, memleaks with NFS.
4244 Details : We used to require that call to ll_revalidate_it was always
4245 followed by ll_lookup_it. Also with revalidate_special() it is
4246 possible to call ll_revalidate_it() twice for the same dentry
4247 even if first occurence returned success. This fix changes semantic
4248 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4249 reference on a request referred from the intent.
4250 ll_intent_release() then releases such a request.
4253 Frequency : rare, normally benchmark loads only
4255 Description: unlinked inodes were kept in memory on the client
4256 Details : If a client is repeatedly creating and unlinking files it
4257 can accumulate a lot of stale inodes in the inode slab cache.
4258 If there is no other client load running this can cause the
4259 client node to run out of memory. Instead flush old inodes
4260 from client cache that have the same inode number as a new inode.
4263 Frequency : SLES9 2.6.5 kernel and long filenames only
4264 Bugzilla : 9969, 10379
4265 Description: utime reports stale NFS file handle
4266 Details : SLES9 uses out-of-dentry names in some cases, which confused
4267 the lustre dentry revalidation. Change it to always use the
4271 Frequency : rare, unless heavy write-truncate concurrency is continuous
4272 Bugzilla : 4180, 6984, 7171, 9963, 9331
4273 Description: OST becomes very slow and/or deadlocked during object unlink
4274 Details : filter_destroy() was holding onto the parent directory lock
4275 while truncating+unlinking objects. For very large objects this
4276 may block other threads for a long time and slow overall OST
4277 responsiveness. It may also be possible to get a lock ordering
4278 deadlock in this case, or run out of journal credits because of
4279 the combined truncate+unlink. Solution is to do object truncate
4280 first in one transaction without parent lock, and then do the
4281 final unlink in a new transaction with the parent lock. This
4282 reduces the lock hold time dramatically.
4285 Frequency : rare, 2.4 kernels only
4287 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4288 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4289 before unmounting their filesystem in order to avoid deadlock.
4290 The kernel_locked() function in 2.4 kernels only checks whether
4291 the kernel lock is held, not whether it is this process that is
4292 holding it as 2.6 kernels do.
4297 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4298 Details : The client ptlrpc code may be trying to reconnect to a down
4299 server before a previous connection attempt has timed out.
4300 Increase the reconnect interval to be longer than the connection
4301 timeout interval to avoid sending duplicate connections to
4305 Frequency : echo_client brw_test command
4307 Description: fix echo_client to work with OST preallocated code
4308 Details : OST preallocation code (5137) didn't take echo_client IO path
4309 into account: echo_client calls filter methods outside of any
4310 OST thread and, hence, there is no per-thread preallocated
4311 pages and buffers to use. Solution: hijack pga pages for IO. As
4312 a byproduct, this avoids unnecessary data copying.
4316 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4317 Description: Client can oops in mdc_commit_close() after open replay
4318 Details : It was possible for the MDS to return an open request with no
4319 transaction number in mds_finish_transno() if the client was
4320 evicted, but without actually returning an error. Clients
4321 would later try to replay that open and may trip an assertion
4322 Simplify the client close codepath, and always return an error
4323 from the MDS in case the open is not successful.
4326 Frequency : rare, 2.6 OSTs only
4328 Description: OST may deadlock under high load on fragmented files
4329 Details : If there was a heavy load and highly-fragmented OST filesystems
4330 it was possible to have all the OST threads deadlock waiting on
4331 allocation of biovecs, because the biovecs were not released
4332 until the entire RPC IO was completed. Instead, release biovecs
4333 as soon as they are complete to ensure forward IO progress.
4335 Severity : enhancement
4337 Description: Support for specifying external journal device at mount
4338 Details : If an OST or MDS device is formatted with an external journal
4339 device, this device major/minor is stored in the ext3 superblock
4340 and may not be valid for failover. Allow detecting and
4341 specifying the external journal at mount time.
4346 Description: Mounting an MDS with pending unlinked files may cause oops
4347 Details : target_finish_recovery() calls mds_postrecov() which returned
4348 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4349 considers this an error and immediately begins cleaning up the
4350 lov, just after starting the mds_lov process
4352 Severity : enhancement
4354 Description: Implement 'lfs df' to report actual free space on per-OST basis
4355 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4356 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4357 usage of objects; '-h' to report in human readable format.
4359 ------------------------------------------------------------------------------
4361 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4368 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4369 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4370 at startup during orphan unlinking. Instead, skip buddy update
4371 before setup, as it will be regenerated after recovery is complete.
4376 Description: If an OST is inactive, its locks might reference stale inodes.
4377 Details : lov_change_cbdata() must iterate over all namespaces, even if
4378 they are inactive to clear inode references from the lock.
4380 Severity : enhancement
4381 Frequency : occasional, if non-standard max_dirty_mb used
4383 Description: Client will block write RPCs if not enough grant
4384 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4385 then the client will block writes while waiting for another RPC
4386 to complete instead of consuming its dirty limit. With change
4387 we get improved performance when max_dirty_mb is small.
4389 Severity : enhancement
4390 Bugzilla : 3389, 6253
4391 Description: Add support for supplementary groups on the MDS.
4392 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4393 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4394 lookups for user supplementary groups into a cache.
4398 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4399 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4400 for liblustre O_EXCL is enforced by MDS.
4404 Description: readdir never returns NULL in liblustre.
4405 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4406 dentry in struct dirent.
4410 Frequency : liblustre only, depends on application IO pattern
4411 Description: liblustre clients evicted if not contacting servers
4412 Details : Don't put liblustre clients into the ping_evictor list, so
4413 they will not be evicted by the pinger ever.
4415 Severity : enhancement
4417 Description: Add ability to evict clients by NID from MDS.
4418 Details : By echoing "nid:$NID" string into
4419 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4420 $NID would be instantly evicted from this MDS and from all active
4421 OSTs connected to it.
4425 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4426 Details : lookup_it_finish() used to query file size from OSTs that was not
4431 Description: service threads change working directory to that of init
4432 Details : Starting lustre service threads may pin the working directory
4433 of the parent thread, making that filesystem busy. Threads
4434 now change to the working directory of init to avoid this.
4438 Frequency : during shutdown only
4439 Description: shutdown with a failed MDS or OST can cause unmount to hang
4440 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4445 Frequency : 2.6 only
4446 Description: chmod/chown may include an extra supplementary group
4447 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4448 supplementary group and if none is specified this is used.
4451 Bugzilla : 5479 (6816)
4453 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4454 Details : If lookup is in progress on a file that is unlinked we might try
4455 to revalidate the inode and fail in revalidate after lookup is
4456 complete and ll_file_open() enqueues the open again but
4457 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4460 Frequency : always, if lconf --abort_recovery used
4462 Description: lconf --abort_recovery fails with 'Operation not supported'
4463 Details : lconf was attempting to abort recovery on the MDT device and not
4466 ------------------------------------------------------------------------------
4468 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4473 Frequency : rare (only unsupported configurations with a node running as an
4475 Bugzilla : 6514, 5137
4476 Description: Mounting a Lustre file system on a node running as an OST could
4478 Details : OSTs now preallocates memory needed to write out data at
4479 startup, instead of when needed, to avoid having to
4480 allocate memory in possibly low memory situations.
4481 Specifically, if the file system is mounted on on OST,
4482 memory pressure could force it to try to write out data,
4483 which it needed to allocate memory to do. Due to the low
4484 memory, it would be unable to do so and the node would
4485 become unresponsive.
4487 Severity : enhancement
4489 Description: Addition of lconf --service command line option
4490 Details : lconf now accepts a '--service <arg>' option, which is
4491 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4493 Severity : enhancement
4495 Description: Failover mode is now the default for OSTs.
4496 Details : By default, OSTs will now run in failover mode. To return to
4497 the old behaviour, add '--failout' to the lmc line for OSTs.
4499 Severity : enhancement
4501 Description: Health checks are now provided for MDS and OSTs
4502 Details : Additional detailed health check information on MSD and OSTs
4503 is now provided through the procfs health_check value.
4506 Frequency : occasional, depends on IO load
4508 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4509 after numerous create/delete cycles
4510 Details : The ext3 inode allocation policy would not allocate new inodes
4511 very well on the OSTs because there are no new directories
4512 being created. Instead we look for groups with free space if
4513 the parent directories are nearly full.
4518 Description: Network or server problems during mount may cause partially
4519 mounted clients instead of returning an error.
4520 Details : The config llog parsing code may overwrite the error return
4521 code during mount error handling, returning success instead
4527 Description: MDS can fail to allocate large reply buffers
4528 Details : After long uptimes the MDS can fail to allocate large reply
4529 buffers (e.g. zconf client mount config records) due to memory
4530 fragmentation or consumption by the buffer cache. Preallocate
4531 some large reply buffers so that these replies can be sent even
4532 under memory pressure.
4536 Frequency : rare (liblustre)
4537 Description: fsx running with liblustre complained that using truncate() to
4538 extend the file doesn't work. This patch corrects that issue.
4539 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4540 ATTR_SIZE and lsm use in llu_setattr_raw.
4544 Frequency : rare, only 2.6 kernels
4545 Description: Unusual file access patterns on the MDS may result in inode
4546 data being lost in very rare circumstances.
4547 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4548 code in the 2.6 kernel caused the inode and/or EA data not to
4549 be read from disk, causing single-file corruption.
4553 Frequency : rare, only 2.6 filesystems using extents
4554 Description: Heavy concurrent write and delete load may cause data corruption.
4555 Details : It was possible under high-load situations to have an extent
4556 metadata block in the block device cache from a just-unlinked
4557 file overwrite a newly-allocated data block. We now unmap any
4558 metadata buffers that alias just-allocated data blocks.
4562 Frequency : filesystems with default stripe_count larger than 77
4563 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4564 Details : lconf specifies an inode size of 4096 bytes when the default
4565 stripe_count is larger than 77. This conflicts with the default
4566 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4567 this case to avoid pinning too much memory for large EAs.
4569 ------------------------------------------------------------------------------
4571 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4576 Frequency : rare (extremely heavy IO load with hundreds of clients)
4578 Description: Client is evicted, gets IO error writing to file
4579 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4580 caused clients to be evicted for AST timeouts. The fixes in
4581 bug 5192 mean we no longer need to have such short AST timeouts
4582 so ldlm_timeout has been increased.
4585 Frequency : occasional during --force or --failover shutdown under load
4586 Bugzilla : 5949, 4834
4587 Description: Server oops/LBUG if stopped with --force or --failover under load
4588 Details : a collection of import/export refcount and cleanup ordering
4589 issues fixed for safer force cleanup
4592 Frequency : only filesystems larger than 120 OSTs
4593 Bugzilla : 5990, 6223
4594 Description: lfs getstripe would oops on a very large filesystem
4595 Details : lov_getconfig used kfree on vmalloc'd memory
4598 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4599 Bugzilla : 6242, 6243
4600 Description: reading from files that had been truncated to a non-zero size
4601 but never opened returned no data
4602 Details : ll_file_read() reads zeros from no-object files to EOF
4607 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4608 Details : There's little harm in aborting MDS/OSS recovery and letting it
4609 try again, so I removed the LASSERT and return an error instead.
4611 Severity : enhancement
4613 Description: New debugging infrastructure for tracking down data corruption
4614 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4615 (b) cover more of the client-side code path, and (c) try to narrow
4616 down where problems occurred
4620 Bugzilla : 3819, 4364, 4397, 6313
4621 Description: Racing close and eviction MDS could cause assertion in mds_close
4622 Details : It was possible to get multiple mfd references during close and
4623 client eviction, leading to one thread referencing a freed mfd.
4625 Severity: : enhancement
4626 Bugzilla : 3262, 6359
4627 Description: Attempts to reconnect to servers are now more aggressive.
4628 Details : This builds on the enhanced upcall-less recovery that was added
4629 in 1.4.2. When trying to reconnect to servers, clients will
4630 now try each server in the failover group every 10 seconds. By
4631 default, clients would previously try one server every 25 seconds.
4636 Description: After recovery, certain operations trigger a failed
4637 assertion on a client.
4638 Details : Failing over an mds, using lconf -d --failover, while a
4639 client was doing a readdir() call would cause the client to
4640 LBUG after recovery completed and the readdir() was resent.
4642 Severity : enhancement
4644 Description: Default groups are now added by lconf
4645 Details : You can now run lconf --group <servicename> without having to
4646 manually add groups with lmc.
4649 Frequency : occasional
4651 Description: Nodes with an elan id of 0 trigger a failed assertion
4654 Frequency : always when accessing e.g. tty/console device nodes
4656 Description: tty and some other devices nodes cannot be used on lustre
4657 Details : file's private_data field is used by device data and lustre
4658 values in there got lost. New field was added to struct file to
4659 store fs-specific private data.
4662 Frequency : when exporting Lustre via NFS
4664 Description: NFSD failed occasionally when looking up a path component
4665 Details : NFSD is looking up ".." which was broken in ext3 directories
4666 that had grown large enough to become hashed.
4669 Frequency : Clusters with multiple interfaces not on the same subnet
4671 Description: Nodes will repeatedly try to reconnect to an interface which it
4672 cannot reach and report an error to the log.
4673 Details : Extra peer list entries will be created by lconf with some peers
4674 unreachable. lconf now validates the peer before adding it.
4677 Frequency : Only if a default stripe is set on the filesystem root.
4679 Description: Setting a default stripe on the filesystem root prevented the
4680 filesystem from being remounted.
4681 Details : The client was sending extra request flags in the root getattr
4682 request and did not allocate a reply buffer for the dir EA.
4685 Frequency : occasional, higher if lots of files are accessed by one client
4686 Bugzilla : 6159, 6097
4687 Description: Client trips assertion regarding lsm mismatch/magic
4688 Details : While revalidating inodes the VFS looks up inodes with ifind()
4689 and in rare cases can find an inode that is being freed.
4690 The ll_test_inode() code will free the lsm during ifind()
4691 when it finds an existing inode and then the VFS later attaches
4692 this free lsm to a new inode.
4696 Bugzilla : 6422, 7030
4697 Description: MDS deadlock between mkdir and client eviction
4698 Details : Creating a new file via mkdir or mknod (starting a transaction
4699 and getting the ns lock) can deadlock with client eviction
4700 (gets ns lock and trying to finish a synchronous transaction).
4703 Frequency : occasional
4704 Description: While starting a server, the fsfilt_ext3 module could not be
4706 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4707 kernels. Previously, lconf would still use the ext3 name
4708 when trying to load modules. Now, it will correctly use
4709 ext3 on 2.4 and ldiskfs on 2.6.
4711 Severity : enhancement
4712 Description: The default stripe count has been changed to 1
4713 Details : The interpretation of the default stripe count (0, to lfs
4714 or lmc) has been changed to mean striping across a single
4715 OST, rather than all available. For general usage we have
4716 found a stripe count of 1 or 2 works best.
4718 Severity : enhancement
4719 Description: Add support for compiling against Cray portals.
4720 Details : Conditional compiling for some areas that are different
4724 Frequency : occasional
4725 Bugzilla : 6409, 6834
4726 Description: Creating files with an explicit stripe count may lead to
4727 a failed assertion on the MDS
4728 Details : If some OSTs are full or unavailable, creating files may
4729 trigger a failed assertion on the MDS. Now, Lustre will
4730 try to use other servers or return an error to the
4734 Frequency : occasional
4736 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4737 caused lock timeout during readahead (since 1.4.2).
4738 Details : Processes doing readahead might match a lock that hasn't been
4739 granted yet if there are overlapping and conflicting lock
4740 requests. The readahead process waits on ungranted lock
4741 (original lock is CBPENDING), while OST waits for that process
4742 to cancel CBPENDING read lock and eventually evicts client.
4744 Severity : enhancement
4746 Description: Initial enabling of flock support for clients
4747 Details : Implements fcntl advisory locking and file status functions.
4748 This feature is provided as an optional mount flag (default
4749 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4750 locking are implemented yet, and those that are are not guaranteed
4751 to be completely correct in production environments.
4752 mount -t lustre -o [flock|noflock] ...
4755 Frequency : occasional
4757 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4758 assertion in the ext3 JBD (journaling) layer.
4759 Details : The b_committed_data struct is protected by the big kernel lock
4760 in 2.4 kernels, serializing journal_commit_transaction() and
4761 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4762 access to this struct. In 2.6 kernels there is finer grained
4763 locking to improve SMP performance of the JBD layer.
4767 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4772 Description: merge and cleanup kernel patches.
4775 -----------------------------------------------------------------------------
4777 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4779 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4780 versions of Lustre. Please update all servers and clients to
4781 version 1.4.2 or later at the same time. You must also run
4782 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4783 to update the configuration logs.
4785 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4786 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4787 - swab llog records of type '0' so we get proper header size/idx (5861)
4788 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4789 - fix rename of one directory over another leaking an inode (5953)
4790 - avoid SetPageDirty on 2.6 (5981)
4791 - don't re-add just-being-destroyed locks to the waiting list (5653)
4792 - when creating new directories, inherit the parent's custom
4793 striping settings if present parent (3048)
4794 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4795 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4796 - don't hold client locks on temporary worklist from l_lru (5666)
4797 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4798 - automatically evict dead clients (5921)
4799 - Update file size properly in create+truncate+fstat case (6196)
4800 - Do not unhash mountpoint dentries, do not allow removal of
4802 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4803 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4804 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4805 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4806 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4807 - Correctly handle reads of files with no objects (6243)
4808 - lctl recover will also mark a device active if deactivate used (5933)
4810 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4811 - allow --write-conf on an MDS with different nettype than client (5619)
4812 - don't write config llogs to MDS for mounts not from that MDS (5617)
4813 - lconf should create multiple TCP connections from a client (5201)
4814 - init scripts are now turned off by default; run chkconfig --on
4815 lustre and chkconfig --on lustrefs to use them
4816 - upcalls are no longer needed for clients to recover to failover
4818 - add --abort-recovery option to lconf to abort recovery on device
4820 - add support for an arbitrary number of OSTs (3026)
4821 - Quota support protocol changes.
4822 - forward compatibility changes to wire structs (6007)
4823 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4824 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4825 - improved "lustre status" script
4826 - initialize blocksize for non-regular files (6062)
4827 - added --disable-server and --disable-client configure options (5782)
4828 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4829 - Vanilla 2.4.29 support
4830 - increase maximum number of obd devices to 520 (6242)
4831 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4832 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4833 SLES 9 2.6.5 kernels
4834 - limit stripes per file to 160 (the maximum EA size) (6093)
4836 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4839 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4840 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4841 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4842 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4843 - small ext3 extents cleanups and fixes (5733)
4844 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4845 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4846 - avoid needless client->OST connect, fix handle mismatch (5317)
4847 - fix DLM error path that led to out-of-sync client, long delays (5779)
4848 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4849 - fix several locking issues related to i_size (5492,5624,5654,5672)
4850 - don't move pending lock onto export if it is already evicted (5683)
4851 - fix kernel oops when creating .foo in unlinked directory (5548)
4852 - fix deadlock in obdfilter statistics vs. object create (5811)
4853 - use time_{before,after} to avoid timer jiffies wrap (5882)
4854 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4855 - Do not leak request if server was not able to process it (5154)
4856 - If mds_open unable to find parent dir, make that negative lookup(5154)
4857 - don't create new directories with extent-mapping (5909, 5936)
4859 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4860 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4861 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4862 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4863 - support for RHEL4 kernel on i686 (5773)
4864 - provide error messages when incompatible logs are encountered (5898)
4866 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4867 * version 1.4.0.10 (1.4.1 release candidate 1)
4869 - don't keep a lock reference when lock is not granted (4238)
4870 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4871 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4872 - drop import inflight refcount on signal_completed_replay error (5255)
4873 - unlock page after async write error during send (3677)
4874 - handle missing objects in filter_preprw_read properly (5265)
4875 - no transno return for symlink open, don't save no-trasno open (3440)
4876 - don't try to complete elan receive that already failed (4012)
4877 - free RPC server reply state on error (5406)
4878 - clean up thread from ptlrpc_start_thread() on error (5160)
4879 - readahead could read extra page into cache that wasn't ejected (5388)
4880 - prevent races in class_attach/setup/cleanup/detach (5260)
4881 - don't dereference de->d_inode after l_dput of de (5458)
4882 - use "int" for stripe value returned from lock_to_stripe (5544)
4883 - mballoc allocation and error-checking fixes in 2.6 (5504)
4884 - block device patches to fix I/O request sizes in 2.6 (5482)
4885 - look up hostnames for IB nals (5602)
4886 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4887 - don't start multiple acceptors for the same port (5277)
4888 - fix incorrect LASSERT in mds_getattr_name (5635)
4889 - export a proc file for general "ping" checking (5628)
4890 - fix "lfs check" to not block when the MDS is down (5628)
4892 - service request history (4965)
4893 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4894 - create an "evict_client" /proc entry on OSTs, like the MDS has
4895 - fix mount usage message, return errors per mount(8) (5168)
4896 - change grep [] to grep "[]" in tests so they work in more UMLs
4897 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4898 - remove ancient LOV_MAGIC_V0 EA support (5047)
4899 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4900 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4901 - allow readahead allocations to fail when low on memory (5383)
4902 - mmap locking landed again, after considerable improvement (2828)
4903 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4905 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4908 - send OST transaction number in read/write reply to free req (4966)
4909 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4910 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4911 - the watchdog thread now runs as interruptible (5246)
4912 - flock/lockf fixes (but it's still disabled, pending 5135)
4913 - don't use EXT3 constants in llite code (5094)
4914 - memory shortage at startup could cause assertion (5176)
4916 - reorganization of lov code
4917 - single portals codebase
4919 - add extents/mballoc support (5025)
4920 - direct I/O reads in the obdfilter (4048)
4921 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4923 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4926 - send OST transaction number in read/write reply to free req (4966)
4927 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4928 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4929 - the watchdog thread now runs as interruptible (5246)
4930 - handle missing objects in filter_preprw_read properly (5265)
4931 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4932 - drop import inflight refcount on signal_completed_replay error (5255)
4933 - unlock page after async write error during send (3677)
4934 - return original error code on reconstructed replies (3761)
4935 - no transno return for symlink open, don't save no-trasno open (3440)
4937 - add pid to ldlm debugging output (4922)
4938 - bump the watchdog timeouts -- we can't handle 30sec yet
4939 - extra debugging for orphan dentry/inode bug (5259)
4941 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4944 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4945 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4946 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4947 - fix NULL dereference in /proc/sys/portals/routes (4827)
4948 - allow failed mdc_close() operations to be interrupted (4561)
4949 - stop precreate on OST before MDS would time out on it (4778)
4950 - don't send partial-page writes before EOF from client (4410)
4951 - discard client grant for sub-page writes on large-page clients (4520)
4952 - don't free dentries not owned by NFS code, check generation (4806)
4953 - fix lsm leak if mds_create_objects() fails (4801)
4954 - limit debug_daemon file size, always print CERROR messages (4789)
4955 - use transno after validating reply (3892)
4956 - process timed out requests if import state changes (3754)
4957 - update mtime on OST during writes, return in glimpse (4829)
4958 - add mkfsoptions to LDAP (4679)
4959 - use ->max_readahead method instead of zapping global ra (5039)
4960 - don't interrupt __l_wait_event() during strace
4962 - add software watchdogs to catch hung threads quickly (4941)
4963 - make lustrefs init script start after nfs is mounted
4964 - fix CWARN/ERROR duplication (4930)
4965 - return async write errors to application if possible (2248)
4966 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4967 - print NAL number in %x format (4645)
4968 - update barely-supported suse-2.4.21-171 series (4842)
4969 - support for sles 9 %post scripts
4970 - support for building 2.6 kernel-source packages
4971 - support for sles km_* packages
4973 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4976 - ignore -ENOENT errors in osc_destroy (3639)
4977 - notify osc create thread that OSC is being cleaned up (4600)
4978 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4979 - reconstruct ost_handle() like mds_handle() (4657)
4980 - create a new thread to do import eviction to avoid deadlock (3969)
4981 - let lconf resolve symlinked-to devices (4629)
4982 - don't unlink "objects" from directory with default EA (4554)
4983 - hold socknal file ref over connect in case target is down (4394)
4984 - allow more than 32000 subdirectories in a single directory (3244)
4985 - fix blocks count for O_DIRECT writes (3751)
4986 - OST returns ENOSPC from object create when no space left (4539)
4987 - don't send truncate RPC if file size isn't changing (4410)
4988 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4989 - bind to privileged port in socknal and tcpnal (3689)
4991 - rate limit CERROR/CWARN console message to avoid overload (4519)
4992 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4993 - basic mmap support (3918)
4994 - kernel patch series update from b1_4 (4711)
4996 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4999 - avoid crash during MDS cleanup with OST shut down (2775)
5000 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5001 - don't use bad inodes on the MDS (3744)
5002 - dynamic object preallocation to improve recovery speed (4236)
5003 - don't hold spinlock over lock dumping or change debug flags (4401)
5004 - don't zero obd_dev when it is force cleaned (3651)
5005 - print grants to console if they go negative (4431)
5006 - "lctl deactivate" will stop automatic recovery attempts (3406)
5007 - look for existing locks in ldlm_handle_enqueue() (3764)
5008 - don't resolve lock handle twice in recovery avoiding race (4401)
5009 - revalidate should check working dir is a directory (4134)
5011 - don't always mark "slow" obdfilter messages as errors (4418)
5013 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5016 - don't close LustreDB during write_conf until it is done (3860)
5017 - fix typo in lconf for_each_profile (3821)
5018 - allow dumping logs from multiple threads at one time (3820)
5019 - don't allow multiple threads in OSC recovery (3812)
5020 - fix debug_size parameters (3864)
5021 - fix mds_postrecov to initialize import for llog ctxt (3121)
5022 - replace config semaphore with spinlock (3306)
5023 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5024 - don't allow enqueue to complete on a destroyed export (3822)
5025 - down write_lock before checking llog header bitmap (3825)
5026 - recover from lock replay timeout (3764)
5027 - up llog sem before sending rpc (3652)
5028 - reduce ns lock hold times when setting kms (3267)
5029 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5030 - fix NULL deref and obd_dev leak on setup error (3312)
5031 - replace some LBUG about llog ops with error handling (3841)
5032 - don't match INVALID dentries from d_lookup and spin (3784)
5033 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5034 - fix invalid assertion in ptlrpc_set_wait (3880)
5036 - add libwrap support for the TCP acceptor (3996)
5037 - add /proc/sys/portals/routes for non-root route listing (3994)
5038 - allow setting MDS UUID in .xml (2580)
5039 - print the stack of a process that LBUGs (4228)
5041 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5044 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5045 - make sure to unset replay flag from failed open requests (3440)
5046 - if default stripe count is 0, use OST count for inode size (3636)
5047 - update parent mtime/ctime on client for create/unlink (2611)
5048 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5049 - free recovery state on server during a forced cleanup (3571)
5050 - unregister_reply for resent reqs (3063)
5051 - loop back devices mounting and status check on 2.6 (3563)
5052 - fix resource-creation race that can provoke i_size == 0 (3513)
5053 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5054 - more debugging for page-accounting assertion (3746)
5055 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5056 - avoid deadlock after precreation failure (3758)
5057 - fix race and lock order deadlock in orphan handling (3450, 3750)
5058 - add validity checks when grabbing inodes from l_ast_data (3599)
5060 - add /proc/.../recovery_status to obdfilter (3428)
5061 - lightweight CDEBUG infrastructure, debug daemon (3668)
5062 - change default OSC RPC parameters to be better on small clusters
5063 - turn off OST read cache for files smaller than 32MB
5064 - install man pages and include them in rpms (3100)
5065 - add new init script for (un)mounting lustre filesystems (2593)
5066 - run chkconfig in %post for init scripts (3701)
5067 - drop scimac NAL (unmaintained)
5069 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5072 - clean kiobufs before and after use (3485)
5073 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5074 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5075 - change init script to fail more clearly if not run as root (1528)
5076 - allow clients to reconnect during replay (1742)
5077 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5078 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5079 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5080 - fix cond_resched (3554)
5081 - only evict unfinished clients after recovery (3515)
5082 - allow bulk resend, prevent data loss (3570)
5083 - dynamic ptlrpc request buffer allocation (2102)
5084 - don't allow unlinking open directory if it isn't empty (2904)
5085 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5086 - remove extraneous obd dereference causing LASSERT failure (3334)
5087 - don't use get_cycles() when creating temp. files on the mds (3156)
5088 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5089 - handle EEXIST for set-stripe, set proper directory name (3336)
5091 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5092 - fix ksocknal_fmb_callback() error messages (2918)
5094 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5097 - don't copy lvb into (possibly NULL) reply on error (2983)
5098 - don't deref dentry after dput, don't free lvb on error (2922)
5099 - use the kms to determine writeback rpc length (2947)
5100 - increment oti_logcookies when osc is inactive (2948)
5101 - update client's i_blocks count via lvb messages (2543)
5102 - handle intent open/close of special files properly (1557)
5103 - mount MDS with errors=remount-ro, like obdfilter (2009)
5104 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5105 - don't use cancelling-locks' kms values (2947)
5106 - use highest lock extent for kms, not last one (2925)
5107 - don't dereference ERR_PTR() dentry in error handling path (3107)
5108 - fix thread race in portals_debug_dumplog() (3122)
5109 - create lprocfs device entries at setup instead of at attach (1519)
5110 - common AST error handler, don't evict client on completion race (3145)
5111 - zero nameidata in detach_mnt in 2.6 (3118)
5112 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5113 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5114 - initialize RPC timeout timer earlier for 2.6 (3219)
5115 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5116 - print nal/nid for unknown nid (3258)
5117 - additional checks for oscc recovery before doing precreate (3284)
5118 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5119 - don't crash in mdc_close for bad permissions on open (3285)
5120 - zero i_rdev for non-device files (3147)
5121 - clear page->private before handing to FS, better assertion (3119)
5122 - tune the read pipeline (3236)
5123 - fix incorrect decref of invalidated dentry (2350)
5124 - provide read-ahead stats and refine rpc in flight stats (3328)
5125 - don't hold journal transaction open across create RPC (3313)
5126 - update atime on MDS at close time (3265)
5127 - close LDAP connection when recovering to avoid server load (3315)
5128 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5129 - don't leak open file on MDS after open resend (3325)
5130 - serialize filter_precreate and filter_destroy_precreated (3329)
5131 - loop device shouldn't call sync_dev() for nul device (3092)
5132 - clear page cache after eviction (2766)
5133 - resynchronize MDS->OST in background (2824)
5134 - refuse to mount the same filesystem twice on same mountpoint (3394)
5135 - allow llmount to create routes for mounting behind routers (3320)
5136 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5137 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5138 - fix rare problem with rename on htree directories (3417)
5140 - allow default OST striping configuration per directory (1414)
5141 - fix compilation for qswnal for 2.6 kernels (3125)
5142 - increase maximum number of MDS request buffers for large systems
5143 - change liblustreapi to be useful for external progs like lfsck (3098)
5144 - increase local configuration timeout for slow disks (3353)
5145 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5147 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5150 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5151 - don't overwrite extent policy data in reply if lock was blocked (2901)
5152 - drop filter export grants atomically with removal from device (2663)
5153 - del obd_self_export from work_list in class_disconnect_exports (2908)
5154 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5155 - swab reply message in mdc_close, other PPC fixes (2464)
5156 - fix destroying of named logs (2325)
5157 - overwrite old logs when running lconf --write_conf (2264)
5158 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5159 - fix race in target_handle_connect (2898)
5160 - mds_reint_create() should take same inode create lock (2926)
5161 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5162 - don't close files for self_export to avoid uninitialized obd (2936)
5163 - allow MDS with the same name as client node (2939)
5164 - hold dentry reference for closed log files for unlink (2325)
5165 - reserve space for all logs during transactions (2059)
5166 - don't evict page beyond end of stripe extent (2925)
5167 - don't oops on a deleted current working directory (2399)
5168 - handle hard links to targets without a parent properly (2517)
5169 - don't dereference NULL lock when racing during eviction (2867)
5170 - don't grow lock extents when lots of conflicting locks (2919)
5172 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5175 - account for cache space usage on clients to avoid data loss (974)
5176 - lfsck support in lustre kernel code (2349)
5177 - reduce journal credits needed for BRW writes (2370)
5178 - orphan handling to avoid losing space on client/server crashes
5179 - ptlrpcd can be blocked, stopping ALL progress (2477)
5180 - use lock value blocks to assist in proper KMS, faster stat (1021)
5181 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5182 - recovery for initial connections (2355)
5183 - fixes for mds_cleanup_orphans (1934)
5184 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5185 - block all file creations until orphan recovery completes (1901)
5186 - client remove rq_connection from request struct (2423)
5187 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5188 - recovery timer race (2670)
5189 - mdc_close recovey bug (2532)
5190 - ptlrpc cleanup bug (2710)
5191 - mds timeout on local locks (2588)
5192 - namespace lock held during RPCs (2431)
5193 - handle interrupted sync write properly (2503)
5194 - don't try to handle a message that hasn't been replied to (2699)
5195 - client assert failure during cleanup after abort recovery (2701)
5196 - leak mdc device after failed mount (2712)
5197 - ptlrpc_check_set allows timedout requests to complete (2714)
5198 - wait for inflight reqs when ptlrpcd finishes (2710)
5199 - make sure unregistered services are removed from the srv_list
5200 - reset bulk XID's when resending them (caught by 1138 test)
5201 - unregister_bulk after timeout
5202 - fix lconf error (2694)
5203 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5204 - readahead locks pages, leaves pending causing memory pressure (2673)
5205 - increase OST request buffers to 4096 on large machines (2729)
5206 - fix up permission of existing directories in simple_mkdir (2661)
5207 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5208 - don't assign transno to errored transactions (2742)
5209 - don't delete objects on OST if given a bogus objid from MDS (2751)
5210 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5211 - if rq_no_resend, then timeout request after recovery (2432)
5212 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5213 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5214 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5215 - MDS recovery shouldn't delete objects that it has given out (2730)
5216 - if enqueue arrives after completion, don't clobber LVB (2819)
5217 - don't unlock pages twice when trigger_group_io returns error (2814)
5218 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5219 - don't write pages to disk if there was an error (1450)
5220 - don't ping imports that have recovery disabled (2676)
5221 - take buffered bytes into account when balancing socknal conn (2817)
5222 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5223 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5224 - remove little-endian swabbing of llog records (1987)
5225 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5226 - retry reposting request buffers if they fail (1191)
5227 - grow extent at grant time to avoid granting a revoked lock (2809)
5228 - lock revoke doesn't evict page if covered by a second lock (2765)
5229 - disable VM readahead to avoid reading outside lock extents (2805)
5231 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5232 - updated kernel patches for hp-2.4.20 kernel (2681)
5234 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5237 - fix truncated write corruption (2366)
5238 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5240 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5241 - ensure len(uuid) < 37 in lmc (1171)
5242 - fix ia64 OOPS in llog_test (2255)
5243 - zero end of page at obdfilter for partial page writes (2648)
5244 - don't leave stale dentries around after renames (bug 2428)
5245 - fix timeouts when evicting a client with a single lock held (2642)
5246 - set deadline for the initial HELLO message to drain (2634)
5247 - print out dotted-quad IP addresses in the socknal (2302)
5249 - additional debugging for MDS client eviction problem (2443)
5250 - fix mkfsoptions support for osts (2603, 2604)
5252 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5255 - add series for the vanilla 2.6.0 kernel
5256 - add series for the vanilla 2.4.24 kernel
5257 - add series for a cray x86/64 UL kernel drop
5258 - fix xattr patches for the vanilla 2.4.19 series
5260 - generate true UUIDs in lmc (1171)
5261 - have portals stack dumping break in UML (2466)
5262 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5263 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5264 - wake callback waiting threads on client eviction (2460)
5265 - Add --ptldebug and --subsystem to lmc (1719)
5266 - update assertion to allow safe interrupt allocation
5267 - set rq_no_resend for cancel requests (2432)
5268 - recalculate ptlrpcd timeout after resend (2494)
5269 - call vfs_rmdir when removing pending directories (2368)
5270 - fix renaming a file to itself (2429)
5271 - lmc creates a default one-stripe lov (2454)
5272 - expand procfs space to handle large clusters (2326)
5273 - increase UML stack to avoid overflow
5274 - update lconf's list of debug and subsystem masks
5275 - fix lfs find --obd (2510)
5276 - /proc tunable for disabling filter read caching (2591)
5277 - stop rpm packages from altering slapd.conf (2301)
5278 - disable nagle in the socknal under 0conf (2578)
5279 - choose mds inode size based on stripe count (2572)
5280 - fix kernel-source rpm problems (2516)
5282 - add --disable-doc to avoid pdf generation (2421)
5283 - update documentation, tests, type-os, comments
5284 - avoid format warnings on ia64
5285 - remove the TOE NAL
5286 - tiny code cleanups by removing unused fields
5288 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5291 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5292 - remove the most blatant lies from BUILDING file (2371)
5293 - change default debug level to reasonable production setting
5294 - reduce client side cache size to reduce cache flush time
5295 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5296 - make TCP zerocopy and pinger support enabled by default (2476)
5297 - sync writes completed after process exits caused crashes (2319)
5298 - maintain correct mount count on the MDS (2356)
5299 - backout 1557, because 2316 wasn't really fixed
5300 - better file I/O statistics gathering in /proc
5301 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5302 - another kernel patch to fix zero-copy TCP function export
5303 - don't take duplicate lock when processing re-sent getattr (2420)
5304 - lctl uses obd_self_export instead of creating new conn (2353)
5305 - MDS/OST recovery case which requires object creation asserted (2425)
5306 - move lfs from /usr/sbin to /usr/bin in packages
5307 - fix race between mds_client_add and mds_client_free (2417)
5308 - use kmalloc instead of slabs in portals (2430)
5309 - don't create duplicate records when a failover MDS is present (2442)
5310 - remove unnecessary mount age check (2332)
5311 - don't remove directory inodes from locks prematurely (2451)
5312 - don't break if MDS service name is the same as hostname (2103)
5313 - fix races in client write RPC generation when cache full (2482)
5315 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5318 - remove now-unused request->rq_obd (278)
5319 - if an allocation fails, print out how much memory we've used (1933)
5320 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5321 - add the "configurable stack size" patch to most series files (1256)
5322 - ability to write large log records, for 100+ OST configs (2306)
5323 - fix NULL deref when filter_prep fails (2314)
5324 - fix operator precedence error in filter_sync
5325 - dynamic allocation of socknal TX descriptors (2315)
5326 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5327 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5328 - ensure that CWARN messages in Portals always get to the syslog
5329 - __init/__exit are not for prototype decls (ldlm_init/exit)
5330 - x86-64 compile warning fixes
5331 - fix gateway LMC keyword conflict (2318)
5332 - fix MDS lock inversions in getattr/reint paths (1844)
5333 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5334 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5335 - DLM race condition prevented some lock evictions (2328)
5336 - ENOMEM detection and retry on socknal sends (2230)
5337 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5338 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5339 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5340 - better error messages when a client is rejected during recovery (1505)
5341 - avoid cancelling locks which were never granted, after failure (2330)
5342 - fix i_sem/journal inversion in mds_client_add (2333)
5343 - fix truncate/getattr lock cycle deadlock (2334)
5344 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5345 - fix two rare exit paths which could leak an l_lock() ref (2321)
5346 - fencepost error in MDS/OST orphan recovery (2226)
5347 - make log record alignment 8 bytes (1988)
5348 - lstripe now fails when requested offset > ost_count (2237)
5349 - ensure that all kernel series have a complete list.h (1607)
5350 - fix crashes in special-file operations (2316)
5351 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5353 - allow configurable automake binary, for testing new versions
5354 - small update to the lfs documentation
5356 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5358 * fix negative export reference count in fsfilt_sync (2312)
5360 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5361 * release candidate 0.9.1
5363 - orphans are moved into the PENDING directory for possible recovery
5364 - replayed opens now open by fid for orphan/rename safety (1042)
5365 - last close of an orphan inode generates a transno (683)
5366 - chdir() and mount() now pin the directory entry (1020)
5367 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5368 - discard very old requests without processing them (1502)
5369 - remove some common, well-understood CERRORs (1505)
5370 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5371 - clear "grant" flags in OST replies until OST grant code lands (1644)
5372 - fix read performance by not clobbering i_blksize on client (1598)
5373 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5374 - make LRU size a /proc tunable, clears locks when reduced (707)
5375 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5376 - server locks take references on exports now (1558)
5377 - build fixes for 2.4.20-rh trees (1663)
5378 - return an error from lov_create if all OSCs are inactive (1751)
5379 - fix import levels when a reconnect happens without a timeout (1597)
5380 - exit early from mds_open if we get a lookup error (1749)
5381 - partial page read at EOF wouldn't wait for disk before sending (1642)
5382 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5383 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5384 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5385 - write/truncate lock inversion (1639)
5386 - Don't auto-load obdclass, portals modules during cleanup (1495)
5387 - fix timestamps from jumping to "now" (1763)
5388 - extra journal assertions (1648)
5389 - add an extra multiunlink test (1771)
5390 - fix read_record/write_record API (1776)
5391 - fix leak of offset_extent, possible incorrect i_size later (1772)
5392 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5393 - Debugging for the kqswnal_get_idle_tx problems (1820)
5394 - Allow recovery to be attempted multiple times (1536)
5395 - Write out MDS last_rcvd file after it is first created (1600)
5396 - Fix tx_descriptor leak in failed transmit situations (1827)
5397 - ext3 journaling fixes for assertion failure after IO error (1871)
5398 - class_export_put() on freed export after completion AST error (1896)
5399 - Fix revalidate looping in VFS (1322)
5400 - Don't access a freed export during MDS_REINT timeout (1521)
5401 - Add open-unlink recovery support on the MDS (1673,1764)
5402 - Return an error if no MDS data was read from last_rcvd (1946)
5403 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5404 - Don't setup a disk device that doesn't match exported UUID (317)
5405 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5406 - avoid committing NULL handle in force close
5407 - local.sh is now a one-stripe LOV configuration
5408 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5409 - fix ext3 htree duplicate directory entry corruption (1516)
5410 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5411 - update to new LOV EA format (2097)
5412 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5414 2003-06-15 Phil Schwan <phil@clusterfs.com>
5417 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5418 - per-import/export recovery handling (958, 931, 959)
5419 - multiple last-rcvd slots, for serving multiple FSes (949)
5420 - connections are again shared between multiple imp/exports (963, 964)
5421 - "umount -f" would hang if any requests needed to be sent (393, 978)
5422 - avoid pinning large req buffer by copying for queued messages (989)
5423 - add "uuid" to "lctl device" command to help upcalls (991)
5424 - "open" RPCs with transnos would confuse recovery counters (1037)
5425 - do proper endian conversion of all wire messages (288, 340, 891)
5426 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5427 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5428 - fix ptlrpc_connection leak in target_handle_connect (1174)
5429 - fix import refcounting bug in OST and MDS cleanup (1134)
5430 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5431 - fix very unlikely obd_types race condition (501)
5432 - remove osc_open hack for echo_client (1187)
5433 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5434 - a failure in read_inode2 leads to deadlock (1139)
5435 - cancel ack-locks as soon as transaction is committed (1072)
5436 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5437 - make sure to commitrw after any preprw to avoid deadlock (1162)
5438 - failing to execute a file in a lustre FS would lock inode (1203)
5439 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5440 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5441 - obd_uuid2tgt would walk off the end of the list (1255)
5442 - on IA64 the osc would give portals incorrect bulk size (1258)
5443 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5444 - fix lock inversion caused by new llite matching code (1282)
5445 - limit the number of dirty pages on a client to 10MB (1286)
5446 - timed out locks were not being corrected cancelled (1289)
5447 - fix O_DIRECT above 4GB on IA-32 (1292)
5448 * major user-visible changes
5449 - fail out/fail over policy now controlled by the upcall (993)
5451 - add OBD_PING to check server availability and failure (954)
5452 - lustre messages are now sent in sending host order (288, 340, 891)
5453 - add eadatalen to MDS getattr reply (340)
5454 - OST read replies may contain second buffer, with per-page status (593)
5456 2003-03-11 Phil Schwan <phil@clusterfs.com>
5459 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5460 - failed open()s could cause deadlock; fixed (867, 869)
5461 - stop cancelling OST locks when files are closed (481)
5462 - overlapping XID spaces caused network corruption (851, 853)
5463 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5464 - setattr_raw added, to do single-RPC, server-side setattrs
5465 - lmc/lconf syntax change for OST UUIDs
5466 - fix crashy race condition between ptlrpc_free_req and osc_close
5467 - don't use request in mdc_enqueue if we hit a timeout (889)
5468 - don't set the inode i_size for regular files from the MDS (896)
5469 - handle out of order completion AST (842)
5470 - don't LBUG if a lock request times out after receiving AST (913)
5471 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5472 - if a bad lock AST arrives, send an error instead of dropping entirely
5473 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5474 - fix leak in bulk IO when only partially completed (899, 900, 926)
5475 - fix O_DIRECT for ia64 (55)
5476 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5477 - C-z after timeout could hang a process forever; fixed (977)
5479 - client-side I/O cache (678, 924, 929, 941, 970)
5481 - READPAGE and SETATTRs which don't take server-side locks get
5484 2003-02-11 Phil Schwan <phil@clusterfs.com>
5487 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5488 - Fix fsfilt_extN_readpage() to read a full page of directory
5489 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5490 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5491 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5492 - Most intent operations were removed, in favour of a new RPC mode
5493 that does a single RPC to the server and bypasses most of the VFS
5494 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5495 - Aggressively cancel local locks on DLM servers
5496 - mds_reint_unlink sends EA to the client if it's the last nlink.
5497 client uses that EA to unlink OST objects.
5498 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5499 - recursive symlinks were fixed (439)
5500 - fixed NULL deref in DEBUG_REQ
5501 - filter_update_lastobjid no longer calls sync, which annoyed extN
5502 - fixed multi-client small-writes to a single file problem (445)
5503 - fixed mtime updates during file writes (607)
5504 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5505 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5506 - fixed Linux OST size reporting problem (444, 656)
5507 - OST now updates object mtime with writes or setattr (607, 619)
5508 - client verifies file size before zeroing page past EOF (445)
5509 - OST now writes last allocated objid to disk with allocation (108)
5510 - LOV on echo now works (409)
5512 - mds_reint_unlink sends a new buffer, with the EA included. this
5513 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5514 set if a regular file was being unlinked, and it was the last link
5515 - use PtlGet from the target for bulk writes (315)
5516 - OST now updates object mtime with writes or setattr (607, 619)
5517 - LDLM now has a grant-time callback to revalidate locked items, if
5519 - Many MDS operations were reorganized to combat race conditions
5521 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5522 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5524 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5527 - Fully reactivate OST imports after reconnection (512, others)
5528 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5529 - More graceful error handling for truncating on dead OST (515)
5530 - Don't error out unless we're actually accessing dead stripes (474)
5531 - Fix garbage sizes when stripes are missing (410)
5532 - LRU counters were broken, causing constant lock purge (433, 432)
5533 - garbage on read from stripes with failed OSTs (441)
5534 - mark OSCs as active before reconnecting during recovery (438)
5535 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5536 - lfind did not preserve OST order in output (443)
5537 - symlinks cause hung clients, incorrect data (439)
5538 - stop dereferencing request after dropping refcount (457)
5539 - don't LASSERT(spin_is_locked) on non-SMP (455)
5540 - fixes for many rename() bugs
5541 - fstat didn't correctly synchronize attributes (399)
5542 - server must handle lock cancellation during blocking AST prep (487)
5543 - bulk descriptors were free()d too soon (511)
5544 - fix paths in lconf, which would load incorrect modules (451, 507)
5545 - fix confusing lconf 'host not found' error message (386)
5546 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5547 - fix race condition in mdc_blocking_ast() for inode access (526)
5548 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5549 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5550 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5551 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5552 - fix lstripe in several ways (kernel side) (527)
5553 - fix request leak in ldlm_cli_enqueue (262)
5554 - incorrect OSC was marked inactive after OST failure
5555 - call mds_fs_cleanup before unmounting filesystem (524)
5556 - fix races between taking ns_lock and ldlm_lock_change_resource
5557 - fix races updating LOV export open file list
5558 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5559 - fix recovery NULL deref in ldlm_cli_cancel_unused
5560 - fix some DLM races by using new hash table for lock handles (419)
5561 - permit the client to specify desired inodes, at replay
5562 - duplicate requests when we queue them for replay reintegration
5563 - fix last_rcvd offset calculation
5564 - sync after each recovered transaction, so we always make progress
5565 - never, not always, ERESTART requests without transnos
5566 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5568 - skip replay if the MDS says that the client is already connected
5569 - don't check for a recovery-enabled export to match lctl's UUID
5570 - don't INC_USE_COUNT for phantom exports
5571 - don't crash when cleaning up phantom exports (567)
5572 - don't double-finish or set replay data for errored mdc_open requests
5573 - abort requests when they time out, so we don't get old replies
5574 - send/receive replies for AST messages again
5575 - if the client says that it doesn't have the lock, cancel it on the
5577 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5578 mark it as destroyed, it will all work out when decref is called
5579 - fix module use counts (22, 581)
5581 - ASTs now expect a reply (server cancels lock on error reply)
5583 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5586 - fix many simultaneous client startup (392)
5587 - fix dentry->d_it clobbering
5588 - credentials weren't being shipped for readdir/getattr operations
5589 - remove invalid assertions triggered during some concurrent MD
5591 - proper Lustre versions added (336, 389)
5592 - fix memory leak for create error case (398)
5593 - fix LOV locking bug that would get cli/srv out of sync
5594 - fix echo client over LOV (409)
5595 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5596 - fix double-O_EXCL intent crash (424)
5597 - avoid sending multiple lock CANCELs (352)
5599 - MDS can do multi-client recovery (modulo bugs in new code)
5601 - many updates, edits, cleanups
5603 2002-11-18 Phil Schwan <phil@clusterfs.com>
5606 - fix null d_it dereference (346)
5607 - fix full OST/dbench hang (333)
5608 - fix permission problem with file removal (286)
5609 - fix removal of OSCs from LOV when they fail
5610 - fix NULL deref during bulk timeout (214)
5611 - fix problems related to multiple filesystems on one MDS (241)
5612 - fixed serious subtle metadata locking bugs
5613 - free locks on clients when inodes are removed due to memory
5615 - fix inode pointer in lock data (285)
5616 - partial support for multiple MDS on a single host (241)
5617 - data locks weren't cancelled at clear_inode time (290, 311)
5618 - intent locks could lead to unbounded lock growth (205)
5619 - added a maximum lock count, an LRU list, and a flusher
5620 - fix multiple rename (365)
5621 - properly abstracted the echo client
5622 - OSC locked 1 byte too many; fixed
5623 - rewrote brw callback code:
5624 - fixed recovery bugs related to LOVs (306)
5625 - fixed too-many-pages-in-one-write crash (191)
5626 - fixed (again) crash in sync_io_timeout (214)
5627 - probably fixed callback-related race (385)
5629 - Add capability to MDS protocol
5630 - LDLM cancellations and callbacks on different portals
5632 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5635 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5636 - defer open object destruction to close time (601981, 138)
5637 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5638 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5639 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5640 - don't call obd_setattr in ll_file_release for destroyed objects
5641 * protocol change to lustre_msg: move |version| and add |flags|
5642 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5643 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5644 * added replay of create, unlink, link and rename operations during
5645 MDS failover; recovery should be much more robust now
5646 * remove failed OSCs from LOVs (only lov_create uses this so far)
5647 * the lustre-HOWTO was brought (more) up to date (582544)
5649 2002-10-23 Phil Schwan <phil@clusterfs.com>
5652 - in-use dentries weren't being reused properly (617851)
5653 - prevent multiple LDLM setup (599178)
5654 - fix LOV size calculations for truncate (617853)
5655 - fix client handling of MDS intent errors (POSIX)
5656 - fix permission bug in lovstripe.c test (624321)
5657 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5658 - truncate past end of file could corrupt data
5659 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5660 - a race in recovery could return ETIMEDOUT to apps (623947)
5661 - building outside the source directory was fixed
5662 * the lustre-HOWTO was brought (more) up to date (582544)
5663 * major progress was made on recovery functionality
5665 2002-10-10 Phil Schwan <phil@clusterfs.com>
5668 - recovery deadlock fix
5669 - rm -rf causes LBUG fix (617817)
5670 - file open by multiple tasks fix (618962)
5671 - directory permissions bugs (602707 and 620007)
5672 - journal_stop fixed with locking (611313)
5673 - O_APPEND failures resolved (618273, perhaps 614459)
5674 - lconf PATH fix (619770)
5675 - IA64 build fix (621450)
5676 - RPC buffer sizes scale with amount of memory
5678 2002-10-01 Phil Schwan <phil@clusterfs.com>
5681 - locks would be cancelled without throwing away data pages,
5682 resulting in inconsistent data (605627)
5683 - inode attributes were not always being refreshed (605627, 612449)
5684 - lconf now continues to cleanup after lctl reports an error
5685 - MDS now enforces user permissions (602707)
5686 - lprocfs cleanup fixed, but not yet enabled (614157)
5687 - fixed infinite server hang, should a client not respond to an AST
5688 - avoid going into recovery if user calls readlink() with a buffer
5689 that's too small (613941)
5690 - AST RPCs no longer require replies (614867) -- this may be changed
5691 - don't crash server if client sends an IOV that's too big (611336)
5692 - fixed lock conversion deadlock (611892)
5693 - fixed the following of symlinks (614622)
5694 * recovery: the server can remove locks from a client that dies, other
5695 clients can make progress
5696 * more extN patch fixes
5697 * compile-time configurable ptlrpc buffer allocations
5699 - collaborative read cache document
5700 - Lustre Lite Performance CDR document-in-progress
5702 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5705 - fix typo in patch-2.4.18
5707 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5710 - clear ptlrpc request each time in handle_incoming_request()
5711 - unlink of files now destroys the object on the OST
5713 2002-09-19 Peter Braam <braam@clusterfs.com>
5715 * add hard link support
5716 * change obdfile creation method
5717 * kernel patch changed
5719 2002-09-19 Peter Braam <braam@clusterfs.com>
5722 - stack overflow bug in extN fixed
5724 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5726 * documentation updates
5727 - add man pages for config tools
5728 - update tests/README to describe testing with new config tools
5729 - finish metadata API descriptions
5730 * bug fixes and cleanups
5731 - statfs workaround for 16TB limit
5732 - LOV stripe allocation improved, can stripe on subset of OSTs
5733 - LOV file size/IO offset was wrong for files > 4GB in size
5734 - object EA data was being dropped, caused files to be unreadable
5735 - memory overflow with non-LOV OST caused memory corruption
5736 - fixed regression tests to work with new config tools, obdfilter
5737 - fixed bug when directory size became larger than 1 block
5738 - fixed bug (for single client case) when PWD was deleted
5739 - invalidate local directory pages when doing intent-based ops
5740 - avoid LDLM oops when lock callback contained bad data
5742 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5744 * documentation updates
5745 * bug fixes and cleanups
5746 - configuration tools
5749 - 64-bit compile warnings
5750 - 64-bit internal statfs data
5752 * test_brw on persistent OST devices
5754 * lprocfs (disabled)
5756 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5758 * documentation updates
5759 * bug fixes and cleanups
5760 * configuration tools
5762 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5765 * many small fixes to 0_5_4
5766 * io/network handling
5767 * thinkos in MDS operations
5769 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5772 * crucial basic fixes to 0.5.3
5773 * IOR, Iozone work over Elan
5776 2002-08-07 Phil Schwan <phil@clusterfs.com>
5777 * version 0_5_3, our first alpha
5778 * we use the new Portals iovs
5779 * documentation updates
5780 * bug fixes and cleanups
5781 * small changes in the DLM wire protocol
5783 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5784 * version 0_5_1 with some initial stability,
5785 * locking on MD and file I/O.
5786 * documentation updates
5787 * several bug fixes since 0.5.0
5788 * small changes in wire protocol
5790 2002-07-18 Phil Schwan <phil@clusterfs.com>
5792 * delivered as Lustre Light Alpha
5793 * fixed a crash after handling invalid MDS requests
5794 * fixed directory pages for architectures with non-4k pages sizes
5796 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5797 * release version v0_4_4
5798 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5800 * Fixes a number of interruption problems with OST operations.
5801 * Update documentation for portals header changes
5802 * Move all wire protocol structs/defines to lustre_idl.h
5803 * Fixes symlink length bug.
5804 * Add tcpdump to repository.
5806 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5807 * release version v0_4_3
5808 * Fixes statfs for inodes on extN.
5809 * Fixes bug in runtests which would delete /etc/hosts.
5810 * Use 64-bit object IDs wherever possible (not into VFS though)
5811 Remove ost_get_info, which is unused by lustre, and out of date.
5813 2002-07-03 Peter Braam <braam@clusterfs.com>
5814 * release version v0_4_2 Fixes a lookup error (type not passed)
5815 * move forward to head of Portals
5816 * move forward to latest Lustre kernel
5818 2002-06-25 Peter Braam <braam@clusterfs.com>
5819 * release version v0_4_1. Hopefully stable on single node use.