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: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
19 Description: fix racy locking of mballoc block bitmaps causing BUG
20 Details : The locking of the mballoc buddy bitmap and the in-memory
21 block bitmap was using two different spin locks in some
22 cases. This made it possible to incorrectly access the
23 mballoc bitmap while another process was modifying it,
24 causing a sanity assertion to fail. While no on-disk corruption
25 was reported, there was some risk of this happening.
28 Frequency : rare, on llog test 6
30 Descriptoin: don't allow connect to already connected import
31 Details : allowing connect to already connected import is hide connecting problem.
36 Descriptoin: allow kill process which wait statahead result
37 Details : for some reasons 'ls' can stick in waiting result from statahead,
38 in this case need way for kill this process.
40 Severity : enhancement
42 Description: Add state history info file, enhance import info file
43 Details : Track import connection state changes in a new osc/mdc proc file;
44 add overview-type data to the osc/mdc import proc file.
46 Severity : enhancement
48 Description: MDS create should not wait for statfs RPC while holding DLM lock.
50 Severity : enhancement
52 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
57 Descriptoin: OST grant too much space to client even there are not enough space.
58 Details : Client will shrink its grant cache to OST if there are no write
59 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
60 this grant cache if there are already not enough avaible space
61 (left_space < total_clients * 32M).
64 Frequency : start MDS on uncleanly shutdowned MDS device
66 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
67 Details : stay in waiting mds<>ost recovery finished produce random bugs
68 due race between two ll_sync thread for one lov target. send
69 ACTIVATE event only if connect realy finished and import have
73 Frequency : rare, connect and disconnect target at same time
75 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
76 Details : don't call obd_disconnect under lov_lock. this long time
77 operation and can block ptlrpcd which answer to connect request.
82 Descriptoin: don't lose wakeup for imp_recovery_waitq
83 Details : recover_import_no_retry or invalidate_import and import_close can
84 both sleep on imp_recovery_waitq, but we was send only one wakeup
88 Frequency : always with long access acl
90 Descriptoin: mds can't pack reply with long acl.
91 Details : mds don't control size of acl but they limited by reint/getattr
94 Severity : enhancement
96 Description: Update to SLES10 kernel-2.6.16.60-0.33.
98 Severity : enhancement
100 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
103 Frequency : start MDS on uncleanly shutdowned MDS device
105 Descriptoin: aborting recovery hang on MDS
106 Details : don't throttle destroy RPCs for the MDT.
109 Frequency : on remount
111 Description: external journal device not working after the remount
112 Details : clear dev_rdonly flag for external journal devices in
118 Description: shutdown vs evict race
119 Details : client_disconnect_export vs connect request race.
120 if client will evicted at this time - we start invalidate
121 thread without referece to import and import can be freed
125 Frequency : rare, need acl's on inode.
127 Description: client can't handle ost additional correctly
128 Details : if ost was added after client connected to mds client can have
129 hit lnet_try_match_md ... to big messages to wide striped files.
130 in this case need teach client to handle config events about add
131 lov target and update client max ea size at that event.
133 Severity : enhancement
135 Description: Changelogs
136 Details : Changelogs are a lightweight mechanism to track filesystem
137 metadata and namespace changes. The changelog is recorded
138 permanently on the MDTs, and is periodically "consumed" / purged
139 when records are no longer needed.
141 Severity : enhancement
143 Description: compact fld format with extents
144 Details : Store range of seq rather than every seq in FLD. Seq
145 controller update FLD rather than clients. In Case of CMD, mdt0
146 has FLD, all other metadata server act as non persistent proxy
147 for FLD queries and cache fld entries in fld cache.
152 Description: don't skip ost target if they assigned to file
153 Details : Drop slow OSCs if we can, but not for requested start idx.
154 This means "if OSC is slow and it is not the requested
155 start OST, then it can be skipped, otherwise skip it only
156 if it is inactive/recovering/out-of-space.
160 Description: more cleanup in mds_lov
161 Details : not send LOV EA under replay, we can't know about they size at this
162 time. Don't allow client connect to mds before any ost connected,
163 for avoid problems with LOV EA size and returning EIO to client.
165 Severity : enhancement
167 Description: Interoperability at server side (Disk interoperability)
169 Severity : enhancement
171 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
173 Severity : enhancement
175 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
177 Severity : enhancement
179 Description: New client IO stack (CLIO).
181 Severity : enhancement
183 Description: Commit on sharing. Eliminate inter-client dependencies between
184 uncommitted transactions by doing transaction commits.
185 Thereby clients may recovery independently.
188 Frequency : Create a symlink file with a very long name
190 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
191 Details : If there is no extra space in the request for early cancels,
192 ldlm_req_handles_avail() returns 0 instead of a negative value.
194 Severity : enhancement
196 Description: Add /proc entry for import status
197 Details : The mdc, osc, and mgc import directories now have
198 an import directory that contains useful import data for debugging
201 Severity : enhancement
203 Description: Re-disable certain /proc logging
204 Details : Enable and disable client's offset_stats, extents_stats and
205 extents_stats_per_process stats logging on the fly.
208 Frequency : Only on FC kernels 2.6.22+
210 Description: oops in statahead
211 Details : Do not drop reference count for the dentry from VFS when lookup,
212 VFS will do that by itself.
214 Severity : enhancement
216 Description: Generic /proc file permissions
217 Details : Set /Proc file permissions in a more generic way to enable non-
218 root users operate on some /proc files.
222 Description: Hitting mdc_commit_close() ASSERTION
223 Details : Properly handle request reference release in
224 ll_release_openhandle().
228 Description: quota recovery deadlock during mds failover
229 Details : This patch includes att18982, att18236, att18237 in bz14840.
231 1. fix osts hang when mds does failover with quotaon
232 2. prevent watchdog storm when osts threads wait for the
237 Frequency : only patchless client
238 Description: add workaround for race between add/remove dentry from hash
240 Severity : enhancement
242 Description: Allow OST glimpses to return PW locks
246 Description: LBUG when llog conf file is full
247 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
251 Description: Prevent import from entering FULL state when server in recovery
255 Description: service mount cannot take device name with ":"
256 Details : Only when device name contains ":/" will mount treat it as
262 Description: replace ptlrpcd with the statahead thread to interpret the async
263 statahead RPC callback
267 Frequency : on recovery
268 Description: I/O failures after umount during fail back
269 Details : if client reconnected to restarted server we need join to recovery
270 instead of find server handler is changed and process self
271 eviction with cancel all locks.
273 Severity : enhancement
275 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
277 Severity : enhancement
279 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
281 Severity : enhancement
283 Description: Upcall on Lustre log has been dumped
284 Details : Allow for a user mode script to be called once a Lustre log has
285 been dumped. It passes the filename of the dumped log to the
286 script, the location of the script can be specified via
287 /proc/sys/lnet/debug_log_upcall.
292 Description: avoid idr_remove called for id which is not allocated.
293 Details : Move assigment s_dev for clustered nfs to end of initialization,
294 to avoid problem with error handling.
299 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
300 Details : When connection is reused this not moved from CONN_UNUSED_HASH
301 into CONN_USED_HASH and this prodice warning when put connection
302 again in unused hash.
305 Severity : enhancement
307 Description: File striping can now be set to use an arbitrary pool of OSTs.
309 Severity : enhancement
311 Description: Export bytes_read/bytes_write count on OSC/OST.
315 Description: Early reply size mismatch, MGC loses connection
316 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
317 the connect flags are properly negotiated.
322 Description: quotas are not honored with O_DIRECT
323 Details : all writes with the flag O_DIRECT will use grants which leads to
324 this problem. Now using OBD_BRW_SYNC to guard this.
328 Description: add quota statistics
329 Details : 1. sort out quota proc entries and proc code.
330 2. add quota statistics
332 Severity : enhancement
334 Description: enable quota support for HEAD.
338 Description: Properly propagate oinfo flags from lov to osc for statfs
339 Details : restore missing copy oi_flags to lov requests.
341 Severity : enhancement
343 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
344 : l_getgroups(8), lst(8), routerstat(8)
345 Details : included man pages for llobdstat(8), llstat(8),
346 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
348 Severity : enhancement
350 Description: configure's --enable-quota should check the
351 : kernel .config for CONFIG_QUOTA
352 Details : configure is terminated if --enable-quota is passed but
353 : no quota support is in kernel
357 Description: Remove portals compatibility
358 Details : Remove portals compatibility, not interoperable with releases
363 Description: Resolve device initialization race
364 Details : Prevent proc handler from accessing devices added to the
365 obd_devs array but yet be intialized.
367 Severity : enhancement
369 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
371 Severity : enhancement
373 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
378 Description: Using wrong pointer in osc_brw_prep_request
379 Details : Access to array[-1] can produce panic if kernel compiled with
380 CONFIG_PAGE_ALLOC enabled
384 Description: Client runs out of low memory
385 Details : Consider only lowmem when counting initial number of llap pages
389 Description: *optional* service tags registration
390 Details : if the "service tags" package is installed on a Lustre node
391 When the filesystem is mounted, a local-node service tag will
392 be created. See http://inventory.sun.com/ for more information
393 about the Service Tags asset management system.
397 Description: Kernel BUG tries to release flock
398 Details : Lustre does not destroy flock lock before last reference goes
399 away. So always drop flock locks when client is evicted and
400 perform unlock regardless of successfulness of speaking to MDS.
404 Description: add refcount protection for osc callbacks, avoid panic on shutdown
408 Description: sanity test 65a fails if stripecount of -1 is set
409 Details : handle -1 striping on filesystem in ll_dirstripe_verify
414 Description: ASSERTION(CheckWriteback(page,cmd)) failed
415 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
418 Severity : enhancement
420 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
424 Description: do not process already freed flock
425 Details : flock can possibly be freed by another thread before it reaches
426 to ldlm_flock_completion_ast.
430 Description: LBUG during stress test
431 Details : Need properly lock accesses the flock deadlock detection list.
435 Description: oops in page fault handler
436 Details : kernel page fault handler can return two special 'pages' in error
437 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
441 Description: timeout with invalidate import.
442 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
443 should be handled by ptlrpcd. This produce long age waiting and
444 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
446 Severity : enhancement
448 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
452 Description: enable MGS and MDT services start separately
453 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
454 a co-located MGS without starting the MGS, which is a complement
455 to 'nosvc' mount option.
459 Frequency : after recovery
460 Description: precreate to many object's after del orphan.
461 Details : del orphan st in oscc last_id == next_id and this triger growing
462 count of precreated objects. Set flag LOW to skip increase count
463 of precreated objects.
467 Frequency : rare, on clear nid stats
468 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
469 Details : when clean nid stats sometimes try destroy live entry,
470 and this produce panic in free.
474 Description: Stack overflow during MDS log replay
475 ease stack pressure by using a thread dealing llog_process.
479 Description: wait until IO finished before start new when do lock cancel.
480 Details : VM protocol want old IO finished before start new, in this case
481 need wait until PG_writeback is cleared until check dirty flag and
482 call writepages in lock cancel callback.
484 Severity : enhancement
486 Description: using special macro for print time and cleanup in includes.
490 Description: mds_mfd_close() ASSERTION(rc == 0)
491 Details : In mds_mfd_close(), we need protect inode's writecount change
492 within its orphan write semaphore to prevent possible races.
496 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
500 Frequency : rare, on shutdown ost
501 Description: don't hit live lock with umount ost.
502 Details : shrink_dcache_parent can be in long loop with destroy dentries,
503 use shrink_dcache_sb instead.
507 Description: don't panic with use echo client
508 Details : echo client pass NULL as client nid pointer and this produce null
513 Description: fix build on ppc32
514 Details : compile code with -m64 flag produce wrong object file for ppc32.
518 Description: add message levels for liblustreapi
522 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
523 Details : Move the 'good_osts' check before the 'total_bavail' check. This
524 will result in an -EAGAIN and in the exit call path we call
525 alloc_rr() which will with increasing aggressiveness attempt to
526 aquire precreated objects on the minimum number of required OSCs.
530 Description: Use old size assignment to avoid deadlock
531 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
532 the scheduling while holding a spinlock. We do not need locking
533 for size in ll_update_inode() because size is only updated from
534 the MDS for directories or files without objects, so there is no
535 other place to do the update, and concurrent access to such inodes
536 are protected by the inode lock.
540 Description: resolve "_IOWR redefined" build error on SLES10
544 Description: dump the memory debugging after all modules are unloaded to
545 suppress false negative in conf_sanity test 39
547 Severity : enhancement
549 Description: build kernel-ib packages for OFED 1.3 in our release cycle
554 Description: fix SLES kernel versioning
555 Details : the kernel version for our SLES 10 kernel did not include a "-"
556 before the "smp" at the end. while this was not a problem in
557 general, it did mean that software trying to use the kernel
558 version to try to detect a vendor specific kernel would fail.
559 this was most evident by the OFED build scripts.
563 Description: Don't update lov_desc members until making sure they are valid
564 Details : When updating lov_desc members via proc fs, need fix their
565 validities before doing the real update.
569 Description: don't put request into delay list while invalidate in flight.
570 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
571 import in flight. this produce timeout for invalidate and sometimes
572 can cause stale data.
576 Frequency : on ppc only
577 Description: not convert ost objects for directory because it's not exist.
578 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
582 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
584 Severity : enhancement
586 Description: Update to support 2.6.22.14 vanilla kernel.
590 Frequency : rare, on recovery
591 Description: read procfs can produce deadlock in some situation
592 Details : Holding lprocfs lock which send rpc can produce block for destroy
593 obd objects and this also block reconnect with -EALREADY. This
594 isn't fix all lprocfs bugs - but make it rare.
596 Severity : enhancement
598 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
601 Frequency : frequent on X2 node
603 Description: mdc_set_open_replay_data LBUG
604 Details : Set replay data for requests that are eligible for replay.
608 Description: lustre_mgs: operation 101 on unconnected MGS
609 Details : When MGC is disconnected from MGS long enough, MGS will evict the
610 MGC, and late on MGC cannot successfully connect to MGS and a lot
611 of the error messages complaining that MGS is not connected.
615 Frequency : on network error
616 Description: panic with double free request if network error
617 Details : mdc_finish_enqueue is finish request if any network error ocuring,
618 but it's true only for synchronus enqueue, for async enqueue
619 (via ptlrpcd) this incorrect and ptlrpcd want finish request
622 Severity : enhancement
624 Description: client-side metadata stat-ahead during readdir(directory readahead)
625 Details : perform client-side metadata stat-ahead when the client detects
626 readdir and sequential stat of dir entries therein
629 Frequency : on start mds
631 Description: Implement get_info(last_id) in obdfilter.
634 Frequency : occasional
636 Description: Correctly check stale fid, not start epoch if ost not support SOM
637 Details : open with flag O_CREATE need set old fid in op_fid3 because
638 op_fid2 was overwritten with new generated fid, but mds can answer
639 with one of these two fids and both is not stale. Setattr
640 incorrectly started an epoch and assume will be called
641 done_writing, but without SOM done_writing ever being called.
644 Frequency : rare, depends on device drivers and load
646 Description: MDS or OSS nodes crash due to stack overflow
647 Details : Code changes in 1.8.0 increased the stack usage of some functions.
648 In some cases, in conjunction with device drivers that use a lot
649 of stack the MDS (or possibly OSS) service threads could overflow
650 the stack. One change which was identified to consume additional
651 stack has been reworked to avoid the extra stack usage.
654 Frequency : occasional
656 Description: Do not fail import if osc_interpret_create gets -EAGAIN
657 Details : If osc_interpret_create got -EAGAIN it immediately exits and
658 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
659 oscc_has_objects and see OSC has no objests and call
660 oscc_internal_create to resend create request.
662 Severity : enhancement
664 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
666 Severity : enhancement
668 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
671 Frequency : very rare
673 Description: Fix replay if there is an un-replied request and open
674 Details : In some cases, older replay request will revert the
675 mcd->mcd_last_xid on MDS which is used to record the client's
678 Severity : enhancement
680 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
682 Severity : enhancement
684 Description: Add rhel5 support to HEAD.
686 Serverity : enhancement
688 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
693 Description: Don't allow skipping OSTs if index has been specified.
694 Details : Don't allow skipping OSTs if index has been specified, make
695 locking in internal create lots better.
699 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
700 Details : access to bitfield in structure is always rounded to long
701 and this produce problem with not atomic change any bit.
705 Description: Lustre make rpms failed.
706 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
707 builds Lustre rpms from tarball.
710 Frequency : rare on shutdown ost
712 Description: If llog cancel was not send before clean_exports phase, this can
713 produce deadlock in llog code.
714 Details : If llog thread has last reference to obd and call class_import_put
715 this produce deadlock because llog_cleanup_commit_master wait when
716 last llog_commit_thread exited, but this never success because was
717 called from llog_commit_thread.
721 Description: allow userland application know is lost one of stripes.
722 Details : fill lvb_blocks with error code on ost and return it to
723 application if error flag found.
727 Description: NULL lov_tgts causing MDS oops
728 Details : more safe checks for NULL lov_tgts for avoid oops.
730 Severity : enhancement
732 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
736 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
737 Details : If an OST is down the MDS will hang indefinitely in
738 obd_statfs_rqset() waiting for the statfs data. While for
739 MDS QOS usage of statfs, it should not stuck in waiting.
741 Severity : enhancement
743 Description: remote_acl support
744 Details : Support ACL-based permission check for remote user.
745 Support setfacl/getfacl for remote user with the utils
746 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
747 the system "{s,g}etfacl" utils.
749 Severity : enhancement
751 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
753 Severity : enhancement
755 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
759 Description: make mgs_setparam() handle fsname containing dash
760 Details : fsname containing a dash does not work with lctl conf_param
762 Severity : enhancement
764 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
766 Severity : enhancement
768 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
772 Frequency : rare, at shutdown
773 Description: access already free / zero obd_namespace.
774 Details : if client_disconnect_export was called without force flag set,
775 and exist connect request in flight, this can produce access to
776 NULL pointer (or already free pointer) when connect_interpret
777 store ocd flags in obd_namespace.
781 Frequency : only at startup
782 Description: not alloc memory with spinlock held.
783 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
784 if any spinlock held.
786 Severity : enhancement
788 Description: make lustre randomly fail allocating memory
789 Details : Make lustre randomly failed allocating memory for testing purpose.
791 Severity : enhancement
793 Description: lost problems with lov objid file
794 Details : Fixes some scability and access to not inited memory problems
795 in work with lov objdid file.
800 Description: lfs find does not continue on file error
801 Details : Continue other files processing when a file/dir is absent.
805 Description: Inconsistent usage of lustre_pack_reply()
806 Details : Standardize the usage of lustre_pack_reply() such that it
807 always generate a CERROR on failure.
810 Frequency : occasional
812 Description: MDS hang or stay in waiting lock
813 Details : If client receive lock with CBPENDING flag ldlm need send lock
814 cancel as separate rpc, to avoid situation when cancel request
815 can't processed due all i/o threads stay in wait lock.
819 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
823 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
827 Description: Update to SLES9 kernel-2.6.5-7.287.3.
831 Description: Update to RHEL4 latest kernel.
833 Severity : enhancement
835 Description: Build SLES10 patchless client fails
836 Details : The configure was broken by run ./configure with
837 --with-linux-obj=.... argument for patchless client. When the
838 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
839 can't search header adequately. Use absolute path such as
840 -I($LINUX)/include instead.
844 Description: interrupt oig_wait produce painc on resend.
845 Details : brw_redo_request can be used for resend requests from ptlrpcd and
846 private set, and this produce situation when rq_ptlrpcd_data not
847 copyed to new allocated request and triggered LBUG on assert
848 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
849 ptlrpcd set if request is changed and can be safety changed to use
854 Description: LASSERT_{REQ,REP}SWAB macros are buggy
855 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
856 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
857 these macros and replace them with their difinitions instead.
861 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
862 Details : Update which_patch & target file for SLES10 latest kernel.
865 Description: Remove client patches from SLES 10 kernel.
866 Details : This causes SLES 10 clients to behave as patchless clients
867 even on a Lustre-patched (server) kernel.
869 Severity : enhancement
871 Description: self-adjustable client's lru lists
872 Details : use adaptive algorithm for managing client cached locks lru
873 lists according to current server load, other client's work
874 pattern, memory activities, etc. Both, server and client
875 side namespaces provide number of proc tunables for controlling
878 Severity : enhancement
880 Description: light-weight GSS support
881 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
882 and reduce performance overhead.
884 Severity : enhancement
886 Description: Linux keyring support
887 Details : Support using service of Linux keyring for Lustre GSS internal
888 context refresh/cache mechanism.
892 Description: Fix errors in lfs documentation
893 Details : Fixes man pages
897 Description: don't use GFP_* in generic Lustre code.
898 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
902 Description: obdclass is limited by single OBD_ALLOC(idarray)
903 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
907 Description: warnings with build patchless client with vanila 2.6.19 and up
908 Details : change old ctl_table style and replace ctl_table/ctl_table_header
909 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
913 Frequency : Only for SLES
914 Description: sanity_quota fail test_1
915 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
916 file, which makes TSTID[2] inunique.
920 Frequency : Only for SLES9
921 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
922 Details : Update target/ChangeLog/which_patch .
926 Frequency : Only for test_10 in sanity-quota.sh
927 Description: a bug in quota test code
928 Details : checking a test flag in a wrong place
932 Frequency : Only for test_13 in sanity-quota.sh
933 Description: a bug in quota test script
934 Details : shouldn't check overfull allocated quota
938 Description: jbd statistics
939 Details : Port older jbd statistics patch for sles10
941 Severity : enhancement
943 Description: Add printing inode info into message about error in writepage.
948 Description: reply_lock_interpret crash due to race with it and lock cancel.
949 Details : Do not replay locks that are being cancelled. Do not reference
950 locks by their address during replay, just by their handle.
954 Frequency : When flocks are used.
955 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
956 Details : Flock locks might destroy just granted lock if it could be merged
957 with another existing flock, this is done in completion handler,
958 so teach ldlm_cli_enquque_fini that this is a valid case for
964 Description: Oops in read and write path when failing to allocate lock.
965 Details : Check if lock allocation failed and return error back.
969 Description: lstripe command fails for valid OST index
970 Details : The stripe offset is compared to lov->desc.ld_tgt_count
971 instead of lov->desc.ld_active_tgt_count.
975 Description: sanity.sh failed test 103
976 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
981 Description: df doesn't work properly if diskfs blocksize != 4K
982 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
986 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
987 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
989 Severity : enhancement
991 Description: Adaptive timeouts
992 Details : RPC timeouts adapt to changing server load and network
993 conditions to reduce resend attempts and improve recovery time.
997 Description: llapi_file_create() does not allow some changes
998 Details : add llapi_file_open() that allows specifying the mode and
999 open flags, and also returns an open file handle.
1003 Description: merge and cleanup kernel patches.
1004 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1008 Description: Add journal checksum support.(Kernel part)
1009 Details : The journal checksum feature adds two new flags i.e
1010 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1011 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1012 indicates that the commit block contains the checksum for
1013 the blocks described by the descriptor blocks. Now commit
1014 record can be sent to disk without waiting for descriptor
1015 blocks to be written to disk. This behavior is controlled
1016 using JBD2_FEATURE_ASYNC_COMMIT flag.
1020 Description: OSS needs mutliple precreate threads
1021 Details : Add ability to start more than one create thread per OSS.
1025 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1026 Details : execute lfs setstripe on client
1030 Description: mds_obd_create error creating tmp object
1031 Details : When the user sets quota on root, llog will be affected and can't
1032 create files and write files.
1035 Frequency : Always on ia64 patchless client, and possibly others.
1037 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1038 Details : This allows the patchless client to be loaded on architectures
1039 without this export.
1043 Description: RedHat Update kernel for RHEL5
1044 Details : Add the kernel config file for RHEL5.
1048 Description: RedHat Update kernel for RHEL5
1049 Details : Modify the kernel config file more closer RHEL5.
1053 Description: Build failure against Centos5 (RHEL5)
1054 Details : Define PAGE_SIZE when it isn't present.
1058 Description: test_7 on sanity-quota.sh can't be used on separate machines
1059 Details : add facet to handle it
1063 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1064 Details : with NFS, the anon dentry's parent was set to itself in
1065 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1066 corresponding dentry other than use rec->ur_name.
1068 Severity : enhancement
1070 Description: lfs setstripe enhancement
1071 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1075 Description: enable data checksumming by default
1076 Details : enable checksum by default, allow --disable-checksum
1077 configure option and "-o nochecksum" mount option. Checksums
1078 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1082 Description: lustre support for RHEL5
1083 Details : Add support for RHEL5.
1087 Description: Client eviction due to failover config
1088 Details : after a connection loss, the lustre client should attempt to
1089 reconnect to the last active server first before trying the
1090 other potential connections.
1094 Description: when mds and osts use different quota unit(32bit and 64bit),
1095 quota will be released repeatly.
1096 Details : void sending multiple quota reqs to mds, which will keep the
1097 status between the reqs.
1101 Description: osts not allocated evenly to files
1102 Details : change the condition to increase offset_idx
1105 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1107 Description: Data corruption for OSTs that are formatted larger than 2TB
1109 Details : When generating the bio request for lustre file writes the
1110 sector number would overflow a temporary variable before being
1111 used for the IO. The data reads correctly from Lustre (which
1112 will overflow in a similar manner) but other file data or
1113 filesystem metadata may be corrupted in some cases.
1117 Description: Tune the kernel for good SCSI performance.
1118 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1119 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1124 Description: rewrite ext2-derived code in obdclass/uuid.c
1125 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1126 from scratch preserving functionality.
1130 Description: "lfs find -obd UUID" prints directories
1131 Details : "lfs find -obd UUID" will return all directory names instead
1132 of just file names. It is incorrect because the directories
1133 do not reside on the OSTs.
1137 Description: use i_size_read and i_size_write in 2.6 port
1138 Details : replace inode->i_size access with i_size_read/write()
1141 Frequency : when removing large files
1143 Description: scheduling issue during removal of large Lustre files
1144 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1145 It causes scheduling issues when removing large files (17TB in the
1149 Frequency : only with liblustre clients on XT3
1151 Description: evictions taking too long
1152 Details : allow llrd to evict clients directly on OSTs
1155 Frequency : only on ppc
1157 Description: /proc/fs/lustre/devices broken on ppc
1158 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1159 We should make sure the type of 'index' is loff_t and then cast
1160 explicitly as needed below. Do not assign an explicitly cast
1164 Frequency : only for rhel5
1166 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1167 Details : Modify the target file & which_kernel.
1169 Severity : enhancement
1171 Description: omit set fsid for export NFS
1172 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1173 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1177 Frequency : Always, for kernels after 2.6.16
1178 Description: Fix warning idr_remove called for id=.. which is not allocated.
1179 Details : Last kernels save old s_dev before kill super and not allow
1180 to restore from callback - restore it before call kill_anon_super.
1184 Description: buffer overruns could theoretically occur
1185 Details : llapi_semantic_traverse() modifies the "path" argument by
1186 appending values to the end of the origin string, and a
1187 overrun may occur. Adding buffer overrun check in liblustreapi.
1191 Description: Fix error on 'ls .' at the top of the Lustre mount.
1192 Details : Don't revalidate dentry if it is a root dentry.
1196 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1197 Details : Modify vm-tunables-rhel4.patch.
1201 Description: Kernel config for 2.6.18-vanilla.
1202 Details : Modify targets/2.6-vanilla.target.in.
1203 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1204 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1205 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1206 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1210 Description: improve handling recoverable errors
1211 Details : if request processig with error which can be recoverable on server
1212 request should be resend, otherwise page released from cache and
1217 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1218 Details : Modify target file & which_patch.
1219 A flaw was found in the IA32 system call emulation provided
1220 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1221 value could be stored in the %RAX register, which could trigger an
1222 out-of-bounds system call table access. An untrusted local user
1223 could exploit this flaw to run code in the kernel
1224 (ie a root privilege escalation). (CVE-2007-4573).
1228 Description: O_DIRECT bypasses client statistics.
1229 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1230 read_ahead_stats were updated. copied the stats section from
1231 osc_send_oap_rpc() into async_internal().
1235 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1239 Description: change order of libsysio includes
1240 Details : '#include sysio.h' should always come before '#include xtio.h'
1244 Description: handle "serious error: objid * already exists" more gracefully
1245 Details : If LAST_ID value on disk is smaller than the objects existing in
1246 the O/0/d* directories, it indicates disk corruption and causes an
1247 LBUG(). If the object is 0-length, then we should use the existing
1248 object. This will help to avoid a full fsck in most cases.
1250 Severity : enhancement
1252 Description: adapt the lustre_config script to support the upgrade case
1253 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1254 server targets to 1.6 in parallel.
1258 Description: To avoid grant space > avaible space when the disk is almost
1259 full. Without this patch you might see the error "grant XXXX >
1260 available" or some LBUG about grant, when the disk is almost
1262 Details : In filter_check_grant, for non_grant cache write, we should
1263 check the left space by if (*left > ungranted + bytes), instead
1264 of (*left > ungranted), because only we are sure the left space
1265 is enough for another "bytes", then the ungrant space should be
1266 increase. In client, we should update cl_avail_grant only there
1267 is OBD_MD_FLGRANT in the reply.
1271 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1272 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1273 described in CVE-2007-4573. This problem could allow untrusted
1274 local users to gain root access.
1277 Frequency : when using O_DIRECT and quotas
1279 Description: Incorrect file ownership on O_DIRECT output files
1280 Details : block usage reported by 'lfs quota' does not take into account
1281 files that have been written with O_DIRECT.
1286 Description: touch file failed when fs is not full
1287 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1288 otherwise we can get ENOSP while fs is not full.
1292 Description: parallel lock callbacks
1293 Details : Instead of sending blocking and completion callbacks as separated
1294 requests, adding them to a set and sending in parallel.
1297 Frequency : only for Cray XT3
1298 Bugzilla : 12829/13455
1299 Description: Changing primary group doesn't change the group lustre assigns to
1301 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1302 overridden with the primary group provided by the group upcall,
1303 whereas the supplied fsgid can be trusted if it is in the list of
1304 supplementary groups returned by the group upcall.
1306 Severity : enhancement
1308 Description: Allow masking D_WARNING, D_ERROR messages from console
1309 Details : Console messages can now be disabled via lnet.printk.
1313 Description: User code with malformed file open parameter crashes client node
1314 Details : Before packing join_file req, all the related reference should be
1315 checked carefully in case some malformed flags cause fake
1316 join_file req on client.
1320 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1321 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1322 mode is equal to the granted mode. However, it should also check
1323 that LDLM_FL_CP_REQD is not set.
1327 Description: Heavy nfs access might result in deadlocks
1328 Details : After ELC code landed, it is now improper to enqueue any mds
1329 locks under och_sem, because enqueue might want to decide to
1330 cancel open locks for same inode we are holding och_sem for.
1334 Description: Client eviction while running blogbench
1335 Details : A lot of unlink operations with concurrent I/O can lead to a
1336 deadlock causing evictions. To address the problem, the number of
1337 oustanding OST_DESTROY requests is now throttled to
1338 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1339 callbacks are processed in priority.
1343 Description: enable ACLs on MDS by default
1344 Details : ACLs must be enabled on MDS by default.
1347 Frequency : PPC/PPC64 only
1349 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1350 Details : fix duplicated definitions between asm-ppc64/types.h and
1351 lustre_types.h on PPC.
1354 Frequency : PPC/PPC64 only
1356 Description: asm-ppc/segment.h does not exist
1357 Details : fix compile issue on PPC.
1361 Description: better handle error messages in extents code
1364 Frequency : RHEL4 only
1366 Description: mkfs is very slow on IA64/RHEL4
1367 Details : A performance regression has been discovered in the MPT Fusion
1368 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1369 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1370 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1372 Severity : enhancement
1374 Description: SNMP support enhancement
1375 Details : Adding total number of sampled request for an MDS node in snmp
1378 Severity : enhancement
1380 Description: Optimize ldlm waiting list processing for PR extent locks
1381 Details : When processing waiting list for read extent lock and meeting read
1382 lock that is same or wider to it that is not contended, skip
1383 processing rest of the list and immediatelly return current
1384 status of conflictness, since we are guaranteed there are no
1385 conflicting locks in the rest of the list.
1389 Description: Time out and refuse to reconnect
1390 Details : When the failover node is the primary node, it is possible
1391 to have two identical connections in imp_conn_list. We must
1392 compare not conn's pointers but NIDs, otherwise we can defeat
1393 connection throttling.
1397 Description: port llog fixes from b1_6 into HEAD
1398 Details : Port llog reference couting and some llog cleanups from b1_6
1399 (bug 10800) into HEAD, for protect from panic and access to already
1400 free llog structures.
1404 Description: Detect stride IO mode in read-ahead
1405 Details : When a client does stride read, read-ahead should detect that and
1406 read-ahead pages according to the detected stride pattern.
1410 Description: data checksumming impacts single node performance
1411 Details : add support for several checksum algorithm. Currently, only CRC32
1412 and Adler-32 are supported. The checksum type can be changed on
1413 the fly via /proc/fs/lustre/osc/*/checksum_type.
1417 Description: use adler32 for page checksums
1418 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1423 Description: build for x2 fails
1424 Details : fix compile issue on Cray systems.
1428 Description: Properly match for duplicate locks
1429 Details : Due to different lock order from skiplists code, we need to
1430 traverse entire list for now
1433 Frequency : only on PPC/SLES10
1435 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1436 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1437 be defined. Add a hack in mkfs_lustre.c to work around this compile
1442 Description: LASSERT on MDS when client holding flock lock dies
1443 Details : ldlm pool logic depends on number of granted locks equal to
1444 number of released locks which is not true for flock locks, so
1445 just exclude such locks from consideration.
1449 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1450 Details : Use fsfilt_sync() for both the whole filesystem sync and
1451 individual file sync to eliminate dangerous inode locking
1452 with I_LOCK that can lead to a deadlock.
1456 Description: performance in 1.6.3
1457 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1458 sure that 1MB requests are not fragmented by the block layer.
1460 Severity : enhancement
1462 Description: organize the server-side client stats on per-nid basis
1463 Details : Change the structure of stats under obdfilter and mds to
1471 The "uuid"s file would list the uuids of _active_ exports.
1472 And the clear entry is to clear all stats and stale nids.
1474 Severity : enhancement
1476 Description: eliminate client locks in face of contention
1477 Details : file contention detection and lockless i/o implementation
1478 for contended files.
1482 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1485 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1487 Description: MDS may lose file striping (and hence file data) in some cases
1488 Details : If there are additional extended attributes stored on the MDS,
1489 in particular ACLs, SELinux, or user attributes (if user_xattr
1490 is specified for the client mount options) then there is a risk
1491 of attribute loss. Additionally, the Lustre file striping
1492 needs to be larger than default (e.g. striped over all OSTs),
1493 and an additional attribute must be stored initially in the
1494 inode and then increase in size enough to be moved to the
1495 external attribute block (e.g. ACL growing in size) for file
1500 Description: skiplist implementation simplification
1501 Details : skiplists are used to group compatible locks on granted list
1502 that was implemented as tracking first and last lock of each
1503 lock group the patch changes that to using doubly linked lists
1507 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1508 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1509 because LASSERT on both the data supplied by a client, and the
1510 data on disk is dangerous and incorrect.
1512 Severity : enhancement
1514 Description: Slow truncate/writes to huge files at high offsets.
1515 Details : Directly associate cached pages to lock that protect those pages,
1516 this allows us to quickly find what pages to write and remove
1517 once lock callback is received.
1521 Description: more ldlm soft lockups
1522 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1523 starve other threads from the resource lock for a long time in
1524 case of long waiting queue, so change the debug level from
1525 D_OTHER to the less frequently used D_INFO.
1527 Severity : enhancement
1529 Description: add -gid, -group, -uid, -user options to lfs find
1533 Description: Hung threads in invalidate_inode_pages2_range
1534 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1535 once one is completed. As a result, some RPCs are stuck in the
1536 queue and are never sent.
1540 Description: filter threads hungs on waiting journal commit
1541 Details : Cleanup filter group llog code, then only filter group llog will
1542 be only created in the MDS/OST syncing process.
1546 Description: Procfs and llog threads access destoryed import sometimes.
1547 Details : Sync the import destoryed process with procfs and llog threads by
1548 the import refcount and semaphore.
1550 Severity : enhancement
1552 Description: openlock cache of b1_6 port to HEAD
1557 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1558 Details : If insertion of an extent fails, then discard the inode
1559 preallocation and free data blocks else it can lead to duplicate
1564 Description: don't always update ctime in ext3_xattr_set_handle()
1565 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1566 In some cases the ctime should not be updated, for example for
1567 2.0->1.8 compatibility it is necessary to delete an xattr and it
1568 should not update the ctime.
1572 Bugzilla : 15713/16362
1573 Description: Assertion in iopen_connect_dentry in 1.6.3
1574 Details : looking up an inode via iopen with the wrong generation number can
1575 populate the dcache with a disconneced dentry while the inode
1576 number is in the process of being reallocated. This causes an
1577 assertion failure in iopen since the inode's dentry list contains
1578 both a connected and disconnected dentry.
1582 Description: assertion failure in ldlm_handle2lock()
1583 Details : fix a race between class_handle_unhash() and class_handle2object()
1584 introduced in lustre 1.6.5 by bug 13622.
1589 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1590 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1595 Description: oops in sd_iostats_seq_show()
1596 Details : unloading/reloading the scsi low level driver triggers a kernel
1597 bug when trying to access the sd iostat file.
1602 Description: Kernel panics during QLogic driver reload
1603 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1604 patch, causing memory corruption.
1609 Description: journal_dev option does not work in b1_6
1610 Details : pass mount option during pre-mount.
1612 Severity : enhancement
1614 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1615 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1616 extent information of a file. It can be used to map logical blocks
1617 in a file to physical blocks in the block device.
1621 Description: LDLM soft lockups - improvement
1622 Details : It is be possible to send the lock handle along with each read
1623 or write request because the client is already doing a lock match
1624 itself so there isn't any reason the OST should have to re-do that
1628 Frequency : only on Cray X2
1630 Description: X2 build failures
1631 Details : fix build failures on Cray X2.
1635 Description: xid & resent requests
1636 Details : Initialize RPC XID from clock at startup (randomly if clock is
1639 Severity : enhancement
1641 Description: Add lustre_start utility to start or stop multiple Lustre servers
1646 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1647 Details : In case of memory pressure, list_del() can be called twice on
1648 req->rq_history_list, causing a kernel oops.
1652 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1653 Details : fix stack overflow in the distributed lock manager by defering
1654 export eviction after a failed AST to the elt thread instead of
1655 handling it in the dlm interpret routine.
1659 Description: Convert some comments to new format.
1660 Details : Update documenting comments to match doxygen conventions.
1664 Description: Grammar fixes.
1665 Details : A couple of trivial spelling fixes.
1669 Description: OSD_COUNTERS-mandatory
1670 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1671 to simplify the code.
1675 Description: simplify cmm_device freeing logic.
1676 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1680 Description: Add lockdep support to dt_object_operations locking interface.
1681 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1682 parameter indicating lock ordering. Update mdd code to use new
1687 Description: Introduce failloc constants for lockless IO tests.
1688 Details : Add two new failloc constants to test lockless IO. Only one of
1689 them in implemented---another is checked in yet to be landed
1694 Description: Add lockdep support for inode mutex.
1695 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1696 in the situations where Lustre has to lock more than one inode
1701 Description: Add optional invariants checking support.
1702 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1703 configure switch. This macro is to be used for consistency and
1704 sanity checks that are too expensive to be left in `production'
1709 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1710 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1715 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1716 Details : Use spin_lock_nested() in (the only) situation where more than
1717 one ldlm_lock is locked simultaneously. Also, fix possible
1718 dead-lock in ldlm_lock_change_resource() by enforcing particular
1723 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1724 Details : Instead of specifying each ldlm_lock call-back through separate
1725 parameter, wrap them into struct ldlm_callback_suite.
1729 Description: Kill join_lru obd method and its callers.
1730 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1731 in memory---a requirement implemented through ->o_join_lru() obd
1732 method in HEAD. Remove this method and its users.
1736 Description: Add asynchronous ldlm ENQUEUE completion handler.
1737 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1738 case that stock ldlm_completion_ast() cannot handle as it waits
1739 until lock is granted. Introduce new ldlm_completion_ast_async()
1740 for this. Also comment ldlm_completion_ast().
1744 Description: ldlm_error <-> errno conversion.
1745 Details : Add functions to map (rather arbitrary) between LDLM error codes
1746 and standard errno values. CLIO needs this to prevent LDLM specific
1747 constants from escaping ldlm and osc.
1751 Description: Kill unused ldlm_handle2lock_ns() function.
1752 Details : Kill unused ldlm_handle2lock_ns() function.
1756 Description: Add lu_ref support to ldlm_lock
1757 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1758 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1759 and ldlm_resource. LDLM interface has to be changed, because code that
1760 releases a reference on a lock, has to "know" what reference this is.
1761 In the most frequent case
1763 lock = ldlm_handle2lock(handle);
1765 LDLM_LOCK_PUT(lock);
1767 no changes are required. When any other reference (received _not_ from
1768 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1769 instead of LDLM_LOCK_PUT().
1771 Arguably, changes are pervasive, and interface requires some discipline
1772 for proper use. On the other hand, it was very instrumental in finding
1773 a few leaked lock references.
1777 Description: Add ldlm_lock_addref_try().
1778 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1779 attempts to addref a lock that might be being canceled
1784 Description: Add ldlm_weigh_callback().
1785 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1786 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1787 hard-coded `number of pages' logic.
1791 Description: Add lockdep annotations to llog code.
1792 Details : Use appropriately tagged _nested() locking calls in the places
1793 where llog takes more than one ->lgh_lock lock.
1797 Description: Add loi_kms_set().
1798 Details : Wrap kms updates into a helper function.
1802 Description: Constify instances of struct lsm_operations.
1803 Details : Constify instances of struct lsm_operations.
1807 Description: lu_conf support.
1808 Details : On a server, a file system object is uniquely identified
1809 by a fid, which is sufficient to locate and load all object
1810 state (inode). On a client, on the other hand, more data are
1811 necessary instantiate an object. Change lu_object_find() and
1812 friends to take additional `lu_conf' argument describing object.
1813 Typically this includes layout information.
1817 Description: lu_context fixes.
1818 Details : Introduce new lu_context functions that are needed on the client
1819 side, where some system threads (ptlrpcd) are shared by multiple
1820 modules, and so cannot be stopped during module shutdown.
1824 Description: Add start and stop methods to lu_device_type_operations.
1825 Details : Introduce two new methods in lu_device_type_operations, that are
1826 invoked when first instance of a given type is created and last one
1827 is destroyed respectively. This is need by CLIO.
1831 Description: Add lu_ref support to struct lu_device.
1832 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1833 track leaked references.
1837 Description: Introduce lu_kmem_descr.
1838 Details : lu_kmem_descr and its companion interface allow to create
1839 and destroy a number of kmem caches at once.
1843 Description: Fix lu_object finalization race.
1844 Details : Fix a race between lu_object_find() finding an object and its
1845 concurrent finalization. This race is (most likely) not possible
1846 on the server, but might happen on the client.
1850 Description: Introduce lu_ref interface.
1851 Details : lu_ref is a debugging module allowing to track references to
1852 a given object. It is quite cpu expensive, and has to be
1853 explicitly enabled with --enable-lu_ref. See usage description
1858 Description: Factor lu_site procfs stats into a separate function.
1859 Details : Separate lu_site stats printing code into a separate function
1860 that can be reused on a client.
1864 Description: Constify instances of struct {lu,dt,md}_device_operations.
1865 Details : Constify instances of struct {lu,dt,md}_device_operations.
1869 Description: Introduce struct md_site and move meta-data specific parts of
1870 struct lu_site here.
1871 Details : Move md-specific fields out of struct lu_site into special struct
1872 md_site, so that lu_site can be used on a client.
1876 Description: Kill mdd_lov_destroy().
1877 Details : Remove unused mdd code.
1881 Description: Add st_block checking to multistat.c.
1882 Details : Add st_block checking to multistat.c.
1886 Description: Add lu_ref support to struct obd_device.
1887 Details : Add lu_ref tracking to obd_device.
1891 Description: Kill obd_set_fail_loc().
1892 Details : Remove unused code.
1896 Description: Add special type for ptlrpc_request interpret functions.
1897 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1898 there. Actual usage will be in CLIO.
1902 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1903 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1904 doesn't work with lockdep.
1908 Description: Add rwv.c test program.
1909 Details : New testing program exercising readv(2) and writev(2) (Qian).
1913 Description: Add sendfile.c test program.
1914 Details : New testing program exercising sendfile(2) (Jay).
1918 Description: Ratelimit a message that can be very frequent.
1919 Details : Ratelimit a memory allocation failure message that can
1924 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1925 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1929 Description: Liblustre build fixes.
1930 Details : Liblustre build fixes.
1934 Description: libcfs: add cfs_{need,cond}_resched() interface.
1935 Details : libcfs: add cfs_{need,cond}_resched() definition and
1936 implementations for Linux, NT, and liblustre.
1938 Severity : enhancement
1940 Description: More exported tunables for mballoc
1941 Details : Add support for tunable preallocation window and new tunables for
1942 large/small requests
1946 Description: Detect corruption of block bitmap and checking for preallocations
1947 Details : Checks validity of on-disk block bitmap. Also it does better
1948 checking of number of applied preallocations. When corruption is
1949 found, it turns filesystem readonly to prevent further corruptions.
1953 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1954 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1955 window to 0. If it does hit cache miss, and read-pattern is still
1956 stride-io mode, does not reset the stride window, but also does not
1957 increase the stride window length in this case.
1961 Frequency : only for big-endian servers
1962 Description: Check if system is big-endian while mounting fs with extents feature
1963 Details : Mounting a filesystem with extents feature will fail on big-endian
1964 systems since ext3-based ldiskfs is not supported on big-endian
1965 systems. This can be over-riden with "bigendian_extents" mount option.
1967 Severity : enhancement
1969 Description: The root squash functionality
1970 Details : A security feature, which is to prevent users from being able
1971 to mount lustre on their desktop, run as root, and delete
1972 all of the files in the filesystem. The goal is accomplished by
1973 remapping user id (UID) and group id (GID) of the root user to
1974 a UID and GID specified by the system administartor via Lustre
1975 configuration management server (MGS). The functionality also
1976 allows to specify sets of clients for which the remapping does
1981 Description: Excessive recovery window
1982 Details : With AT enabled, the recovery window can be excessively long (6000+
1983 seconds). To address this problem, we no longer use
1984 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1985 timeout no longer depends on the service time, it is set to
1986 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1987 time via pb_service_time.
1991 Description: Watchdog triggered on MDS failover
1992 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1993 the OSTs know that the MDS "UUID" can be reused for the same export
1994 from a different NID, so we do not need to wait for the export to be
1998 Frequency : rare, only if using MMP with Linux RAID
2000 Description: MMP doesn't work with Linux RAID
2001 Details : While using HA for Lustre servers with Linux RAID, it is possible
2002 that MMP will not detect multiple mounts. To make this work we
2003 need to unplug the device queue in RAID when the MMP block is being
2004 written. Also while reading the MMP block, we should read it from
2005 disk and not the cached one.
2007 Severity : enhancement
2009 Description: open file using fid
2010 Details : A file can be opened using just its fid, like
2011 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2013 --------------------------------------------------------------------------------
2015 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2017 * Support for kernels:
2018 2.6.5-7.283 (SLES 9),
2019 2.6.9-55.EL (RHEL 4),
2020 2.6.16.46-0.14 (SLES 10),
2021 2.6.18.8 vanilla (kernel.org)
2022 * Client support for unpatched kernels:
2023 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2024 2.6.16 - 2.6.22 vanilla (kernel.org)
2025 * Due to recently discovered recovery problems, we do not recommend
2026 using patchless RHEL 4 clients with this or any earlier release.
2027 * Recommended e2fsprogs version: 1.39.cfs8
2028 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2029 * Starting with this release, the ldiskfs backing filesystem required
2030 by Lustre is now in its own package, lustre-ldiskfs. This package
2031 should be installed. It is versioned separately from Lustre and
2032 may be released separately in future.
2036 Description: block reactivating mgc import until all deactivates complete
2037 Details : Fix race when failing back MDT/MGS to itself (testing)
2040 Frequency : at statup only
2042 Description: mds_lov_synchronize race leads to various problems
2043 Details : simultaneous MDT->OST connections at startup can cause the
2044 sync to abort, leaving the OSC in a bad state.
2046 Severity : enhancement
2048 Description: add optional extra BUILD_VERSION info
2049 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2050 to override the lustre version.
2053 Frequency : 2.6.18 servers only
2055 Description: ll_kern_mount() doesn't release the module reference
2056 Details : The ldiskfs module reference count never drops down to 0
2057 because ll_kern_mount() doesn't release the module reference.
2062 Description: server LBUG when using old ost_num_threads parameter
2063 Details : Accept the old ost_num_threads parameter but warn that it
2064 is deprecated, and fix an off-by-one error that caused an LBUG.
2069 Description: Transient SCSI error results in persistent IO issue
2070 Details : iobuf->dr_error is not reinitialized to 0 between two
2074 Frequency : sometimes when underlying device returns I/O errors
2076 Description: OSTs not going read-only during write failures
2077 Details : OSTs are not remounted read-only when the journal commit threads
2078 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2079 instead of the ext3 wrappers.
2083 Description: poor connect scaling with increasing client count
2084 Details : Don't run filter_grant_sanity_check for more than 100 exports
2085 to improve scaling for large numbers of clients.
2088 Frequency : SLES10 only
2090 Description: sanity-quota.sh quotacheck failed: rc = -22
2091 Details : Quotas cannot be enabled on SLES10.
2094 Frequency : liblustre clients only
2096 Description: getdirentries does not give error when run on compute nodes
2097 Details : getdirentries does not fail when the size specified as an argument
2098 is too small to contain at least one entry
2100 Severity : enhancement
2102 Description: Add LNET router traceability for debug purposes
2103 Details : If a checksum failure occurs with a router as part of the
2104 IO path, the NID of the last router that forwarded the bulk data
2105 is printed so it can be identified.
2110 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2111 Details : Due to a race condition, liblustre clients were occasionally
2112 evicted incorrectly.
2114 Severity : enhancement
2116 Description: lfs setstripe use optional parameters instead of postional
2119 Severity : enhancement
2121 Description: Nanosecond timestamp support for ldiskfs
2122 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2123 resolution timestamps. There is not yet support for this at
2124 the Lustre filesystem level.
2127 Frequency : during server recovery
2129 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2130 Details : request with rq_no_resend flag not awake l_wait_event if they get
2134 Frequency : nfs export on patchless client
2136 Description: connectathon hang when test nfs export over patchless client
2137 Details : Disconnected dentry cannot be found with lookup, so we do not need
2138 to unhash it or make it invalid
2141 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2142 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2143 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2148 Description: open req refcounting wrong on reconnect
2149 Details : If reconnect happened between getting open reply from server and
2150 call to mdc_set_replay_data in ll_file_open, we will schedule
2151 replay for unreferenced request that we are about to free.
2152 Subsequent close will crash in variety of ways.
2153 Check that request is still eligible for replay in
2154 mdc_set_replay_data().
2159 Description: disable writes to filesystem when reading health_check file
2160 Details : the default for reading the health_check proc file has changed
2161 to NOT do a journal transaction and write to disk, because this
2162 can cause reads of the /proc file to hang and block HA state
2163 checking on a healthy but otherwise heavily loaded system. It
2164 is possible to return to the previous behaviour during configure
2165 with --enable-health-write.
2167 Severity : enhancement
2169 Description: 64-bit inode version
2170 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2171 to the inode. This will be required for version-based recovery.
2176 Description: MDS fails to start if a duplicate client export is detected
2177 Details : in some rare cases it was possible for a client to connect to
2178 an MDS multiple times. Upon recovery the MDS would detect this
2179 and fail during startup. Handle this more gracefully.
2181 Severity : enhancement
2183 Description: Add -o localflock option to simulate old noflock
2185 Details : This will achieve local-only flock/fcntl locks
2191 Description: log_commit_thread vs filter_destroy race leads to crash
2192 Details : Take import reference before releasing llog record semaphore
2197 Description: Wrong request locking in request set processing
2198 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2199 imp_delayed_list, in this place should be used imp_lock.
2202 Frequency : when reconnection
2204 Description: Grant Leak when osc reconnect to OST
2205 Details : When osc reconnect ost, OST(filter) should check whether it
2206 should grant more space to client by comparing fed_grant and
2207 cl_avail_grant, and return the granted space to client instead
2208 of "new granted" space, because client will call osc_init_grant
2209 to update the client grant space info.
2212 Frequency : when client reconnect to OST
2214 Description: Grant Leak when osc do resend and replay bulk write
2215 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2216 bulk write request, because the grant info while be sync between
2217 OSC and OST when reconnect, and we should ignore the grant info
2218 these of resend/replay write req.
2223 Description: Grant space more than avaiable left space sometimes.
2224 Details : When then OST is about to be full, if two bulk writing from
2225 different clients came to OST. Accord the avaliable space of the
2226 OST, the first req should be permitted, and the second one
2227 should be denied by ENOSPC. But if the seconde arrived before
2228 the first one is commited. The OST might wrongly permit second
2229 writing, which will cause grant space > avaiable space.
2232 Frequency : when client is evicted
2234 Description: Grant might be wrongly erased when osc is evicted by OST
2235 Details : when the import is evicted by server, it will fork another
2236 thread ptlrpc_invalidate_import_thread to invalidate the
2237 import, where the grant will be set to 0. While the original
2238 thread will update the grant it got when connecting. So if
2239 the former happened latter, the grant will be wrongly errased
2240 because of this race.
2245 Description: Checking Stale with correct fid
2246 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2247 to check whether it is stale, because sometimes, we want the
2248 enqueue happened anyway, and op_data.fid2 will not be initialized.
2250 Severity : enhancement
2252 Description: update patchless client
2253 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2256 Frequency : only with 2.4 kernel
2258 Description: random memory corruption
2259 Details : size of struct ll_inode_info is to big for union inode.u and this
2260 can be cause of random memory corruption.
2265 Description: Memory leak in recovery
2266 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2267 It should also check obd_fail before fsfilt_start, otherwise if
2268 fsfilt_start return -EROFS,(failover mds during mds recovery).
2269 then the req will return with repmsg->transno = 0 and rc = EROFS.
2270 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2271 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2272 be freed no matter whether fsfilt_commit success or not.
2275 Frequency : only with huge count clients
2277 Description: Prevents from taking the superblock lock in llap_from_page for
2279 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2280 for llap_from_page call in ll_removepage prevents from taking the
2281 superblock lock for a soon died page.
2286 Description: Not check open intent error before release open handle
2287 Details : in some rare cases, the open intent error is not checked before
2288 release open handle, which may cause
2289 ASSERTION(open_req->rq_transno != 0), because it tries to release
2290 the failed open handle.
2295 Description: Set cat log bitmap only after create log success.
2296 Details : in some rare cases, the cat log bitmap is set too early. and it
2297 should be set only after create log success.
2301 Description: Accessing a block bevice can re-enable I/O when Lustre is
2302 tearing down a device.
2303 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2308 Description: service threads may hog cpus when there are a lot of requests
2309 Details : Insert cond_resched to give other threads a chance to use some CPU
2314 Description: the cat log was not initialized in recovery
2315 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2316 unlink log on mds will not be initialized until mds post
2317 recovery. And also in mds post recovery, the unlink log will
2318 initialization will be done asynchronausly, so there will be race
2319 between add unlink log and unlink log initialization.
2323 Description: brw_stats were being printed incorrectly
2324 Details : brw_stats were being printed as log2 but all of them were not
2325 recorded as log2. Also remove some code duplication arising from
2326 filter_tally_{read,write}.
2330 Frequency : rare, only in recovery.
2331 Description: ASSERTION(req->rq_type != LI_POISON) failed
2332 Details : imp_lock should be held while iterating over imp_sending_list for
2333 prevent destroy request after get timeout in ptlrpc_queue_wait.
2337 Description: replay-single.sh test 52 fails
2338 Details : A lock's skiplist need to be cleanup when it being unlinked
2339 from its resource list.
2343 Description: Short directio read returns full requested size rather than
2345 Details : Direct I/O operations should return actual amount of bytes
2346 transferred rather than requested size.
2348 Severity : enhancement
2350 Description: metadata RPC reduction (e.g. for rm performance)
2351 Details : decrease the amount of synchronous RPC between clients and servers
2352 by canceling conflicing lock before the operation on the client
2353 and packing thier handles into the main operation RPC to server.
2355 Severity : enhancement
2357 Description: Async OSC create to avoid the blocking unnecessarily.
2358 Details : If a OST has no remain object, system will block on the creating
2359 when need to create a new object on this OST. Now, ways use
2360 pre-created objects when available, instead of blocking on an
2361 empty osc while others are not empty. If we must block, we block
2362 for the shortest possible period of time.
2366 Description: improve handling recoverable errors
2367 Details : if request processig with error which can be recoverable on server
2368 request should be resend, otherwise page released from cache and
2371 Severity : enhancement
2373 Description: refine locking for avoid write wrong info into lov_objid file
2374 Details : fix possible races with add new target and write/update data in
2377 --------------------------------------------------------------------------------
2379 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2384 Frequency : on some architectures
2386 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2387 Details : Uninitialized flags sometimes cause configuration commands to
2391 Frequency : patchless clients only
2393 Description: missing __iget() symbol export
2394 Details : The __iget() symbol export is missing. To avoid the need for
2395 this on patchless clients the deathrow inode reaper is turned
2396 off, and we depend on the VM to clean up old inodes. This
2397 dependency was during via the fix for bug 12181.
2399 --------------------------------------------------------------------------------
2401 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2403 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2404 INTEROPERATE with older versions automatically. In many cases a
2405 special upgrade step is needed. Please read the
2406 user documentation before upgrading any part of a live system.
2407 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2408 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2409 * WARNING: Lustre configuration and startup changes are required with
2410 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2414 Severity : enhancement
2416 Description: MountConf
2417 Details : Lustre configuration is now managed via mkfs and mount
2418 commands instead of lmc and lconf. New obd types (MGS, MGC)
2419 are added for dynamic configuration management. See
2420 https://mail.clusterfs.com/wikis/lustre/MountConf for
2423 Severity : enhancement
2425 Description: dynamic OST addition
2426 Details : OSTs can now be added to a live filesystem
2428 Severity : enhancement
2430 Description: startup order invariance
2431 Details : MDTs and OSTs can be started in any order. Clients only
2432 require the MDT to complete startup.
2434 Severity : enhancement
2436 Description: parallel, asynchronous orphan cleanup
2437 Details : orphan cleanup is now performed in separate threads for each
2438 OST, allowing parallel non-blocking operation.
2440 Severity : enhancement
2442 Description: optimized stripe assignment
2443 Details : stripe assignments are now made based on ost space available,
2444 ost previous usage, and OSS previous usage, in order to try
2445 to optimize storage space and networking resources.
2447 Severity : enhancement
2449 Description: Permanently set tunables
2450 Details : All writable /proc/fs/lustre tunables can now be permanently
2451 set on a per-server basis, at mkfs time or on a live system.
2453 Severity : enhancement
2455 Description: Lustre message v2
2456 Details : Add lustre message format v2.
2458 Severity : enhancement
2460 Description: client OST exclusion list
2461 Details : Clients can be started with a list of OSTs that should be
2462 declared "inactive" for known non-responsive OSTs.
2466 Description: SPEC SFS validation failure on NFS v2 over lustre.
2467 Details : Changes the blocksize for regular files to be 2x RPC size,
2468 and not depend on stripe size.
2470 Severity : enhancement
2472 Description: Multiple MD RPCs in flight.
2473 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2474 To avoid overly-overloading MDS, introduce a limit on number of
2475 MDS RPCs in flight for a single client and add /proc controls
2476 to adjust this limit.
2478 Severity : enhancement
2480 Description: client read/write statistics
2481 Details : Add client read/write call usage stats for performance
2482 analysis of user processes.
2483 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2484 file access. extents_stats shows chunk size distribution.
2485 extents_stats_per_process show chunk size distribution per
2488 Severity : enhancement
2490 Description: mds statistics
2491 Details : Add detailed mds operations statistics in
2492 /proc/fs/lustre/mds/*/stats.
2496 Description: Failure of copying files with lustre special EAs.
2497 Details : Client side always return success for setxattr call for lustre
2498 special xattr (currently only "trusted.lov").
2503 Description: Refcount LNET uuids
2504 Details : The global LNET uuid list grew linearly with every startup;
2505 refcount repeated list entries instead of always adding to
2508 Severity : enhancement
2510 Description: Dynamic service threads
2511 Details : Within a small range, start extra service threads
2512 automatically when the request queue builds up.
2515 Frequency : mixed-endian client/server environments
2517 Description: mixed-endian crashes
2518 Details : The new msg_v2 system had some failures in mixed-endian
2521 Severity : enhancement
2523 Description: Easy OST removal
2524 Details : OSTs can be permanently deactivated with e.g. 'lctl
2525 conf_param lustre-OST0001.osc.active=0'
2527 Severity : enhancement
2529 Description: MGS proc entries
2530 Details : Added basic proc entries for the MGS showing what filesystems
2533 Severity : enhancement
2535 Description: provide MGS failover
2536 Details : Added config lock reacquisition after MGS server failover.
2538 Severity : enhancement
2540 Description: add Linux 2.4 support
2541 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2545 Description: a large application tries to do I/O to the same resource and dies
2546 in the middle of it.
2547 Details : Check the req->rq_arrival time after the call to
2548 ost_brw_lock_get(), but before we do anything about
2549 processing it & sending the BULK transfer request. This
2550 should help move old stale pending locks off the queue as
2551 quickly as obd_timeout.
2554 Frequency : when an incorrect nid is specified during startup
2556 Description: ptlrpc connect to non-existant node causes kernel crash
2557 Details : LNET can't be re-entered from an event callback, which
2558 happened when we expire a message after the export has been
2559 cleaned up. Instead, hand the zombie cleanup off to another
2562 Severity : enhancement
2564 Description: plain/inodebits lock performance improvement
2565 Details : Grouping plain/inodebits in granted list by their request modes
2566 and bits policy, thus improving the performance of search through
2570 Frequency : only if OST filesystem is corrupted
2572 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2573 Details : for a short time RPCs with bulk IO are in the replay list,
2574 but replay of bulk IOs is unimplemented. If the OST filesystem
2575 is corrupted due to disk cache incoherency and then replay is
2576 started it is possible to trip an assertion. Avoid putting
2577 committed RPCs into the replay list at all to avoid this issue.
2580 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2582 Description: System hang on startup
2583 Details : This bug allowed the liblustre (e.g. catamount) client to
2584 return to the app before handling all startup RPCs. This
2585 could leave the node unresponsive to lustre network traffic
2586 and manifested as a server ptllnd timeout.
2588 Severity : enhancement
2590 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2591 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2592 boolean to control whether to print peer debug info when a
2593 client's RPC times out.
2596 Frequency : only for kernels with patches from Lustre below 1.4.3
2598 Description: Remove old rdonly API
2599 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2602 Frequency : only for devices with external journals
2604 Description: Set external device read-only also
2605 Details : During a commanded failover stop, we set the disk device
2606 read-only while the server shuts down. We now also set any
2607 external journal device read-only at the same time.
2610 Frequency : when upgrading from 1.4 while trying to change parameters
2612 Description: The wrong (new) MDC name was used when setting parameters for
2613 upgraded MDT's. Also allows changing of OSC (and MDC)
2614 parameters if --writeconf is specified at tunefs upgrade time.
2617 Frequency : when setting specific ost indicies
2619 Description: QOS code breaks on skipped indicies
2620 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2621 created with --index need not be sequential.
2623 Severity : enhancement
2625 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2626 Details : The uninit_groups feature works in conjunction with the kernel
2627 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2628 up the pass1 processing of e2fsck. This is a read-only feature
2629 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2630 mount filesystems that have had this feature enabled.
2632 Severity : enhancement
2634 Description: Improve multi-block allocation algorithm to avoid fragmentation
2635 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2636 allocation locality and avoid filesystem fragmentation.
2638 ------------------------------------------------------------------------------
2640 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2642 * Support for kernels:
2643 2.4.21-47.0.1.EL (RHEL 3)
2644 2.6.5-7.283 (SLES 9)
2645 2.6.9-42.0.10.EL (RHEL 4)
2646 2.6.12.6 vanilla (kernel.org)
2647 2.6.16.27-0.9 (SLES 10)
2648 * Recommended e2fsprogs version: 1.39.cfs5
2650 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2654 Frequency : occasional, depends on client load and configuration
2655 Bugzilla : 12181, 12203
2656 Description: data loss for recently-modified files
2658 Details : In some cases it is possible that recently written or created
2659 files may not be written to disk in a timely manner (this should
2660 normally be within 30s unless client IO load is very high).
2661 The problem appears as zero-length files or files that are a
2662 multiple of 1MB in size after a client crash or client eviction
2663 that are missing data at the end of the file.
2665 This problem is more likely to be hit on clients where files are
2666 repeatedly created and unlinked in the same directory, clients
2667 have a large amount of RAM, have many CPUs, the filesystem has
2668 many OSTs, the clients are rebooted frequently, and/or the files
2669 are not accessed by other nodes after being written.
2671 The presence of the problem can be detected by looking at
2672 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2673 smaller than the second (nr_unused) then dirty files will not
2674 be flushed automatically to disk. "sync; sleep 10" should be
2675 run several times on the node before unmounting it to update
2676 Lustre (this is also safe to run on nodes without this problem).
2678 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2679 that can cause this same problem, so customers using that kernel
2680 also need to update the kernel in addition to Lustre. In order
2681 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2683 It is normal that files written just before a client crash (less
2684 than 30s) may not yet have been flushed to disk, even for local
2688 Frequency : frequent on thin XT3 nodes
2690 Description: UUID collision on thin XT3 Linux nodes
2691 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2692 randomly, since we relied on an insufficiently-seeded PRNG.
2697 Description: OSS hangs after "All ost request buffers busy"
2698 Details : A deadlock between quota and journal operations caused OSS
2699 hangs after printing "All ost request buffers busy."
2702 Frequency : always on liblustre builds
2704 Description: Cleanup compiler warnings on liblustre
2707 Frequency : always on liblustre builds on XT3
2709 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2713 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2714 > 32000 subdirectories
2715 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2716 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2717 subdirectory count crosses 32000. This will aid e2fsck to
2718 correctly handle more than 32000 subdirectories.
2721 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2723 Description: System hang on startup
2724 Details : This bug allowed the liblustre (e.g. catamount) client to
2725 return to the app before handling all startup RPCs. This
2726 could leave the node unresponsive to lustre network traffic
2727 and manifested as a server ptllnd timeout.
2729 Severity : enhancement
2731 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2732 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2733 boolean to control whether to print peer debug info when a
2734 client's RPC times out.
2739 Description: make O_SYNC working on 2.6 kernels
2740 Details : 2.6 kernels use different method for mark pages for write,
2741 so need add a code to lustre for O_SYNC work.
2746 Description: Failure to close file and release space on NFS
2747 Details : Put inode details into lock acquired in ll_intent_file_open.
2748 Use mdc_intent_lock in ll_intent_open to properly
2749 detect all kind of errors unhandled by mdc_enqueue.
2754 Description: proc file read during shutdown sometimes raced obd removal,
2756 Details : Add lock to prevent obd access after proc file removal.
2759 Frequency : Only for files larger than 4GB on 32-bit clients.
2761 Description: improperly doing page alignment of locks
2762 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2763 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2768 Description: RPCs being resent when they shouldn't be
2769 Details : Some RPCs that should not be resent are being resent. This
2770 can cause inconsistencies in the RPC state machine. Do not
2771 resend such requests.
2774 Frequency : rare, only with NFS export
2776 Description: Crash on NFS re-export node
2777 Details : under very unusual load conditions an assertion is hit in
2778 ll_intent_file_open()
2781 Frequency : only if OST filesystem is corrupted
2783 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2784 Details : for a short time RPCs with bulk IO are in the replay list,
2785 but replay of bulk IOs is unimplemented. If the OST filesystem
2786 is corrupted due to disk cache incoherency and then replay is
2787 started it is possible to trip an assertion. Avoid putting
2788 committed RPCs into the replay list at all to avoid this issue.
2793 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2794 Details : Large single O_DIRECT read and write calls can fail to allocate
2795 a sufficiently large buffer to process the request. In case of
2796 allocation failure the allocation is retried with a smaller
2797 buffer and broken into smaller requests.
2799 Severity : enhancement
2801 Description: Add -o localflock option to simulate old noflock behaviour.
2802 Details : This will achieve local-only flock/fcntl locks coherentness.
2807 Description: versioning check is incomplete
2808 Details : Checking the version difference of client vs. server, report
2809 error if the gap is too big.
2814 Description: add support PG_writeback bit
2815 Details : add support for PG_writeback bit for Lustre, for more carefull
2816 work with page cache in 2.6 kernel. This also fix some deadlocks
2817 and remove hack for work O_SYNC with 2.6 kernel.
2819 Severity : enhancement
2821 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2822 Details : The uninit_groups feature works in conjunction with the kernel
2823 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2824 up the pass1 processing of e2fsck. This is a read-only feature
2825 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2826 mount filesystems that have had this feature enabled.
2828 Severity : enhancement
2830 Description: Improve multi-block allocation algorithm to avoid fragmentation
2831 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2832 allocation locality and avoid filesystem fragmentation.
2834 ------------------------------------------------------------------------------
2836 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2838 * Support for kernels:
2839 2.6.9-42.0.3.EL (RHEL 4)
2840 2.6.5-7.276 (SLES 9)
2841 2.4.21-47.0.1.EL (RHEL 3)
2842 2.6.12.6 vanilla (kernel.org)
2843 2.6.16.21-0.8 (SLES10)
2844 * Recommended e2fsprogs version: 1.39.cfs2-0
2846 * The backwards-compatible /proc/sys/portals symlink has been removed
2847 in this release. Before upgrading, please ensure that you change
2848 any configuration scripts or /etc/sysctl.conf files that access
2849 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2850 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2851 in advance of the upgrade on any system running Lustre 1.4.6 or
2852 newer, since /proc/sys/lnet was added in that version.
2853 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2857 Frequency : only when quota is used
2859 Description: avoid scanning export list for quota master
2860 Details : Change the algorithms to avoid scanning export list in order
2861 to improve the efficiency.
2864 Frequency : MDS failover only, very rarely
2866 Description: "went back in time" messages on mds failover
2867 Details : The greatest transno may be lost when the current operation
2868 finishes with an error (transno==0) and the client's last_rcvd
2869 record is over-written. Save the greatest transno in the
2870 mds_last_transno for this case.
2873 Frequency : always for specific kernels and striping counts
2875 Description: client may get "Matching packet too big" without ACL support
2876 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2877 when trying to access files in certain configurations. The
2878 clients should in fact be denied when mounting because they do
2879 not understand ACLs.
2882 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2884 Description: many clients connecting with IO in progress causes connect timeouts
2885 Details : Avoid synchronous journal commits to avoid delays caused by many
2886 clients connecting/disconnecting when bulk IO is in progress.
2887 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2888 OST_IO_PORTAL to avoid delays behind potentially many pending
2892 Frequency : occasionally with multiple writers to a single file
2894 Description: shared writes to file may result in wrong size reported by stat()
2895 Details : Allow growing of kms when extent lock is cancelled
2898 Frequency : always with random mmap IO to multi-striped file
2900 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2901 Details : Only if the hole is at the end of OST object so that kms is too
2902 small. Fix is to increase kms accordingly in ll_nopage.
2905 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2907 Description: writes to a missing object would leak memory on the OST
2908 Details : If there is an inconsistency between the MDS and OST filesystems,
2909 such that the MDS references an object that doesn't exist, writes
2910 to that object will leak memory due to incorrect cleanup in the
2911 error handling path, eventually running out of memory on the OST.
2916 Description: Creating too long symlink causes lustre errors
2917 Details : Check symlink and name lengths before sending requests to MDS.
2920 Frequency : only if flock is enabled (not on by default)
2922 Description: posix locks not released on fd closure on 2.6.9+
2923 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2924 kernels, this caused such locks not to be released on fd close and
2925 then assertions on fs unmount about still used locks.
2928 Frequency : MDS failover only, very rarely
2930 Description: clients may get ASSERTION(granted_lock != NULL)
2931 Details : When request was taking a long time, and a client was resending
2932 a getattr by name lock request. The were multiple lock requests
2933 with the same client lock handle and
2934 mds_getattr_name->fixup_handle_for_resent_request found one of the
2935 lock handles but later failed with ASSERTION(granted_lock != NULL).
2940 Description: handle->h_buffer_credits > 0, assertion failure
2941 Details : h_buffer_credits is zero after truncate, causing assertion
2942 failure. This patch extends the transaction or creates a new
2946 Frequency : NFS re-export or patchless client
2947 Bugzilla : 11179, 10796
2948 Description: Crash on NFS re-export node (__d_move)
2949 Details : We do not want to hash the dentry if we don't have a lock.
2950 But if this dentry is later used in d_move, we'd hit uninitialised
2951 list head d_hash, so we just do this to init d_hash field but
2952 leave dentry unhashed.
2955 Frequency : NFS re-export or patchless client
2957 Description: NFS exports has problem with symbolic link
2958 Details : lustre client didn't properly install dentry when re-exported
2959 to NFS or running patchless client.
2962 Frequency : NFS re-export or patchless client
2964 Description: Various nfs/patchless fixes.
2965 Details : fixes reuse disconected alias for lookup process - this fixes
2966 warning "find_exported_dentry: npd != pd",
2967 fix permission error with open files at nfs.
2968 fix apply umask when do revalidate.
2971 Frequency : occasional
2973 Description: Crash on NFS re-export node
2974 Details : calling clear_page() on the wrong pointer triggered oops in
2975 generic_mapping_read().
2978 Frequency : rarely, using O_DIRECT IO
2980 Description: unaligned directio crashes client with LASSERT
2981 Details : check for unaligned buffers before trying any requests.
2984 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2986 Description: stale data returned from RAID cache
2987 Details : If only a small amount of IO is done to the RAID device before
2988 reading it again it is possible to get stale data from the RAID
2989 cache instead of reading it from disk.
2992 Frequency : always for sles10 kernel
2994 Description: sles10 support
2995 Details : ll_follow_link: compile fixes and using of nd_set_link
2996 under newer kernels.
2999 Frequency : depends on arch, kernel and compiler version, always on sles10
3002 Description: recursive or deep enough symlinks cause stack overflow
3003 Details : getting rid of large stack-allocated variable in
3007 Frequency : depends on hardware
3009 Description: lustre write performance loss in the SLES10 kernel
3010 Details : the performance loss is caused by using of write barriers in the
3011 ext3 code. The SLES10 kernel turns barrier support on by
3012 default. The fix is to undo that change for ldiskfs.
3014 ------------------------------------------------------------------------------
3016 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3018 * Support for kernels:
3019 2.6.9-42.0.3EL (RHEL 4)
3020 2.6.5-7.276 (SLES 9)
3021 2.4.21-47.0.1.EL (RHEL 3)
3022 2.6.12.6 vanilla (kernel.org)
3026 Frequency : quota enabled and large files being deleted
3028 Description: releasing more than 4GB of quota at once hangs OST
3029 Details : If a user deletes more than 4GB of files on a single OST it
3030 will cause the OST to spin in an infinite loop. Release
3031 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3036 Description: statfs data retrieved from /proc may be stale or zero
3037 Details : When reading per-device statfs data from /proc, in the
3038 {kbytes,files}_{total,free,avail} files, it may appear
3039 as zero or be out of date.
3042 Frequency : systems with MD RAID1 external journal devices
3044 Description: lconf's call to blkid is confused by RAID1 journal devices
3045 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3046 of returning all block devices that match the journal UUID.
3049 Frequency : always, for aggregate stripe size over 4GB
3051 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3052 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3053 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3056 Frequency : always if "lfs find" used on a local file/directory
3058 Description: "lfs find" segfaults if used on a local file/directory
3059 Details : The case where a directory component was not specified wasn't
3060 handled correctly. Handle this properly.
3063 Frequency : always on ppc64
3065 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3066 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3067 perform endianness conversion.
3070 Frequency : rarely (truncate to non-zero file size after write under load)
3071 Bugzilla : 10730, 10687
3072 Description: Files padded with zeros to next 4K multiple
3073 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3074 it is possible that files that are truncated to a non-zero size
3075 immediately after being written are filled with zero bytes beyond
3076 the truncated size. No file data is lost.
3078 Severity : enhancement
3080 Description: Allow recovery/failover for liblustre clients.
3081 Details : liblustre clients were unaware of failover configurations until
3084 Severity : enhancement
3086 Description: user file locks should fail when not mounting with flock option
3087 Details : Set up an error-returning stub in ll_file_operations.lock field
3088 to prevent incorrect behaviour when client is mounted without
3089 flock option. Also, set up properly f_op->flock field for
3093 Frequency : always on ia64
3095 Description: "lfs df" loops on printing out MDS statfs information
3096 Details : The obd_ioctl_data was not initialized and in some systems
3097 this caused a failure during the ioctl that did not return
3098 an error. Initialize the struct and return an error on failure.
3101 Frequency : SLES 9 only
3103 Description: Error of copying files with lustre special EAs as root
3104 Details : Client side always return success for setxattr call for lustre
3105 special xattr (currently only "trusted.lov").
3108 Frequency : rarely on clusters with both ia64+i386 clients
3110 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3111 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3112 clients are doing concurrent writes to the same file it is
3113 possible that the ia64 clients may LASSERT because the OST
3114 extent locks are not PAGE_SIZE aligned. Ensure that grown
3115 locks are always aligned on the request boundary.
3118 Frequency : specific use, occasional
3120 Description: Overwriting in use executable truncates on-disk binary image
3121 Details : If one node attempts to overwrite an executable in use by
3122 another node, we now correctly return ETXTBSY instead of
3123 truncating the file.
3125 Severity : enhancement
3127 Description: Async OSC create to avoid the blocking unnecessarily.
3128 Details : If a OST has no remain object, system will block on the creating
3129 when need to create a new object on this OST. Now, ways use
3130 pre-created objects when available, instead of blocking on an
3131 empty osc while others are not empty. If we must block, we block
3132 for the shortest possible period of time.
3137 Description: chmod on Lustre root is propagated to other clients
3138 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3139 invalid by the follow_mount time.
3142 Frequency : liblustre clients only
3144 Description: Race in 'instant cancel' lock handling could lead to such locks
3145 never to be granted in case of SMP MDS
3146 Details : Do not destroy not yet granted but cbpending locks in
3150 Frequency : replay/resend of open
3152 Description: non null lock assetion failure in mds_intent_policy
3153 Details : Trying to replay/resend lockless open requests resulted in
3154 mds_open() returning 0 with no lock. Now it sets a flag if
3155 a lock is going to be returned.
3157 Severity : enhancement
3159 Description: Checksum enhancements
3160 Details : New checksum enhancements allow for resending RPCs that failed
3163 Severity : enhancement
3165 Description: Tunables on number of dirty pages in cacche
3166 Details : Allow to set limit on number of dirty pages cached.
3171 Description: client crash on unmount - lock still has references
3172 Details : In some error handling cases it was possible to leak a lock
3173 reference on a client while accessing a file. This was not
3174 harmful to the client during operation, but would cause the
3175 client to crash when the filesystem is unmounted.
3178 Frequency : specific case, rare
3180 Description: ETXTBSY on mds though file not in use by client
3181 Details : ETXTBSY is no longer incorrectly returned when attempting to
3182 chmod or chown a directory that the user previously tried to
3183 execute or a currently-executing binary.
3186 Frequency : extremely rare except on liblustre-based clients
3188 Description: Lustre space not freed when files are deleted
3189 Details : Clean up open-unlinked files after client eviction. Previously
3190 the unlink was skipped and the files remained as orphans.
3195 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3196 Details : In some cases it was possible to send a blocking callback to
3197 a client doing a glimpse, even though that client didn't get
3198 a lock granted. When the glimpse lock is cancelled on the OST
3199 the freed lock is left on the waiting list and corrupted the list.
3202 Frequency : all core dumps
3204 Description: Broke core dumps to lustre
3205 Details : Negative dentry may be unhashed if parent does not have UPDATE
3206 lock, but some callers, e.g. do_coredump, expect dentry to be
3207 hashed after successful create, hash it in ll_create_it.
3209 ------------------------------------------------------------------------------
3211 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3213 * Support for kernels:
3214 2.6.9-42.0.2.EL (RHEL 4)
3215 2.6.5-7.276 (SLES 9)
3216 2.4.21-40.EL (RHEL 3)
3217 2.6.12.6 vanilla (kernel.org)
3221 Frequency : always on RHEL 3
3223 Description: Number of open files grows over time
3224 Details : The number of open files grows over time, whether or not
3225 Lustre is started. This was due to a filp leak introduced
3226 by one of our kernel patches.
3228 ------------------------------------------------------------------------------
3230 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3232 * Support for kernels:
3233 2.6.9-42.EL (RHEL 4)
3234 2.6.5-7.276 (SLES 9)
3235 2.4.21-40.EL (RHEL 3)
3236 2.6.12.6 vanilla (kernel.org)
3241 Bugzilla : 5719, 9635, 9792, 9684
3242 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3243 Details : If a server is under heavy load and cannot reply to new
3244 connection requests before the client resends the (re)connect,
3245 the connection handling code can behave badly if two service
3246 threads are concurrently handing separate (re)connections from
3247 the same client. Add better locking to the connection handling
3248 code, and ensure that only a single connection will be processed
3249 for a given client UUID, even if the lock is dropped.
3251 Severity : enhancement
3253 Description: add TCP zero-copy support to kernel
3254 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3255 sends if the hardware supports scatter-gather and checksumming.
3256 This allows socklnd to do client-write and server-read more
3257 efficiently and reduce CPU utilization from skbuf copying.
3260 Frequency : only if NFS exporting from client
3262 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3263 Details : If setting attributes on a file created under NFS that had
3264 never been opened it would be possible to oops the client
3265 if the file had no objects.
3268 Frequency : always for liblustre
3270 Description: liblustre client does MDS+OSTs setattr RPC for each write
3271 Details : When doing a write from a liblustre client, the client
3272 incorrectly issued an RPC to the MDS and each OST the file was
3273 striped over in order to update the timestamps. When writing
3274 with small chunks and many clients this could overwhelm the MDS
3275 with RPCs. In all cases it would slow down the write because
3276 these RPCs are unnecessary.
3278 Severity : enhancement
3280 Description: allow number of MDS service threads to be changed at module load
3281 Details : It is now possible to change the number of MDS service threads
3282 running. Adding "options mds mds_num_threads={N}" to the MDS's
3283 /etc/modprobe.conf will set the number of threads for the next
3284 time Lustre is restarted (assuming the "mds" module is also
3285 reloaded at that time). The default number of threads will
3286 stay the same, 32 for most systems.
3291 Description: OST crash if filesystem is unformatted or corrupt
3292 Details : If an OST is started on a device that has never been formatted
3293 or if the filesystem is corrupt and cannot even mount then the
3294 error handling cleanup routines would dereference a NULL pointer.
3299 Description: NULL pointer deref in llap_from_page.
3300 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3301 incorrect) mapping if the page was truncated/reclaimed while it was
3302 searched for. Check for this condition and skip such pages when
3303 doing readahead. Introduce extra check to llap_from_page() to
3304 verify page->mapping->host is non-NULL (so page is not anonymous).
3307 Frequency : Sometimes when using sys_sendfile
3309 Description: "page not covered by a lock" warnings from ll_readpage
3310 Details : sendfile called ll_readpage without right page locks present.
3311 Now we introduced ll_file_sendfile that does necessary locking
3312 around call to generic_file_sendfile() much like we do in
3316 Frequency : with certain MDS communication failures at client mount time
3318 Description: NULL pointer deref after failed client mount
3319 Details : a client connection request may delayed by the network layer
3320 and not be sent until after the PTLRPC layer has timed out the
3321 request. If the client fails the mount immediately it will try
3322 to clean up before the network times out the request. Add a
3323 reference from the request import to the obd device and delay
3324 the cleanup until the network drops the request.
3327 Frequency : occasionally during client (re)connect
3329 Description: assertion failure during client (re)connect
3330 Details : processing a client connection request may be delayed by the
3331 client or server longer than the client connect timeout. This
3332 causes the client to resend the connection request. If the
3333 original connection request is replied in this interval, the
3334 client may trip an assertion failure in ptlrpc_connect_interpret()
3335 which thought it would be the only running connect process.
3338 Frequency : only with obd_echo servers and clients that are rebooted
3340 Description: kernel BUG accessing uninitialized data structure
3341 Details : When running an obd_echo server it did not start the ping_evictor
3342 thread, and when a client was evicted an uninitialized data
3343 structure was accessed. Start the ping_evictor in the RPC
3344 service startup instead of the OBD startup.
3346 Severity : enhancement
3347 Bugzilla : 10193 (patchless)
3348 Description: Remove dependency on various unexported kernel interfaces.
3349 Details : No longer need reparent_to_init, exit_mm, exit_files,
3350 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3353 Frequency : rare (only users of deprecated and unsupported LDAP config)
3355 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3356 Details : LDAP apparently contains 'lustreName' attributes instead of
3357 'name'. A simple remapping of the name is sufficient.
3360 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3362 Description: waiting_locks_callback trips kernel BUG if client is evicted
3363 Details : Running with the dump_on_timeout debug flag turned on makes
3364 it possible that the waiting_locks_callback() can try to dump
3365 the Lustre kernel debug logs from an interrupt handler. Defer
3366 this log dumping to the expired_lock_main() thread.
3368 Severity : enhancement
3370 Description: Support NFS exporting on 2.6 kernels.
3371 Details : Implement non-rawops metadata methods for NFS server to use without
3372 changing NFS server code.
3375 Frequency : very rare (synthetic metadata workload only)
3377 Description: two racing renames might cause an MDS thread to deadlock
3378 Details : Running the "racer" program may cause one MDS thread to rename
3379 a file from being the source of a rename to being the target of
3380 a rename at exactly the same time that another thread is doing
3381 so, and the second thread has already enqueued these locks after
3382 doing a lookup of the target and is trying to relock them in
3383 order. Ensure that we don't try to re-lock the same resource.
3386 Frequency : only very large systems with liblustre clients
3388 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3389 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3390 all OSTs in parallel. This allows the eviction of stale liblustre
3391 clients to proceed much faster than if they were done in series,
3392 and also offers similar improvements for other set_info RPCs.
3397 Description: excessive CPU usage during initial read phase on client
3398 Details : During the initial read phase on a client, it would agressively
3399 retry readahead on the file, consuming too much CPU and impacting
3400 performance (since 1.4.5.8). Improve the readahead algorithm
3401 to avoid this, and also improve some other common cases (read
3402 of small files in particular, where "small" is files smaller than
3403 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3408 Description: MDS crash when receiving packet with unknown intent.
3409 Details : Do not LBUG in unknown intent case, just return -EFAULT
3411 Severity : enhancement
3412 Bugzilla : 9293, 9385
3413 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3414 Details : Do not serialize getattr (non-intent version) and statfs.
3417 Frequency : occasional, when OST network is overloaded/intermittent
3419 Description: client evicted by OST after bulk IO timeout
3420 Details : If a client sends a bulk IO request (read or write) the OST
3421 may evict the client if it is unresposive to its data GET/PUT
3422 request. This is incorrect if the network is overloaded (takes
3423 too long to transfer the RPC data) or dropped the OST GET/PUT
3424 request. There is no need to evict the client at all, since
3425 the pinger and/or lock callbacks will handle this, and the
3426 client can restart the bulk request.
3429 Frequency : Always when mmapping file with no objects
3431 Description: client crashes when mmapping file with no objects
3432 Details : Check that we actually have objects in a file before doing any
3433 operations on objects in ll_vm_open, ll_vm_close and
3439 Description: Request leak when working with deleted CWD
3440 Details : Introduce advanced request refcount tracking for requests
3441 referenced from lustre intent.
3443 Severity : Enhancement
3445 Description: Cache open file handles on client.
3446 Details : MDS now will return special lock along with openhandle, if
3447 requested and client is allowed to hold openhandle, even if unused,
3448 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3449 closing files for every read/write openration.
3451 Severity : Enhancement
3453 Description: Cache open negative dentries on client when possible.
3454 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3455 negative dentries on lock revocation.
3460 Description: Remounting a client read-only wasn't possible with a zconf mount
3461 Details : It wasn't possible to remount a client read-only with llmount.
3463 Severity : enhancement
3464 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3465 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3466 MPICH is included to add Lustre-specific ADIO interfaces.
3467 This is based closely on the UFS ADIO layer and only differs
3468 in file creation, in order to allow the OST striping to be set.
3469 This is user-contributed code and not supported by CFS.
3474 Description: extended inode attributes (immutable, append-only) work improperly
3475 when 2.4 and 2.6 kernels are used on client/server or vice versa
3476 Details : Introduce kernel-independent values for these flags.
3478 Severity : enhancement
3481 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3482 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3483 granularity. Now, Fractional MB granularity is be supported,
3484 this is very useful for low memory system.
3486 Severity : enhancement
3488 Description: Getattr by fid
3489 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3490 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3491 have locks and avoids getattr rpc requests for it on every stat.
3494 Frequency : Always, for filesystems larger than 2TB
3496 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3497 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3498 filesystems larger than 2TB to be mounted with mballoc enabled.
3501 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3503 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3504 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3505 filesystems larger than 2TB is mounted then the kernel will
3506 silently corrupt the start of the filesystem. CONFIG_LBD is
3507 enabled for all CFS-supported kernels, but the possibility of
3508 this happening with a modified kernel config exists.
3510 Severity : enhancement
3512 Description: add client O_DIRECT support for 2.6 kernels
3513 Details : It is now possible to do O_DIRECT reads and writes to files
3514 in the Lustre client mountpoint on 2.6 kernel clients.
3516 Severity : enhancement
3518 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3519 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3520 parallel, not waiting for response from every OST before sending
3521 a rpc to the next OST.
3526 Description: setattr vs write race when updating file timestamps
3527 Details : Client processes that update a file timestamp into the past
3528 right after writing to the file (e.g. tar) it is possible that
3529 the updated file modification time can be reset to the current
3530 time due to a race between processing the setattr and write RPC.
3532 Severity : enhancement
3534 Description: Bring 'lfs find' closer in line with regular Linux find.
3535 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3536 -print0 options and obtains all the needed info through the lustre
3539 Severity : enhancement
3541 Description: support up to 1024 configured devices on one node
3542 Details : change obd_dev array from statically allocated to dynamically
3543 allocated structs as they are first used to reduce memory usage
3548 Description: Flush dirty partially truncated pages during truncate
3549 Details : Immediatelly flush partially truncated pages in filter_setattr,
3550 this way we completely avoid having any pages in page cache on OST
3551 and can retire ugly workarounds during writes to flush such pages.
3556 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3557 Details : protect inode from truncation within vfs_unlink() context
3558 just take a reference before calling vfs_unlink() and release it
3559 when parent's i_sem is free.
3564 Description: last_id value checked outside lock on OST caused LASSERT failure
3565 Details : If there were multiple MDS->OST object precreate requests in
3566 flight, it was possible that the OST's last object id was checked
3567 outside a lock and incorrectly tripped an assertion. Move checks
3568 inside locks, and discard old precreate requests.
3571 Frequency : always, if extents are used on OSTs
3573 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3574 Details : OSTs using the extents format would not zero the high 16 bits of
3575 the index physical block number. This is not a problem for any
3576 OST filesystems smaller than 16TB, and no kernels support ext3
3577 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3578 new/modified files) and can be fixed for existing filesystems
3579 with e2fsprogs-1.39-cfs1.
3584 Description: import connection selection may be incorrect if timer wraps
3585 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3586 import connections to be ignored if the 32-bit jiffies counter
3587 wraps. Use a 64-bit jiffies counter.
3590 Frequency : very large clusters immediately after boot
3592 Description: LNET request buffers exhausted under heavy short-term load
3593 Details : If a large number of client requests are generated on a service
3594 that has previously never seen so many requests it is possible
3595 that the request buffer growth cannot keep up with the spike in
3596 demand. Instead of dropping incoming requests, they are held in
3597 the LND until the RPC service can accept more requests.
3600 Frequency : Sometimes during replay
3602 Description: Assertion failure in ll_local_open after replay.
3603 Details : If replay happened on an open request reply before we were able
3604 to set replay handler, reply will become not swabbed tripping the
3605 assertion in ll_local_open. Now we set the handler right after
3606 recognising of open request
3609 Frequency : very rare
3611 Description: kernel reports "badness in vsnprintf"
3612 Details : Reading from the "recovery_status" /proc file in small chunks
3613 may cause a negative length in lprocfs_obd_rd_recovery_status()
3614 call to vsnprintf() (which is otherwise harmless). Exit early
3615 if there is no more space in the output buffer.
3617 Severity : enhancement
3619 Description: clear OBD RPC statistics by writing to them
3620 Details : It is now possible to clear the OBD RPC statistics by writing
3621 to the "stats" file.
3626 Description: Client mtime is not the same on different clients after utimes
3627 Details : In some cases, the client was using the utimes() syscall on
3628 a file cached on another node. The clients now validate the
3629 ctime from the MDS + OSTs to determine which one is right.
3634 Description: Inability to activate failout mode
3635 Details : lconf script incorrectly assumed that in pythong string's numeric
3636 value is used in comparisons.
3639 Frequency : always with multiple stripes per file
3641 Description: Inefficient object allocation for mutli-stripe files
3642 Details : When selecting which OSTs to stripe files over, for files with
3643 a stripe count that divides evenly into the number of OSTs,
3644 the MDS is always picking the same starting OST for each file.
3645 Return the OST selection heuristic to the original design.
3650 Description: mount failures may take full timeout to return an error
3651 Details : Under some heavy load conditions it is possible that a
3652 failed mount can wait for the full obd_timeout interval,
3653 possibly several minutes, before reporting an error.
3654 Instead return an error as soon as the status is known.
3656 Frequency : quota enabled and large files being deleted
3658 Description: releasing more than 4GB of quota at once hangs OST
3659 Details : If a user deletes more than 4GB of files on a single OST it
3660 will cause the OST to spin in an infinite loop. Release
3661 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3666 Description: statfs data retrieved from /proc may be stale or zero
3667 Details : When reading per-device statfs data from /proc, in the
3668 {kbytes,files}_{total,free,avail} files, it may appear
3669 as zero or be out of date.
3672 Frequency : systems with MD RAID1 external journal devices
3674 Description: lconf's call to blkid is confused by RAID1 journal devices
3675 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3676 of returning all block devices that match the journal UUID.
3679 Frequency : always, for aggregate stripe size over 4GB
3681 Description: assertion fails when trying to use 4GB stripe size
3682 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3683 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3686 Frequency : always on ppc64
3688 Description: the first write on an ext3 filesystem with mballoc got stuck
3689 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3690 perform endianness conversion.
3692 ------------------------------------------------------------------------------
3694 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3696 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3697 INTEROPERATE with older versions automatically. Please read the
3698 user documentation before upgrading any part of a live system.
3699 * WARNING: Lustre networking configuration changes are required with
3700 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3703 * Support for kernels:
3704 2.6.9-22.0.2.EL (RHEL 4)
3705 2.6.5-7.244 (SLES 9)
3706 2.6.12.6 vanilla (kernel.org)
3709 Severity : enhancement
3710 Bugzilla : 7981/8208
3711 Description: Introduced Lustre Networking (LNET)
3712 Details : LNET is new networking infrastructure for Lustre, it includes
3713 a reorganized network configuration mode (see the user
3714 documentation for full details) as well as support for routing
3715 between different network fabrics. Lustre Networking Devices
3716 (LNDs) for the supported network fabrics have also been
3717 created for this new infrastructure.
3719 Severity : enhancement
3720 Description: Introduced Access control lists
3721 Details : clients can set ACLs on files and directories in order to have
3722 more fine-grained permissions than the standard Unix UGO+RWX.
3723 The MDS must be started with the "-o acl" mount option.
3725 Severity : enhancement
3726 Description: Introduced filesystem quotas
3727 Details : Administrators may now establish per-user quotas on the
3730 Severity : enhancement
3732 Description: Configuration change for the XT3
3733 The PTLLND is now used to run Lustre over Portals on the XT3
3734 The configure option(s) --with-cray-portals are no longer used.
3735 Rather --with-portals=<path-to-portals-includes> is used to
3736 enable building on the XT3. In addition to enable XT3 specific
3737 features the option --enable-cray-xt3 must be used.
3742 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3743 Details : A race between allocating a new llcd and re-getting the llcd_lock
3744 allowed another thread to grab newly-allocated llcd.
3746 Severity : enhancement
3748 Description: 2.6 OST async journal commit and locking fix to improve performance
3749 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3750 2.6 kernels are now async as they already were in 2.4 kernels so
3751 that they can commit concurrently with the network bulk transfer.
3752 For block-allocated files the filter allocation semaphore is held
3753 to avoid filesystem fragmentation during allocation. BKL lock
3754 removed for 2.6 xattr operations where it is no longer needed.
3759 Description: lconf incorrectly determined whether two IP networks could talk
3760 Details : In some more complicated routing and multiple-network
3761 configurations, lconf will avoid trying to make a network
3762 connection to a disjoint part of the IP space. It was doing the
3763 math incorrectly for one set of cases.
3768 Description: Fix for potential infinite loop processing records in an llog.
3769 Details : If an llog record is corrupted/zeroed, it is possible to loop
3770 forever in llog_process(). Validate the llog record length
3771 and skip the remainder of the block on error.
3774 Frequency : occasional (liblustre only)
3776 Description: liblustre could not open files whose last component is a symlink
3777 Details : sysio_path_walk() would incorrectly pass the open intent to
3778 intermediate path components.
3781 Frequency : rare (liblustre only with non-standard tuning)
3782 Bugzilla : 7201 (7350)
3783 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3784 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3785 after it has been released, only do this for referenced locks
3787 Severity : enhancement
3789 Description: specifying an (invalid) directory default stripe_size of -1
3790 would reset the directory default striping
3791 Details : stripe_size -1 was used internally to signal directory stripe
3792 removal, now use "all default" to signal dir stripe removal
3793 as a directory striping of "all default" is not useful
3796 Frequency : common for large clusters running liblustre clients
3798 Description: doing an ls when liblustre clients are running is slow
3799 Details : sending a glimpse AST to a liblustre client waits for every AST
3800 to time out, as liblustre clients will not respond. Since they
3801 cannot cache data we refresh the OST lock LVB from disk instead.
3803 Severity : enhancement
3805 Description: doing an ls at the same time as file IO can be slow
3806 Details : enqueue and other "small" requests can be blocked behind many
3807 large IO requests. Create a new OST IO portal for non-IO
3808 requests so they can be processed faster.
3811 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3813 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3814 created file with mode 000
3815 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3816 allow an ftruncate() as a non-root user to a file with mode 000.
3817 Lustre now respects this flag to disable mode checks when
3818 truncating a file owned by the user
3821 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3824 Description: Revoking locks from clients that went dead or catatonic might take
3826 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3827 cancellation of such locks instant on servers without waiting for
3828 any reply from clients. Clients drops these locks when cancel
3829 notification from server is received without replying.
3832 Frequency : liblustre-only, when liblustre client dies or becomes busy
3834 Description: Doing ls on Linux clients can take a long time with active
3836 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3837 granting such locks to it in the first place if possible. Locks
3838 are taken by proxy on the OST during the read or write and
3839 dropped immediately afterward. Add connect flags handling, do
3840 not grant locks to liblustre clients for glimpse ASTs.
3842 Severity : enhancement
3844 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3845 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3846 sequential reads and behaves badly with applications doing random
3847 reads. Improve it by reading ahead at least read region, and
3848 avoiding excessive large RPC for small reads.
3850 Severity : enhancement
3852 Description: Creating more than 1000 files for a single job may cause a load
3853 imbalance on the OSTs if there are also a large number of OSTs.
3854 Details : qos_prep_create() uses an OST index reseed value that is an
3855 even multiple of the number of available OSTs so that if the
3856 reseed happens in the middle of the object allocation it will
3857 still utilize the OSTs as uniformly as possible.
3862 Description: OST or MDS may oops in ping_evictor_main()
3863 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3864 but doesn't restart at beginning of obd_exports_timed list
3867 Severity : enhancement
3869 Description: improve by-nid export eviction on the MDS and OST
3870 Details : allow multiple exports with the same NID to be evicted at one
3871 time without re-searching the exports list.
3874 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3876 Description: MDS may oops in groups_free()
3877 Details : in rare race conditions a newly allocated group_info struct is
3878 freed again, and this can be NULL. The 2.4 compatibility code
3879 for groups_free() checked for a NULL pointer, but 2.6 did not.
3882 Frequency : common for liblustre clients doing little filesystem IO
3883 Bugzilla : 9352, 7313
3884 Description: server may evict liblustre clients accessing contended locks
3885 Details : if a client is granted a lock or receives a completion AST
3886 with a blocking AST already set it would not reply to the AST
3887 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3890 Frequency : lfs setstripe, only systems with more than 160 OSTs
3892 Description: unable to set striping with a starting offset beyond OST 160
3893 Details : llapi_create_file() incorrectly limited the starting stripe
3894 index to the maximum single-file stripe count.
3897 Frequency : LDAP users only
3899 Description: lconf did not handle in-kernel recovery with LDAP properly
3900 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3902 Severity : enhancement
3904 Description: bind OST threads to NUMA nodes to improve performance
3905 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3906 node and do their allocations there to localize memory access
3908 Severity : enhancement
3910 Description: llmount can determine client NID directly from Myrinet (GM)
3911 Details : the client NID code from gmnalnid was moved directly into
3912 llmount, removing the need to use this or specifying the
3913 client NID explicitly when mounting GM clients with zeroconf
3916 Frequency : if client is started with down MDS
3918 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3919 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3920 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3921 ptlrpc_import_delayed_req() because we didn't check if the
3922 request was interrupted, and we also didn't break out of the
3923 event loop if there was no timeout
3928 Description: data loss during non-page-aligned writes to a single file from
3929 both multiple nodes and multiple threads on one node at same time
3930 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3931 inconsistency led to false short-reads, that were cached and later
3932 used by ->prepare_write() to fill in partially written page,
3933 leading to data loss.
3936 Frequency : always, if lconf --abort_recovery used
3938 Description: lconf --abort_recovery fails with 'Operation not supported'
3939 Details : lconf was attempting to abort recovery on the MDT device and not
3942 Severity : enhancement
3944 Description: remove cleanup logs
3945 Details : replace lconf-generated cleanup logs with lustre internal
3946 cleanup routines. Eliminates the need for client-cleanup and
3949 Severity : enhancement
3951 Description: add support for EAs (user and system) on lustre filesystems
3952 Details : it is now possible to store extended attributes in the Lustre
3953 client filesystem, and with the user_xattr mount option it
3954 is possible to allow users to store EAs on their files also
3956 Severity : enhancement
3958 Description: Add possibility (config option) to show minimal available OST free
3960 Details : When compiled with --enable-mindf configure option, statfs(2)
3961 (and so, df) will return least minimal free space available from
3962 all OSTs as amount of free space on FS, instead of summary of
3963 free spaces of all OSTs.
3965 Severity : enhancement
3967 Description: do not expand extent locks acquired on OST-side
3968 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3969 by server: they are not cached anyway.
3972 Frequency : when mmap is used/binaries executed from Lustre
3974 Description: Unmmap pages before throwing them away from read cache.
3975 Details : llap_shrink cache now attempts to unmap pages before discarding
3976 them (if unmapping failed - do not discard). SLES9 kernel has
3977 extra checks that trigger if this unmapping is not done first.
3982 Description: lconf didn't resolve symlinks before checking to see whether a
3983 given mountpoint was already in use
3986 Frequency : when migrating failover services
3987 Bugzilla : 6395, 9514
3988 Description: When migrating a subset of services from a node (e.g. failback
3989 from a failover service node) the remaining services would
3990 time out and evict clients.
3991 Details : lconf --force (implied by --failover) sets the global obd_timeout
3992 to 5 seconds in order to quickly disconnect, but this caused
3993 other RPCs to time out too quickly. Do not change the global
3994 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3996 Severity : enhancement
3997 Frequency : if MDS is started with down OST
3998 Bugzilla : 9439,5706
3999 Description: Allow startup/shutdown of an MDS without depending on the
4000 availability of the OSTs.
4001 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4002 Add appropriate locking and lov-osc refcounts for safe
4003 cleaning. Add osc abort_inflight calls in case the
4004 synchronize never started.
4007 Frequency : occasional (Cray XT3 only)
4009 Description: root not authorized to access files in CRAY_PORTALS environment
4010 Details : The client process capabilities were not honoured on the MDS in
4011 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4012 been accessed by an authorized user then root was able to access
4013 the file on the local client also. The root user capabilities
4014 are now allowed on the MDS, as this environment has secure UID.
4017 Frequency : occasional
4019 Description: ldiskfs "too long searching" message happens too often
4020 Details : A debugging message (otherwise harmless) prints too often on
4021 the OST console. This has been reduced to only happen when
4022 there are fragmentation problems on the filesystem.
4027 Description: Division by zero in statfs when all OSCs are inactive
4028 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4029 lov_statfs divides by value returned by lov_get_stripecnt().
4033 Bugzilla : 9489, 3273
4034 Description: First write from each client to each OST was only 4kB in size,
4035 to initialize client writeback cache, which caused sub-optimal
4036 RPCs and poor layout on disk for the first writen file.
4037 Details : Clients now request an initial cache grant at (re)connect time
4038 and so that they can start streaming writes to the cache right
4039 away and always do full-sized RPCs if there is enough data.
4040 If the OST is rebooted the client also re-establishes its grant
4041 so that client cached writes will be honoured under the grant.
4046 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4047 Details : Now I/O RPCs go to different portal number and (presumably) fast
4048 lock requests (and glimses) and other RPCs get their own service
4049 threads pool that should be able to service those RPCs
4052 Severity : enhancement
4054 Description: Ability to exchange lustre version between client and servers and
4055 issue warnings at client side if client is too old. Also for
4056 liblustre clients there is ability to refuse connection of too old
4058 Details : New 'version' field is added to connect data structure that is
4059 filled with version info. That info is later checked by server and
4063 Frequency : rare, liblustre only.
4064 Bugzilla : 9296, 9581
4065 Description: Two simultaneous writes from liblustre at offset within same page
4066 might proceed at the same time overwriting eachother with stale
4068 Details : I/O lock withing llu_file_prwv was released too early, before data
4069 actually was hitting the wire. Extended lock-holding time until
4070 server acknowledges receiving data.
4073 Frequency : extremely rare. Never observed in practice.
4075 Description: avoid generating lustre_handle cookie of 0.
4076 Details : class_handle_hash() generates handle cookies by incrementing
4077 global counter, and can hit 0 occasionaly (this is unlikely, but
4078 not impossible, because initial value of cookie counter is
4079 selected randonly). Value of 0 is used as a sentinel meaning
4080 "unassigned handle" --- avoid it. Also coalesce two critical
4081 sections in this function into one.
4083 Severity : enhancement
4085 Description: allow liblustre clients to delegate truncate locking to OST
4086 Details : To avoid overhead of locking, liblustre client instructs OST to
4087 take extent lock in ost_punch() on client's behalf. New connection
4088 flag is added to handle backward compatibility.
4090 Severity : enhancement
4091 Bugzilla : 4928, 7341, 9758
4092 Description: allow number of OST service threads to be specified
4093 Details : a module parameter allows the number of OST service threads
4094 to be specified via "options ost ost_num_threads={N}" in the
4095 OSS's /etc/modules.conf or /etc/modprobe.conf.
4099 Bugzilla : 6146, 9635, 9895
4100 Description: servers crash with bad pointer in target_handle_connect()
4101 Details : In rare cases when a client is reconnecting it was possible that
4102 the connection request was the last reference for that export.
4103 We would temporarily drop the export reference and get a new
4104 one, but this may have been the last reference and the export
4105 was just destroyed. Get new reference before dropping old one.
4107 Severity : enhancement
4108 Frequency : if client is started with failover MDS
4110 Description: Allow multiple MDS hostnames in the mount command
4111 Details : Try to read the configuration from all specified MDS
4112 hostnames during a client mount in case the "primary"
4115 Severity : enhancement
4117 Description: Stop sending data to evicted clients as soon as possible.
4118 Details : Check if the client we are about to send or are sending data to
4119 was evicted already. (Check is done every second of waiting,
4120 for which l_wait_event interface was extended to allow checking
4121 of exit condition at specified intervals).
4124 Frequency : rare, normally only when NFS exporting is done from client
4126 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4128 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4131 Frequency : always, if srand() is called before liblustre initialization
4133 Description: Liblustre uses system PRNG disturbing its usage by user application
4134 Details : Introduce internal to lustre fast and high-quality PRNG for
4135 lustre usage and make liblustre and some other places in generic
4136 lustre code to use it.
4138 Severity : enhancement
4139 Bugzilla : 9477, 9557, 9870
4140 Description: Verify that the MDS configuration logs are updated when xml is
4141 Details : Check if the .xml configuration logs are newer than the config
4142 logs stored on the MDS and report an error if this is the case.
4143 Request --write-conf, or allow starting with --old_conf.
4145 Severity : enhancement
4147 Description: Handle symlinks in the path when checking if Lustre is mounted.
4148 Details : Resolve intermediate symlinks when checking if a client has
4149 mounted a filesystem to avoid duplicate client mounts.
4154 Description: lconf can hit an error exception but still return success.
4155 Details : The lconf command catches the Command error exception at the top
4156 level script context and will exit with the associated exit
4157 status, but doesn't ensure that this exit status is non-zero.
4162 Description: failure of ptlrpc thread startup can cause oops
4163 Details : Starting a ptlrpc service thread can fail if there are a large
4164 number of threads or the server memory is very fragmented.
4165 Handle this without oopsing.
4168 Frequency : always, only if liblustre and non-default acceptor port was used
4170 Description: liblustre cannot connect to servers with non-default acceptor port
4171 Details : tcpnal_set_default_params() was not called and was therefore
4172 ignoring the environment varaible TCPNAL_PORT, as well as other
4173 TCPNAL_ environment variables
4178 Description: two objects could be created on the same OST for a single file
4179 Details : If an OST is down, in some cases it was possible to create two
4180 objects on a single OST for a single file. No problems other
4181 than potential performance impact and spurious error messages.
4185 Bugzilla : 5681, 9562
4186 Description: Client may oops in ll_unhash_aliases
4187 Details : Client dcache may become inconsistent in race condition.
4188 In some cases "getcwd" can fail if the current directory is
4194 Description: Inode refcounting problems in NFS export code
4195 Details : link_raw functions used to call d_instantiate without obtaining
4196 extra inode reference first.
4200 Bugzilla : 9942, 9903
4201 Description: Referencing freed requests leading to crash, memleaks with NFS.
4202 Details : We used to require that call to ll_revalidate_it was always
4203 followed by ll_lookup_it. Also with revalidate_special() it is
4204 possible to call ll_revalidate_it() twice for the same dentry
4205 even if first occurence returned success. This fix changes semantic
4206 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4207 reference on a request referred from the intent.
4208 ll_intent_release() then releases such a request.
4211 Frequency : rare, normally benchmark loads only
4213 Description: unlinked inodes were kept in memory on the client
4214 Details : If a client is repeatedly creating and unlinking files it
4215 can accumulate a lot of stale inodes in the inode slab cache.
4216 If there is no other client load running this can cause the
4217 client node to run out of memory. Instead flush old inodes
4218 from client cache that have the same inode number as a new inode.
4221 Frequency : SLES9 2.6.5 kernel and long filenames only
4222 Bugzilla : 9969, 10379
4223 Description: utime reports stale NFS file handle
4224 Details : SLES9 uses out-of-dentry names in some cases, which confused
4225 the lustre dentry revalidation. Change it to always use the
4229 Frequency : rare, unless heavy write-truncate concurrency is continuous
4230 Bugzilla : 4180, 6984, 7171, 9963, 9331
4231 Description: OST becomes very slow and/or deadlocked during object unlink
4232 Details : filter_destroy() was holding onto the parent directory lock
4233 while truncating+unlinking objects. For very large objects this
4234 may block other threads for a long time and slow overall OST
4235 responsiveness. It may also be possible to get a lock ordering
4236 deadlock in this case, or run out of journal credits because of
4237 the combined truncate+unlink. Solution is to do object truncate
4238 first in one transaction without parent lock, and then do the
4239 final unlink in a new transaction with the parent lock. This
4240 reduces the lock hold time dramatically.
4243 Frequency : rare, 2.4 kernels only
4245 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4246 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4247 before unmounting their filesystem in order to avoid deadlock.
4248 The kernel_locked() function in 2.4 kernels only checks whether
4249 the kernel lock is held, not whether it is this process that is
4250 holding it as 2.6 kernels do.
4255 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4256 Details : The client ptlrpc code may be trying to reconnect to a down
4257 server before a previous connection attempt has timed out.
4258 Increase the reconnect interval to be longer than the connection
4259 timeout interval to avoid sending duplicate connections to
4263 Frequency : echo_client brw_test command
4265 Description: fix echo_client to work with OST preallocated code
4266 Details : OST preallocation code (5137) didn't take echo_client IO path
4267 into account: echo_client calls filter methods outside of any
4268 OST thread and, hence, there is no per-thread preallocated
4269 pages and buffers to use. Solution: hijack pga pages for IO. As
4270 a byproduct, this avoids unnecessary data copying.
4274 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4275 Description: Client can oops in mdc_commit_close() after open replay
4276 Details : It was possible for the MDS to return an open request with no
4277 transaction number in mds_finish_transno() if the client was
4278 evicted, but without actually returning an error. Clients
4279 would later try to replay that open and may trip an assertion
4280 Simplify the client close codepath, and always return an error
4281 from the MDS in case the open is not successful.
4284 Frequency : rare, 2.6 OSTs only
4286 Description: OST may deadlock under high load on fragmented files
4287 Details : If there was a heavy load and highly-fragmented OST filesystems
4288 it was possible to have all the OST threads deadlock waiting on
4289 allocation of biovecs, because the biovecs were not released
4290 until the entire RPC IO was completed. Instead, release biovecs
4291 as soon as they are complete to ensure forward IO progress.
4293 Severity : enhancement
4295 Description: Support for specifying external journal device at mount
4296 Details : If an OST or MDS device is formatted with an external journal
4297 device, this device major/minor is stored in the ext3 superblock
4298 and may not be valid for failover. Allow detecting and
4299 specifying the external journal at mount time.
4304 Description: Mounting an MDS with pending unlinked files may cause oops
4305 Details : target_finish_recovery() calls mds_postrecov() which returned
4306 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4307 considers this an error and immediately begins cleaning up the
4308 lov, just after starting the mds_lov process
4310 Severity : enhancement
4312 Description: Implement 'lfs df' to report actual free space on per-OST basis
4313 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4314 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4315 usage of objects; '-h' to report in human readable format.
4317 ------------------------------------------------------------------------------
4319 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4326 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4327 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4328 at startup during orphan unlinking. Instead, skip buddy update
4329 before setup, as it will be regenerated after recovery is complete.
4334 Description: If an OST is inactive, its locks might reference stale inodes.
4335 Details : lov_change_cbdata() must iterate over all namespaces, even if
4336 they are inactive to clear inode references from the lock.
4338 Severity : enhancement
4339 Frequency : occasional, if non-standard max_dirty_mb used
4341 Description: Client will block write RPCs if not enough grant
4342 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4343 then the client will block writes while waiting for another RPC
4344 to complete instead of consuming its dirty limit. With change
4345 we get improved performance when max_dirty_mb is small.
4347 Severity : enhancement
4348 Bugzilla : 3389, 6253
4349 Description: Add support for supplementary groups on the MDS.
4350 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4351 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4352 lookups for user supplementary groups into a cache.
4356 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4357 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4358 for liblustre O_EXCL is enforced by MDS.
4362 Description: readdir never returns NULL in liblustre.
4363 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4364 dentry in struct dirent.
4368 Frequency : liblustre only, depends on application IO pattern
4369 Description: liblustre clients evicted if not contacting servers
4370 Details : Don't put liblustre clients into the ping_evictor list, so
4371 they will not be evicted by the pinger ever.
4373 Severity : enhancement
4375 Description: Add ability to evict clients by NID from MDS.
4376 Details : By echoing "nid:$NID" string into
4377 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4378 $NID would be instantly evicted from this MDS and from all active
4379 OSTs connected to it.
4383 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4384 Details : lookup_it_finish() used to query file size from OSTs that was not
4389 Description: service threads change working directory to that of init
4390 Details : Starting lustre service threads may pin the working directory
4391 of the parent thread, making that filesystem busy. Threads
4392 now change to the working directory of init to avoid this.
4396 Frequency : during shutdown only
4397 Description: shutdown with a failed MDS or OST can cause unmount to hang
4398 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4403 Frequency : 2.6 only
4404 Description: chmod/chown may include an extra supplementary group
4405 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4406 supplementary group and if none is specified this is used.
4409 Bugzilla : 5479 (6816)
4411 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4412 Details : If lookup is in progress on a file that is unlinked we might try
4413 to revalidate the inode and fail in revalidate after lookup is
4414 complete and ll_file_open() enqueues the open again but
4415 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4418 Frequency : always, if lconf --abort_recovery used
4420 Description: lconf --abort_recovery fails with 'Operation not supported'
4421 Details : lconf was attempting to abort recovery on the MDT device and not
4424 ------------------------------------------------------------------------------
4426 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4431 Frequency : rare (only unsupported configurations with a node running as an
4433 Bugzilla : 6514, 5137
4434 Description: Mounting a Lustre file system on a node running as an OST could
4436 Details : OSTs now preallocates memory needed to write out data at
4437 startup, instead of when needed, to avoid having to
4438 allocate memory in possibly low memory situations.
4439 Specifically, if the file system is mounted on on OST,
4440 memory pressure could force it to try to write out data,
4441 which it needed to allocate memory to do. Due to the low
4442 memory, it would be unable to do so and the node would
4443 become unresponsive.
4445 Severity : enhancement
4447 Description: Addition of lconf --service command line option
4448 Details : lconf now accepts a '--service <arg>' option, which is
4449 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4451 Severity : enhancement
4453 Description: Failover mode is now the default for OSTs.
4454 Details : By default, OSTs will now run in failover mode. To return to
4455 the old behaviour, add '--failout' to the lmc line for OSTs.
4457 Severity : enhancement
4459 Description: Health checks are now provided for MDS and OSTs
4460 Details : Additional detailed health check information on MSD and OSTs
4461 is now provided through the procfs health_check value.
4464 Frequency : occasional, depends on IO load
4466 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4467 after numerous create/delete cycles
4468 Details : The ext3 inode allocation policy would not allocate new inodes
4469 very well on the OSTs because there are no new directories
4470 being created. Instead we look for groups with free space if
4471 the parent directories are nearly full.
4476 Description: Network or server problems during mount may cause partially
4477 mounted clients instead of returning an error.
4478 Details : The config llog parsing code may overwrite the error return
4479 code during mount error handling, returning success instead
4485 Description: MDS can fail to allocate large reply buffers
4486 Details : After long uptimes the MDS can fail to allocate large reply
4487 buffers (e.g. zconf client mount config records) due to memory
4488 fragmentation or consumption by the buffer cache. Preallocate
4489 some large reply buffers so that these replies can be sent even
4490 under memory pressure.
4494 Frequency : rare (liblustre)
4495 Description: fsx running with liblustre complained that using truncate() to
4496 extend the file doesn't work. This patch corrects that issue.
4497 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4498 ATTR_SIZE and lsm use in llu_setattr_raw.
4502 Frequency : rare, only 2.6 kernels
4503 Description: Unusual file access patterns on the MDS may result in inode
4504 data being lost in very rare circumstances.
4505 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4506 code in the 2.6 kernel caused the inode and/or EA data not to
4507 be read from disk, causing single-file corruption.
4511 Frequency : rare, only 2.6 filesystems using extents
4512 Description: Heavy concurrent write and delete load may cause data corruption.
4513 Details : It was possible under high-load situations to have an extent
4514 metadata block in the block device cache from a just-unlinked
4515 file overwrite a newly-allocated data block. We now unmap any
4516 metadata buffers that alias just-allocated data blocks.
4520 Frequency : filesystems with default stripe_count larger than 77
4521 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4522 Details : lconf specifies an inode size of 4096 bytes when the default
4523 stripe_count is larger than 77. This conflicts with the default
4524 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4525 this case to avoid pinning too much memory for large EAs.
4527 ------------------------------------------------------------------------------
4529 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4534 Frequency : rare (extremely heavy IO load with hundreds of clients)
4536 Description: Client is evicted, gets IO error writing to file
4537 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4538 caused clients to be evicted for AST timeouts. The fixes in
4539 bug 5192 mean we no longer need to have such short AST timeouts
4540 so ldlm_timeout has been increased.
4543 Frequency : occasional during --force or --failover shutdown under load
4544 Bugzilla : 5949, 4834
4545 Description: Server oops/LBUG if stopped with --force or --failover under load
4546 Details : a collection of import/export refcount and cleanup ordering
4547 issues fixed for safer force cleanup
4550 Frequency : only filesystems larger than 120 OSTs
4551 Bugzilla : 5990, 6223
4552 Description: lfs getstripe would oops on a very large filesystem
4553 Details : lov_getconfig used kfree on vmalloc'd memory
4556 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4557 Bugzilla : 6242, 6243
4558 Description: reading from files that had been truncated to a non-zero size
4559 but never opened returned no data
4560 Details : ll_file_read() reads zeros from no-object files to EOF
4565 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4566 Details : There's little harm in aborting MDS/OSS recovery and letting it
4567 try again, so I removed the LASSERT and return an error instead.
4569 Severity : enhancement
4571 Description: New debugging infrastructure for tracking down data corruption
4572 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4573 (b) cover more of the client-side code path, and (c) try to narrow
4574 down where problems occurred
4578 Bugzilla : 3819, 4364, 4397, 6313
4579 Description: Racing close and eviction MDS could cause assertion in mds_close
4580 Details : It was possible to get multiple mfd references during close and
4581 client eviction, leading to one thread referencing a freed mfd.
4583 Severity: : enhancement
4584 Bugzilla : 3262, 6359
4585 Description: Attempts to reconnect to servers are now more aggressive.
4586 Details : This builds on the enhanced upcall-less recovery that was added
4587 in 1.4.2. When trying to reconnect to servers, clients will
4588 now try each server in the failover group every 10 seconds. By
4589 default, clients would previously try one server every 25 seconds.
4594 Description: After recovery, certain operations trigger a failed
4595 assertion on a client.
4596 Details : Failing over an mds, using lconf -d --failover, while a
4597 client was doing a readdir() call would cause the client to
4598 LBUG after recovery completed and the readdir() was resent.
4600 Severity : enhancement
4602 Description: Default groups are now added by lconf
4603 Details : You can now run lconf --group <servicename> without having to
4604 manually add groups with lmc.
4607 Frequency : occasional
4609 Description: Nodes with an elan id of 0 trigger a failed assertion
4612 Frequency : always when accessing e.g. tty/console device nodes
4614 Description: tty and some other devices nodes cannot be used on lustre
4615 Details : file's private_data field is used by device data and lustre
4616 values in there got lost. New field was added to struct file to
4617 store fs-specific private data.
4620 Frequency : when exporting Lustre via NFS
4622 Description: NFSD failed occasionally when looking up a path component
4623 Details : NFSD is looking up ".." which was broken in ext3 directories
4624 that had grown large enough to become hashed.
4627 Frequency : Clusters with multiple interfaces not on the same subnet
4629 Description: Nodes will repeatedly try to reconnect to an interface which it
4630 cannot reach and report an error to the log.
4631 Details : Extra peer list entries will be created by lconf with some peers
4632 unreachable. lconf now validates the peer before adding it.
4635 Frequency : Only if a default stripe is set on the filesystem root.
4637 Description: Setting a default stripe on the filesystem root prevented the
4638 filesystem from being remounted.
4639 Details : The client was sending extra request flags in the root getattr
4640 request and did not allocate a reply buffer for the dir EA.
4643 Frequency : occasional, higher if lots of files are accessed by one client
4644 Bugzilla : 6159, 6097
4645 Description: Client trips assertion regarding lsm mismatch/magic
4646 Details : While revalidating inodes the VFS looks up inodes with ifind()
4647 and in rare cases can find an inode that is being freed.
4648 The ll_test_inode() code will free the lsm during ifind()
4649 when it finds an existing inode and then the VFS later attaches
4650 this free lsm to a new inode.
4654 Bugzilla : 6422, 7030
4655 Description: MDS deadlock between mkdir and client eviction
4656 Details : Creating a new file via mkdir or mknod (starting a transaction
4657 and getting the ns lock) can deadlock with client eviction
4658 (gets ns lock and trying to finish a synchronous transaction).
4661 Frequency : occasional
4662 Description: While starting a server, the fsfilt_ext3 module could not be
4664 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4665 kernels. Previously, lconf would still use the ext3 name
4666 when trying to load modules. Now, it will correctly use
4667 ext3 on 2.4 and ldiskfs on 2.6.
4669 Severity : enhancement
4670 Description: The default stripe count has been changed to 1
4671 Details : The interpretation of the default stripe count (0, to lfs
4672 or lmc) has been changed to mean striping across a single
4673 OST, rather than all available. For general usage we have
4674 found a stripe count of 1 or 2 works best.
4676 Severity : enhancement
4677 Description: Add support for compiling against Cray portals.
4678 Details : Conditional compiling for some areas that are different
4682 Frequency : occasional
4683 Bugzilla : 6409, 6834
4684 Description: Creating files with an explicit stripe count may lead to
4685 a failed assertion on the MDS
4686 Details : If some OSTs are full or unavailable, creating files may
4687 trigger a failed assertion on the MDS. Now, Lustre will
4688 try to use other servers or return an error to the
4692 Frequency : occasional
4694 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4695 caused lock timeout during readahead (since 1.4.2).
4696 Details : Processes doing readahead might match a lock that hasn't been
4697 granted yet if there are overlapping and conflicting lock
4698 requests. The readahead process waits on ungranted lock
4699 (original lock is CBPENDING), while OST waits for that process
4700 to cancel CBPENDING read lock and eventually evicts client.
4702 Severity : enhancement
4704 Description: Initial enabling of flock support for clients
4705 Details : Implements fcntl advisory locking and file status functions.
4706 This feature is provided as an optional mount flag (default
4707 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4708 locking are implemented yet, and those that are are not guaranteed
4709 to be completely correct in production environments.
4710 mount -t lustre -o [flock|noflock] ...
4713 Frequency : occasional
4715 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4716 assertion in the ext3 JBD (journaling) layer.
4717 Details : The b_committed_data struct is protected by the big kernel lock
4718 in 2.4 kernels, serializing journal_commit_transaction() and
4719 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4720 access to this struct. In 2.6 kernels there is finer grained
4721 locking to improve SMP performance of the JBD layer.
4725 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4730 Description: merge and cleanup kernel patches.
4733 -----------------------------------------------------------------------------
4735 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4737 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4738 versions of Lustre. Please update all servers and clients to
4739 version 1.4.2 or later at the same time. You must also run
4740 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4741 to update the configuration logs.
4743 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4744 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4745 - swab llog records of type '0' so we get proper header size/idx (5861)
4746 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4747 - fix rename of one directory over another leaking an inode (5953)
4748 - avoid SetPageDirty on 2.6 (5981)
4749 - don't re-add just-being-destroyed locks to the waiting list (5653)
4750 - when creating new directories, inherit the parent's custom
4751 striping settings if present parent (3048)
4752 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4753 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4754 - don't hold client locks on temporary worklist from l_lru (5666)
4755 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4756 - automatically evict dead clients (5921)
4757 - Update file size properly in create+truncate+fstat case (6196)
4758 - Do not unhash mountpoint dentries, do not allow removal of
4760 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4761 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4762 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4763 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4764 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4765 - Correctly handle reads of files with no objects (6243)
4766 - lctl recover will also mark a device active if deactivate used (5933)
4768 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4769 - allow --write-conf on an MDS with different nettype than client (5619)
4770 - don't write config llogs to MDS for mounts not from that MDS (5617)
4771 - lconf should create multiple TCP connections from a client (5201)
4772 - init scripts are now turned off by default; run chkconfig --on
4773 lustre and chkconfig --on lustrefs to use them
4774 - upcalls are no longer needed for clients to recover to failover
4776 - add --abort-recovery option to lconf to abort recovery on device
4778 - add support for an arbitrary number of OSTs (3026)
4779 - Quota support protocol changes.
4780 - forward compatibility changes to wire structs (6007)
4781 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4782 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4783 - improved "lustre status" script
4784 - initialize blocksize for non-regular files (6062)
4785 - added --disable-server and --disable-client configure options (5782)
4786 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4787 - Vanilla 2.4.29 support
4788 - increase maximum number of obd devices to 520 (6242)
4789 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4790 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4791 SLES 9 2.6.5 kernels
4792 - limit stripes per file to 160 (the maximum EA size) (6093)
4794 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4797 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4798 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4799 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4800 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4801 - small ext3 extents cleanups and fixes (5733)
4802 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4803 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4804 - avoid needless client->OST connect, fix handle mismatch (5317)
4805 - fix DLM error path that led to out-of-sync client, long delays (5779)
4806 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4807 - fix several locking issues related to i_size (5492,5624,5654,5672)
4808 - don't move pending lock onto export if it is already evicted (5683)
4809 - fix kernel oops when creating .foo in unlinked directory (5548)
4810 - fix deadlock in obdfilter statistics vs. object create (5811)
4811 - use time_{before,after} to avoid timer jiffies wrap (5882)
4812 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4813 - Do not leak request if server was not able to process it (5154)
4814 - If mds_open unable to find parent dir, make that negative lookup(5154)
4815 - don't create new directories with extent-mapping (5909, 5936)
4817 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4818 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4819 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4820 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4821 - support for RHEL4 kernel on i686 (5773)
4822 - provide error messages when incompatible logs are encountered (5898)
4824 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4825 * version 1.4.0.10 (1.4.1 release candidate 1)
4827 - don't keep a lock reference when lock is not granted (4238)
4828 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4829 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4830 - drop import inflight refcount on signal_completed_replay error (5255)
4831 - unlock page after async write error during send (3677)
4832 - handle missing objects in filter_preprw_read properly (5265)
4833 - no transno return for symlink open, don't save no-trasno open (3440)
4834 - don't try to complete elan receive that already failed (4012)
4835 - free RPC server reply state on error (5406)
4836 - clean up thread from ptlrpc_start_thread() on error (5160)
4837 - readahead could read extra page into cache that wasn't ejected (5388)
4838 - prevent races in class_attach/setup/cleanup/detach (5260)
4839 - don't dereference de->d_inode after l_dput of de (5458)
4840 - use "int" for stripe value returned from lock_to_stripe (5544)
4841 - mballoc allocation and error-checking fixes in 2.6 (5504)
4842 - block device patches to fix I/O request sizes in 2.6 (5482)
4843 - look up hostnames for IB nals (5602)
4844 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4845 - don't start multiple acceptors for the same port (5277)
4846 - fix incorrect LASSERT in mds_getattr_name (5635)
4847 - export a proc file for general "ping" checking (5628)
4848 - fix "lfs check" to not block when the MDS is down (5628)
4850 - service request history (4965)
4851 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4852 - create an "evict_client" /proc entry on OSTs, like the MDS has
4853 - fix mount usage message, return errors per mount(8) (5168)
4854 - change grep [] to grep "[]" in tests so they work in more UMLs
4855 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4856 - remove ancient LOV_MAGIC_V0 EA support (5047)
4857 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4858 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4859 - allow readahead allocations to fail when low on memory (5383)
4860 - mmap locking landed again, after considerable improvement (2828)
4861 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4863 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4866 - send OST transaction number in read/write reply to free req (4966)
4867 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4868 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4869 - the watchdog thread now runs as interruptible (5246)
4870 - flock/lockf fixes (but it's still disabled, pending 5135)
4871 - don't use EXT3 constants in llite code (5094)
4872 - memory shortage at startup could cause assertion (5176)
4874 - reorganization of lov code
4875 - single portals codebase
4877 - add extents/mballoc support (5025)
4878 - direct I/O reads in the obdfilter (4048)
4879 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4881 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4884 - send OST transaction number in read/write reply to free req (4966)
4885 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4886 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4887 - the watchdog thread now runs as interruptible (5246)
4888 - handle missing objects in filter_preprw_read properly (5265)
4889 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4890 - drop import inflight refcount on signal_completed_replay error (5255)
4891 - unlock page after async write error during send (3677)
4892 - return original error code on reconstructed replies (3761)
4893 - no transno return for symlink open, don't save no-trasno open (3440)
4895 - add pid to ldlm debugging output (4922)
4896 - bump the watchdog timeouts -- we can't handle 30sec yet
4897 - extra debugging for orphan dentry/inode bug (5259)
4899 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4902 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4903 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4904 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4905 - fix NULL dereference in /proc/sys/portals/routes (4827)
4906 - allow failed mdc_close() operations to be interrupted (4561)
4907 - stop precreate on OST before MDS would time out on it (4778)
4908 - don't send partial-page writes before EOF from client (4410)
4909 - discard client grant for sub-page writes on large-page clients (4520)
4910 - don't free dentries not owned by NFS code, check generation (4806)
4911 - fix lsm leak if mds_create_objects() fails (4801)
4912 - limit debug_daemon file size, always print CERROR messages (4789)
4913 - use transno after validating reply (3892)
4914 - process timed out requests if import state changes (3754)
4915 - update mtime on OST during writes, return in glimpse (4829)
4916 - add mkfsoptions to LDAP (4679)
4917 - use ->max_readahead method instead of zapping global ra (5039)
4918 - don't interrupt __l_wait_event() during strace
4920 - add software watchdogs to catch hung threads quickly (4941)
4921 - make lustrefs init script start after nfs is mounted
4922 - fix CWARN/ERROR duplication (4930)
4923 - return async write errors to application if possible (2248)
4924 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4925 - print NAL number in %x format (4645)
4926 - update barely-supported suse-2.4.21-171 series (4842)
4927 - support for sles 9 %post scripts
4928 - support for building 2.6 kernel-source packages
4929 - support for sles km_* packages
4931 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4934 - ignore -ENOENT errors in osc_destroy (3639)
4935 - notify osc create thread that OSC is being cleaned up (4600)
4936 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4937 - reconstruct ost_handle() like mds_handle() (4657)
4938 - create a new thread to do import eviction to avoid deadlock (3969)
4939 - let lconf resolve symlinked-to devices (4629)
4940 - don't unlink "objects" from directory with default EA (4554)
4941 - hold socknal file ref over connect in case target is down (4394)
4942 - allow more than 32000 subdirectories in a single directory (3244)
4943 - fix blocks count for O_DIRECT writes (3751)
4944 - OST returns ENOSPC from object create when no space left (4539)
4945 - don't send truncate RPC if file size isn't changing (4410)
4946 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4947 - bind to privileged port in socknal and tcpnal (3689)
4949 - rate limit CERROR/CWARN console message to avoid overload (4519)
4950 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4951 - basic mmap support (3918)
4952 - kernel patch series update from b1_4 (4711)
4954 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4957 - avoid crash during MDS cleanup with OST shut down (2775)
4958 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4959 - don't use bad inodes on the MDS (3744)
4960 - dynamic object preallocation to improve recovery speed (4236)
4961 - don't hold spinlock over lock dumping or change debug flags (4401)
4962 - don't zero obd_dev when it is force cleaned (3651)
4963 - print grants to console if they go negative (4431)
4964 - "lctl deactivate" will stop automatic recovery attempts (3406)
4965 - look for existing locks in ldlm_handle_enqueue() (3764)
4966 - don't resolve lock handle twice in recovery avoiding race (4401)
4967 - revalidate should check working dir is a directory (4134)
4969 - don't always mark "slow" obdfilter messages as errors (4418)
4971 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4974 - don't close LustreDB during write_conf until it is done (3860)
4975 - fix typo in lconf for_each_profile (3821)
4976 - allow dumping logs from multiple threads at one time (3820)
4977 - don't allow multiple threads in OSC recovery (3812)
4978 - fix debug_size parameters (3864)
4979 - fix mds_postrecov to initialize import for llog ctxt (3121)
4980 - replace config semaphore with spinlock (3306)
4981 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4982 - don't allow enqueue to complete on a destroyed export (3822)
4983 - down write_lock before checking llog header bitmap (3825)
4984 - recover from lock replay timeout (3764)
4985 - up llog sem before sending rpc (3652)
4986 - reduce ns lock hold times when setting kms (3267)
4987 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4988 - fix NULL deref and obd_dev leak on setup error (3312)
4989 - replace some LBUG about llog ops with error handling (3841)
4990 - don't match INVALID dentries from d_lookup and spin (3784)
4991 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4992 - fix invalid assertion in ptlrpc_set_wait (3880)
4994 - add libwrap support for the TCP acceptor (3996)
4995 - add /proc/sys/portals/routes for non-root route listing (3994)
4996 - allow setting MDS UUID in .xml (2580)
4997 - print the stack of a process that LBUGs (4228)
4999 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5002 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5003 - make sure to unset replay flag from failed open requests (3440)
5004 - if default stripe count is 0, use OST count for inode size (3636)
5005 - update parent mtime/ctime on client for create/unlink (2611)
5006 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5007 - free recovery state on server during a forced cleanup (3571)
5008 - unregister_reply for resent reqs (3063)
5009 - loop back devices mounting and status check on 2.6 (3563)
5010 - fix resource-creation race that can provoke i_size == 0 (3513)
5011 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5012 - more debugging for page-accounting assertion (3746)
5013 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5014 - avoid deadlock after precreation failure (3758)
5015 - fix race and lock order deadlock in orphan handling (3450, 3750)
5016 - add validity checks when grabbing inodes from l_ast_data (3599)
5018 - add /proc/.../recovery_status to obdfilter (3428)
5019 - lightweight CDEBUG infrastructure, debug daemon (3668)
5020 - change default OSC RPC parameters to be better on small clusters
5021 - turn off OST read cache for files smaller than 32MB
5022 - install man pages and include them in rpms (3100)
5023 - add new init script for (un)mounting lustre filesystems (2593)
5024 - run chkconfig in %post for init scripts (3701)
5025 - drop scimac NAL (unmaintained)
5027 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5030 - clean kiobufs before and after use (3485)
5031 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5032 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5033 - change init script to fail more clearly if not run as root (1528)
5034 - allow clients to reconnect during replay (1742)
5035 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5036 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5037 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5038 - fix cond_resched (3554)
5039 - only evict unfinished clients after recovery (3515)
5040 - allow bulk resend, prevent data loss (3570)
5041 - dynamic ptlrpc request buffer allocation (2102)
5042 - don't allow unlinking open directory if it isn't empty (2904)
5043 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5044 - remove extraneous obd dereference causing LASSERT failure (3334)
5045 - don't use get_cycles() when creating temp. files on the mds (3156)
5046 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5047 - handle EEXIST for set-stripe, set proper directory name (3336)
5049 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5050 - fix ksocknal_fmb_callback() error messages (2918)
5052 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5055 - don't copy lvb into (possibly NULL) reply on error (2983)
5056 - don't deref dentry after dput, don't free lvb on error (2922)
5057 - use the kms to determine writeback rpc length (2947)
5058 - increment oti_logcookies when osc is inactive (2948)
5059 - update client's i_blocks count via lvb messages (2543)
5060 - handle intent open/close of special files properly (1557)
5061 - mount MDS with errors=remount-ro, like obdfilter (2009)
5062 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5063 - don't use cancelling-locks' kms values (2947)
5064 - use highest lock extent for kms, not last one (2925)
5065 - don't dereference ERR_PTR() dentry in error handling path (3107)
5066 - fix thread race in portals_debug_dumplog() (3122)
5067 - create lprocfs device entries at setup instead of at attach (1519)
5068 - common AST error handler, don't evict client on completion race (3145)
5069 - zero nameidata in detach_mnt in 2.6 (3118)
5070 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5071 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5072 - initialize RPC timeout timer earlier for 2.6 (3219)
5073 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5074 - print nal/nid for unknown nid (3258)
5075 - additional checks for oscc recovery before doing precreate (3284)
5076 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5077 - don't crash in mdc_close for bad permissions on open (3285)
5078 - zero i_rdev for non-device files (3147)
5079 - clear page->private before handing to FS, better assertion (3119)
5080 - tune the read pipeline (3236)
5081 - fix incorrect decref of invalidated dentry (2350)
5082 - provide read-ahead stats and refine rpc in flight stats (3328)
5083 - don't hold journal transaction open across create RPC (3313)
5084 - update atime on MDS at close time (3265)
5085 - close LDAP connection when recovering to avoid server load (3315)
5086 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5087 - don't leak open file on MDS after open resend (3325)
5088 - serialize filter_precreate and filter_destroy_precreated (3329)
5089 - loop device shouldn't call sync_dev() for nul device (3092)
5090 - clear page cache after eviction (2766)
5091 - resynchronize MDS->OST in background (2824)
5092 - refuse to mount the same filesystem twice on same mountpoint (3394)
5093 - allow llmount to create routes for mounting behind routers (3320)
5094 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5095 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5096 - fix rare problem with rename on htree directories (3417)
5098 - allow default OST striping configuration per directory (1414)
5099 - fix compilation for qswnal for 2.6 kernels (3125)
5100 - increase maximum number of MDS request buffers for large systems
5101 - change liblustreapi to be useful for external progs like lfsck (3098)
5102 - increase local configuration timeout for slow disks (3353)
5103 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5105 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5108 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5109 - don't overwrite extent policy data in reply if lock was blocked (2901)
5110 - drop filter export grants atomically with removal from device (2663)
5111 - del obd_self_export from work_list in class_disconnect_exports (2908)
5112 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5113 - swab reply message in mdc_close, other PPC fixes (2464)
5114 - fix destroying of named logs (2325)
5115 - overwrite old logs when running lconf --write_conf (2264)
5116 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5117 - fix race in target_handle_connect (2898)
5118 - mds_reint_create() should take same inode create lock (2926)
5119 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5120 - don't close files for self_export to avoid uninitialized obd (2936)
5121 - allow MDS with the same name as client node (2939)
5122 - hold dentry reference for closed log files for unlink (2325)
5123 - reserve space for all logs during transactions (2059)
5124 - don't evict page beyond end of stripe extent (2925)
5125 - don't oops on a deleted current working directory (2399)
5126 - handle hard links to targets without a parent properly (2517)
5127 - don't dereference NULL lock when racing during eviction (2867)
5128 - don't grow lock extents when lots of conflicting locks (2919)
5130 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5133 - account for cache space usage on clients to avoid data loss (974)
5134 - lfsck support in lustre kernel code (2349)
5135 - reduce journal credits needed for BRW writes (2370)
5136 - orphan handling to avoid losing space on client/server crashes
5137 - ptlrpcd can be blocked, stopping ALL progress (2477)
5138 - use lock value blocks to assist in proper KMS, faster stat (1021)
5139 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5140 - recovery for initial connections (2355)
5141 - fixes for mds_cleanup_orphans (1934)
5142 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5143 - block all file creations until orphan recovery completes (1901)
5144 - client remove rq_connection from request struct (2423)
5145 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5146 - recovery timer race (2670)
5147 - mdc_close recovey bug (2532)
5148 - ptlrpc cleanup bug (2710)
5149 - mds timeout on local locks (2588)
5150 - namespace lock held during RPCs (2431)
5151 - handle interrupted sync write properly (2503)
5152 - don't try to handle a message that hasn't been replied to (2699)
5153 - client assert failure during cleanup after abort recovery (2701)
5154 - leak mdc device after failed mount (2712)
5155 - ptlrpc_check_set allows timedout requests to complete (2714)
5156 - wait for inflight reqs when ptlrpcd finishes (2710)
5157 - make sure unregistered services are removed from the srv_list
5158 - reset bulk XID's when resending them (caught by 1138 test)
5159 - unregister_bulk after timeout
5160 - fix lconf error (2694)
5161 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5162 - readahead locks pages, leaves pending causing memory pressure (2673)
5163 - increase OST request buffers to 4096 on large machines (2729)
5164 - fix up permission of existing directories in simple_mkdir (2661)
5165 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5166 - don't assign transno to errored transactions (2742)
5167 - don't delete objects on OST if given a bogus objid from MDS (2751)
5168 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5169 - if rq_no_resend, then timeout request after recovery (2432)
5170 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5171 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5172 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5173 - MDS recovery shouldn't delete objects that it has given out (2730)
5174 - if enqueue arrives after completion, don't clobber LVB (2819)
5175 - don't unlock pages twice when trigger_group_io returns error (2814)
5176 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5177 - don't write pages to disk if there was an error (1450)
5178 - don't ping imports that have recovery disabled (2676)
5179 - take buffered bytes into account when balancing socknal conn (2817)
5180 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5181 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5182 - remove little-endian swabbing of llog records (1987)
5183 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5184 - retry reposting request buffers if they fail (1191)
5185 - grow extent at grant time to avoid granting a revoked lock (2809)
5186 - lock revoke doesn't evict page if covered by a second lock (2765)
5187 - disable VM readahead to avoid reading outside lock extents (2805)
5189 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5190 - updated kernel patches for hp-2.4.20 kernel (2681)
5192 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5195 - fix truncated write corruption (2366)
5196 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5198 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5199 - ensure len(uuid) < 37 in lmc (1171)
5200 - fix ia64 OOPS in llog_test (2255)
5201 - zero end of page at obdfilter for partial page writes (2648)
5202 - don't leave stale dentries around after renames (bug 2428)
5203 - fix timeouts when evicting a client with a single lock held (2642)
5204 - set deadline for the initial HELLO message to drain (2634)
5205 - print out dotted-quad IP addresses in the socknal (2302)
5207 - additional debugging for MDS client eviction problem (2443)
5208 - fix mkfsoptions support for osts (2603, 2604)
5210 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5213 - add series for the vanilla 2.6.0 kernel
5214 - add series for the vanilla 2.4.24 kernel
5215 - add series for a cray x86/64 UL kernel drop
5216 - fix xattr patches for the vanilla 2.4.19 series
5218 - generate true UUIDs in lmc (1171)
5219 - have portals stack dumping break in UML (2466)
5220 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5221 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5222 - wake callback waiting threads on client eviction (2460)
5223 - Add --ptldebug and --subsystem to lmc (1719)
5224 - update assertion to allow safe interrupt allocation
5225 - set rq_no_resend for cancel requests (2432)
5226 - recalculate ptlrpcd timeout after resend (2494)
5227 - call vfs_rmdir when removing pending directories (2368)
5228 - fix renaming a file to itself (2429)
5229 - lmc creates a default one-stripe lov (2454)
5230 - expand procfs space to handle large clusters (2326)
5231 - increase UML stack to avoid overflow
5232 - update lconf's list of debug and subsystem masks
5233 - fix lfs find --obd (2510)
5234 - /proc tunable for disabling filter read caching (2591)
5235 - stop rpm packages from altering slapd.conf (2301)
5236 - disable nagle in the socknal under 0conf (2578)
5237 - choose mds inode size based on stripe count (2572)
5238 - fix kernel-source rpm problems (2516)
5240 - add --disable-doc to avoid pdf generation (2421)
5241 - update documentation, tests, type-os, comments
5242 - avoid format warnings on ia64
5243 - remove the TOE NAL
5244 - tiny code cleanups by removing unused fields
5246 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5249 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5250 - remove the most blatant lies from BUILDING file (2371)
5251 - change default debug level to reasonable production setting
5252 - reduce client side cache size to reduce cache flush time
5253 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5254 - make TCP zerocopy and pinger support enabled by default (2476)
5255 - sync writes completed after process exits caused crashes (2319)
5256 - maintain correct mount count on the MDS (2356)
5257 - backout 1557, because 2316 wasn't really fixed
5258 - better file I/O statistics gathering in /proc
5259 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5260 - another kernel patch to fix zero-copy TCP function export
5261 - don't take duplicate lock when processing re-sent getattr (2420)
5262 - lctl uses obd_self_export instead of creating new conn (2353)
5263 - MDS/OST recovery case which requires object creation asserted (2425)
5264 - move lfs from /usr/sbin to /usr/bin in packages
5265 - fix race between mds_client_add and mds_client_free (2417)
5266 - use kmalloc instead of slabs in portals (2430)
5267 - don't create duplicate records when a failover MDS is present (2442)
5268 - remove unnecessary mount age check (2332)
5269 - don't remove directory inodes from locks prematurely (2451)
5270 - don't break if MDS service name is the same as hostname (2103)
5271 - fix races in client write RPC generation when cache full (2482)
5273 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5276 - remove now-unused request->rq_obd (278)
5277 - if an allocation fails, print out how much memory we've used (1933)
5278 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5279 - add the "configurable stack size" patch to most series files (1256)
5280 - ability to write large log records, for 100+ OST configs (2306)
5281 - fix NULL deref when filter_prep fails (2314)
5282 - fix operator precedence error in filter_sync
5283 - dynamic allocation of socknal TX descriptors (2315)
5284 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5285 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5286 - ensure that CWARN messages in Portals always get to the syslog
5287 - __init/__exit are not for prototype decls (ldlm_init/exit)
5288 - x86-64 compile warning fixes
5289 - fix gateway LMC keyword conflict (2318)
5290 - fix MDS lock inversions in getattr/reint paths (1844)
5291 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5292 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5293 - DLM race condition prevented some lock evictions (2328)
5294 - ENOMEM detection and retry on socknal sends (2230)
5295 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5296 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5297 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5298 - better error messages when a client is rejected during recovery (1505)
5299 - avoid cancelling locks which were never granted, after failure (2330)
5300 - fix i_sem/journal inversion in mds_client_add (2333)
5301 - fix truncate/getattr lock cycle deadlock (2334)
5302 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5303 - fix two rare exit paths which could leak an l_lock() ref (2321)
5304 - fencepost error in MDS/OST orphan recovery (2226)
5305 - make log record alignment 8 bytes (1988)
5306 - lstripe now fails when requested offset > ost_count (2237)
5307 - ensure that all kernel series have a complete list.h (1607)
5308 - fix crashes in special-file operations (2316)
5309 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5311 - allow configurable automake binary, for testing new versions
5312 - small update to the lfs documentation
5314 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5316 * fix negative export reference count in fsfilt_sync (2312)
5318 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5319 * release candidate 0.9.1
5321 - orphans are moved into the PENDING directory for possible recovery
5322 - replayed opens now open by fid for orphan/rename safety (1042)
5323 - last close of an orphan inode generates a transno (683)
5324 - chdir() and mount() now pin the directory entry (1020)
5325 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5326 - discard very old requests without processing them (1502)
5327 - remove some common, well-understood CERRORs (1505)
5328 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5329 - clear "grant" flags in OST replies until OST grant code lands (1644)
5330 - fix read performance by not clobbering i_blksize on client (1598)
5331 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5332 - make LRU size a /proc tunable, clears locks when reduced (707)
5333 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5334 - server locks take references on exports now (1558)
5335 - build fixes for 2.4.20-rh trees (1663)
5336 - return an error from lov_create if all OSCs are inactive (1751)
5337 - fix import levels when a reconnect happens without a timeout (1597)
5338 - exit early from mds_open if we get a lookup error (1749)
5339 - partial page read at EOF wouldn't wait for disk before sending (1642)
5340 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5341 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5342 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5343 - write/truncate lock inversion (1639)
5344 - Don't auto-load obdclass, portals modules during cleanup (1495)
5345 - fix timestamps from jumping to "now" (1763)
5346 - extra journal assertions (1648)
5347 - add an extra multiunlink test (1771)
5348 - fix read_record/write_record API (1776)
5349 - fix leak of offset_extent, possible incorrect i_size later (1772)
5350 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5351 - Debugging for the kqswnal_get_idle_tx problems (1820)
5352 - Allow recovery to be attempted multiple times (1536)
5353 - Write out MDS last_rcvd file after it is first created (1600)
5354 - Fix tx_descriptor leak in failed transmit situations (1827)
5355 - ext3 journaling fixes for assertion failure after IO error (1871)
5356 - class_export_put() on freed export after completion AST error (1896)
5357 - Fix revalidate looping in VFS (1322)
5358 - Don't access a freed export during MDS_REINT timeout (1521)
5359 - Add open-unlink recovery support on the MDS (1673,1764)
5360 - Return an error if no MDS data was read from last_rcvd (1946)
5361 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5362 - Don't setup a disk device that doesn't match exported UUID (317)
5363 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5364 - avoid committing NULL handle in force close
5365 - local.sh is now a one-stripe LOV configuration
5366 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5367 - fix ext3 htree duplicate directory entry corruption (1516)
5368 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5369 - update to new LOV EA format (2097)
5370 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5372 2003-06-15 Phil Schwan <phil@clusterfs.com>
5375 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5376 - per-import/export recovery handling (958, 931, 959)
5377 - multiple last-rcvd slots, for serving multiple FSes (949)
5378 - connections are again shared between multiple imp/exports (963, 964)
5379 - "umount -f" would hang if any requests needed to be sent (393, 978)
5380 - avoid pinning large req buffer by copying for queued messages (989)
5381 - add "uuid" to "lctl device" command to help upcalls (991)
5382 - "open" RPCs with transnos would confuse recovery counters (1037)
5383 - do proper endian conversion of all wire messages (288, 340, 891)
5384 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5385 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5386 - fix ptlrpc_connection leak in target_handle_connect (1174)
5387 - fix import refcounting bug in OST and MDS cleanup (1134)
5388 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5389 - fix very unlikely obd_types race condition (501)
5390 - remove osc_open hack for echo_client (1187)
5391 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5392 - a failure in read_inode2 leads to deadlock (1139)
5393 - cancel ack-locks as soon as transaction is committed (1072)
5394 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5395 - make sure to commitrw after any preprw to avoid deadlock (1162)
5396 - failing to execute a file in a lustre FS would lock inode (1203)
5397 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5398 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5399 - obd_uuid2tgt would walk off the end of the list (1255)
5400 - on IA64 the osc would give portals incorrect bulk size (1258)
5401 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5402 - fix lock inversion caused by new llite matching code (1282)
5403 - limit the number of dirty pages on a client to 10MB (1286)
5404 - timed out locks were not being corrected cancelled (1289)
5405 - fix O_DIRECT above 4GB on IA-32 (1292)
5406 * major user-visible changes
5407 - fail out/fail over policy now controlled by the upcall (993)
5409 - add OBD_PING to check server availability and failure (954)
5410 - lustre messages are now sent in sending host order (288, 340, 891)
5411 - add eadatalen to MDS getattr reply (340)
5412 - OST read replies may contain second buffer, with per-page status (593)
5414 2003-03-11 Phil Schwan <phil@clusterfs.com>
5417 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5418 - failed open()s could cause deadlock; fixed (867, 869)
5419 - stop cancelling OST locks when files are closed (481)
5420 - overlapping XID spaces caused network corruption (851, 853)
5421 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5422 - setattr_raw added, to do single-RPC, server-side setattrs
5423 - lmc/lconf syntax change for OST UUIDs
5424 - fix crashy race condition between ptlrpc_free_req and osc_close
5425 - don't use request in mdc_enqueue if we hit a timeout (889)
5426 - don't set the inode i_size for regular files from the MDS (896)
5427 - handle out of order completion AST (842)
5428 - don't LBUG if a lock request times out after receiving AST (913)
5429 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5430 - if a bad lock AST arrives, send an error instead of dropping entirely
5431 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5432 - fix leak in bulk IO when only partially completed (899, 900, 926)
5433 - fix O_DIRECT for ia64 (55)
5434 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5435 - C-z after timeout could hang a process forever; fixed (977)
5437 - client-side I/O cache (678, 924, 929, 941, 970)
5439 - READPAGE and SETATTRs which don't take server-side locks get
5442 2003-02-11 Phil Schwan <phil@clusterfs.com>
5445 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5446 - Fix fsfilt_extN_readpage() to read a full page of directory
5447 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5448 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5449 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5450 - Most intent operations were removed, in favour of a new RPC mode
5451 that does a single RPC to the server and bypasses most of the VFS
5452 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5453 - Aggressively cancel local locks on DLM servers
5454 - mds_reint_unlink sends EA to the client if it's the last nlink.
5455 client uses that EA to unlink OST objects.
5456 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5457 - recursive symlinks were fixed (439)
5458 - fixed NULL deref in DEBUG_REQ
5459 - filter_update_lastobjid no longer calls sync, which annoyed extN
5460 - fixed multi-client small-writes to a single file problem (445)
5461 - fixed mtime updates during file writes (607)
5462 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5463 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5464 - fixed Linux OST size reporting problem (444, 656)
5465 - OST now updates object mtime with writes or setattr (607, 619)
5466 - client verifies file size before zeroing page past EOF (445)
5467 - OST now writes last allocated objid to disk with allocation (108)
5468 - LOV on echo now works (409)
5470 - mds_reint_unlink sends a new buffer, with the EA included. this
5471 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5472 set if a regular file was being unlinked, and it was the last link
5473 - use PtlGet from the target for bulk writes (315)
5474 - OST now updates object mtime with writes or setattr (607, 619)
5475 - LDLM now has a grant-time callback to revalidate locked items, if
5477 - Many MDS operations were reorganized to combat race conditions
5479 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5480 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5482 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5485 - Fully reactivate OST imports after reconnection (512, others)
5486 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5487 - More graceful error handling for truncating on dead OST (515)
5488 - Don't error out unless we're actually accessing dead stripes (474)
5489 - Fix garbage sizes when stripes are missing (410)
5490 - LRU counters were broken, causing constant lock purge (433, 432)
5491 - garbage on read from stripes with failed OSTs (441)
5492 - mark OSCs as active before reconnecting during recovery (438)
5493 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5494 - lfind did not preserve OST order in output (443)
5495 - symlinks cause hung clients, incorrect data (439)
5496 - stop dereferencing request after dropping refcount (457)
5497 - don't LASSERT(spin_is_locked) on non-SMP (455)
5498 - fixes for many rename() bugs
5499 - fstat didn't correctly synchronize attributes (399)
5500 - server must handle lock cancellation during blocking AST prep (487)
5501 - bulk descriptors were free()d too soon (511)
5502 - fix paths in lconf, which would load incorrect modules (451, 507)
5503 - fix confusing lconf 'host not found' error message (386)
5504 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5505 - fix race condition in mdc_blocking_ast() for inode access (526)
5506 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5507 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5508 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5509 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5510 - fix lstripe in several ways (kernel side) (527)
5511 - fix request leak in ldlm_cli_enqueue (262)
5512 - incorrect OSC was marked inactive after OST failure
5513 - call mds_fs_cleanup before unmounting filesystem (524)
5514 - fix races between taking ns_lock and ldlm_lock_change_resource
5515 - fix races updating LOV export open file list
5516 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5517 - fix recovery NULL deref in ldlm_cli_cancel_unused
5518 - fix some DLM races by using new hash table for lock handles (419)
5519 - permit the client to specify desired inodes, at replay
5520 - duplicate requests when we queue them for replay reintegration
5521 - fix last_rcvd offset calculation
5522 - sync after each recovered transaction, so we always make progress
5523 - never, not always, ERESTART requests without transnos
5524 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5526 - skip replay if the MDS says that the client is already connected
5527 - don't check for a recovery-enabled export to match lctl's UUID
5528 - don't INC_USE_COUNT for phantom exports
5529 - don't crash when cleaning up phantom exports (567)
5530 - don't double-finish or set replay data for errored mdc_open requests
5531 - abort requests when they time out, so we don't get old replies
5532 - send/receive replies for AST messages again
5533 - if the client says that it doesn't have the lock, cancel it on the
5535 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5536 mark it as destroyed, it will all work out when decref is called
5537 - fix module use counts (22, 581)
5539 - ASTs now expect a reply (server cancels lock on error reply)
5541 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5544 - fix many simultaneous client startup (392)
5545 - fix dentry->d_it clobbering
5546 - credentials weren't being shipped for readdir/getattr operations
5547 - remove invalid assertions triggered during some concurrent MD
5549 - proper Lustre versions added (336, 389)
5550 - fix memory leak for create error case (398)
5551 - fix LOV locking bug that would get cli/srv out of sync
5552 - fix echo client over LOV (409)
5553 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5554 - fix double-O_EXCL intent crash (424)
5555 - avoid sending multiple lock CANCELs (352)
5557 - MDS can do multi-client recovery (modulo bugs in new code)
5559 - many updates, edits, cleanups
5561 2002-11-18 Phil Schwan <phil@clusterfs.com>
5564 - fix null d_it dereference (346)
5565 - fix full OST/dbench hang (333)
5566 - fix permission problem with file removal (286)
5567 - fix removal of OSCs from LOV when they fail
5568 - fix NULL deref during bulk timeout (214)
5569 - fix problems related to multiple filesystems on one MDS (241)
5570 - fixed serious subtle metadata locking bugs
5571 - free locks on clients when inodes are removed due to memory
5573 - fix inode pointer in lock data (285)
5574 - partial support for multiple MDS on a single host (241)
5575 - data locks weren't cancelled at clear_inode time (290, 311)
5576 - intent locks could lead to unbounded lock growth (205)
5577 - added a maximum lock count, an LRU list, and a flusher
5578 - fix multiple rename (365)
5579 - properly abstracted the echo client
5580 - OSC locked 1 byte too many; fixed
5581 - rewrote brw callback code:
5582 - fixed recovery bugs related to LOVs (306)
5583 - fixed too-many-pages-in-one-write crash (191)
5584 - fixed (again) crash in sync_io_timeout (214)
5585 - probably fixed callback-related race (385)
5587 - Add capability to MDS protocol
5588 - LDLM cancellations and callbacks on different portals
5590 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5593 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5594 - defer open object destruction to close time (601981, 138)
5595 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5596 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5597 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5598 - don't call obd_setattr in ll_file_release for destroyed objects
5599 * protocol change to lustre_msg: move |version| and add |flags|
5600 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5601 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5602 * added replay of create, unlink, link and rename operations during
5603 MDS failover; recovery should be much more robust now
5604 * remove failed OSCs from LOVs (only lov_create uses this so far)
5605 * the lustre-HOWTO was brought (more) up to date (582544)
5607 2002-10-23 Phil Schwan <phil@clusterfs.com>
5610 - in-use dentries weren't being reused properly (617851)
5611 - prevent multiple LDLM setup (599178)
5612 - fix LOV size calculations for truncate (617853)
5613 - fix client handling of MDS intent errors (POSIX)
5614 - fix permission bug in lovstripe.c test (624321)
5615 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5616 - truncate past end of file could corrupt data
5617 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5618 - a race in recovery could return ETIMEDOUT to apps (623947)
5619 - building outside the source directory was fixed
5620 * the lustre-HOWTO was brought (more) up to date (582544)
5621 * major progress was made on recovery functionality
5623 2002-10-10 Phil Schwan <phil@clusterfs.com>
5626 - recovery deadlock fix
5627 - rm -rf causes LBUG fix (617817)
5628 - file open by multiple tasks fix (618962)
5629 - directory permissions bugs (602707 and 620007)
5630 - journal_stop fixed with locking (611313)
5631 - O_APPEND failures resolved (618273, perhaps 614459)
5632 - lconf PATH fix (619770)
5633 - IA64 build fix (621450)
5634 - RPC buffer sizes scale with amount of memory
5636 2002-10-01 Phil Schwan <phil@clusterfs.com>
5639 - locks would be cancelled without throwing away data pages,
5640 resulting in inconsistent data (605627)
5641 - inode attributes were not always being refreshed (605627, 612449)
5642 - lconf now continues to cleanup after lctl reports an error
5643 - MDS now enforces user permissions (602707)
5644 - lprocfs cleanup fixed, but not yet enabled (614157)
5645 - fixed infinite server hang, should a client not respond to an AST
5646 - avoid going into recovery if user calls readlink() with a buffer
5647 that's too small (613941)
5648 - AST RPCs no longer require replies (614867) -- this may be changed
5649 - don't crash server if client sends an IOV that's too big (611336)
5650 - fixed lock conversion deadlock (611892)
5651 - fixed the following of symlinks (614622)
5652 * recovery: the server can remove locks from a client that dies, other
5653 clients can make progress
5654 * more extN patch fixes
5655 * compile-time configurable ptlrpc buffer allocations
5657 - collaborative read cache document
5658 - Lustre Lite Performance CDR document-in-progress
5660 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5663 - fix typo in patch-2.4.18
5665 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5668 - clear ptlrpc request each time in handle_incoming_request()
5669 - unlink of files now destroys the object on the OST
5671 2002-09-19 Peter Braam <braam@clusterfs.com>
5673 * add hard link support
5674 * change obdfile creation method
5675 * kernel patch changed
5677 2002-09-19 Peter Braam <braam@clusterfs.com>
5680 - stack overflow bug in extN fixed
5682 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5684 * documentation updates
5685 - add man pages for config tools
5686 - update tests/README to describe testing with new config tools
5687 - finish metadata API descriptions
5688 * bug fixes and cleanups
5689 - statfs workaround for 16TB limit
5690 - LOV stripe allocation improved, can stripe on subset of OSTs
5691 - LOV file size/IO offset was wrong for files > 4GB in size
5692 - object EA data was being dropped, caused files to be unreadable
5693 - memory overflow with non-LOV OST caused memory corruption
5694 - fixed regression tests to work with new config tools, obdfilter
5695 - fixed bug when directory size became larger than 1 block
5696 - fixed bug (for single client case) when PWD was deleted
5697 - invalidate local directory pages when doing intent-based ops
5698 - avoid LDLM oops when lock callback contained bad data
5700 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5702 * documentation updates
5703 * bug fixes and cleanups
5704 - configuration tools
5707 - 64-bit compile warnings
5708 - 64-bit internal statfs data
5710 * test_brw on persistent OST devices
5712 * lprocfs (disabled)
5714 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5716 * documentation updates
5717 * bug fixes and cleanups
5718 * configuration tools
5720 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5723 * many small fixes to 0_5_4
5724 * io/network handling
5725 * thinkos in MDS operations
5727 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5730 * crucial basic fixes to 0.5.3
5731 * IOR, Iozone work over Elan
5734 2002-08-07 Phil Schwan <phil@clusterfs.com>
5735 * version 0_5_3, our first alpha
5736 * we use the new Portals iovs
5737 * documentation updates
5738 * bug fixes and cleanups
5739 * small changes in the DLM wire protocol
5741 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5742 * version 0_5_1 with some initial stability,
5743 * locking on MD and file I/O.
5744 * documentation updates
5745 * several bug fixes since 0.5.0
5746 * small changes in wire protocol
5748 2002-07-18 Phil Schwan <phil@clusterfs.com>
5750 * delivered as Lustre Light Alpha
5751 * fixed a crash after handling invalid MDS requests
5752 * fixed directory pages for architectures with non-4k pages sizes
5754 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5755 * release version v0_4_4
5756 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5758 * Fixes a number of interruption problems with OST operations.
5759 * Update documentation for portals header changes
5760 * Move all wire protocol structs/defines to lustre_idl.h
5761 * Fixes symlink length bug.
5762 * Add tcpdump to repository.
5764 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5765 * release version v0_4_3
5766 * Fixes statfs for inodes on extN.
5767 * Fixes bug in runtests which would delete /etc/hosts.
5768 * Use 64-bit object IDs wherever possible (not into VFS though)
5769 Remove ost_get_info, which is unused by lustre, and out of date.
5771 2002-07-03 Peter Braam <braam@clusterfs.com>
5772 * release version v0_4_2 Fixes a lookup error (type not passed)
5773 * move forward to head of Portals
5774 * move forward to latest Lustre kernel
5776 2002-06-25 Peter Braam <braam@clusterfs.com>
5777 * release version v0_4_1. Hopefully stable on single node use.