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 Descriptoin: allow kill process which wait statahead result
20 Details : for some reasons 'ls' can stick in waiting result from statahead,
21 in this case need way for kill this process.
23 Severity : enhancement
25 Description: Add state history info file, enhance import info file
26 Details : Track import connection state changes in a new osc/mdc proc file;
27 add overview-type data to the osc/mdc import proc file.
29 Severity : enhancement
31 Description: MDS create should not wait for statfs RPC while holding DLM lock.
33 Severity : enhancement
35 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
40 Descriptoin: OST grant too much space to client even there are not enough space.
41 Details : Client will shrink its grant cache to OST if there are no write
42 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
43 this grant cache if there are already not enough avaible space
44 (left_space < total_clients * 32M).
47 Frequency : start MDS on uncleanly shutdowned MDS device
49 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
50 Details : stay in waiting mds<>ost recovery finished produce random bugs
51 due race between two ll_sync thread for one lov target. send
52 ACTIVATE event only if connect realy finished and import have
56 Frequency : rare, connect and disconnect target at same time
58 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
59 Details : don't call obd_disconnect under lov_lock. this long time
60 operation and can block ptlrpcd which answer to connect request.
65 Descriptoin: don't lose wakeup for imp_recovery_waitq
66 Details : recover_import_no_retry or invalidate_import and import_close can
67 both sleep on imp_recovery_waitq, but we was send only one wakeup
71 Frequency : always with long access acl
73 Descriptoin: mds can't pack reply with long acl.
74 Details : mds don't control size of acl but they limited by reint/getattr
77 Severity : enhancement
79 Description: Update to SLES10 kernel-2.6.16.60-0.33.
81 Severity : enhancement
83 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
86 Frequency : start MDS on uncleanly shutdowned MDS device
88 Descriptoin: aborting recovery hang on MDS
89 Details : don't throttle destroy RPCs for the MDT.
92 Frequency : on remount
94 Description: external journal device not working after the remount
95 Details : clear dev_rdonly flag for external journal devices in
101 Description: shutdown vs evict race
102 Details : client_disconnect_export vs connect request race.
103 if client will evicted at this time - we start invalidate
104 thread without referece to import and import can be freed
108 Frequency : rare, need acl's on inode.
110 Description: client can't handle ost additional correctly
111 Details : if ost was added after client connected to mds client can have
112 hit lnet_try_match_md ... to big messages to wide striped files.
113 in this case need teach client to handle config events about add
114 lov target and update client max ea size at that event.
116 Severity : enhancement
118 Description: Changelogs
119 Details : Changelogs are a lightweight mechanism to track filesystem
120 metadata and namespace changes. The changelog is recorded
121 permanently on the MDTs, and is periodically "consumed" / purged
122 when records are no longer needed.
124 Severity : enhancement
126 Description: compact fld format with extents
127 Details : Store range of seq rather than every seq in FLD. Seq
128 controller update FLD rather than clients. In Case of CMD, mdt0
129 has FLD, all other metadata server act as non persistent proxy
130 for FLD queries and cache fld entries in fld cache.
135 Description: don't skip ost target if they assigned to file
136 Details : Drop slow OSCs if we can, but not for requested start idx.
137 This means "if OSC is slow and it is not the requested
138 start OST, then it can be skipped, otherwise skip it only
139 if it is inactive/recovering/out-of-space.
143 Description: more cleanup in mds_lov
144 Details : not send LOV EA under replay, we can't know about they size at this
145 time. Don't allow client connect to mds before any ost connected,
146 for avoid problems with LOV EA size and returning EIO to client.
148 Severity : enhancement
150 Description: Interoperability at server side (Disk interoperability)
152 Severity : enhancement
154 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
156 Severity : enhancement
158 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
160 Severity : enhancement
162 Description: New client IO stack (CLIO).
164 Severity : enhancement
166 Description: Commit on sharing. Eliminate inter-client dependencies between
167 uncommitted transactions by doing transaction commits.
168 Thereby clients may recovery independently.
171 Frequency : Create a symlink file with a very long name
173 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
174 Details : If there is no extra space in the request for early cancels,
175 ldlm_req_handles_avail() returns 0 instead of a negative value.
177 Severity : enhancement
179 Description: Add /proc entry for import status
180 Details : The mdc, osc, and mgc import directories now have
181 an import directory that contains useful import data for debugging
184 Severity : enhancement
186 Description: Re-disable certain /proc logging
187 Details : Enable and disable client's offset_stats, extents_stats and
188 extents_stats_per_process stats logging on the fly.
191 Frequency : Only on FC kernels 2.6.22+
193 Description: oops in statahead
194 Details : Do not drop reference count for the dentry from VFS when lookup,
195 VFS will do that by itself.
197 Severity : enhancement
199 Description: Generic /proc file permissions
200 Details : Set /Proc file permissions in a more generic way to enable non-
201 root users operate on some /proc files.
205 Description: Hitting mdc_commit_close() ASSERTION
206 Details : Properly handle request reference release in
207 ll_release_openhandle().
211 Description: quota recovery deadlock during mds failover
212 Details : This patch includes att18982, att18236, att18237 in bz14840.
214 1. fix osts hang when mds does failover with quotaon
215 2. prevent watchdog storm when osts threads wait for the
220 Frequency : only patchless client
221 Description: add workaround for race between add/remove dentry from hash
223 Severity : enhancement
225 Description: Allow OST glimpses to return PW locks
229 Description: LBUG when llog conf file is full
230 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
234 Description: Prevent import from entering FULL state when server in recovery
238 Description: service mount cannot take device name with ":"
239 Details : Only when device name contains ":/" will mount treat it as
245 Description: replace ptlrpcd with the statahead thread to interpret the async
246 statahead RPC callback
250 Frequency : on recovery
251 Description: I/O failures after umount during fail back
252 Details : if client reconnected to restarted server we need join to recovery
253 instead of find server handler is changed and process self
254 eviction with cancel all locks.
256 Severity : enhancement
258 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
260 Severity : enhancement
262 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
264 Severity : enhancement
266 Description: Upcall on Lustre log has been dumped
267 Details : Allow for a user mode script to be called once a Lustre log has
268 been dumped. It passes the filename of the dumped log to the
269 script, the location of the script can be specified via
270 /proc/sys/lnet/debug_log_upcall.
275 Description: avoid idr_remove called for id which is not allocated.
276 Details : Move assigment s_dev for clustered nfs to end of initialization,
277 to avoid problem with error handling.
282 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
283 Details : When connection is reused this not moved from CONN_UNUSED_HASH
284 into CONN_USED_HASH and this prodice warning when put connection
285 again in unused hash.
288 Severity : enhancement
290 Description: File striping can now be set to use an arbitrary pool of OSTs.
292 Severity : enhancement
294 Description: Export bytes_read/bytes_write count on OSC/OST.
298 Description: Early reply size mismatch, MGC loses connection
299 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
300 the connect flags are properly negotiated.
305 Description: quotas are not honored with O_DIRECT
306 Details : all writes with the flag O_DIRECT will use grants which leads to
307 this problem. Now using OBD_BRW_SYNC to guard this.
311 Description: add quota statistics
312 Details : 1. sort out quota proc entries and proc code.
313 2. add quota statistics
315 Severity : enhancement
317 Description: enable quota support for HEAD.
321 Description: Properly propagate oinfo flags from lov to osc for statfs
322 Details : restore missing copy oi_flags to lov requests.
324 Severity : enhancement
326 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
327 : l_getgroups(8), lst(8), routerstat(8)
328 Details : included man pages for llobdstat(8), llstat(8),
329 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
331 Severity : enhancement
333 Description: configure's --enable-quota should check the
334 : kernel .config for CONFIG_QUOTA
335 Details : configure is terminated if --enable-quota is passed but
336 : no quota support is in kernel
340 Description: Remove portals compatibility
341 Details : Remove portals compatibility, not interoperable with releases
346 Description: Resolve device initialization race
347 Details : Prevent proc handler from accessing devices added to the
348 obd_devs array but yet be intialized.
350 Severity : enhancement
352 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
354 Severity : enhancement
356 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
361 Description: Using wrong pointer in osc_brw_prep_request
362 Details : Access to array[-1] can produce panic if kernel compiled with
363 CONFIG_PAGE_ALLOC enabled
367 Description: Client runs out of low memory
368 Details : Consider only lowmem when counting initial number of llap pages
372 Description: *optional* service tags registration
373 Details : if the "service tags" package is installed on a Lustre node
374 When the filesystem is mounted, a local-node service tag will
375 be created. See http://inventory.sun.com/ for more information
376 about the Service Tags asset management system.
380 Description: Kernel BUG tries to release flock
381 Details : Lustre does not destroy flock lock before last reference goes
382 away. So always drop flock locks when client is evicted and
383 perform unlock regardless of successfulness of speaking to MDS.
387 Description: add refcount protection for osc callbacks, avoid panic on shutdown
391 Description: sanity test 65a fails if stripecount of -1 is set
392 Details : handle -1 striping on filesystem in ll_dirstripe_verify
397 Description: ASSERTION(CheckWriteback(page,cmd)) failed
398 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
401 Severity : enhancement
403 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
407 Description: do not process already freed flock
408 Details : flock can possibly be freed by another thread before it reaches
409 to ldlm_flock_completion_ast.
413 Description: LBUG during stress test
414 Details : Need properly lock accesses the flock deadlock detection list.
418 Description: oops in page fault handler
419 Details : kernel page fault handler can return two special 'pages' in error
420 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
424 Description: timeout with invalidate import.
425 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
426 should be handled by ptlrpcd. This produce long age waiting and
427 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
429 Severity : enhancement
431 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
435 Description: enable MGS and MDT services start separately
436 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
437 a co-located MGS without starting the MGS, which is a complement
438 to 'nosvc' mount option.
442 Frequency : after recovery
443 Description: precreate to many object's after del orphan.
444 Details : del orphan st in oscc last_id == next_id and this triger growing
445 count of precreated objects. Set flag LOW to skip increase count
446 of precreated objects.
450 Frequency : rare, on clear nid stats
451 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
452 Details : when clean nid stats sometimes try destroy live entry,
453 and this produce panic in free.
457 Description: Stack overflow during MDS log replay
458 ease stack pressure by using a thread dealing llog_process.
462 Description: wait until IO finished before start new when do lock cancel.
463 Details : VM protocol want old IO finished before start new, in this case
464 need wait until PG_writeback is cleared until check dirty flag and
465 call writepages in lock cancel callback.
467 Severity : enhancement
469 Description: using special macro for print time and cleanup in includes.
473 Description: mds_mfd_close() ASSERTION(rc == 0)
474 Details : In mds_mfd_close(), we need protect inode's writecount change
475 within its orphan write semaphore to prevent possible races.
479 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
483 Frequency : rare, on shutdown ost
484 Description: don't hit live lock with umount ost.
485 Details : shrink_dcache_parent can be in long loop with destroy dentries,
486 use shrink_dcache_sb instead.
490 Description: don't panic with use echo client
491 Details : echo client pass NULL as client nid pointer and this produce null
496 Description: fix build on ppc32
497 Details : compile code with -m64 flag produce wrong object file for ppc32.
501 Description: add message levels for liblustreapi
505 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
506 Details : Move the 'good_osts' check before the 'total_bavail' check. This
507 will result in an -EAGAIN and in the exit call path we call
508 alloc_rr() which will with increasing aggressiveness attempt to
509 aquire precreated objects on the minimum number of required OSCs.
513 Description: Use old size assignment to avoid deadlock
514 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
515 the scheduling while holding a spinlock. We do not need locking
516 for size in ll_update_inode() because size is only updated from
517 the MDS for directories or files without objects, so there is no
518 other place to do the update, and concurrent access to such inodes
519 are protected by the inode lock.
523 Description: resolve "_IOWR redefined" build error on SLES10
527 Description: dump the memory debugging after all modules are unloaded to
528 suppress false negative in conf_sanity test 39
530 Severity : enhancement
532 Description: build kernel-ib packages for OFED 1.3 in our release cycle
537 Description: fix SLES kernel versioning
538 Details : the kernel version for our SLES 10 kernel did not include a "-"
539 before the "smp" at the end. while this was not a problem in
540 general, it did mean that software trying to use the kernel
541 version to try to detect a vendor specific kernel would fail.
542 this was most evident by the OFED build scripts.
546 Description: Don't update lov_desc members until making sure they are valid
547 Details : When updating lov_desc members via proc fs, need fix their
548 validities before doing the real update.
552 Description: don't put request into delay list while invalidate in flight.
553 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
554 import in flight. this produce timeout for invalidate and sometimes
555 can cause stale data.
559 Frequency : on ppc only
560 Description: not convert ost objects for directory because it's not exist.
561 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
565 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
567 Severity : enhancement
569 Description: Update to support 2.6.22.14 vanilla kernel.
573 Frequency : rare, on recovery
574 Description: read procfs can produce deadlock in some situation
575 Details : Holding lprocfs lock which send rpc can produce block for destroy
576 obd objects and this also block reconnect with -EALREADY. This
577 isn't fix all lprocfs bugs - but make it rare.
579 Severity : enhancement
581 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
584 Frequency : frequent on X2 node
586 Description: mdc_set_open_replay_data LBUG
587 Details : Set replay data for requests that are eligible for replay.
591 Description: lustre_mgs: operation 101 on unconnected MGS
592 Details : When MGC is disconnected from MGS long enough, MGS will evict the
593 MGC, and late on MGC cannot successfully connect to MGS and a lot
594 of the error messages complaining that MGS is not connected.
598 Frequency : on network error
599 Description: panic with double free request if network error
600 Details : mdc_finish_enqueue is finish request if any network error ocuring,
601 but it's true only for synchronus enqueue, for async enqueue
602 (via ptlrpcd) this incorrect and ptlrpcd want finish request
605 Severity : enhancement
607 Description: client-side metadata stat-ahead during readdir(directory readahead)
608 Details : perform client-side metadata stat-ahead when the client detects
609 readdir and sequential stat of dir entries therein
612 Frequency : on start mds
614 Description: Implement get_info(last_id) in obdfilter.
617 Frequency : occasional
619 Description: Correctly check stale fid, not start epoch if ost not support SOM
620 Details : open with flag O_CREATE need set old fid in op_fid3 because
621 op_fid2 was overwritten with new generated fid, but mds can answer
622 with one of these two fids and both is not stale. Setattr
623 incorrectly started an epoch and assume will be called
624 done_writing, but without SOM done_writing ever being called.
627 Frequency : rare, depends on device drivers and load
629 Description: MDS or OSS nodes crash due to stack overflow
630 Details : Code changes in 1.8.0 increased the stack usage of some functions.
631 In some cases, in conjunction with device drivers that use a lot
632 of stack the MDS (or possibly OSS) service threads could overflow
633 the stack. One change which was identified to consume additional
634 stack has been reworked to avoid the extra stack usage.
637 Frequency : occasional
639 Description: Do not fail import if osc_interpret_create gets -EAGAIN
640 Details : If osc_interpret_create got -EAGAIN it immediately exits and
641 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
642 oscc_has_objects and see OSC has no objests and call
643 oscc_internal_create to resend create request.
645 Severity : enhancement
647 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
649 Severity : enhancement
651 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
654 Frequency : very rare
656 Description: Fix replay if there is an un-replied request and open
657 Details : In some cases, older replay request will revert the
658 mcd->mcd_last_xid on MDS which is used to record the client's
661 Severity : enhancement
663 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
665 Severity : enhancement
667 Description: Add rhel5 support to HEAD.
669 Serverity : enhancement
671 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
676 Description: Don't allow skipping OSTs if index has been specified.
677 Details : Don't allow skipping OSTs if index has been specified, make
678 locking in internal create lots better.
682 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
683 Details : access to bitfield in structure is always rounded to long
684 and this produce problem with not atomic change any bit.
688 Description: Lustre make rpms failed.
689 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
690 builds Lustre rpms from tarball.
693 Frequency : rare on shutdown ost
695 Description: If llog cancel was not send before clean_exports phase, this can
696 produce deadlock in llog code.
697 Details : If llog thread has last reference to obd and call class_import_put
698 this produce deadlock because llog_cleanup_commit_master wait when
699 last llog_commit_thread exited, but this never success because was
700 called from llog_commit_thread.
704 Description: allow userland application know is lost one of stripes.
705 Details : fill lvb_blocks with error code on ost and return it to
706 application if error flag found.
710 Description: NULL lov_tgts causing MDS oops
711 Details : more safe checks for NULL lov_tgts for avoid oops.
713 Severity : enhancement
715 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
719 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
720 Details : If an OST is down the MDS will hang indefinitely in
721 obd_statfs_rqset() waiting for the statfs data. While for
722 MDS QOS usage of statfs, it should not stuck in waiting.
724 Severity : enhancement
726 Description: remote_acl support
727 Details : Support ACL-based permission check for remote user.
728 Support setfacl/getfacl for remote user with the utils
729 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
730 the system "{s,g}etfacl" utils.
732 Severity : enhancement
734 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
736 Severity : enhancement
738 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
742 Description: make mgs_setparam() handle fsname containing dash
743 Details : fsname containing a dash does not work with lctl conf_param
745 Severity : enhancement
747 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
749 Severity : enhancement
751 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
755 Frequency : rare, at shutdown
756 Description: access already free / zero obd_namespace.
757 Details : if client_disconnect_export was called without force flag set,
758 and exist connect request in flight, this can produce access to
759 NULL pointer (or already free pointer) when connect_interpret
760 store ocd flags in obd_namespace.
764 Frequency : only at startup
765 Description: not alloc memory with spinlock held.
766 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
767 if any spinlock held.
769 Severity : enhancement
771 Description: make lustre randomly fail allocating memory
772 Details : Make lustre randomly failed allocating memory for testing purpose.
774 Severity : enhancement
776 Description: lost problems with lov objid file
777 Details : Fixes some scability and access to not inited memory problems
778 in work with lov objdid file.
783 Description: lfs find does not continue on file error
784 Details : Continue other files processing when a file/dir is absent.
788 Description: Inconsistent usage of lustre_pack_reply()
789 Details : Standardize the usage of lustre_pack_reply() such that it
790 always generate a CERROR on failure.
793 Frequency : occasional
795 Description: MDS hang or stay in waiting lock
796 Details : If client receive lock with CBPENDING flag ldlm need send lock
797 cancel as separate rpc, to avoid situation when cancel request
798 can't processed due all i/o threads stay in wait lock.
802 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
806 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
810 Description: Update to SLES9 kernel-2.6.5-7.287.3.
814 Description: Update to RHEL4 latest kernel.
816 Severity : enhancement
818 Description: Build SLES10 patchless client fails
819 Details : The configure was broken by run ./configure with
820 --with-linux-obj=.... argument for patchless client. When the
821 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
822 can't search header adequately. Use absolute path such as
823 -I($LINUX)/include instead.
827 Description: interrupt oig_wait produce painc on resend.
828 Details : brw_redo_request can be used for resend requests from ptlrpcd and
829 private set, and this produce situation when rq_ptlrpcd_data not
830 copyed to new allocated request and triggered LBUG on assert
831 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
832 ptlrpcd set if request is changed and can be safety changed to use
837 Description: LASSERT_{REQ,REP}SWAB macros are buggy
838 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
839 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
840 these macros and replace them with their difinitions instead.
844 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
845 Details : Update which_patch & target file for SLES10 latest kernel.
848 Description: Remove client patches from SLES 10 kernel.
849 Details : This causes SLES 10 clients to behave as patchless clients
850 even on a Lustre-patched (server) kernel.
852 Severity : enhancement
854 Description: self-adjustable client's lru lists
855 Details : use adaptive algorithm for managing client cached locks lru
856 lists according to current server load, other client's work
857 pattern, memory activities, etc. Both, server and client
858 side namespaces provide number of proc tunables for controlling
861 Severity : enhancement
863 Description: light-weight GSS support
864 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
865 and reduce performance overhead.
867 Severity : enhancement
869 Description: Linux keyring support
870 Details : Support using service of Linux keyring for Lustre GSS internal
871 context refresh/cache mechanism.
875 Description: Fix errors in lfs documentation
876 Details : Fixes man pages
880 Description: don't use GFP_* in generic Lustre code.
881 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
885 Description: obdclass is limited by single OBD_ALLOC(idarray)
886 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
890 Description: warnings with build patchless client with vanila 2.6.19 and up
891 Details : change old ctl_table style and replace ctl_table/ctl_table_header
892 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
896 Frequency : Only for SLES
897 Description: sanity_quota fail test_1
898 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
899 file, which makes TSTID[2] inunique.
903 Frequency : Only for SLES9
904 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
905 Details : Update target/ChangeLog/which_patch .
909 Frequency : Only for test_10 in sanity-quota.sh
910 Description: a bug in quota test code
911 Details : checking a test flag in a wrong place
915 Frequency : Only for test_13 in sanity-quota.sh
916 Description: a bug in quota test script
917 Details : shouldn't check overfull allocated quota
921 Description: jbd statistics
922 Details : Port older jbd statistics patch for sles10
924 Severity : enhancement
926 Description: Add printing inode info into message about error in writepage.
931 Description: reply_lock_interpret crash due to race with it and lock cancel.
932 Details : Do not replay locks that are being cancelled. Do not reference
933 locks by their address during replay, just by their handle.
937 Frequency : When flocks are used.
938 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
939 Details : Flock locks might destroy just granted lock if it could be merged
940 with another existing flock, this is done in completion handler,
941 so teach ldlm_cli_enquque_fini that this is a valid case for
947 Description: Oops in read and write path when failing to allocate lock.
948 Details : Check if lock allocation failed and return error back.
952 Description: lstripe command fails for valid OST index
953 Details : The stripe offset is compared to lov->desc.ld_tgt_count
954 instead of lov->desc.ld_active_tgt_count.
958 Description: sanity.sh failed test 103
959 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
964 Description: df doesn't work properly if diskfs blocksize != 4K
965 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
969 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
970 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
972 Severity : enhancement
974 Description: Adaptive timeouts
975 Details : RPC timeouts adapt to changing server load and network
976 conditions to reduce resend attempts and improve recovery time.
980 Description: llapi_file_create() does not allow some changes
981 Details : add llapi_file_open() that allows specifying the mode and
982 open flags, and also returns an open file handle.
986 Description: merge and cleanup kernel patches.
987 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
991 Description: Add journal checksum support.(Kernel part)
992 Details : The journal checksum feature adds two new flags i.e
993 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
994 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
995 indicates that the commit block contains the checksum for
996 the blocks described by the descriptor blocks. Now commit
997 record can be sent to disk without waiting for descriptor
998 blocks to be written to disk. This behavior is controlled
999 using JBD2_FEATURE_ASYNC_COMMIT flag.
1003 Description: OSS needs mutliple precreate threads
1004 Details : Add ability to start more than one create thread per OSS.
1008 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1009 Details : execute lfs setstripe on client
1013 Description: mds_obd_create error creating tmp object
1014 Details : When the user sets quota on root, llog will be affected and can't
1015 create files and write files.
1018 Frequency : Always on ia64 patchless client, and possibly others.
1020 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1021 Details : This allows the patchless client to be loaded on architectures
1022 without this export.
1026 Description: RedHat Update kernel for RHEL5
1027 Details : Add the kernel config file for RHEL5.
1031 Description: RedHat Update kernel for RHEL5
1032 Details : Modify the kernel config file more closer RHEL5.
1036 Description: Build failure against Centos5 (RHEL5)
1037 Details : Define PAGE_SIZE when it isn't present.
1041 Description: test_7 on sanity-quota.sh can't be used on separate machines
1042 Details : add facet to handle it
1046 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1047 Details : with NFS, the anon dentry's parent was set to itself in
1048 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1049 corresponding dentry other than use rec->ur_name.
1051 Severity : enhancement
1053 Description: lfs setstripe enhancement
1054 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1058 Description: enable data checksumming by default
1059 Details : enable checksum by default, allow --disable-checksum
1060 configure option and "-o nochecksum" mount option. Checksums
1061 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1065 Description: lustre support for RHEL5
1066 Details : Add support for RHEL5.
1070 Description: Client eviction due to failover config
1071 Details : after a connection loss, the lustre client should attempt to
1072 reconnect to the last active server first before trying the
1073 other potential connections.
1077 Description: when mds and osts use different quota unit(32bit and 64bit),
1078 quota will be released repeatly.
1079 Details : void sending multiple quota reqs to mds, which will keep the
1080 status between the reqs.
1084 Description: osts not allocated evenly to files
1085 Details : change the condition to increase offset_idx
1088 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1090 Description: Data corruption for OSTs that are formatted larger than 2TB
1092 Details : When generating the bio request for lustre file writes the
1093 sector number would overflow a temporary variable before being
1094 used for the IO. The data reads correctly from Lustre (which
1095 will overflow in a similar manner) but other file data or
1096 filesystem metadata may be corrupted in some cases.
1100 Description: Tune the kernel for good SCSI performance.
1101 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1102 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1107 Description: rewrite ext2-derived code in obdclass/uuid.c
1108 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1109 from scratch preserving functionality.
1113 Description: "lfs find -obd UUID" prints directories
1114 Details : "lfs find -obd UUID" will return all directory names instead
1115 of just file names. It is incorrect because the directories
1116 do not reside on the OSTs.
1120 Description: use i_size_read and i_size_write in 2.6 port
1121 Details : replace inode->i_size access with i_size_read/write()
1124 Frequency : when removing large files
1126 Description: scheduling issue during removal of large Lustre files
1127 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1128 It causes scheduling issues when removing large files (17TB in the
1132 Frequency : only with liblustre clients on XT3
1134 Description: evictions taking too long
1135 Details : allow llrd to evict clients directly on OSTs
1138 Frequency : only on ppc
1140 Description: /proc/fs/lustre/devices broken on ppc
1141 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1142 We should make sure the type of 'index' is loff_t and then cast
1143 explicitly as needed below. Do not assign an explicitly cast
1147 Frequency : only for rhel5
1149 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1150 Details : Modify the target file & which_kernel.
1152 Severity : enhancement
1154 Description: omit set fsid for export NFS
1155 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1156 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1160 Frequency : Always, for kernels after 2.6.16
1161 Description: Fix warning idr_remove called for id=.. which is not allocated.
1162 Details : Last kernels save old s_dev before kill super and not allow
1163 to restore from callback - restore it before call kill_anon_super.
1167 Description: buffer overruns could theoretically occur
1168 Details : llapi_semantic_traverse() modifies the "path" argument by
1169 appending values to the end of the origin string, and a
1170 overrun may occur. Adding buffer overrun check in liblustreapi.
1174 Description: Fix error on 'ls .' at the top of the Lustre mount.
1175 Details : Don't revalidate dentry if it is a root dentry.
1179 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1180 Details : Modify vm-tunables-rhel4.patch.
1184 Description: Kernel config for 2.6.18-vanilla.
1185 Details : Modify targets/2.6-vanilla.target.in.
1186 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1187 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1188 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1189 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1193 Description: improve handling recoverable errors
1194 Details : if request processig with error which can be recoverable on server
1195 request should be resend, otherwise page released from cache and
1200 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1201 Details : Modify target file & which_patch.
1202 A flaw was found in the IA32 system call emulation provided
1203 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1204 value could be stored in the %RAX register, which could trigger an
1205 out-of-bounds system call table access. An untrusted local user
1206 could exploit this flaw to run code in the kernel
1207 (ie a root privilege escalation). (CVE-2007-4573).
1211 Description: O_DIRECT bypasses client statistics.
1212 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1213 read_ahead_stats were updated. copied the stats section from
1214 osc_send_oap_rpc() into async_internal().
1218 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1222 Description: change order of libsysio includes
1223 Details : '#include sysio.h' should always come before '#include xtio.h'
1227 Description: handle "serious error: objid * already exists" more gracefully
1228 Details : If LAST_ID value on disk is smaller than the objects existing in
1229 the O/0/d* directories, it indicates disk corruption and causes an
1230 LBUG(). If the object is 0-length, then we should use the existing
1231 object. This will help to avoid a full fsck in most cases.
1233 Severity : enhancement
1235 Description: adapt the lustre_config script to support the upgrade case
1236 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1237 server targets to 1.6 in parallel.
1241 Description: To avoid grant space > avaible space when the disk is almost
1242 full. Without this patch you might see the error "grant XXXX >
1243 available" or some LBUG about grant, when the disk is almost
1245 Details : In filter_check_grant, for non_grant cache write, we should
1246 check the left space by if (*left > ungranted + bytes), instead
1247 of (*left > ungranted), because only we are sure the left space
1248 is enough for another "bytes", then the ungrant space should be
1249 increase. In client, we should update cl_avail_grant only there
1250 is OBD_MD_FLGRANT in the reply.
1254 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1255 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1256 described in CVE-2007-4573. This problem could allow untrusted
1257 local users to gain root access.
1260 Frequency : when using O_DIRECT and quotas
1262 Description: Incorrect file ownership on O_DIRECT output files
1263 Details : block usage reported by 'lfs quota' does not take into account
1264 files that have been written with O_DIRECT.
1269 Description: touch file failed when fs is not full
1270 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1271 otherwise we can get ENOSP while fs is not full.
1275 Description: parallel lock callbacks
1276 Details : Instead of sending blocking and completion callbacks as separated
1277 requests, adding them to a set and sending in parallel.
1280 Frequency : only for Cray XT3
1281 Bugzilla : 12829/13455
1282 Description: Changing primary group doesn't change the group lustre assigns to
1284 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1285 overridden with the primary group provided by the group upcall,
1286 whereas the supplied fsgid can be trusted if it is in the list of
1287 supplementary groups returned by the group upcall.
1289 Severity : enhancement
1291 Description: Allow masking D_WARNING, D_ERROR messages from console
1292 Details : Console messages can now be disabled via lnet.printk.
1296 Description: User code with malformed file open parameter crashes client node
1297 Details : Before packing join_file req, all the related reference should be
1298 checked carefully in case some malformed flags cause fake
1299 join_file req on client.
1303 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1304 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1305 mode is equal to the granted mode. However, it should also check
1306 that LDLM_FL_CP_REQD is not set.
1310 Description: Heavy nfs access might result in deadlocks
1311 Details : After ELC code landed, it is now improper to enqueue any mds
1312 locks under och_sem, because enqueue might want to decide to
1313 cancel open locks for same inode we are holding och_sem for.
1317 Description: Client eviction while running blogbench
1318 Details : A lot of unlink operations with concurrent I/O can lead to a
1319 deadlock causing evictions. To address the problem, the number of
1320 oustanding OST_DESTROY requests is now throttled to
1321 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1322 callbacks are processed in priority.
1326 Description: enable ACLs on MDS by default
1327 Details : ACLs must be enabled on MDS by default.
1330 Frequency : PPC/PPC64 only
1332 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1333 Details : fix duplicated definitions between asm-ppc64/types.h and
1334 lustre_types.h on PPC.
1337 Frequency : PPC/PPC64 only
1339 Description: asm-ppc/segment.h does not exist
1340 Details : fix compile issue on PPC.
1344 Description: better handle error messages in extents code
1347 Frequency : RHEL4 only
1349 Description: mkfs is very slow on IA64/RHEL4
1350 Details : A performance regression has been discovered in the MPT Fusion
1351 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1352 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1353 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1355 Severity : enhancement
1357 Description: SNMP support enhancement
1358 Details : Adding total number of sampled request for an MDS node in snmp
1361 Severity : enhancement
1363 Description: Optimize ldlm waiting list processing for PR extent locks
1364 Details : When processing waiting list for read extent lock and meeting read
1365 lock that is same or wider to it that is not contended, skip
1366 processing rest of the list and immediatelly return current
1367 status of conflictness, since we are guaranteed there are no
1368 conflicting locks in the rest of the list.
1372 Description: Time out and refuse to reconnect
1373 Details : When the failover node is the primary node, it is possible
1374 to have two identical connections in imp_conn_list. We must
1375 compare not conn's pointers but NIDs, otherwise we can defeat
1376 connection throttling.
1380 Description: port llog fixes from b1_6 into HEAD
1381 Details : Port llog reference couting and some llog cleanups from b1_6
1382 (bug 10800) into HEAD, for protect from panic and access to already
1383 free llog structures.
1387 Description: Detect stride IO mode in read-ahead
1388 Details : When a client does stride read, read-ahead should detect that and
1389 read-ahead pages according to the detected stride pattern.
1393 Description: data checksumming impacts single node performance
1394 Details : add support for several checksum algorithm. Currently, only CRC32
1395 and Adler-32 are supported. The checksum type can be changed on
1396 the fly via /proc/fs/lustre/osc/*/checksum_type.
1400 Description: use adler32 for page checksums
1401 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1406 Description: build for x2 fails
1407 Details : fix compile issue on Cray systems.
1411 Description: Properly match for duplicate locks
1412 Details : Due to different lock order from skiplists code, we need to
1413 traverse entire list for now
1416 Frequency : only on PPC/SLES10
1418 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1419 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1420 be defined. Add a hack in mkfs_lustre.c to work around this compile
1425 Description: LASSERT on MDS when client holding flock lock dies
1426 Details : ldlm pool logic depends on number of granted locks equal to
1427 number of released locks which is not true for flock locks, so
1428 just exclude such locks from consideration.
1432 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1433 Details : Use fsfilt_sync() for both the whole filesystem sync and
1434 individual file sync to eliminate dangerous inode locking
1435 with I_LOCK that can lead to a deadlock.
1439 Description: performance in 1.6.3
1440 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1441 sure that 1MB requests are not fragmented by the block layer.
1443 Severity : enhancement
1445 Description: organize the server-side client stats on per-nid basis
1446 Details : Change the structure of stats under obdfilter and mds to
1454 The "uuid"s file would list the uuids of _active_ exports.
1455 And the clear entry is to clear all stats and stale nids.
1457 Severity : enhancement
1459 Description: eliminate client locks in face of contention
1460 Details : file contention detection and lockless i/o implementation
1461 for contended files.
1465 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1468 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1470 Description: MDS may lose file striping (and hence file data) in some cases
1471 Details : If there are additional extended attributes stored on the MDS,
1472 in particular ACLs, SELinux, or user attributes (if user_xattr
1473 is specified for the client mount options) then there is a risk
1474 of attribute loss. Additionally, the Lustre file striping
1475 needs to be larger than default (e.g. striped over all OSTs),
1476 and an additional attribute must be stored initially in the
1477 inode and then increase in size enough to be moved to the
1478 external attribute block (e.g. ACL growing in size) for file
1483 Description: skiplist implementation simplification
1484 Details : skiplists are used to group compatible locks on granted list
1485 that was implemented as tracking first and last lock of each
1486 lock group the patch changes that to using doubly linked lists
1490 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1491 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1492 because LASSERT on both the data supplied by a client, and the
1493 data on disk is dangerous and incorrect.
1495 Severity : enhancement
1497 Description: Slow truncate/writes to huge files at high offsets.
1498 Details : Directly associate cached pages to lock that protect those pages,
1499 this allows us to quickly find what pages to write and remove
1500 once lock callback is received.
1504 Description: more ldlm soft lockups
1505 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1506 starve other threads from the resource lock for a long time in
1507 case of long waiting queue, so change the debug level from
1508 D_OTHER to the less frequently used D_INFO.
1510 Severity : enhancement
1512 Description: add -gid, -group, -uid, -user options to lfs find
1516 Description: Hung threads in invalidate_inode_pages2_range
1517 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1518 once one is completed. As a result, some RPCs are stuck in the
1519 queue and are never sent.
1523 Description: filter threads hungs on waiting journal commit
1524 Details : Cleanup filter group llog code, then only filter group llog will
1525 be only created in the MDS/OST syncing process.
1529 Description: Procfs and llog threads access destoryed import sometimes.
1530 Details : Sync the import destoryed process with procfs and llog threads by
1531 the import refcount and semaphore.
1533 Severity : enhancement
1535 Description: openlock cache of b1_6 port to HEAD
1540 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1541 Details : If insertion of an extent fails, then discard the inode
1542 preallocation and free data blocks else it can lead to duplicate
1547 Description: don't always update ctime in ext3_xattr_set_handle()
1548 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1549 In some cases the ctime should not be updated, for example for
1550 2.0->1.8 compatibility it is necessary to delete an xattr and it
1551 should not update the ctime.
1555 Bugzilla : 15713/16362
1556 Description: Assertion in iopen_connect_dentry in 1.6.3
1557 Details : looking up an inode via iopen with the wrong generation number can
1558 populate the dcache with a disconneced dentry while the inode
1559 number is in the process of being reallocated. This causes an
1560 assertion failure in iopen since the inode's dentry list contains
1561 both a connected and disconnected dentry.
1565 Description: assertion failure in ldlm_handle2lock()
1566 Details : fix a race between class_handle_unhash() and class_handle2object()
1567 introduced in lustre 1.6.5 by bug 13622.
1572 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1573 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1578 Description: oops in sd_iostats_seq_show()
1579 Details : unloading/reloading the scsi low level driver triggers a kernel
1580 bug when trying to access the sd iostat file.
1585 Description: Kernel panics during QLogic driver reload
1586 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1587 patch, causing memory corruption.
1592 Description: journal_dev option does not work in b1_6
1593 Details : pass mount option during pre-mount.
1595 Severity : enhancement
1597 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1598 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1599 extent information of a file. It can be used to map logical blocks
1600 in a file to physical blocks in the block device.
1604 Description: LDLM soft lockups - improvement
1605 Details : It is be possible to send the lock handle along with each read
1606 or write request because the client is already doing a lock match
1607 itself so there isn't any reason the OST should have to re-do that
1611 Frequency : only on Cray X2
1613 Description: X2 build failures
1614 Details : fix build failures on Cray X2.
1618 Description: xid & resent requests
1619 Details : Initialize RPC XID from clock at startup (randomly if clock is
1622 Severity : enhancement
1624 Description: Add lustre_start utility to start or stop multiple Lustre servers
1629 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1630 Details : In case of memory pressure, list_del() can be called twice on
1631 req->rq_history_list, causing a kernel oops.
1635 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1636 Details : fix stack overflow in the distributed lock manager by defering
1637 export eviction after a failed AST to the elt thread instead of
1638 handling it in the dlm interpret routine.
1642 Description: Convert some comments to new format.
1643 Details : Update documenting comments to match doxygen conventions.
1647 Description: Grammar fixes.
1648 Details : A couple of trivial spelling fixes.
1652 Description: OSD_COUNTERS-mandatory
1653 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1654 to simplify the code.
1658 Description: simplify cmm_device freeing logic.
1659 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1663 Description: Add lockdep support to dt_object_operations locking interface.
1664 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1665 parameter indicating lock ordering. Update mdd code to use new
1670 Description: Introduce failloc constants for lockless IO tests.
1671 Details : Add two new failloc constants to test lockless IO. Only one of
1672 them in implemented---another is checked in yet to be landed
1677 Description: Add lockdep support for inode mutex.
1678 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1679 in the situations where Lustre has to lock more than one inode
1684 Description: Add optional invariants checking support.
1685 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1686 configure switch. This macro is to be used for consistency and
1687 sanity checks that are too expensive to be left in `production'
1692 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1693 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1698 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1699 Details : Use spin_lock_nested() in (the only) situation where more than
1700 one ldlm_lock is locked simultaneously. Also, fix possible
1701 dead-lock in ldlm_lock_change_resource() by enforcing particular
1706 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1707 Details : Instead of specifying each ldlm_lock call-back through separate
1708 parameter, wrap them into struct ldlm_callback_suite.
1712 Description: Kill join_lru obd method and its callers.
1713 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1714 in memory---a requirement implemented through ->o_join_lru() obd
1715 method in HEAD. Remove this method and its users.
1719 Description: Add asynchronous ldlm ENQUEUE completion handler.
1720 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1721 case that stock ldlm_completion_ast() cannot handle as it waits
1722 until lock is granted. Introduce new ldlm_completion_ast_async()
1723 for this. Also comment ldlm_completion_ast().
1727 Description: ldlm_error <-> errno conversion.
1728 Details : Add functions to map (rather arbitrary) between LDLM error codes
1729 and standard errno values. CLIO needs this to prevent LDLM specific
1730 constants from escaping ldlm and osc.
1734 Description: Kill unused ldlm_handle2lock_ns() function.
1735 Details : Kill unused ldlm_handle2lock_ns() function.
1739 Description: Add lu_ref support to ldlm_lock
1740 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1741 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1742 and ldlm_resource. LDLM interface has to be changed, because code that
1743 releases a reference on a lock, has to "know" what reference this is.
1744 In the most frequent case
1746 lock = ldlm_handle2lock(handle);
1748 LDLM_LOCK_PUT(lock);
1750 no changes are required. When any other reference (received _not_ from
1751 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1752 instead of LDLM_LOCK_PUT().
1754 Arguably, changes are pervasive, and interface requires some discipline
1755 for proper use. On the other hand, it was very instrumental in finding
1756 a few leaked lock references.
1760 Description: Add ldlm_lock_addref_try().
1761 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1762 attempts to addref a lock that might be being canceled
1767 Description: Add ldlm_weigh_callback().
1768 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1769 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1770 hard-coded `number of pages' logic.
1774 Description: Add lockdep annotations to llog code.
1775 Details : Use appropriately tagged _nested() locking calls in the places
1776 where llog takes more than one ->lgh_lock lock.
1780 Description: Add loi_kms_set().
1781 Details : Wrap kms updates into a helper function.
1785 Description: Constify instances of struct lsm_operations.
1786 Details : Constify instances of struct lsm_operations.
1790 Description: lu_conf support.
1791 Details : On a server, a file system object is uniquely identified
1792 by a fid, which is sufficient to locate and load all object
1793 state (inode). On a client, on the other hand, more data are
1794 necessary instantiate an object. Change lu_object_find() and
1795 friends to take additional `lu_conf' argument describing object.
1796 Typically this includes layout information.
1800 Description: lu_context fixes.
1801 Details : Introduce new lu_context functions that are needed on the client
1802 side, where some system threads (ptlrpcd) are shared by multiple
1803 modules, and so cannot be stopped during module shutdown.
1807 Description: Add start and stop methods to lu_device_type_operations.
1808 Details : Introduce two new methods in lu_device_type_operations, that are
1809 invoked when first instance of a given type is created and last one
1810 is destroyed respectively. This is need by CLIO.
1814 Description: Add lu_ref support to struct lu_device.
1815 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1816 track leaked references.
1820 Description: Introduce lu_kmem_descr.
1821 Details : lu_kmem_descr and its companion interface allow to create
1822 and destroy a number of kmem caches at once.
1826 Description: Fix lu_object finalization race.
1827 Details : Fix a race between lu_object_find() finding an object and its
1828 concurrent finalization. This race is (most likely) not possible
1829 on the server, but might happen on the client.
1833 Description: Introduce lu_ref interface.
1834 Details : lu_ref is a debugging module allowing to track references to
1835 a given object. It is quite cpu expensive, and has to be
1836 explicitly enabled with --enable-lu_ref. See usage description
1841 Description: Factor lu_site procfs stats into a separate function.
1842 Details : Separate lu_site stats printing code into a separate function
1843 that can be reused on a client.
1847 Description: Constify instances of struct {lu,dt,md}_device_operations.
1848 Details : Constify instances of struct {lu,dt,md}_device_operations.
1852 Description: Introduce struct md_site and move meta-data specific parts of
1853 struct lu_site here.
1854 Details : Move md-specific fields out of struct lu_site into special struct
1855 md_site, so that lu_site can be used on a client.
1859 Description: Kill mdd_lov_destroy().
1860 Details : Remove unused mdd code.
1864 Description: Add st_block checking to multistat.c.
1865 Details : Add st_block checking to multistat.c.
1869 Description: Add lu_ref support to struct obd_device.
1870 Details : Add lu_ref tracking to obd_device.
1874 Description: Kill obd_set_fail_loc().
1875 Details : Remove unused code.
1879 Description: Add special type for ptlrpc_request interpret functions.
1880 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1881 there. Actual usage will be in CLIO.
1885 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1886 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1887 doesn't work with lockdep.
1891 Description: Add rwv.c test program.
1892 Details : New testing program exercising readv(2) and writev(2) (Qian).
1896 Description: Add sendfile.c test program.
1897 Details : New testing program exercising sendfile(2) (Jay).
1901 Description: Ratelimit a message that can be very frequent.
1902 Details : Ratelimit a memory allocation failure message that can
1907 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1908 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1912 Description: Liblustre build fixes.
1913 Details : Liblustre build fixes.
1917 Description: libcfs: add cfs_{need,cond}_resched() interface.
1918 Details : libcfs: add cfs_{need,cond}_resched() definition and
1919 implementations for Linux, NT, and liblustre.
1921 Severity : enhancement
1923 Description: More exported tunables for mballoc
1924 Details : Add support for tunable preallocation window and new tunables for
1925 large/small requests
1929 Description: Detect corruption of block bitmap and checking for preallocations
1930 Details : Checks validity of on-disk block bitmap. Also it does better
1931 checking of number of applied preallocations. When corruption is
1932 found, it turns filesystem readonly to prevent further corruptions.
1936 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1937 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1938 window to 0. If it does hit cache miss, and read-pattern is still
1939 stride-io mode, does not reset the stride window, but also does not
1940 increase the stride window length in this case.
1944 Frequency : only for big-endian servers
1945 Description: Check if system is big-endian while mounting fs with extents feature
1946 Details : Mounting a filesystem with extents feature will fail on big-endian
1947 systems since ext3-based ldiskfs is not supported on big-endian
1948 systems. This can be over-riden with "bigendian_extents" mount option.
1950 Severity : enhancement
1952 Description: The root squash functionality
1953 Details : A security feature, which is to prevent users from being able
1954 to mount lustre on their desktop, run as root, and delete
1955 all of the files in the filesystem. The goal is accomplished by
1956 remapping user id (UID) and group id (GID) of the root user to
1957 a UID and GID specified by the system administartor via Lustre
1958 configuration management server (MGS). The functionality also
1959 allows to specify sets of clients for which the remapping does
1964 Description: Excessive recovery window
1965 Details : With AT enabled, the recovery window can be excessively long (6000+
1966 seconds). To address this problem, we no longer use
1967 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1968 timeout no longer depends on the service time, it is set to
1969 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1970 time via pb_service_time.
1974 Description: Watchdog triggered on MDS failover
1975 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1976 the OSTs know that the MDS "UUID" can be reused for the same export
1977 from a different NID, so we do not need to wait for the export to be
1981 Frequency : rare, only if using MMP with Linux RAID
1983 Description: MMP doesn't work with Linux RAID
1984 Details : While using HA for Lustre servers with Linux RAID, it is possible
1985 that MMP will not detect multiple mounts. To make this work we
1986 need to unplug the device queue in RAID when the MMP block is being
1987 written. Also while reading the MMP block, we should read it from
1988 disk and not the cached one.
1990 Severity : enhancement
1992 Description: open file using fid
1993 Details : A file can be opened using just its fid, like
1994 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
1996 --------------------------------------------------------------------------------
1998 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2000 * Support for kernels:
2001 2.6.5-7.283 (SLES 9),
2002 2.6.9-55.EL (RHEL 4),
2003 2.6.16.46-0.14 (SLES 10),
2004 2.6.18.8 vanilla (kernel.org)
2005 * Client support for unpatched kernels:
2006 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2007 2.6.16 - 2.6.22 vanilla (kernel.org)
2008 * Due to recently discovered recovery problems, we do not recommend
2009 using patchless RHEL 4 clients with this or any earlier release.
2010 * Recommended e2fsprogs version: 1.39.cfs8
2011 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2012 * Starting with this release, the ldiskfs backing filesystem required
2013 by Lustre is now in its own package, lustre-ldiskfs. This package
2014 should be installed. It is versioned separately from Lustre and
2015 may be released separately in future.
2019 Description: block reactivating mgc import until all deactivates complete
2020 Details : Fix race when failing back MDT/MGS to itself (testing)
2023 Frequency : at statup only
2025 Description: mds_lov_synchronize race leads to various problems
2026 Details : simultaneous MDT->OST connections at startup can cause the
2027 sync to abort, leaving the OSC in a bad state.
2029 Severity : enhancement
2031 Description: add optional extra BUILD_VERSION info
2032 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2033 to override the lustre version.
2036 Frequency : 2.6.18 servers only
2038 Description: ll_kern_mount() doesn't release the module reference
2039 Details : The ldiskfs module reference count never drops down to 0
2040 because ll_kern_mount() doesn't release the module reference.
2045 Description: server LBUG when using old ost_num_threads parameter
2046 Details : Accept the old ost_num_threads parameter but warn that it
2047 is deprecated, and fix an off-by-one error that caused an LBUG.
2052 Description: Transient SCSI error results in persistent IO issue
2053 Details : iobuf->dr_error is not reinitialized to 0 between two
2057 Frequency : sometimes when underlying device returns I/O errors
2059 Description: OSTs not going read-only during write failures
2060 Details : OSTs are not remounted read-only when the journal commit threads
2061 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2062 instead of the ext3 wrappers.
2066 Description: poor connect scaling with increasing client count
2067 Details : Don't run filter_grant_sanity_check for more than 100 exports
2068 to improve scaling for large numbers of clients.
2071 Frequency : SLES10 only
2073 Description: sanity-quota.sh quotacheck failed: rc = -22
2074 Details : Quotas cannot be enabled on SLES10.
2077 Frequency : liblustre clients only
2079 Description: getdirentries does not give error when run on compute nodes
2080 Details : getdirentries does not fail when the size specified as an argument
2081 is too small to contain at least one entry
2083 Severity : enhancement
2085 Description: Add LNET router traceability for debug purposes
2086 Details : If a checksum failure occurs with a router as part of the
2087 IO path, the NID of the last router that forwarded the bulk data
2088 is printed so it can be identified.
2093 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2094 Details : Due to a race condition, liblustre clients were occasionally
2095 evicted incorrectly.
2097 Severity : enhancement
2099 Description: lfs setstripe use optional parameters instead of postional
2102 Severity : enhancement
2104 Description: Nanosecond timestamp support for ldiskfs
2105 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2106 resolution timestamps. There is not yet support for this at
2107 the Lustre filesystem level.
2110 Frequency : during server recovery
2112 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2113 Details : request with rq_no_resend flag not awake l_wait_event if they get
2117 Frequency : nfs export on patchless client
2119 Description: connectathon hang when test nfs export over patchless client
2120 Details : Disconnected dentry cannot be found with lookup, so we do not need
2121 to unhash it or make it invalid
2124 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2125 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2126 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2131 Description: open req refcounting wrong on reconnect
2132 Details : If reconnect happened between getting open reply from server and
2133 call to mdc_set_replay_data in ll_file_open, we will schedule
2134 replay for unreferenced request that we are about to free.
2135 Subsequent close will crash in variety of ways.
2136 Check that request is still eligible for replay in
2137 mdc_set_replay_data().
2142 Description: disable writes to filesystem when reading health_check file
2143 Details : the default for reading the health_check proc file has changed
2144 to NOT do a journal transaction and write to disk, because this
2145 can cause reads of the /proc file to hang and block HA state
2146 checking on a healthy but otherwise heavily loaded system. It
2147 is possible to return to the previous behaviour during configure
2148 with --enable-health-write.
2150 Severity : enhancement
2152 Description: 64-bit inode version
2153 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2154 to the inode. This will be required for version-based recovery.
2159 Description: MDS fails to start if a duplicate client export is detected
2160 Details : in some rare cases it was possible for a client to connect to
2161 an MDS multiple times. Upon recovery the MDS would detect this
2162 and fail during startup. Handle this more gracefully.
2164 Severity : enhancement
2166 Description: Add -o localflock option to simulate old noflock
2168 Details : This will achieve local-only flock/fcntl locks
2174 Description: log_commit_thread vs filter_destroy race leads to crash
2175 Details : Take import reference before releasing llog record semaphore
2180 Description: Wrong request locking in request set processing
2181 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2182 imp_delayed_list, in this place should be used imp_lock.
2185 Frequency : when reconnection
2187 Description: Grant Leak when osc reconnect to OST
2188 Details : When osc reconnect ost, OST(filter) should check whether it
2189 should grant more space to client by comparing fed_grant and
2190 cl_avail_grant, and return the granted space to client instead
2191 of "new granted" space, because client will call osc_init_grant
2192 to update the client grant space info.
2195 Frequency : when client reconnect to OST
2197 Description: Grant Leak when osc do resend and replay bulk write
2198 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2199 bulk write request, because the grant info while be sync between
2200 OSC and OST when reconnect, and we should ignore the grant info
2201 these of resend/replay write req.
2206 Description: Grant space more than avaiable left space sometimes.
2207 Details : When then OST is about to be full, if two bulk writing from
2208 different clients came to OST. Accord the avaliable space of the
2209 OST, the first req should be permitted, and the second one
2210 should be denied by ENOSPC. But if the seconde arrived before
2211 the first one is commited. The OST might wrongly permit second
2212 writing, which will cause grant space > avaiable space.
2215 Frequency : when client is evicted
2217 Description: Grant might be wrongly erased when osc is evicted by OST
2218 Details : when the import is evicted by server, it will fork another
2219 thread ptlrpc_invalidate_import_thread to invalidate the
2220 import, where the grant will be set to 0. While the original
2221 thread will update the grant it got when connecting. So if
2222 the former happened latter, the grant will be wrongly errased
2223 because of this race.
2228 Description: Checking Stale with correct fid
2229 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2230 to check whether it is stale, because sometimes, we want the
2231 enqueue happened anyway, and op_data.fid2 will not be initialized.
2233 Severity : enhancement
2235 Description: update patchless client
2236 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2239 Frequency : only with 2.4 kernel
2241 Description: random memory corruption
2242 Details : size of struct ll_inode_info is to big for union inode.u and this
2243 can be cause of random memory corruption.
2248 Description: Memory leak in recovery
2249 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2250 It should also check obd_fail before fsfilt_start, otherwise if
2251 fsfilt_start return -EROFS,(failover mds during mds recovery).
2252 then the req will return with repmsg->transno = 0 and rc = EROFS.
2253 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2254 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2255 be freed no matter whether fsfilt_commit success or not.
2258 Frequency : only with huge count clients
2260 Description: Prevents from taking the superblock lock in llap_from_page for
2262 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2263 for llap_from_page call in ll_removepage prevents from taking the
2264 superblock lock for a soon died page.
2269 Description: Not check open intent error before release open handle
2270 Details : in some rare cases, the open intent error is not checked before
2271 release open handle, which may cause
2272 ASSERTION(open_req->rq_transno != 0), because it tries to release
2273 the failed open handle.
2278 Description: Set cat log bitmap only after create log success.
2279 Details : in some rare cases, the cat log bitmap is set too early. and it
2280 should be set only after create log success.
2284 Description: Accessing a block bevice can re-enable I/O when Lustre is
2285 tearing down a device.
2286 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2291 Description: service threads may hog cpus when there are a lot of requests
2292 Details : Insert cond_resched to give other threads a chance to use some CPU
2297 Description: the cat log was not initialized in recovery
2298 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2299 unlink log on mds will not be initialized until mds post
2300 recovery. And also in mds post recovery, the unlink log will
2301 initialization will be done asynchronausly, so there will be race
2302 between add unlink log and unlink log initialization.
2306 Description: brw_stats were being printed incorrectly
2307 Details : brw_stats were being printed as log2 but all of them were not
2308 recorded as log2. Also remove some code duplication arising from
2309 filter_tally_{read,write}.
2313 Frequency : rare, only in recovery.
2314 Description: ASSERTION(req->rq_type != LI_POISON) failed
2315 Details : imp_lock should be held while iterating over imp_sending_list for
2316 prevent destroy request after get timeout in ptlrpc_queue_wait.
2320 Description: replay-single.sh test 52 fails
2321 Details : A lock's skiplist need to be cleanup when it being unlinked
2322 from its resource list.
2326 Description: Short directio read returns full requested size rather than
2328 Details : Direct I/O operations should return actual amount of bytes
2329 transferred rather than requested size.
2331 Severity : enhancement
2333 Description: metadata RPC reduction (e.g. for rm performance)
2334 Details : decrease the amount of synchronous RPC between clients and servers
2335 by canceling conflicing lock before the operation on the client
2336 and packing thier handles into the main operation RPC to server.
2338 Severity : enhancement
2340 Description: Async OSC create to avoid the blocking unnecessarily.
2341 Details : If a OST has no remain object, system will block on the creating
2342 when need to create a new object on this OST. Now, ways use
2343 pre-created objects when available, instead of blocking on an
2344 empty osc while others are not empty. If we must block, we block
2345 for the shortest possible period of time.
2349 Description: improve handling recoverable errors
2350 Details : if request processig with error which can be recoverable on server
2351 request should be resend, otherwise page released from cache and
2354 Severity : enhancement
2356 Description: refine locking for avoid write wrong info into lov_objid file
2357 Details : fix possible races with add new target and write/update data in
2360 --------------------------------------------------------------------------------
2362 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2367 Frequency : on some architectures
2369 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2370 Details : Uninitialized flags sometimes cause configuration commands to
2374 Frequency : patchless clients only
2376 Description: missing __iget() symbol export
2377 Details : The __iget() symbol export is missing. To avoid the need for
2378 this on patchless clients the deathrow inode reaper is turned
2379 off, and we depend on the VM to clean up old inodes. This
2380 dependency was during via the fix for bug 12181.
2382 --------------------------------------------------------------------------------
2384 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2386 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2387 INTEROPERATE with older versions automatically. In many cases a
2388 special upgrade step is needed. Please read the
2389 user documentation before upgrading any part of a live system.
2390 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2391 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2392 * WARNING: Lustre configuration and startup changes are required with
2393 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2397 Severity : enhancement
2399 Description: MountConf
2400 Details : Lustre configuration is now managed via mkfs and mount
2401 commands instead of lmc and lconf. New obd types (MGS, MGC)
2402 are added for dynamic configuration management. See
2403 https://mail.clusterfs.com/wikis/lustre/MountConf for
2406 Severity : enhancement
2408 Description: dynamic OST addition
2409 Details : OSTs can now be added to a live filesystem
2411 Severity : enhancement
2413 Description: startup order invariance
2414 Details : MDTs and OSTs can be started in any order. Clients only
2415 require the MDT to complete startup.
2417 Severity : enhancement
2419 Description: parallel, asynchronous orphan cleanup
2420 Details : orphan cleanup is now performed in separate threads for each
2421 OST, allowing parallel non-blocking operation.
2423 Severity : enhancement
2425 Description: optimized stripe assignment
2426 Details : stripe assignments are now made based on ost space available,
2427 ost previous usage, and OSS previous usage, in order to try
2428 to optimize storage space and networking resources.
2430 Severity : enhancement
2432 Description: Permanently set tunables
2433 Details : All writable /proc/fs/lustre tunables can now be permanently
2434 set on a per-server basis, at mkfs time or on a live system.
2436 Severity : enhancement
2438 Description: Lustre message v2
2439 Details : Add lustre message format v2.
2441 Severity : enhancement
2443 Description: client OST exclusion list
2444 Details : Clients can be started with a list of OSTs that should be
2445 declared "inactive" for known non-responsive OSTs.
2449 Description: SPEC SFS validation failure on NFS v2 over lustre.
2450 Details : Changes the blocksize for regular files to be 2x RPC size,
2451 and not depend on stripe size.
2453 Severity : enhancement
2455 Description: Multiple MD RPCs in flight.
2456 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2457 To avoid overly-overloading MDS, introduce a limit on number of
2458 MDS RPCs in flight for a single client and add /proc controls
2459 to adjust this limit.
2461 Severity : enhancement
2463 Description: client read/write statistics
2464 Details : Add client read/write call usage stats for performance
2465 analysis of user processes.
2466 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2467 file access. extents_stats shows chunk size distribution.
2468 extents_stats_per_process show chunk size distribution per
2471 Severity : enhancement
2473 Description: mds statistics
2474 Details : Add detailed mds operations statistics in
2475 /proc/fs/lustre/mds/*/stats.
2479 Description: Failure of copying files with lustre special EAs.
2480 Details : Client side always return success for setxattr call for lustre
2481 special xattr (currently only "trusted.lov").
2486 Description: Refcount LNET uuids
2487 Details : The global LNET uuid list grew linearly with every startup;
2488 refcount repeated list entries instead of always adding to
2491 Severity : enhancement
2493 Description: Dynamic service threads
2494 Details : Within a small range, start extra service threads
2495 automatically when the request queue builds up.
2498 Frequency : mixed-endian client/server environments
2500 Description: mixed-endian crashes
2501 Details : The new msg_v2 system had some failures in mixed-endian
2504 Severity : enhancement
2506 Description: Easy OST removal
2507 Details : OSTs can be permanently deactivated with e.g. 'lctl
2508 conf_param lustre-OST0001.osc.active=0'
2510 Severity : enhancement
2512 Description: MGS proc entries
2513 Details : Added basic proc entries for the MGS showing what filesystems
2516 Severity : enhancement
2518 Description: provide MGS failover
2519 Details : Added config lock reacquisition after MGS server failover.
2521 Severity : enhancement
2523 Description: add Linux 2.4 support
2524 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2528 Description: a large application tries to do I/O to the same resource and dies
2529 in the middle of it.
2530 Details : Check the req->rq_arrival time after the call to
2531 ost_brw_lock_get(), but before we do anything about
2532 processing it & sending the BULK transfer request. This
2533 should help move old stale pending locks off the queue as
2534 quickly as obd_timeout.
2537 Frequency : when an incorrect nid is specified during startup
2539 Description: ptlrpc connect to non-existant node causes kernel crash
2540 Details : LNET can't be re-entered from an event callback, which
2541 happened when we expire a message after the export has been
2542 cleaned up. Instead, hand the zombie cleanup off to another
2545 Severity : enhancement
2547 Description: plain/inodebits lock performance improvement
2548 Details : Grouping plain/inodebits in granted list by their request modes
2549 and bits policy, thus improving the performance of search through
2553 Frequency : only if OST filesystem is corrupted
2555 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2556 Details : for a short time RPCs with bulk IO are in the replay list,
2557 but replay of bulk IOs is unimplemented. If the OST filesystem
2558 is corrupted due to disk cache incoherency and then replay is
2559 started it is possible to trip an assertion. Avoid putting
2560 committed RPCs into the replay list at all to avoid this issue.
2563 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2565 Description: System hang on startup
2566 Details : This bug allowed the liblustre (e.g. catamount) client to
2567 return to the app before handling all startup RPCs. This
2568 could leave the node unresponsive to lustre network traffic
2569 and manifested as a server ptllnd timeout.
2571 Severity : enhancement
2573 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2574 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2575 boolean to control whether to print peer debug info when a
2576 client's RPC times out.
2579 Frequency : only for kernels with patches from Lustre below 1.4.3
2581 Description: Remove old rdonly API
2582 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2585 Frequency : only for devices with external journals
2587 Description: Set external device read-only also
2588 Details : During a commanded failover stop, we set the disk device
2589 read-only while the server shuts down. We now also set any
2590 external journal device read-only at the same time.
2593 Frequency : when upgrading from 1.4 while trying to change parameters
2595 Description: The wrong (new) MDC name was used when setting parameters for
2596 upgraded MDT's. Also allows changing of OSC (and MDC)
2597 parameters if --writeconf is specified at tunefs upgrade time.
2600 Frequency : when setting specific ost indicies
2602 Description: QOS code breaks on skipped indicies
2603 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2604 created with --index need not be sequential.
2606 Severity : enhancement
2608 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2609 Details : The uninit_groups feature works in conjunction with the kernel
2610 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2611 up the pass1 processing of e2fsck. This is a read-only feature
2612 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2613 mount filesystems that have had this feature enabled.
2615 Severity : enhancement
2617 Description: Improve multi-block allocation algorithm to avoid fragmentation
2618 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2619 allocation locality and avoid filesystem fragmentation.
2621 ------------------------------------------------------------------------------
2623 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2625 * Support for kernels:
2626 2.4.21-47.0.1.EL (RHEL 3)
2627 2.6.5-7.283 (SLES 9)
2628 2.6.9-42.0.10.EL (RHEL 4)
2629 2.6.12.6 vanilla (kernel.org)
2630 2.6.16.27-0.9 (SLES 10)
2631 * Recommended e2fsprogs version: 1.39.cfs5
2633 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2637 Frequency : occasional, depends on client load and configuration
2638 Bugzilla : 12181, 12203
2639 Description: data loss for recently-modified files
2641 Details : In some cases it is possible that recently written or created
2642 files may not be written to disk in a timely manner (this should
2643 normally be within 30s unless client IO load is very high).
2644 The problem appears as zero-length files or files that are a
2645 multiple of 1MB in size after a client crash or client eviction
2646 that are missing data at the end of the file.
2648 This problem is more likely to be hit on clients where files are
2649 repeatedly created and unlinked in the same directory, clients
2650 have a large amount of RAM, have many CPUs, the filesystem has
2651 many OSTs, the clients are rebooted frequently, and/or the files
2652 are not accessed by other nodes after being written.
2654 The presence of the problem can be detected by looking at
2655 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2656 smaller than the second (nr_unused) then dirty files will not
2657 be flushed automatically to disk. "sync; sleep 10" should be
2658 run several times on the node before unmounting it to update
2659 Lustre (this is also safe to run on nodes without this problem).
2661 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2662 that can cause this same problem, so customers using that kernel
2663 also need to update the kernel in addition to Lustre. In order
2664 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2666 It is normal that files written just before a client crash (less
2667 than 30s) may not yet have been flushed to disk, even for local
2671 Frequency : frequent on thin XT3 nodes
2673 Description: UUID collision on thin XT3 Linux nodes
2674 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2675 randomly, since we relied on an insufficiently-seeded PRNG.
2680 Description: OSS hangs after "All ost request buffers busy"
2681 Details : A deadlock between quota and journal operations caused OSS
2682 hangs after printing "All ost request buffers busy."
2685 Frequency : always on liblustre builds
2687 Description: Cleanup compiler warnings on liblustre
2690 Frequency : always on liblustre builds on XT3
2692 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2696 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2697 > 32000 subdirectories
2698 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2699 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2700 subdirectory count crosses 32000. This will aid e2fsck to
2701 correctly handle more than 32000 subdirectories.
2704 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2706 Description: System hang on startup
2707 Details : This bug allowed the liblustre (e.g. catamount) client to
2708 return to the app before handling all startup RPCs. This
2709 could leave the node unresponsive to lustre network traffic
2710 and manifested as a server ptllnd timeout.
2712 Severity : enhancement
2714 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2715 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2716 boolean to control whether to print peer debug info when a
2717 client's RPC times out.
2722 Description: make O_SYNC working on 2.6 kernels
2723 Details : 2.6 kernels use different method for mark pages for write,
2724 so need add a code to lustre for O_SYNC work.
2729 Description: Failure to close file and release space on NFS
2730 Details : Put inode details into lock acquired in ll_intent_file_open.
2731 Use mdc_intent_lock in ll_intent_open to properly
2732 detect all kind of errors unhandled by mdc_enqueue.
2737 Description: proc file read during shutdown sometimes raced obd removal,
2739 Details : Add lock to prevent obd access after proc file removal.
2742 Frequency : Only for files larger than 4GB on 32-bit clients.
2744 Description: improperly doing page alignment of locks
2745 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2746 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2751 Description: RPCs being resent when they shouldn't be
2752 Details : Some RPCs that should not be resent are being resent. This
2753 can cause inconsistencies in the RPC state machine. Do not
2754 resend such requests.
2757 Frequency : rare, only with NFS export
2759 Description: Crash on NFS re-export node
2760 Details : under very unusual load conditions an assertion is hit in
2761 ll_intent_file_open()
2764 Frequency : only if OST filesystem is corrupted
2766 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2767 Details : for a short time RPCs with bulk IO are in the replay list,
2768 but replay of bulk IOs is unimplemented. If the OST filesystem
2769 is corrupted due to disk cache incoherency and then replay is
2770 started it is possible to trip an assertion. Avoid putting
2771 committed RPCs into the replay list at all to avoid this issue.
2776 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2777 Details : Large single O_DIRECT read and write calls can fail to allocate
2778 a sufficiently large buffer to process the request. In case of
2779 allocation failure the allocation is retried with a smaller
2780 buffer and broken into smaller requests.
2782 Severity : enhancement
2784 Description: Add -o localflock option to simulate old noflock behaviour.
2785 Details : This will achieve local-only flock/fcntl locks coherentness.
2790 Description: versioning check is incomplete
2791 Details : Checking the version difference of client vs. server, report
2792 error if the gap is too big.
2797 Description: add support PG_writeback bit
2798 Details : add support for PG_writeback bit for Lustre, for more carefull
2799 work with page cache in 2.6 kernel. This also fix some deadlocks
2800 and remove hack for work O_SYNC with 2.6 kernel.
2802 Severity : enhancement
2804 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2805 Details : The uninit_groups feature works in conjunction with the kernel
2806 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2807 up the pass1 processing of e2fsck. This is a read-only feature
2808 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2809 mount filesystems that have had this feature enabled.
2811 Severity : enhancement
2813 Description: Improve multi-block allocation algorithm to avoid fragmentation
2814 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2815 allocation locality and avoid filesystem fragmentation.
2817 ------------------------------------------------------------------------------
2819 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2821 * Support for kernels:
2822 2.6.9-42.0.3.EL (RHEL 4)
2823 2.6.5-7.276 (SLES 9)
2824 2.4.21-47.0.1.EL (RHEL 3)
2825 2.6.12.6 vanilla (kernel.org)
2826 2.6.16.21-0.8 (SLES10)
2827 * Recommended e2fsprogs version: 1.39.cfs2-0
2829 * The backwards-compatible /proc/sys/portals symlink has been removed
2830 in this release. Before upgrading, please ensure that you change
2831 any configuration scripts or /etc/sysctl.conf files that access
2832 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2833 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2834 in advance of the upgrade on any system running Lustre 1.4.6 or
2835 newer, since /proc/sys/lnet was added in that version.
2836 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2840 Frequency : only when quota is used
2842 Description: avoid scanning export list for quota master
2843 Details : Change the algorithms to avoid scanning export list in order
2844 to improve the efficiency.
2847 Frequency : MDS failover only, very rarely
2849 Description: "went back in time" messages on mds failover
2850 Details : The greatest transno may be lost when the current operation
2851 finishes with an error (transno==0) and the client's last_rcvd
2852 record is over-written. Save the greatest transno in the
2853 mds_last_transno for this case.
2856 Frequency : always for specific kernels and striping counts
2858 Description: client may get "Matching packet too big" without ACL support
2859 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2860 when trying to access files in certain configurations. The
2861 clients should in fact be denied when mounting because they do
2862 not understand ACLs.
2865 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2867 Description: many clients connecting with IO in progress causes connect timeouts
2868 Details : Avoid synchronous journal commits to avoid delays caused by many
2869 clients connecting/disconnecting when bulk IO is in progress.
2870 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2871 OST_IO_PORTAL to avoid delays behind potentially many pending
2875 Frequency : occasionally with multiple writers to a single file
2877 Description: shared writes to file may result in wrong size reported by stat()
2878 Details : Allow growing of kms when extent lock is cancelled
2881 Frequency : always with random mmap IO to multi-striped file
2883 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2884 Details : Only if the hole is at the end of OST object so that kms is too
2885 small. Fix is to increase kms accordingly in ll_nopage.
2888 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2890 Description: writes to a missing object would leak memory on the OST
2891 Details : If there is an inconsistency between the MDS and OST filesystems,
2892 such that the MDS references an object that doesn't exist, writes
2893 to that object will leak memory due to incorrect cleanup in the
2894 error handling path, eventually running out of memory on the OST.
2899 Description: Creating too long symlink causes lustre errors
2900 Details : Check symlink and name lengths before sending requests to MDS.
2903 Frequency : only if flock is enabled (not on by default)
2905 Description: posix locks not released on fd closure on 2.6.9+
2906 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2907 kernels, this caused such locks not to be released on fd close and
2908 then assertions on fs unmount about still used locks.
2911 Frequency : MDS failover only, very rarely
2913 Description: clients may get ASSERTION(granted_lock != NULL)
2914 Details : When request was taking a long time, and a client was resending
2915 a getattr by name lock request. The were multiple lock requests
2916 with the same client lock handle and
2917 mds_getattr_name->fixup_handle_for_resent_request found one of the
2918 lock handles but later failed with ASSERTION(granted_lock != NULL).
2923 Description: handle->h_buffer_credits > 0, assertion failure
2924 Details : h_buffer_credits is zero after truncate, causing assertion
2925 failure. This patch extends the transaction or creates a new
2929 Frequency : NFS re-export or patchless client
2930 Bugzilla : 11179, 10796
2931 Description: Crash on NFS re-export node (__d_move)
2932 Details : We do not want to hash the dentry if we don't have a lock.
2933 But if this dentry is later used in d_move, we'd hit uninitialised
2934 list head d_hash, so we just do this to init d_hash field but
2935 leave dentry unhashed.
2938 Frequency : NFS re-export or patchless client
2940 Description: NFS exports has problem with symbolic link
2941 Details : lustre client didn't properly install dentry when re-exported
2942 to NFS or running patchless client.
2945 Frequency : NFS re-export or patchless client
2947 Description: Various nfs/patchless fixes.
2948 Details : fixes reuse disconected alias for lookup process - this fixes
2949 warning "find_exported_dentry: npd != pd",
2950 fix permission error with open files at nfs.
2951 fix apply umask when do revalidate.
2954 Frequency : occasional
2956 Description: Crash on NFS re-export node
2957 Details : calling clear_page() on the wrong pointer triggered oops in
2958 generic_mapping_read().
2961 Frequency : rarely, using O_DIRECT IO
2963 Description: unaligned directio crashes client with LASSERT
2964 Details : check for unaligned buffers before trying any requests.
2967 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2969 Description: stale data returned from RAID cache
2970 Details : If only a small amount of IO is done to the RAID device before
2971 reading it again it is possible to get stale data from the RAID
2972 cache instead of reading it from disk.
2975 Frequency : always for sles10 kernel
2977 Description: sles10 support
2978 Details : ll_follow_link: compile fixes and using of nd_set_link
2979 under newer kernels.
2982 Frequency : depends on arch, kernel and compiler version, always on sles10
2985 Description: recursive or deep enough symlinks cause stack overflow
2986 Details : getting rid of large stack-allocated variable in
2990 Frequency : depends on hardware
2992 Description: lustre write performance loss in the SLES10 kernel
2993 Details : the performance loss is caused by using of write barriers in the
2994 ext3 code. The SLES10 kernel turns barrier support on by
2995 default. The fix is to undo that change for ldiskfs.
2997 ------------------------------------------------------------------------------
2999 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3001 * Support for kernels:
3002 2.6.9-42.0.3EL (RHEL 4)
3003 2.6.5-7.276 (SLES 9)
3004 2.4.21-47.0.1.EL (RHEL 3)
3005 2.6.12.6 vanilla (kernel.org)
3009 Frequency : quota enabled and large files being deleted
3011 Description: releasing more than 4GB of quota at once hangs OST
3012 Details : If a user deletes more than 4GB of files on a single OST it
3013 will cause the OST to spin in an infinite loop. Release
3014 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3019 Description: statfs data retrieved from /proc may be stale or zero
3020 Details : When reading per-device statfs data from /proc, in the
3021 {kbytes,files}_{total,free,avail} files, it may appear
3022 as zero or be out of date.
3025 Frequency : systems with MD RAID1 external journal devices
3027 Description: lconf's call to blkid is confused by RAID1 journal devices
3028 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3029 of returning all block devices that match the journal UUID.
3032 Frequency : always, for aggregate stripe size over 4GB
3034 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3035 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3036 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3039 Frequency : always if "lfs find" used on a local file/directory
3041 Description: "lfs find" segfaults if used on a local file/directory
3042 Details : The case where a directory component was not specified wasn't
3043 handled correctly. Handle this properly.
3046 Frequency : always on ppc64
3048 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3049 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3050 perform endianness conversion.
3053 Frequency : rarely (truncate to non-zero file size after write under load)
3054 Bugzilla : 10730, 10687
3055 Description: Files padded with zeros to next 4K multiple
3056 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3057 it is possible that files that are truncated to a non-zero size
3058 immediately after being written are filled with zero bytes beyond
3059 the truncated size. No file data is lost.
3061 Severity : enhancement
3063 Description: Allow recovery/failover for liblustre clients.
3064 Details : liblustre clients were unaware of failover configurations until
3067 Severity : enhancement
3069 Description: user file locks should fail when not mounting with flock option
3070 Details : Set up an error-returning stub in ll_file_operations.lock field
3071 to prevent incorrect behaviour when client is mounted without
3072 flock option. Also, set up properly f_op->flock field for
3076 Frequency : always on ia64
3078 Description: "lfs df" loops on printing out MDS statfs information
3079 Details : The obd_ioctl_data was not initialized and in some systems
3080 this caused a failure during the ioctl that did not return
3081 an error. Initialize the struct and return an error on failure.
3084 Frequency : SLES 9 only
3086 Description: Error of copying files with lustre special EAs as root
3087 Details : Client side always return success for setxattr call for lustre
3088 special xattr (currently only "trusted.lov").
3091 Frequency : rarely on clusters with both ia64+i386 clients
3093 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3094 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3095 clients are doing concurrent writes to the same file it is
3096 possible that the ia64 clients may LASSERT because the OST
3097 extent locks are not PAGE_SIZE aligned. Ensure that grown
3098 locks are always aligned on the request boundary.
3101 Frequency : specific use, occasional
3103 Description: Overwriting in use executable truncates on-disk binary image
3104 Details : If one node attempts to overwrite an executable in use by
3105 another node, we now correctly return ETXTBSY instead of
3106 truncating the file.
3108 Severity : enhancement
3110 Description: Async OSC create to avoid the blocking unnecessarily.
3111 Details : If a OST has no remain object, system will block on the creating
3112 when need to create a new object on this OST. Now, ways use
3113 pre-created objects when available, instead of blocking on an
3114 empty osc while others are not empty. If we must block, we block
3115 for the shortest possible period of time.
3120 Description: chmod on Lustre root is propagated to other clients
3121 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3122 invalid by the follow_mount time.
3125 Frequency : liblustre clients only
3127 Description: Race in 'instant cancel' lock handling could lead to such locks
3128 never to be granted in case of SMP MDS
3129 Details : Do not destroy not yet granted but cbpending locks in
3133 Frequency : replay/resend of open
3135 Description: non null lock assetion failure in mds_intent_policy
3136 Details : Trying to replay/resend lockless open requests resulted in
3137 mds_open() returning 0 with no lock. Now it sets a flag if
3138 a lock is going to be returned.
3140 Severity : enhancement
3142 Description: Checksum enhancements
3143 Details : New checksum enhancements allow for resending RPCs that failed
3146 Severity : enhancement
3148 Description: Tunables on number of dirty pages in cacche
3149 Details : Allow to set limit on number of dirty pages cached.
3154 Description: client crash on unmount - lock still has references
3155 Details : In some error handling cases it was possible to leak a lock
3156 reference on a client while accessing a file. This was not
3157 harmful to the client during operation, but would cause the
3158 client to crash when the filesystem is unmounted.
3161 Frequency : specific case, rare
3163 Description: ETXTBSY on mds though file not in use by client
3164 Details : ETXTBSY is no longer incorrectly returned when attempting to
3165 chmod or chown a directory that the user previously tried to
3166 execute or a currently-executing binary.
3169 Frequency : extremely rare except on liblustre-based clients
3171 Description: Lustre space not freed when files are deleted
3172 Details : Clean up open-unlinked files after client eviction. Previously
3173 the unlink was skipped and the files remained as orphans.
3178 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3179 Details : In some cases it was possible to send a blocking callback to
3180 a client doing a glimpse, even though that client didn't get
3181 a lock granted. When the glimpse lock is cancelled on the OST
3182 the freed lock is left on the waiting list and corrupted the list.
3185 Frequency : all core dumps
3187 Description: Broke core dumps to lustre
3188 Details : Negative dentry may be unhashed if parent does not have UPDATE
3189 lock, but some callers, e.g. do_coredump, expect dentry to be
3190 hashed after successful create, hash it in ll_create_it.
3192 ------------------------------------------------------------------------------
3194 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3196 * Support for kernels:
3197 2.6.9-42.0.2.EL (RHEL 4)
3198 2.6.5-7.276 (SLES 9)
3199 2.4.21-40.EL (RHEL 3)
3200 2.6.12.6 vanilla (kernel.org)
3204 Frequency : always on RHEL 3
3206 Description: Number of open files grows over time
3207 Details : The number of open files grows over time, whether or not
3208 Lustre is started. This was due to a filp leak introduced
3209 by one of our kernel patches.
3211 ------------------------------------------------------------------------------
3213 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3215 * Support for kernels:
3216 2.6.9-42.EL (RHEL 4)
3217 2.6.5-7.276 (SLES 9)
3218 2.4.21-40.EL (RHEL 3)
3219 2.6.12.6 vanilla (kernel.org)
3224 Bugzilla : 5719, 9635, 9792, 9684
3225 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3226 Details : If a server is under heavy load and cannot reply to new
3227 connection requests before the client resends the (re)connect,
3228 the connection handling code can behave badly if two service
3229 threads are concurrently handing separate (re)connections from
3230 the same client. Add better locking to the connection handling
3231 code, and ensure that only a single connection will be processed
3232 for a given client UUID, even if the lock is dropped.
3234 Severity : enhancement
3236 Description: add TCP zero-copy support to kernel
3237 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3238 sends if the hardware supports scatter-gather and checksumming.
3239 This allows socklnd to do client-write and server-read more
3240 efficiently and reduce CPU utilization from skbuf copying.
3243 Frequency : only if NFS exporting from client
3245 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3246 Details : If setting attributes on a file created under NFS that had
3247 never been opened it would be possible to oops the client
3248 if the file had no objects.
3251 Frequency : always for liblustre
3253 Description: liblustre client does MDS+OSTs setattr RPC for each write
3254 Details : When doing a write from a liblustre client, the client
3255 incorrectly issued an RPC to the MDS and each OST the file was
3256 striped over in order to update the timestamps. When writing
3257 with small chunks and many clients this could overwhelm the MDS
3258 with RPCs. In all cases it would slow down the write because
3259 these RPCs are unnecessary.
3261 Severity : enhancement
3263 Description: allow number of MDS service threads to be changed at module load
3264 Details : It is now possible to change the number of MDS service threads
3265 running. Adding "options mds mds_num_threads={N}" to the MDS's
3266 /etc/modprobe.conf will set the number of threads for the next
3267 time Lustre is restarted (assuming the "mds" module is also
3268 reloaded at that time). The default number of threads will
3269 stay the same, 32 for most systems.
3274 Description: OST crash if filesystem is unformatted or corrupt
3275 Details : If an OST is started on a device that has never been formatted
3276 or if the filesystem is corrupt and cannot even mount then the
3277 error handling cleanup routines would dereference a NULL pointer.
3282 Description: NULL pointer deref in llap_from_page.
3283 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3284 incorrect) mapping if the page was truncated/reclaimed while it was
3285 searched for. Check for this condition and skip such pages when
3286 doing readahead. Introduce extra check to llap_from_page() to
3287 verify page->mapping->host is non-NULL (so page is not anonymous).
3290 Frequency : Sometimes when using sys_sendfile
3292 Description: "page not covered by a lock" warnings from ll_readpage
3293 Details : sendfile called ll_readpage without right page locks present.
3294 Now we introduced ll_file_sendfile that does necessary locking
3295 around call to generic_file_sendfile() much like we do in
3299 Frequency : with certain MDS communication failures at client mount time
3301 Description: NULL pointer deref after failed client mount
3302 Details : a client connection request may delayed by the network layer
3303 and not be sent until after the PTLRPC layer has timed out the
3304 request. If the client fails the mount immediately it will try
3305 to clean up before the network times out the request. Add a
3306 reference from the request import to the obd device and delay
3307 the cleanup until the network drops the request.
3310 Frequency : occasionally during client (re)connect
3312 Description: assertion failure during client (re)connect
3313 Details : processing a client connection request may be delayed by the
3314 client or server longer than the client connect timeout. This
3315 causes the client to resend the connection request. If the
3316 original connection request is replied in this interval, the
3317 client may trip an assertion failure in ptlrpc_connect_interpret()
3318 which thought it would be the only running connect process.
3321 Frequency : only with obd_echo servers and clients that are rebooted
3323 Description: kernel BUG accessing uninitialized data structure
3324 Details : When running an obd_echo server it did not start the ping_evictor
3325 thread, and when a client was evicted an uninitialized data
3326 structure was accessed. Start the ping_evictor in the RPC
3327 service startup instead of the OBD startup.
3329 Severity : enhancement
3330 Bugzilla : 10193 (patchless)
3331 Description: Remove dependency on various unexported kernel interfaces.
3332 Details : No longer need reparent_to_init, exit_mm, exit_files,
3333 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3336 Frequency : rare (only users of deprecated and unsupported LDAP config)
3338 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3339 Details : LDAP apparently contains 'lustreName' attributes instead of
3340 'name'. A simple remapping of the name is sufficient.
3343 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3345 Description: waiting_locks_callback trips kernel BUG if client is evicted
3346 Details : Running with the dump_on_timeout debug flag turned on makes
3347 it possible that the waiting_locks_callback() can try to dump
3348 the Lustre kernel debug logs from an interrupt handler. Defer
3349 this log dumping to the expired_lock_main() thread.
3351 Severity : enhancement
3353 Description: Support NFS exporting on 2.6 kernels.
3354 Details : Implement non-rawops metadata methods for NFS server to use without
3355 changing NFS server code.
3358 Frequency : very rare (synthetic metadata workload only)
3360 Description: two racing renames might cause an MDS thread to deadlock
3361 Details : Running the "racer" program may cause one MDS thread to rename
3362 a file from being the source of a rename to being the target of
3363 a rename at exactly the same time that another thread is doing
3364 so, and the second thread has already enqueued these locks after
3365 doing a lookup of the target and is trying to relock them in
3366 order. Ensure that we don't try to re-lock the same resource.
3369 Frequency : only very large systems with liblustre clients
3371 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3372 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3373 all OSTs in parallel. This allows the eviction of stale liblustre
3374 clients to proceed much faster than if they were done in series,
3375 and also offers similar improvements for other set_info RPCs.
3380 Description: excessive CPU usage during initial read phase on client
3381 Details : During the initial read phase on a client, it would agressively
3382 retry readahead on the file, consuming too much CPU and impacting
3383 performance (since 1.4.5.8). Improve the readahead algorithm
3384 to avoid this, and also improve some other common cases (read
3385 of small files in particular, where "small" is files smaller than
3386 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3391 Description: MDS crash when receiving packet with unknown intent.
3392 Details : Do not LBUG in unknown intent case, just return -EFAULT
3394 Severity : enhancement
3395 Bugzilla : 9293, 9385
3396 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3397 Details : Do not serialize getattr (non-intent version) and statfs.
3400 Frequency : occasional, when OST network is overloaded/intermittent
3402 Description: client evicted by OST after bulk IO timeout
3403 Details : If a client sends a bulk IO request (read or write) the OST
3404 may evict the client if it is unresposive to its data GET/PUT
3405 request. This is incorrect if the network is overloaded (takes
3406 too long to transfer the RPC data) or dropped the OST GET/PUT
3407 request. There is no need to evict the client at all, since
3408 the pinger and/or lock callbacks will handle this, and the
3409 client can restart the bulk request.
3412 Frequency : Always when mmapping file with no objects
3414 Description: client crashes when mmapping file with no objects
3415 Details : Check that we actually have objects in a file before doing any
3416 operations on objects in ll_vm_open, ll_vm_close and
3422 Description: Request leak when working with deleted CWD
3423 Details : Introduce advanced request refcount tracking for requests
3424 referenced from lustre intent.
3426 Severity : Enhancement
3428 Description: Cache open file handles on client.
3429 Details : MDS now will return special lock along with openhandle, if
3430 requested and client is allowed to hold openhandle, even if unused,
3431 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3432 closing files for every read/write openration.
3434 Severity : Enhancement
3436 Description: Cache open negative dentries on client when possible.
3437 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3438 negative dentries on lock revocation.
3443 Description: Remounting a client read-only wasn't possible with a zconf mount
3444 Details : It wasn't possible to remount a client read-only with llmount.
3446 Severity : enhancement
3447 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3448 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3449 MPICH is included to add Lustre-specific ADIO interfaces.
3450 This is based closely on the UFS ADIO layer and only differs
3451 in file creation, in order to allow the OST striping to be set.
3452 This is user-contributed code and not supported by CFS.
3457 Description: extended inode attributes (immutable, append-only) work improperly
3458 when 2.4 and 2.6 kernels are used on client/server or vice versa
3459 Details : Introduce kernel-independent values for these flags.
3461 Severity : enhancement
3464 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3465 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3466 granularity. Now, Fractional MB granularity is be supported,
3467 this is very useful for low memory system.
3469 Severity : enhancement
3471 Description: Getattr by fid
3472 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3473 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3474 have locks and avoids getattr rpc requests for it on every stat.
3477 Frequency : Always, for filesystems larger than 2TB
3479 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3480 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3481 filesystems larger than 2TB to be mounted with mballoc enabled.
3484 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3486 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3487 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3488 filesystems larger than 2TB is mounted then the kernel will
3489 silently corrupt the start of the filesystem. CONFIG_LBD is
3490 enabled for all CFS-supported kernels, but the possibility of
3491 this happening with a modified kernel config exists.
3493 Severity : enhancement
3495 Description: add client O_DIRECT support for 2.6 kernels
3496 Details : It is now possible to do O_DIRECT reads and writes to files
3497 in the Lustre client mountpoint on 2.6 kernel clients.
3499 Severity : enhancement
3501 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3502 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3503 parallel, not waiting for response from every OST before sending
3504 a rpc to the next OST.
3509 Description: setattr vs write race when updating file timestamps
3510 Details : Client processes that update a file timestamp into the past
3511 right after writing to the file (e.g. tar) it is possible that
3512 the updated file modification time can be reset to the current
3513 time due to a race between processing the setattr and write RPC.
3515 Severity : enhancement
3517 Description: Bring 'lfs find' closer in line with regular Linux find.
3518 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3519 -print0 options and obtains all the needed info through the lustre
3522 Severity : enhancement
3524 Description: support up to 1024 configured devices on one node
3525 Details : change obd_dev array from statically allocated to dynamically
3526 allocated structs as they are first used to reduce memory usage
3531 Description: Flush dirty partially truncated pages during truncate
3532 Details : Immediatelly flush partially truncated pages in filter_setattr,
3533 this way we completely avoid having any pages in page cache on OST
3534 and can retire ugly workarounds during writes to flush such pages.
3539 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3540 Details : protect inode from truncation within vfs_unlink() context
3541 just take a reference before calling vfs_unlink() and release it
3542 when parent's i_sem is free.
3547 Description: last_id value checked outside lock on OST caused LASSERT failure
3548 Details : If there were multiple MDS->OST object precreate requests in
3549 flight, it was possible that the OST's last object id was checked
3550 outside a lock and incorrectly tripped an assertion. Move checks
3551 inside locks, and discard old precreate requests.
3554 Frequency : always, if extents are used on OSTs
3556 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3557 Details : OSTs using the extents format would not zero the high 16 bits of
3558 the index physical block number. This is not a problem for any
3559 OST filesystems smaller than 16TB, and no kernels support ext3
3560 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3561 new/modified files) and can be fixed for existing filesystems
3562 with e2fsprogs-1.39-cfs1.
3567 Description: import connection selection may be incorrect if timer wraps
3568 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3569 import connections to be ignored if the 32-bit jiffies counter
3570 wraps. Use a 64-bit jiffies counter.
3573 Frequency : very large clusters immediately after boot
3575 Description: LNET request buffers exhausted under heavy short-term load
3576 Details : If a large number of client requests are generated on a service
3577 that has previously never seen so many requests it is possible
3578 that the request buffer growth cannot keep up with the spike in
3579 demand. Instead of dropping incoming requests, they are held in
3580 the LND until the RPC service can accept more requests.
3583 Frequency : Sometimes during replay
3585 Description: Assertion failure in ll_local_open after replay.
3586 Details : If replay happened on an open request reply before we were able
3587 to set replay handler, reply will become not swabbed tripping the
3588 assertion in ll_local_open. Now we set the handler right after
3589 recognising of open request
3592 Frequency : very rare
3594 Description: kernel reports "badness in vsnprintf"
3595 Details : Reading from the "recovery_status" /proc file in small chunks
3596 may cause a negative length in lprocfs_obd_rd_recovery_status()
3597 call to vsnprintf() (which is otherwise harmless). Exit early
3598 if there is no more space in the output buffer.
3600 Severity : enhancement
3602 Description: clear OBD RPC statistics by writing to them
3603 Details : It is now possible to clear the OBD RPC statistics by writing
3604 to the "stats" file.
3609 Description: Client mtime is not the same on different clients after utimes
3610 Details : In some cases, the client was using the utimes() syscall on
3611 a file cached on another node. The clients now validate the
3612 ctime from the MDS + OSTs to determine which one is right.
3617 Description: Inability to activate failout mode
3618 Details : lconf script incorrectly assumed that in pythong string's numeric
3619 value is used in comparisons.
3622 Frequency : always with multiple stripes per file
3624 Description: Inefficient object allocation for mutli-stripe files
3625 Details : When selecting which OSTs to stripe files over, for files with
3626 a stripe count that divides evenly into the number of OSTs,
3627 the MDS is always picking the same starting OST for each file.
3628 Return the OST selection heuristic to the original design.
3633 Description: mount failures may take full timeout to return an error
3634 Details : Under some heavy load conditions it is possible that a
3635 failed mount can wait for the full obd_timeout interval,
3636 possibly several minutes, before reporting an error.
3637 Instead return an error as soon as the status is known.
3639 Frequency : quota enabled and large files being deleted
3641 Description: releasing more than 4GB of quota at once hangs OST
3642 Details : If a user deletes more than 4GB of files on a single OST it
3643 will cause the OST to spin in an infinite loop. Release
3644 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3649 Description: statfs data retrieved from /proc may be stale or zero
3650 Details : When reading per-device statfs data from /proc, in the
3651 {kbytes,files}_{total,free,avail} files, it may appear
3652 as zero or be out of date.
3655 Frequency : systems with MD RAID1 external journal devices
3657 Description: lconf's call to blkid is confused by RAID1 journal devices
3658 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3659 of returning all block devices that match the journal UUID.
3662 Frequency : always, for aggregate stripe size over 4GB
3664 Description: assertion fails when trying to use 4GB stripe size
3665 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3666 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3669 Frequency : always on ppc64
3671 Description: the first write on an ext3 filesystem with mballoc got stuck
3672 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3673 perform endianness conversion.
3675 ------------------------------------------------------------------------------
3677 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3679 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3680 INTEROPERATE with older versions automatically. Please read the
3681 user documentation before upgrading any part of a live system.
3682 * WARNING: Lustre networking configuration changes are required with
3683 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3686 * Support for kernels:
3687 2.6.9-22.0.2.EL (RHEL 4)
3688 2.6.5-7.244 (SLES 9)
3689 2.6.12.6 vanilla (kernel.org)
3692 Severity : enhancement
3693 Bugzilla : 7981/8208
3694 Description: Introduced Lustre Networking (LNET)
3695 Details : LNET is new networking infrastructure for Lustre, it includes
3696 a reorganized network configuration mode (see the user
3697 documentation for full details) as well as support for routing
3698 between different network fabrics. Lustre Networking Devices
3699 (LNDs) for the supported network fabrics have also been
3700 created for this new infrastructure.
3702 Severity : enhancement
3703 Description: Introduced Access control lists
3704 Details : clients can set ACLs on files and directories in order to have
3705 more fine-grained permissions than the standard Unix UGO+RWX.
3706 The MDS must be started with the "-o acl" mount option.
3708 Severity : enhancement
3709 Description: Introduced filesystem quotas
3710 Details : Administrators may now establish per-user quotas on the
3713 Severity : enhancement
3715 Description: Configuration change for the XT3
3716 The PTLLND is now used to run Lustre over Portals on the XT3
3717 The configure option(s) --with-cray-portals are no longer used.
3718 Rather --with-portals=<path-to-portals-includes> is used to
3719 enable building on the XT3. In addition to enable XT3 specific
3720 features the option --enable-cray-xt3 must be used.
3725 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3726 Details : A race between allocating a new llcd and re-getting the llcd_lock
3727 allowed another thread to grab newly-allocated llcd.
3729 Severity : enhancement
3731 Description: 2.6 OST async journal commit and locking fix to improve performance
3732 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3733 2.6 kernels are now async as they already were in 2.4 kernels so
3734 that they can commit concurrently with the network bulk transfer.
3735 For block-allocated files the filter allocation semaphore is held
3736 to avoid filesystem fragmentation during allocation. BKL lock
3737 removed for 2.6 xattr operations where it is no longer needed.
3742 Description: lconf incorrectly determined whether two IP networks could talk
3743 Details : In some more complicated routing and multiple-network
3744 configurations, lconf will avoid trying to make a network
3745 connection to a disjoint part of the IP space. It was doing the
3746 math incorrectly for one set of cases.
3751 Description: Fix for potential infinite loop processing records in an llog.
3752 Details : If an llog record is corrupted/zeroed, it is possible to loop
3753 forever in llog_process(). Validate the llog record length
3754 and skip the remainder of the block on error.
3757 Frequency : occasional (liblustre only)
3759 Description: liblustre could not open files whose last component is a symlink
3760 Details : sysio_path_walk() would incorrectly pass the open intent to
3761 intermediate path components.
3764 Frequency : rare (liblustre only with non-standard tuning)
3765 Bugzilla : 7201 (7350)
3766 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3767 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3768 after it has been released, only do this for referenced locks
3770 Severity : enhancement
3772 Description: specifying an (invalid) directory default stripe_size of -1
3773 would reset the directory default striping
3774 Details : stripe_size -1 was used internally to signal directory stripe
3775 removal, now use "all default" to signal dir stripe removal
3776 as a directory striping of "all default" is not useful
3779 Frequency : common for large clusters running liblustre clients
3781 Description: doing an ls when liblustre clients are running is slow
3782 Details : sending a glimpse AST to a liblustre client waits for every AST
3783 to time out, as liblustre clients will not respond. Since they
3784 cannot cache data we refresh the OST lock LVB from disk instead.
3786 Severity : enhancement
3788 Description: doing an ls at the same time as file IO can be slow
3789 Details : enqueue and other "small" requests can be blocked behind many
3790 large IO requests. Create a new OST IO portal for non-IO
3791 requests so they can be processed faster.
3794 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3796 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3797 created file with mode 000
3798 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3799 allow an ftruncate() as a non-root user to a file with mode 000.
3800 Lustre now respects this flag to disable mode checks when
3801 truncating a file owned by the user
3804 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3807 Description: Revoking locks from clients that went dead or catatonic might take
3809 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3810 cancellation of such locks instant on servers without waiting for
3811 any reply from clients. Clients drops these locks when cancel
3812 notification from server is received without replying.
3815 Frequency : liblustre-only, when liblustre client dies or becomes busy
3817 Description: Doing ls on Linux clients can take a long time with active
3819 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3820 granting such locks to it in the first place if possible. Locks
3821 are taken by proxy on the OST during the read or write and
3822 dropped immediately afterward. Add connect flags handling, do
3823 not grant locks to liblustre clients for glimpse ASTs.
3825 Severity : enhancement
3827 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3828 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3829 sequential reads and behaves badly with applications doing random
3830 reads. Improve it by reading ahead at least read region, and
3831 avoiding excessive large RPC for small reads.
3833 Severity : enhancement
3835 Description: Creating more than 1000 files for a single job may cause a load
3836 imbalance on the OSTs if there are also a large number of OSTs.
3837 Details : qos_prep_create() uses an OST index reseed value that is an
3838 even multiple of the number of available OSTs so that if the
3839 reseed happens in the middle of the object allocation it will
3840 still utilize the OSTs as uniformly as possible.
3845 Description: OST or MDS may oops in ping_evictor_main()
3846 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3847 but doesn't restart at beginning of obd_exports_timed list
3850 Severity : enhancement
3852 Description: improve by-nid export eviction on the MDS and OST
3853 Details : allow multiple exports with the same NID to be evicted at one
3854 time without re-searching the exports list.
3857 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3859 Description: MDS may oops in groups_free()
3860 Details : in rare race conditions a newly allocated group_info struct is
3861 freed again, and this can be NULL. The 2.4 compatibility code
3862 for groups_free() checked for a NULL pointer, but 2.6 did not.
3865 Frequency : common for liblustre clients doing little filesystem IO
3866 Bugzilla : 9352, 7313
3867 Description: server may evict liblustre clients accessing contended locks
3868 Details : if a client is granted a lock or receives a completion AST
3869 with a blocking AST already set it would not reply to the AST
3870 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3873 Frequency : lfs setstripe, only systems with more than 160 OSTs
3875 Description: unable to set striping with a starting offset beyond OST 160
3876 Details : llapi_create_file() incorrectly limited the starting stripe
3877 index to the maximum single-file stripe count.
3880 Frequency : LDAP users only
3882 Description: lconf did not handle in-kernel recovery with LDAP properly
3883 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3885 Severity : enhancement
3887 Description: bind OST threads to NUMA nodes to improve performance
3888 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3889 node and do their allocations there to localize memory access
3891 Severity : enhancement
3893 Description: llmount can determine client NID directly from Myrinet (GM)
3894 Details : the client NID code from gmnalnid was moved directly into
3895 llmount, removing the need to use this or specifying the
3896 client NID explicitly when mounting GM clients with zeroconf
3899 Frequency : if client is started with down MDS
3901 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3902 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3903 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3904 ptlrpc_import_delayed_req() because we didn't check if the
3905 request was interrupted, and we also didn't break out of the
3906 event loop if there was no timeout
3911 Description: data loss during non-page-aligned writes to a single file from
3912 both multiple nodes and multiple threads on one node at same time
3913 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3914 inconsistency led to false short-reads, that were cached and later
3915 used by ->prepare_write() to fill in partially written page,
3916 leading to data loss.
3919 Frequency : always, if lconf --abort_recovery used
3921 Description: lconf --abort_recovery fails with 'Operation not supported'
3922 Details : lconf was attempting to abort recovery on the MDT device and not
3925 Severity : enhancement
3927 Description: remove cleanup logs
3928 Details : replace lconf-generated cleanup logs with lustre internal
3929 cleanup routines. Eliminates the need for client-cleanup and
3932 Severity : enhancement
3934 Description: add support for EAs (user and system) on lustre filesystems
3935 Details : it is now possible to store extended attributes in the Lustre
3936 client filesystem, and with the user_xattr mount option it
3937 is possible to allow users to store EAs on their files also
3939 Severity : enhancement
3941 Description: Add possibility (config option) to show minimal available OST free
3943 Details : When compiled with --enable-mindf configure option, statfs(2)
3944 (and so, df) will return least minimal free space available from
3945 all OSTs as amount of free space on FS, instead of summary of
3946 free spaces of all OSTs.
3948 Severity : enhancement
3950 Description: do not expand extent locks acquired on OST-side
3951 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3952 by server: they are not cached anyway.
3955 Frequency : when mmap is used/binaries executed from Lustre
3957 Description: Unmmap pages before throwing them away from read cache.
3958 Details : llap_shrink cache now attempts to unmap pages before discarding
3959 them (if unmapping failed - do not discard). SLES9 kernel has
3960 extra checks that trigger if this unmapping is not done first.
3965 Description: lconf didn't resolve symlinks before checking to see whether a
3966 given mountpoint was already in use
3969 Frequency : when migrating failover services
3970 Bugzilla : 6395, 9514
3971 Description: When migrating a subset of services from a node (e.g. failback
3972 from a failover service node) the remaining services would
3973 time out and evict clients.
3974 Details : lconf --force (implied by --failover) sets the global obd_timeout
3975 to 5 seconds in order to quickly disconnect, but this caused
3976 other RPCs to time out too quickly. Do not change the global
3977 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3979 Severity : enhancement
3980 Frequency : if MDS is started with down OST
3981 Bugzilla : 9439,5706
3982 Description: Allow startup/shutdown of an MDS without depending on the
3983 availability of the OSTs.
3984 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3985 Add appropriate locking and lov-osc refcounts for safe
3986 cleaning. Add osc abort_inflight calls in case the
3987 synchronize never started.
3990 Frequency : occasional (Cray XT3 only)
3992 Description: root not authorized to access files in CRAY_PORTALS environment
3993 Details : The client process capabilities were not honoured on the MDS in
3994 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3995 been accessed by an authorized user then root was able to access
3996 the file on the local client also. The root user capabilities
3997 are now allowed on the MDS, as this environment has secure UID.
4000 Frequency : occasional
4002 Description: ldiskfs "too long searching" message happens too often
4003 Details : A debugging message (otherwise harmless) prints too often on
4004 the OST console. This has been reduced to only happen when
4005 there are fragmentation problems on the filesystem.
4010 Description: Division by zero in statfs when all OSCs are inactive
4011 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4012 lov_statfs divides by value returned by lov_get_stripecnt().
4016 Bugzilla : 9489, 3273
4017 Description: First write from each client to each OST was only 4kB in size,
4018 to initialize client writeback cache, which caused sub-optimal
4019 RPCs and poor layout on disk for the first writen file.
4020 Details : Clients now request an initial cache grant at (re)connect time
4021 and so that they can start streaming writes to the cache right
4022 away and always do full-sized RPCs if there is enough data.
4023 If the OST is rebooted the client also re-establishes its grant
4024 so that client cached writes will be honoured under the grant.
4029 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4030 Details : Now I/O RPCs go to different portal number and (presumably) fast
4031 lock requests (and glimses) and other RPCs get their own service
4032 threads pool that should be able to service those RPCs
4035 Severity : enhancement
4037 Description: Ability to exchange lustre version between client and servers and
4038 issue warnings at client side if client is too old. Also for
4039 liblustre clients there is ability to refuse connection of too old
4041 Details : New 'version' field is added to connect data structure that is
4042 filled with version info. That info is later checked by server and
4046 Frequency : rare, liblustre only.
4047 Bugzilla : 9296, 9581
4048 Description: Two simultaneous writes from liblustre at offset within same page
4049 might proceed at the same time overwriting eachother with stale
4051 Details : I/O lock withing llu_file_prwv was released too early, before data
4052 actually was hitting the wire. Extended lock-holding time until
4053 server acknowledges receiving data.
4056 Frequency : extremely rare. Never observed in practice.
4058 Description: avoid generating lustre_handle cookie of 0.
4059 Details : class_handle_hash() generates handle cookies by incrementing
4060 global counter, and can hit 0 occasionaly (this is unlikely, but
4061 not impossible, because initial value of cookie counter is
4062 selected randonly). Value of 0 is used as a sentinel meaning
4063 "unassigned handle" --- avoid it. Also coalesce two critical
4064 sections in this function into one.
4066 Severity : enhancement
4068 Description: allow liblustre clients to delegate truncate locking to OST
4069 Details : To avoid overhead of locking, liblustre client instructs OST to
4070 take extent lock in ost_punch() on client's behalf. New connection
4071 flag is added to handle backward compatibility.
4073 Severity : enhancement
4074 Bugzilla : 4928, 7341, 9758
4075 Description: allow number of OST service threads to be specified
4076 Details : a module parameter allows the number of OST service threads
4077 to be specified via "options ost ost_num_threads={N}" in the
4078 OSS's /etc/modules.conf or /etc/modprobe.conf.
4082 Bugzilla : 6146, 9635, 9895
4083 Description: servers crash with bad pointer in target_handle_connect()
4084 Details : In rare cases when a client is reconnecting it was possible that
4085 the connection request was the last reference for that export.
4086 We would temporarily drop the export reference and get a new
4087 one, but this may have been the last reference and the export
4088 was just destroyed. Get new reference before dropping old one.
4090 Severity : enhancement
4091 Frequency : if client is started with failover MDS
4093 Description: Allow multiple MDS hostnames in the mount command
4094 Details : Try to read the configuration from all specified MDS
4095 hostnames during a client mount in case the "primary"
4098 Severity : enhancement
4100 Description: Stop sending data to evicted clients as soon as possible.
4101 Details : Check if the client we are about to send or are sending data to
4102 was evicted already. (Check is done every second of waiting,
4103 for which l_wait_event interface was extended to allow checking
4104 of exit condition at specified intervals).
4107 Frequency : rare, normally only when NFS exporting is done from client
4109 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4111 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4114 Frequency : always, if srand() is called before liblustre initialization
4116 Description: Liblustre uses system PRNG disturbing its usage by user application
4117 Details : Introduce internal to lustre fast and high-quality PRNG for
4118 lustre usage and make liblustre and some other places in generic
4119 lustre code to use it.
4121 Severity : enhancement
4122 Bugzilla : 9477, 9557, 9870
4123 Description: Verify that the MDS configuration logs are updated when xml is
4124 Details : Check if the .xml configuration logs are newer than the config
4125 logs stored on the MDS and report an error if this is the case.
4126 Request --write-conf, or allow starting with --old_conf.
4128 Severity : enhancement
4130 Description: Handle symlinks in the path when checking if Lustre is mounted.
4131 Details : Resolve intermediate symlinks when checking if a client has
4132 mounted a filesystem to avoid duplicate client mounts.
4137 Description: lconf can hit an error exception but still return success.
4138 Details : The lconf command catches the Command error exception at the top
4139 level script context and will exit with the associated exit
4140 status, but doesn't ensure that this exit status is non-zero.
4145 Description: failure of ptlrpc thread startup can cause oops
4146 Details : Starting a ptlrpc service thread can fail if there are a large
4147 number of threads or the server memory is very fragmented.
4148 Handle this without oopsing.
4151 Frequency : always, only if liblustre and non-default acceptor port was used
4153 Description: liblustre cannot connect to servers with non-default acceptor port
4154 Details : tcpnal_set_default_params() was not called and was therefore
4155 ignoring the environment varaible TCPNAL_PORT, as well as other
4156 TCPNAL_ environment variables
4161 Description: two objects could be created on the same OST for a single file
4162 Details : If an OST is down, in some cases it was possible to create two
4163 objects on a single OST for a single file. No problems other
4164 than potential performance impact and spurious error messages.
4168 Bugzilla : 5681, 9562
4169 Description: Client may oops in ll_unhash_aliases
4170 Details : Client dcache may become inconsistent in race condition.
4171 In some cases "getcwd" can fail if the current directory is
4177 Description: Inode refcounting problems in NFS export code
4178 Details : link_raw functions used to call d_instantiate without obtaining
4179 extra inode reference first.
4183 Bugzilla : 9942, 9903
4184 Description: Referencing freed requests leading to crash, memleaks with NFS.
4185 Details : We used to require that call to ll_revalidate_it was always
4186 followed by ll_lookup_it. Also with revalidate_special() it is
4187 possible to call ll_revalidate_it() twice for the same dentry
4188 even if first occurence returned success. This fix changes semantic
4189 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4190 reference on a request referred from the intent.
4191 ll_intent_release() then releases such a request.
4194 Frequency : rare, normally benchmark loads only
4196 Description: unlinked inodes were kept in memory on the client
4197 Details : If a client is repeatedly creating and unlinking files it
4198 can accumulate a lot of stale inodes in the inode slab cache.
4199 If there is no other client load running this can cause the
4200 client node to run out of memory. Instead flush old inodes
4201 from client cache that have the same inode number as a new inode.
4204 Frequency : SLES9 2.6.5 kernel and long filenames only
4205 Bugzilla : 9969, 10379
4206 Description: utime reports stale NFS file handle
4207 Details : SLES9 uses out-of-dentry names in some cases, which confused
4208 the lustre dentry revalidation. Change it to always use the
4212 Frequency : rare, unless heavy write-truncate concurrency is continuous
4213 Bugzilla : 4180, 6984, 7171, 9963, 9331
4214 Description: OST becomes very slow and/or deadlocked during object unlink
4215 Details : filter_destroy() was holding onto the parent directory lock
4216 while truncating+unlinking objects. For very large objects this
4217 may block other threads for a long time and slow overall OST
4218 responsiveness. It may also be possible to get a lock ordering
4219 deadlock in this case, or run out of journal credits because of
4220 the combined truncate+unlink. Solution is to do object truncate
4221 first in one transaction without parent lock, and then do the
4222 final unlink in a new transaction with the parent lock. This
4223 reduces the lock hold time dramatically.
4226 Frequency : rare, 2.4 kernels only
4228 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4229 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4230 before unmounting their filesystem in order to avoid deadlock.
4231 The kernel_locked() function in 2.4 kernels only checks whether
4232 the kernel lock is held, not whether it is this process that is
4233 holding it as 2.6 kernels do.
4238 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4239 Details : The client ptlrpc code may be trying to reconnect to a down
4240 server before a previous connection attempt has timed out.
4241 Increase the reconnect interval to be longer than the connection
4242 timeout interval to avoid sending duplicate connections to
4246 Frequency : echo_client brw_test command
4248 Description: fix echo_client to work with OST preallocated code
4249 Details : OST preallocation code (5137) didn't take echo_client IO path
4250 into account: echo_client calls filter methods outside of any
4251 OST thread and, hence, there is no per-thread preallocated
4252 pages and buffers to use. Solution: hijack pga pages for IO. As
4253 a byproduct, this avoids unnecessary data copying.
4257 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4258 Description: Client can oops in mdc_commit_close() after open replay
4259 Details : It was possible for the MDS to return an open request with no
4260 transaction number in mds_finish_transno() if the client was
4261 evicted, but without actually returning an error. Clients
4262 would later try to replay that open and may trip an assertion
4263 Simplify the client close codepath, and always return an error
4264 from the MDS in case the open is not successful.
4267 Frequency : rare, 2.6 OSTs only
4269 Description: OST may deadlock under high load on fragmented files
4270 Details : If there was a heavy load and highly-fragmented OST filesystems
4271 it was possible to have all the OST threads deadlock waiting on
4272 allocation of biovecs, because the biovecs were not released
4273 until the entire RPC IO was completed. Instead, release biovecs
4274 as soon as they are complete to ensure forward IO progress.
4276 Severity : enhancement
4278 Description: Support for specifying external journal device at mount
4279 Details : If an OST or MDS device is formatted with an external journal
4280 device, this device major/minor is stored in the ext3 superblock
4281 and may not be valid for failover. Allow detecting and
4282 specifying the external journal at mount time.
4287 Description: Mounting an MDS with pending unlinked files may cause oops
4288 Details : target_finish_recovery() calls mds_postrecov() which returned
4289 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4290 considers this an error and immediately begins cleaning up the
4291 lov, just after starting the mds_lov process
4293 Severity : enhancement
4295 Description: Implement 'lfs df' to report actual free space on per-OST basis
4296 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4297 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4298 usage of objects; '-h' to report in human readable format.
4300 ------------------------------------------------------------------------------
4302 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4309 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4310 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4311 at startup during orphan unlinking. Instead, skip buddy update
4312 before setup, as it will be regenerated after recovery is complete.
4317 Description: If an OST is inactive, its locks might reference stale inodes.
4318 Details : lov_change_cbdata() must iterate over all namespaces, even if
4319 they are inactive to clear inode references from the lock.
4321 Severity : enhancement
4322 Frequency : occasional, if non-standard max_dirty_mb used
4324 Description: Client will block write RPCs if not enough grant
4325 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4326 then the client will block writes while waiting for another RPC
4327 to complete instead of consuming its dirty limit. With change
4328 we get improved performance when max_dirty_mb is small.
4330 Severity : enhancement
4331 Bugzilla : 3389, 6253
4332 Description: Add support for supplementary groups on the MDS.
4333 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4334 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4335 lookups for user supplementary groups into a cache.
4339 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4340 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4341 for liblustre O_EXCL is enforced by MDS.
4345 Description: readdir never returns NULL in liblustre.
4346 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4347 dentry in struct dirent.
4351 Frequency : liblustre only, depends on application IO pattern
4352 Description: liblustre clients evicted if not contacting servers
4353 Details : Don't put liblustre clients into the ping_evictor list, so
4354 they will not be evicted by the pinger ever.
4356 Severity : enhancement
4358 Description: Add ability to evict clients by NID from MDS.
4359 Details : By echoing "nid:$NID" string into
4360 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4361 $NID would be instantly evicted from this MDS and from all active
4362 OSTs connected to it.
4366 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4367 Details : lookup_it_finish() used to query file size from OSTs that was not
4372 Description: service threads change working directory to that of init
4373 Details : Starting lustre service threads may pin the working directory
4374 of the parent thread, making that filesystem busy. Threads
4375 now change to the working directory of init to avoid this.
4379 Frequency : during shutdown only
4380 Description: shutdown with a failed MDS or OST can cause unmount to hang
4381 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4386 Frequency : 2.6 only
4387 Description: chmod/chown may include an extra supplementary group
4388 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4389 supplementary group and if none is specified this is used.
4392 Bugzilla : 5479 (6816)
4394 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4395 Details : If lookup is in progress on a file that is unlinked we might try
4396 to revalidate the inode and fail in revalidate after lookup is
4397 complete and ll_file_open() enqueues the open again but
4398 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4401 Frequency : always, if lconf --abort_recovery used
4403 Description: lconf --abort_recovery fails with 'Operation not supported'
4404 Details : lconf was attempting to abort recovery on the MDT device and not
4407 ------------------------------------------------------------------------------
4409 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4414 Frequency : rare (only unsupported configurations with a node running as an
4416 Bugzilla : 6514, 5137
4417 Description: Mounting a Lustre file system on a node running as an OST could
4419 Details : OSTs now preallocates memory needed to write out data at
4420 startup, instead of when needed, to avoid having to
4421 allocate memory in possibly low memory situations.
4422 Specifically, if the file system is mounted on on OST,
4423 memory pressure could force it to try to write out data,
4424 which it needed to allocate memory to do. Due to the low
4425 memory, it would be unable to do so and the node would
4426 become unresponsive.
4428 Severity : enhancement
4430 Description: Addition of lconf --service command line option
4431 Details : lconf now accepts a '--service <arg>' option, which is
4432 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4434 Severity : enhancement
4436 Description: Failover mode is now the default for OSTs.
4437 Details : By default, OSTs will now run in failover mode. To return to
4438 the old behaviour, add '--failout' to the lmc line for OSTs.
4440 Severity : enhancement
4442 Description: Health checks are now provided for MDS and OSTs
4443 Details : Additional detailed health check information on MSD and OSTs
4444 is now provided through the procfs health_check value.
4447 Frequency : occasional, depends on IO load
4449 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4450 after numerous create/delete cycles
4451 Details : The ext3 inode allocation policy would not allocate new inodes
4452 very well on the OSTs because there are no new directories
4453 being created. Instead we look for groups with free space if
4454 the parent directories are nearly full.
4459 Description: Network or server problems during mount may cause partially
4460 mounted clients instead of returning an error.
4461 Details : The config llog parsing code may overwrite the error return
4462 code during mount error handling, returning success instead
4468 Description: MDS can fail to allocate large reply buffers
4469 Details : After long uptimes the MDS can fail to allocate large reply
4470 buffers (e.g. zconf client mount config records) due to memory
4471 fragmentation or consumption by the buffer cache. Preallocate
4472 some large reply buffers so that these replies can be sent even
4473 under memory pressure.
4477 Frequency : rare (liblustre)
4478 Description: fsx running with liblustre complained that using truncate() to
4479 extend the file doesn't work. This patch corrects that issue.
4480 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4481 ATTR_SIZE and lsm use in llu_setattr_raw.
4485 Frequency : rare, only 2.6 kernels
4486 Description: Unusual file access patterns on the MDS may result in inode
4487 data being lost in very rare circumstances.
4488 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4489 code in the 2.6 kernel caused the inode and/or EA data not to
4490 be read from disk, causing single-file corruption.
4494 Frequency : rare, only 2.6 filesystems using extents
4495 Description: Heavy concurrent write and delete load may cause data corruption.
4496 Details : It was possible under high-load situations to have an extent
4497 metadata block in the block device cache from a just-unlinked
4498 file overwrite a newly-allocated data block. We now unmap any
4499 metadata buffers that alias just-allocated data blocks.
4503 Frequency : filesystems with default stripe_count larger than 77
4504 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4505 Details : lconf specifies an inode size of 4096 bytes when the default
4506 stripe_count is larger than 77. This conflicts with the default
4507 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4508 this case to avoid pinning too much memory for large EAs.
4510 ------------------------------------------------------------------------------
4512 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4517 Frequency : rare (extremely heavy IO load with hundreds of clients)
4519 Description: Client is evicted, gets IO error writing to file
4520 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4521 caused clients to be evicted for AST timeouts. The fixes in
4522 bug 5192 mean we no longer need to have such short AST timeouts
4523 so ldlm_timeout has been increased.
4526 Frequency : occasional during --force or --failover shutdown under load
4527 Bugzilla : 5949, 4834
4528 Description: Server oops/LBUG if stopped with --force or --failover under load
4529 Details : a collection of import/export refcount and cleanup ordering
4530 issues fixed for safer force cleanup
4533 Frequency : only filesystems larger than 120 OSTs
4534 Bugzilla : 5990, 6223
4535 Description: lfs getstripe would oops on a very large filesystem
4536 Details : lov_getconfig used kfree on vmalloc'd memory
4539 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4540 Bugzilla : 6242, 6243
4541 Description: reading from files that had been truncated to a non-zero size
4542 but never opened returned no data
4543 Details : ll_file_read() reads zeros from no-object files to EOF
4548 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4549 Details : There's little harm in aborting MDS/OSS recovery and letting it
4550 try again, so I removed the LASSERT and return an error instead.
4552 Severity : enhancement
4554 Description: New debugging infrastructure for tracking down data corruption
4555 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4556 (b) cover more of the client-side code path, and (c) try to narrow
4557 down where problems occurred
4561 Bugzilla : 3819, 4364, 4397, 6313
4562 Description: Racing close and eviction MDS could cause assertion in mds_close
4563 Details : It was possible to get multiple mfd references during close and
4564 client eviction, leading to one thread referencing a freed mfd.
4566 Severity: : enhancement
4567 Bugzilla : 3262, 6359
4568 Description: Attempts to reconnect to servers are now more aggressive.
4569 Details : This builds on the enhanced upcall-less recovery that was added
4570 in 1.4.2. When trying to reconnect to servers, clients will
4571 now try each server in the failover group every 10 seconds. By
4572 default, clients would previously try one server every 25 seconds.
4577 Description: After recovery, certain operations trigger a failed
4578 assertion on a client.
4579 Details : Failing over an mds, using lconf -d --failover, while a
4580 client was doing a readdir() call would cause the client to
4581 LBUG after recovery completed and the readdir() was resent.
4583 Severity : enhancement
4585 Description: Default groups are now added by lconf
4586 Details : You can now run lconf --group <servicename> without having to
4587 manually add groups with lmc.
4590 Frequency : occasional
4592 Description: Nodes with an elan id of 0 trigger a failed assertion
4595 Frequency : always when accessing e.g. tty/console device nodes
4597 Description: tty and some other devices nodes cannot be used on lustre
4598 Details : file's private_data field is used by device data and lustre
4599 values in there got lost. New field was added to struct file to
4600 store fs-specific private data.
4603 Frequency : when exporting Lustre via NFS
4605 Description: NFSD failed occasionally when looking up a path component
4606 Details : NFSD is looking up ".." which was broken in ext3 directories
4607 that had grown large enough to become hashed.
4610 Frequency : Clusters with multiple interfaces not on the same subnet
4612 Description: Nodes will repeatedly try to reconnect to an interface which it
4613 cannot reach and report an error to the log.
4614 Details : Extra peer list entries will be created by lconf with some peers
4615 unreachable. lconf now validates the peer before adding it.
4618 Frequency : Only if a default stripe is set on the filesystem root.
4620 Description: Setting a default stripe on the filesystem root prevented the
4621 filesystem from being remounted.
4622 Details : The client was sending extra request flags in the root getattr
4623 request and did not allocate a reply buffer for the dir EA.
4626 Frequency : occasional, higher if lots of files are accessed by one client
4627 Bugzilla : 6159, 6097
4628 Description: Client trips assertion regarding lsm mismatch/magic
4629 Details : While revalidating inodes the VFS looks up inodes with ifind()
4630 and in rare cases can find an inode that is being freed.
4631 The ll_test_inode() code will free the lsm during ifind()
4632 when it finds an existing inode and then the VFS later attaches
4633 this free lsm to a new inode.
4637 Bugzilla : 6422, 7030
4638 Description: MDS deadlock between mkdir and client eviction
4639 Details : Creating a new file via mkdir or mknod (starting a transaction
4640 and getting the ns lock) can deadlock with client eviction
4641 (gets ns lock and trying to finish a synchronous transaction).
4644 Frequency : occasional
4645 Description: While starting a server, the fsfilt_ext3 module could not be
4647 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4648 kernels. Previously, lconf would still use the ext3 name
4649 when trying to load modules. Now, it will correctly use
4650 ext3 on 2.4 and ldiskfs on 2.6.
4652 Severity : enhancement
4653 Description: The default stripe count has been changed to 1
4654 Details : The interpretation of the default stripe count (0, to lfs
4655 or lmc) has been changed to mean striping across a single
4656 OST, rather than all available. For general usage we have
4657 found a stripe count of 1 or 2 works best.
4659 Severity : enhancement
4660 Description: Add support for compiling against Cray portals.
4661 Details : Conditional compiling for some areas that are different
4665 Frequency : occasional
4666 Bugzilla : 6409, 6834
4667 Description: Creating files with an explicit stripe count may lead to
4668 a failed assertion on the MDS
4669 Details : If some OSTs are full or unavailable, creating files may
4670 trigger a failed assertion on the MDS. Now, Lustre will
4671 try to use other servers or return an error to the
4675 Frequency : occasional
4677 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4678 caused lock timeout during readahead (since 1.4.2).
4679 Details : Processes doing readahead might match a lock that hasn't been
4680 granted yet if there are overlapping and conflicting lock
4681 requests. The readahead process waits on ungranted lock
4682 (original lock is CBPENDING), while OST waits for that process
4683 to cancel CBPENDING read lock and eventually evicts client.
4685 Severity : enhancement
4687 Description: Initial enabling of flock support for clients
4688 Details : Implements fcntl advisory locking and file status functions.
4689 This feature is provided as an optional mount flag (default
4690 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4691 locking are implemented yet, and those that are are not guaranteed
4692 to be completely correct in production environments.
4693 mount -t lustre -o [flock|noflock] ...
4696 Frequency : occasional
4698 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4699 assertion in the ext3 JBD (journaling) layer.
4700 Details : The b_committed_data struct is protected by the big kernel lock
4701 in 2.4 kernels, serializing journal_commit_transaction() and
4702 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4703 access to this struct. In 2.6 kernels there is finer grained
4704 locking to improve SMP performance of the JBD layer.
4708 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4713 Description: merge and cleanup kernel patches.
4716 -----------------------------------------------------------------------------
4718 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4720 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4721 versions of Lustre. Please update all servers and clients to
4722 version 1.4.2 or later at the same time. You must also run
4723 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4724 to update the configuration logs.
4726 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4727 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4728 - swab llog records of type '0' so we get proper header size/idx (5861)
4729 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4730 - fix rename of one directory over another leaking an inode (5953)
4731 - avoid SetPageDirty on 2.6 (5981)
4732 - don't re-add just-being-destroyed locks to the waiting list (5653)
4733 - when creating new directories, inherit the parent's custom
4734 striping settings if present parent (3048)
4735 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4736 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4737 - don't hold client locks on temporary worklist from l_lru (5666)
4738 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4739 - automatically evict dead clients (5921)
4740 - Update file size properly in create+truncate+fstat case (6196)
4741 - Do not unhash mountpoint dentries, do not allow removal of
4743 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4744 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4745 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4746 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4747 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4748 - Correctly handle reads of files with no objects (6243)
4749 - lctl recover will also mark a device active if deactivate used (5933)
4751 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4752 - allow --write-conf on an MDS with different nettype than client (5619)
4753 - don't write config llogs to MDS for mounts not from that MDS (5617)
4754 - lconf should create multiple TCP connections from a client (5201)
4755 - init scripts are now turned off by default; run chkconfig --on
4756 lustre and chkconfig --on lustrefs to use them
4757 - upcalls are no longer needed for clients to recover to failover
4759 - add --abort-recovery option to lconf to abort recovery on device
4761 - add support for an arbitrary number of OSTs (3026)
4762 - Quota support protocol changes.
4763 - forward compatibility changes to wire structs (6007)
4764 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4765 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4766 - improved "lustre status" script
4767 - initialize blocksize for non-regular files (6062)
4768 - added --disable-server and --disable-client configure options (5782)
4769 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4770 - Vanilla 2.4.29 support
4771 - increase maximum number of obd devices to 520 (6242)
4772 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4773 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4774 SLES 9 2.6.5 kernels
4775 - limit stripes per file to 160 (the maximum EA size) (6093)
4777 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4780 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4781 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4782 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4783 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4784 - small ext3 extents cleanups and fixes (5733)
4785 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4786 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4787 - avoid needless client->OST connect, fix handle mismatch (5317)
4788 - fix DLM error path that led to out-of-sync client, long delays (5779)
4789 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4790 - fix several locking issues related to i_size (5492,5624,5654,5672)
4791 - don't move pending lock onto export if it is already evicted (5683)
4792 - fix kernel oops when creating .foo in unlinked directory (5548)
4793 - fix deadlock in obdfilter statistics vs. object create (5811)
4794 - use time_{before,after} to avoid timer jiffies wrap (5882)
4795 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4796 - Do not leak request if server was not able to process it (5154)
4797 - If mds_open unable to find parent dir, make that negative lookup(5154)
4798 - don't create new directories with extent-mapping (5909, 5936)
4800 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4801 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4802 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4803 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4804 - support for RHEL4 kernel on i686 (5773)
4805 - provide error messages when incompatible logs are encountered (5898)
4807 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4808 * version 1.4.0.10 (1.4.1 release candidate 1)
4810 - don't keep a lock reference when lock is not granted (4238)
4811 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4812 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4813 - drop import inflight refcount on signal_completed_replay error (5255)
4814 - unlock page after async write error during send (3677)
4815 - handle missing objects in filter_preprw_read properly (5265)
4816 - no transno return for symlink open, don't save no-trasno open (3440)
4817 - don't try to complete elan receive that already failed (4012)
4818 - free RPC server reply state on error (5406)
4819 - clean up thread from ptlrpc_start_thread() on error (5160)
4820 - readahead could read extra page into cache that wasn't ejected (5388)
4821 - prevent races in class_attach/setup/cleanup/detach (5260)
4822 - don't dereference de->d_inode after l_dput of de (5458)
4823 - use "int" for stripe value returned from lock_to_stripe (5544)
4824 - mballoc allocation and error-checking fixes in 2.6 (5504)
4825 - block device patches to fix I/O request sizes in 2.6 (5482)
4826 - look up hostnames for IB nals (5602)
4827 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4828 - don't start multiple acceptors for the same port (5277)
4829 - fix incorrect LASSERT in mds_getattr_name (5635)
4830 - export a proc file for general "ping" checking (5628)
4831 - fix "lfs check" to not block when the MDS is down (5628)
4833 - service request history (4965)
4834 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4835 - create an "evict_client" /proc entry on OSTs, like the MDS has
4836 - fix mount usage message, return errors per mount(8) (5168)
4837 - change grep [] to grep "[]" in tests so they work in more UMLs
4838 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4839 - remove ancient LOV_MAGIC_V0 EA support (5047)
4840 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4841 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4842 - allow readahead allocations to fail when low on memory (5383)
4843 - mmap locking landed again, after considerable improvement (2828)
4844 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4846 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4849 - send OST transaction number in read/write reply to free req (4966)
4850 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4851 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4852 - the watchdog thread now runs as interruptible (5246)
4853 - flock/lockf fixes (but it's still disabled, pending 5135)
4854 - don't use EXT3 constants in llite code (5094)
4855 - memory shortage at startup could cause assertion (5176)
4857 - reorganization of lov code
4858 - single portals codebase
4860 - add extents/mballoc support (5025)
4861 - direct I/O reads in the obdfilter (4048)
4862 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4864 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4867 - send OST transaction number in read/write reply to free req (4966)
4868 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4869 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4870 - the watchdog thread now runs as interruptible (5246)
4871 - handle missing objects in filter_preprw_read properly (5265)
4872 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4873 - drop import inflight refcount on signal_completed_replay error (5255)
4874 - unlock page after async write error during send (3677)
4875 - return original error code on reconstructed replies (3761)
4876 - no transno return for symlink open, don't save no-trasno open (3440)
4878 - add pid to ldlm debugging output (4922)
4879 - bump the watchdog timeouts -- we can't handle 30sec yet
4880 - extra debugging for orphan dentry/inode bug (5259)
4882 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4885 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4886 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4887 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4888 - fix NULL dereference in /proc/sys/portals/routes (4827)
4889 - allow failed mdc_close() operations to be interrupted (4561)
4890 - stop precreate on OST before MDS would time out on it (4778)
4891 - don't send partial-page writes before EOF from client (4410)
4892 - discard client grant for sub-page writes on large-page clients (4520)
4893 - don't free dentries not owned by NFS code, check generation (4806)
4894 - fix lsm leak if mds_create_objects() fails (4801)
4895 - limit debug_daemon file size, always print CERROR messages (4789)
4896 - use transno after validating reply (3892)
4897 - process timed out requests if import state changes (3754)
4898 - update mtime on OST during writes, return in glimpse (4829)
4899 - add mkfsoptions to LDAP (4679)
4900 - use ->max_readahead method instead of zapping global ra (5039)
4901 - don't interrupt __l_wait_event() during strace
4903 - add software watchdogs to catch hung threads quickly (4941)
4904 - make lustrefs init script start after nfs is mounted
4905 - fix CWARN/ERROR duplication (4930)
4906 - return async write errors to application if possible (2248)
4907 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4908 - print NAL number in %x format (4645)
4909 - update barely-supported suse-2.4.21-171 series (4842)
4910 - support for sles 9 %post scripts
4911 - support for building 2.6 kernel-source packages
4912 - support for sles km_* packages
4914 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4917 - ignore -ENOENT errors in osc_destroy (3639)
4918 - notify osc create thread that OSC is being cleaned up (4600)
4919 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4920 - reconstruct ost_handle() like mds_handle() (4657)
4921 - create a new thread to do import eviction to avoid deadlock (3969)
4922 - let lconf resolve symlinked-to devices (4629)
4923 - don't unlink "objects" from directory with default EA (4554)
4924 - hold socknal file ref over connect in case target is down (4394)
4925 - allow more than 32000 subdirectories in a single directory (3244)
4926 - fix blocks count for O_DIRECT writes (3751)
4927 - OST returns ENOSPC from object create when no space left (4539)
4928 - don't send truncate RPC if file size isn't changing (4410)
4929 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4930 - bind to privileged port in socknal and tcpnal (3689)
4932 - rate limit CERROR/CWARN console message to avoid overload (4519)
4933 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4934 - basic mmap support (3918)
4935 - kernel patch series update from b1_4 (4711)
4937 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4940 - avoid crash during MDS cleanup with OST shut down (2775)
4941 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4942 - don't use bad inodes on the MDS (3744)
4943 - dynamic object preallocation to improve recovery speed (4236)
4944 - don't hold spinlock over lock dumping or change debug flags (4401)
4945 - don't zero obd_dev when it is force cleaned (3651)
4946 - print grants to console if they go negative (4431)
4947 - "lctl deactivate" will stop automatic recovery attempts (3406)
4948 - look for existing locks in ldlm_handle_enqueue() (3764)
4949 - don't resolve lock handle twice in recovery avoiding race (4401)
4950 - revalidate should check working dir is a directory (4134)
4952 - don't always mark "slow" obdfilter messages as errors (4418)
4954 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4957 - don't close LustreDB during write_conf until it is done (3860)
4958 - fix typo in lconf for_each_profile (3821)
4959 - allow dumping logs from multiple threads at one time (3820)
4960 - don't allow multiple threads in OSC recovery (3812)
4961 - fix debug_size parameters (3864)
4962 - fix mds_postrecov to initialize import for llog ctxt (3121)
4963 - replace config semaphore with spinlock (3306)
4964 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4965 - don't allow enqueue to complete on a destroyed export (3822)
4966 - down write_lock before checking llog header bitmap (3825)
4967 - recover from lock replay timeout (3764)
4968 - up llog sem before sending rpc (3652)
4969 - reduce ns lock hold times when setting kms (3267)
4970 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4971 - fix NULL deref and obd_dev leak on setup error (3312)
4972 - replace some LBUG about llog ops with error handling (3841)
4973 - don't match INVALID dentries from d_lookup and spin (3784)
4974 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4975 - fix invalid assertion in ptlrpc_set_wait (3880)
4977 - add libwrap support for the TCP acceptor (3996)
4978 - add /proc/sys/portals/routes for non-root route listing (3994)
4979 - allow setting MDS UUID in .xml (2580)
4980 - print the stack of a process that LBUGs (4228)
4982 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4985 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4986 - make sure to unset replay flag from failed open requests (3440)
4987 - if default stripe count is 0, use OST count for inode size (3636)
4988 - update parent mtime/ctime on client for create/unlink (2611)
4989 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4990 - free recovery state on server during a forced cleanup (3571)
4991 - unregister_reply for resent reqs (3063)
4992 - loop back devices mounting and status check on 2.6 (3563)
4993 - fix resource-creation race that can provoke i_size == 0 (3513)
4994 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4995 - more debugging for page-accounting assertion (3746)
4996 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4997 - avoid deadlock after precreation failure (3758)
4998 - fix race and lock order deadlock in orphan handling (3450, 3750)
4999 - add validity checks when grabbing inodes from l_ast_data (3599)
5001 - add /proc/.../recovery_status to obdfilter (3428)
5002 - lightweight CDEBUG infrastructure, debug daemon (3668)
5003 - change default OSC RPC parameters to be better on small clusters
5004 - turn off OST read cache for files smaller than 32MB
5005 - install man pages and include them in rpms (3100)
5006 - add new init script for (un)mounting lustre filesystems (2593)
5007 - run chkconfig in %post for init scripts (3701)
5008 - drop scimac NAL (unmaintained)
5010 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5013 - clean kiobufs before and after use (3485)
5014 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5015 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5016 - change init script to fail more clearly if not run as root (1528)
5017 - allow clients to reconnect during replay (1742)
5018 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5019 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5020 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5021 - fix cond_resched (3554)
5022 - only evict unfinished clients after recovery (3515)
5023 - allow bulk resend, prevent data loss (3570)
5024 - dynamic ptlrpc request buffer allocation (2102)
5025 - don't allow unlinking open directory if it isn't empty (2904)
5026 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5027 - remove extraneous obd dereference causing LASSERT failure (3334)
5028 - don't use get_cycles() when creating temp. files on the mds (3156)
5029 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5030 - handle EEXIST for set-stripe, set proper directory name (3336)
5032 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5033 - fix ksocknal_fmb_callback() error messages (2918)
5035 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5038 - don't copy lvb into (possibly NULL) reply on error (2983)
5039 - don't deref dentry after dput, don't free lvb on error (2922)
5040 - use the kms to determine writeback rpc length (2947)
5041 - increment oti_logcookies when osc is inactive (2948)
5042 - update client's i_blocks count via lvb messages (2543)
5043 - handle intent open/close of special files properly (1557)
5044 - mount MDS with errors=remount-ro, like obdfilter (2009)
5045 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5046 - don't use cancelling-locks' kms values (2947)
5047 - use highest lock extent for kms, not last one (2925)
5048 - don't dereference ERR_PTR() dentry in error handling path (3107)
5049 - fix thread race in portals_debug_dumplog() (3122)
5050 - create lprocfs device entries at setup instead of at attach (1519)
5051 - common AST error handler, don't evict client on completion race (3145)
5052 - zero nameidata in detach_mnt in 2.6 (3118)
5053 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5054 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5055 - initialize RPC timeout timer earlier for 2.6 (3219)
5056 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5057 - print nal/nid for unknown nid (3258)
5058 - additional checks for oscc recovery before doing precreate (3284)
5059 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5060 - don't crash in mdc_close for bad permissions on open (3285)
5061 - zero i_rdev for non-device files (3147)
5062 - clear page->private before handing to FS, better assertion (3119)
5063 - tune the read pipeline (3236)
5064 - fix incorrect decref of invalidated dentry (2350)
5065 - provide read-ahead stats and refine rpc in flight stats (3328)
5066 - don't hold journal transaction open across create RPC (3313)
5067 - update atime on MDS at close time (3265)
5068 - close LDAP connection when recovering to avoid server load (3315)
5069 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5070 - don't leak open file on MDS after open resend (3325)
5071 - serialize filter_precreate and filter_destroy_precreated (3329)
5072 - loop device shouldn't call sync_dev() for nul device (3092)
5073 - clear page cache after eviction (2766)
5074 - resynchronize MDS->OST in background (2824)
5075 - refuse to mount the same filesystem twice on same mountpoint (3394)
5076 - allow llmount to create routes for mounting behind routers (3320)
5077 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5078 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5079 - fix rare problem with rename on htree directories (3417)
5081 - allow default OST striping configuration per directory (1414)
5082 - fix compilation for qswnal for 2.6 kernels (3125)
5083 - increase maximum number of MDS request buffers for large systems
5084 - change liblustreapi to be useful for external progs like lfsck (3098)
5085 - increase local configuration timeout for slow disks (3353)
5086 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5088 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5091 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5092 - don't overwrite extent policy data in reply if lock was blocked (2901)
5093 - drop filter export grants atomically with removal from device (2663)
5094 - del obd_self_export from work_list in class_disconnect_exports (2908)
5095 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5096 - swab reply message in mdc_close, other PPC fixes (2464)
5097 - fix destroying of named logs (2325)
5098 - overwrite old logs when running lconf --write_conf (2264)
5099 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5100 - fix race in target_handle_connect (2898)
5101 - mds_reint_create() should take same inode create lock (2926)
5102 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5103 - don't close files for self_export to avoid uninitialized obd (2936)
5104 - allow MDS with the same name as client node (2939)
5105 - hold dentry reference for closed log files for unlink (2325)
5106 - reserve space for all logs during transactions (2059)
5107 - don't evict page beyond end of stripe extent (2925)
5108 - don't oops on a deleted current working directory (2399)
5109 - handle hard links to targets without a parent properly (2517)
5110 - don't dereference NULL lock when racing during eviction (2867)
5111 - don't grow lock extents when lots of conflicting locks (2919)
5113 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5116 - account for cache space usage on clients to avoid data loss (974)
5117 - lfsck support in lustre kernel code (2349)
5118 - reduce journal credits needed for BRW writes (2370)
5119 - orphan handling to avoid losing space on client/server crashes
5120 - ptlrpcd can be blocked, stopping ALL progress (2477)
5121 - use lock value blocks to assist in proper KMS, faster stat (1021)
5122 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5123 - recovery for initial connections (2355)
5124 - fixes for mds_cleanup_orphans (1934)
5125 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5126 - block all file creations until orphan recovery completes (1901)
5127 - client remove rq_connection from request struct (2423)
5128 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5129 - recovery timer race (2670)
5130 - mdc_close recovey bug (2532)
5131 - ptlrpc cleanup bug (2710)
5132 - mds timeout on local locks (2588)
5133 - namespace lock held during RPCs (2431)
5134 - handle interrupted sync write properly (2503)
5135 - don't try to handle a message that hasn't been replied to (2699)
5136 - client assert failure during cleanup after abort recovery (2701)
5137 - leak mdc device after failed mount (2712)
5138 - ptlrpc_check_set allows timedout requests to complete (2714)
5139 - wait for inflight reqs when ptlrpcd finishes (2710)
5140 - make sure unregistered services are removed from the srv_list
5141 - reset bulk XID's when resending them (caught by 1138 test)
5142 - unregister_bulk after timeout
5143 - fix lconf error (2694)
5144 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5145 - readahead locks pages, leaves pending causing memory pressure (2673)
5146 - increase OST request buffers to 4096 on large machines (2729)
5147 - fix up permission of existing directories in simple_mkdir (2661)
5148 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5149 - don't assign transno to errored transactions (2742)
5150 - don't delete objects on OST if given a bogus objid from MDS (2751)
5151 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5152 - if rq_no_resend, then timeout request after recovery (2432)
5153 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5154 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5155 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5156 - MDS recovery shouldn't delete objects that it has given out (2730)
5157 - if enqueue arrives after completion, don't clobber LVB (2819)
5158 - don't unlock pages twice when trigger_group_io returns error (2814)
5159 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5160 - don't write pages to disk if there was an error (1450)
5161 - don't ping imports that have recovery disabled (2676)
5162 - take buffered bytes into account when balancing socknal conn (2817)
5163 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5164 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5165 - remove little-endian swabbing of llog records (1987)
5166 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5167 - retry reposting request buffers if they fail (1191)
5168 - grow extent at grant time to avoid granting a revoked lock (2809)
5169 - lock revoke doesn't evict page if covered by a second lock (2765)
5170 - disable VM readahead to avoid reading outside lock extents (2805)
5172 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5173 - updated kernel patches for hp-2.4.20 kernel (2681)
5175 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5178 - fix truncated write corruption (2366)
5179 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5181 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5182 - ensure len(uuid) < 37 in lmc (1171)
5183 - fix ia64 OOPS in llog_test (2255)
5184 - zero end of page at obdfilter for partial page writes (2648)
5185 - don't leave stale dentries around after renames (bug 2428)
5186 - fix timeouts when evicting a client with a single lock held (2642)
5187 - set deadline for the initial HELLO message to drain (2634)
5188 - print out dotted-quad IP addresses in the socknal (2302)
5190 - additional debugging for MDS client eviction problem (2443)
5191 - fix mkfsoptions support for osts (2603, 2604)
5193 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5196 - add series for the vanilla 2.6.0 kernel
5197 - add series for the vanilla 2.4.24 kernel
5198 - add series for a cray x86/64 UL kernel drop
5199 - fix xattr patches for the vanilla 2.4.19 series
5201 - generate true UUIDs in lmc (1171)
5202 - have portals stack dumping break in UML (2466)
5203 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5204 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5205 - wake callback waiting threads on client eviction (2460)
5206 - Add --ptldebug and --subsystem to lmc (1719)
5207 - update assertion to allow safe interrupt allocation
5208 - set rq_no_resend for cancel requests (2432)
5209 - recalculate ptlrpcd timeout after resend (2494)
5210 - call vfs_rmdir when removing pending directories (2368)
5211 - fix renaming a file to itself (2429)
5212 - lmc creates a default one-stripe lov (2454)
5213 - expand procfs space to handle large clusters (2326)
5214 - increase UML stack to avoid overflow
5215 - update lconf's list of debug and subsystem masks
5216 - fix lfs find --obd (2510)
5217 - /proc tunable for disabling filter read caching (2591)
5218 - stop rpm packages from altering slapd.conf (2301)
5219 - disable nagle in the socknal under 0conf (2578)
5220 - choose mds inode size based on stripe count (2572)
5221 - fix kernel-source rpm problems (2516)
5223 - add --disable-doc to avoid pdf generation (2421)
5224 - update documentation, tests, type-os, comments
5225 - avoid format warnings on ia64
5226 - remove the TOE NAL
5227 - tiny code cleanups by removing unused fields
5229 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5232 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5233 - remove the most blatant lies from BUILDING file (2371)
5234 - change default debug level to reasonable production setting
5235 - reduce client side cache size to reduce cache flush time
5236 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5237 - make TCP zerocopy and pinger support enabled by default (2476)
5238 - sync writes completed after process exits caused crashes (2319)
5239 - maintain correct mount count on the MDS (2356)
5240 - backout 1557, because 2316 wasn't really fixed
5241 - better file I/O statistics gathering in /proc
5242 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5243 - another kernel patch to fix zero-copy TCP function export
5244 - don't take duplicate lock when processing re-sent getattr (2420)
5245 - lctl uses obd_self_export instead of creating new conn (2353)
5246 - MDS/OST recovery case which requires object creation asserted (2425)
5247 - move lfs from /usr/sbin to /usr/bin in packages
5248 - fix race between mds_client_add and mds_client_free (2417)
5249 - use kmalloc instead of slabs in portals (2430)
5250 - don't create duplicate records when a failover MDS is present (2442)
5251 - remove unnecessary mount age check (2332)
5252 - don't remove directory inodes from locks prematurely (2451)
5253 - don't break if MDS service name is the same as hostname (2103)
5254 - fix races in client write RPC generation when cache full (2482)
5256 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5259 - remove now-unused request->rq_obd (278)
5260 - if an allocation fails, print out how much memory we've used (1933)
5261 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5262 - add the "configurable stack size" patch to most series files (1256)
5263 - ability to write large log records, for 100+ OST configs (2306)
5264 - fix NULL deref when filter_prep fails (2314)
5265 - fix operator precedence error in filter_sync
5266 - dynamic allocation of socknal TX descriptors (2315)
5267 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5268 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5269 - ensure that CWARN messages in Portals always get to the syslog
5270 - __init/__exit are not for prototype decls (ldlm_init/exit)
5271 - x86-64 compile warning fixes
5272 - fix gateway LMC keyword conflict (2318)
5273 - fix MDS lock inversions in getattr/reint paths (1844)
5274 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5275 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5276 - DLM race condition prevented some lock evictions (2328)
5277 - ENOMEM detection and retry on socknal sends (2230)
5278 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5279 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5280 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5281 - better error messages when a client is rejected during recovery (1505)
5282 - avoid cancelling locks which were never granted, after failure (2330)
5283 - fix i_sem/journal inversion in mds_client_add (2333)
5284 - fix truncate/getattr lock cycle deadlock (2334)
5285 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5286 - fix two rare exit paths which could leak an l_lock() ref (2321)
5287 - fencepost error in MDS/OST orphan recovery (2226)
5288 - make log record alignment 8 bytes (1988)
5289 - lstripe now fails when requested offset > ost_count (2237)
5290 - ensure that all kernel series have a complete list.h (1607)
5291 - fix crashes in special-file operations (2316)
5292 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5294 - allow configurable automake binary, for testing new versions
5295 - small update to the lfs documentation
5297 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5299 * fix negative export reference count in fsfilt_sync (2312)
5301 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5302 * release candidate 0.9.1
5304 - orphans are moved into the PENDING directory for possible recovery
5305 - replayed opens now open by fid for orphan/rename safety (1042)
5306 - last close of an orphan inode generates a transno (683)
5307 - chdir() and mount() now pin the directory entry (1020)
5308 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5309 - discard very old requests without processing them (1502)
5310 - remove some common, well-understood CERRORs (1505)
5311 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5312 - clear "grant" flags in OST replies until OST grant code lands (1644)
5313 - fix read performance by not clobbering i_blksize on client (1598)
5314 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5315 - make LRU size a /proc tunable, clears locks when reduced (707)
5316 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5317 - server locks take references on exports now (1558)
5318 - build fixes for 2.4.20-rh trees (1663)
5319 - return an error from lov_create if all OSCs are inactive (1751)
5320 - fix import levels when a reconnect happens without a timeout (1597)
5321 - exit early from mds_open if we get a lookup error (1749)
5322 - partial page read at EOF wouldn't wait for disk before sending (1642)
5323 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5324 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5325 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5326 - write/truncate lock inversion (1639)
5327 - Don't auto-load obdclass, portals modules during cleanup (1495)
5328 - fix timestamps from jumping to "now" (1763)
5329 - extra journal assertions (1648)
5330 - add an extra multiunlink test (1771)
5331 - fix read_record/write_record API (1776)
5332 - fix leak of offset_extent, possible incorrect i_size later (1772)
5333 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5334 - Debugging for the kqswnal_get_idle_tx problems (1820)
5335 - Allow recovery to be attempted multiple times (1536)
5336 - Write out MDS last_rcvd file after it is first created (1600)
5337 - Fix tx_descriptor leak in failed transmit situations (1827)
5338 - ext3 journaling fixes for assertion failure after IO error (1871)
5339 - class_export_put() on freed export after completion AST error (1896)
5340 - Fix revalidate looping in VFS (1322)
5341 - Don't access a freed export during MDS_REINT timeout (1521)
5342 - Add open-unlink recovery support on the MDS (1673,1764)
5343 - Return an error if no MDS data was read from last_rcvd (1946)
5344 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5345 - Don't setup a disk device that doesn't match exported UUID (317)
5346 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5347 - avoid committing NULL handle in force close
5348 - local.sh is now a one-stripe LOV configuration
5349 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5350 - fix ext3 htree duplicate directory entry corruption (1516)
5351 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5352 - update to new LOV EA format (2097)
5353 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5355 2003-06-15 Phil Schwan <phil@clusterfs.com>
5358 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5359 - per-import/export recovery handling (958, 931, 959)
5360 - multiple last-rcvd slots, for serving multiple FSes (949)
5361 - connections are again shared between multiple imp/exports (963, 964)
5362 - "umount -f" would hang if any requests needed to be sent (393, 978)
5363 - avoid pinning large req buffer by copying for queued messages (989)
5364 - add "uuid" to "lctl device" command to help upcalls (991)
5365 - "open" RPCs with transnos would confuse recovery counters (1037)
5366 - do proper endian conversion of all wire messages (288, 340, 891)
5367 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5368 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5369 - fix ptlrpc_connection leak in target_handle_connect (1174)
5370 - fix import refcounting bug in OST and MDS cleanup (1134)
5371 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5372 - fix very unlikely obd_types race condition (501)
5373 - remove osc_open hack for echo_client (1187)
5374 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5375 - a failure in read_inode2 leads to deadlock (1139)
5376 - cancel ack-locks as soon as transaction is committed (1072)
5377 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5378 - make sure to commitrw after any preprw to avoid deadlock (1162)
5379 - failing to execute a file in a lustre FS would lock inode (1203)
5380 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5381 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5382 - obd_uuid2tgt would walk off the end of the list (1255)
5383 - on IA64 the osc would give portals incorrect bulk size (1258)
5384 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5385 - fix lock inversion caused by new llite matching code (1282)
5386 - limit the number of dirty pages on a client to 10MB (1286)
5387 - timed out locks were not being corrected cancelled (1289)
5388 - fix O_DIRECT above 4GB on IA-32 (1292)
5389 * major user-visible changes
5390 - fail out/fail over policy now controlled by the upcall (993)
5392 - add OBD_PING to check server availability and failure (954)
5393 - lustre messages are now sent in sending host order (288, 340, 891)
5394 - add eadatalen to MDS getattr reply (340)
5395 - OST read replies may contain second buffer, with per-page status (593)
5397 2003-03-11 Phil Schwan <phil@clusterfs.com>
5400 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5401 - failed open()s could cause deadlock; fixed (867, 869)
5402 - stop cancelling OST locks when files are closed (481)
5403 - overlapping XID spaces caused network corruption (851, 853)
5404 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5405 - setattr_raw added, to do single-RPC, server-side setattrs
5406 - lmc/lconf syntax change for OST UUIDs
5407 - fix crashy race condition between ptlrpc_free_req and osc_close
5408 - don't use request in mdc_enqueue if we hit a timeout (889)
5409 - don't set the inode i_size for regular files from the MDS (896)
5410 - handle out of order completion AST (842)
5411 - don't LBUG if a lock request times out after receiving AST (913)
5412 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5413 - if a bad lock AST arrives, send an error instead of dropping entirely
5414 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5415 - fix leak in bulk IO when only partially completed (899, 900, 926)
5416 - fix O_DIRECT for ia64 (55)
5417 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5418 - C-z after timeout could hang a process forever; fixed (977)
5420 - client-side I/O cache (678, 924, 929, 941, 970)
5422 - READPAGE and SETATTRs which don't take server-side locks get
5425 2003-02-11 Phil Schwan <phil@clusterfs.com>
5428 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5429 - Fix fsfilt_extN_readpage() to read a full page of directory
5430 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5431 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5432 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5433 - Most intent operations were removed, in favour of a new RPC mode
5434 that does a single RPC to the server and bypasses most of the VFS
5435 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5436 - Aggressively cancel local locks on DLM servers
5437 - mds_reint_unlink sends EA to the client if it's the last nlink.
5438 client uses that EA to unlink OST objects.
5439 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5440 - recursive symlinks were fixed (439)
5441 - fixed NULL deref in DEBUG_REQ
5442 - filter_update_lastobjid no longer calls sync, which annoyed extN
5443 - fixed multi-client small-writes to a single file problem (445)
5444 - fixed mtime updates during file writes (607)
5445 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5446 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5447 - fixed Linux OST size reporting problem (444, 656)
5448 - OST now updates object mtime with writes or setattr (607, 619)
5449 - client verifies file size before zeroing page past EOF (445)
5450 - OST now writes last allocated objid to disk with allocation (108)
5451 - LOV on echo now works (409)
5453 - mds_reint_unlink sends a new buffer, with the EA included. this
5454 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5455 set if a regular file was being unlinked, and it was the last link
5456 - use PtlGet from the target for bulk writes (315)
5457 - OST now updates object mtime with writes or setattr (607, 619)
5458 - LDLM now has a grant-time callback to revalidate locked items, if
5460 - Many MDS operations were reorganized to combat race conditions
5462 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5463 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5465 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5468 - Fully reactivate OST imports after reconnection (512, others)
5469 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5470 - More graceful error handling for truncating on dead OST (515)
5471 - Don't error out unless we're actually accessing dead stripes (474)
5472 - Fix garbage sizes when stripes are missing (410)
5473 - LRU counters were broken, causing constant lock purge (433, 432)
5474 - garbage on read from stripes with failed OSTs (441)
5475 - mark OSCs as active before reconnecting during recovery (438)
5476 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5477 - lfind did not preserve OST order in output (443)
5478 - symlinks cause hung clients, incorrect data (439)
5479 - stop dereferencing request after dropping refcount (457)
5480 - don't LASSERT(spin_is_locked) on non-SMP (455)
5481 - fixes for many rename() bugs
5482 - fstat didn't correctly synchronize attributes (399)
5483 - server must handle lock cancellation during blocking AST prep (487)
5484 - bulk descriptors were free()d too soon (511)
5485 - fix paths in lconf, which would load incorrect modules (451, 507)
5486 - fix confusing lconf 'host not found' error message (386)
5487 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5488 - fix race condition in mdc_blocking_ast() for inode access (526)
5489 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5490 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5491 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5492 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5493 - fix lstripe in several ways (kernel side) (527)
5494 - fix request leak in ldlm_cli_enqueue (262)
5495 - incorrect OSC was marked inactive after OST failure
5496 - call mds_fs_cleanup before unmounting filesystem (524)
5497 - fix races between taking ns_lock and ldlm_lock_change_resource
5498 - fix races updating LOV export open file list
5499 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5500 - fix recovery NULL deref in ldlm_cli_cancel_unused
5501 - fix some DLM races by using new hash table for lock handles (419)
5502 - permit the client to specify desired inodes, at replay
5503 - duplicate requests when we queue them for replay reintegration
5504 - fix last_rcvd offset calculation
5505 - sync after each recovered transaction, so we always make progress
5506 - never, not always, ERESTART requests without transnos
5507 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5509 - skip replay if the MDS says that the client is already connected
5510 - don't check for a recovery-enabled export to match lctl's UUID
5511 - don't INC_USE_COUNT for phantom exports
5512 - don't crash when cleaning up phantom exports (567)
5513 - don't double-finish or set replay data for errored mdc_open requests
5514 - abort requests when they time out, so we don't get old replies
5515 - send/receive replies for AST messages again
5516 - if the client says that it doesn't have the lock, cancel it on the
5518 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5519 mark it as destroyed, it will all work out when decref is called
5520 - fix module use counts (22, 581)
5522 - ASTs now expect a reply (server cancels lock on error reply)
5524 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5527 - fix many simultaneous client startup (392)
5528 - fix dentry->d_it clobbering
5529 - credentials weren't being shipped for readdir/getattr operations
5530 - remove invalid assertions triggered during some concurrent MD
5532 - proper Lustre versions added (336, 389)
5533 - fix memory leak for create error case (398)
5534 - fix LOV locking bug that would get cli/srv out of sync
5535 - fix echo client over LOV (409)
5536 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5537 - fix double-O_EXCL intent crash (424)
5538 - avoid sending multiple lock CANCELs (352)
5540 - MDS can do multi-client recovery (modulo bugs in new code)
5542 - many updates, edits, cleanups
5544 2002-11-18 Phil Schwan <phil@clusterfs.com>
5547 - fix null d_it dereference (346)
5548 - fix full OST/dbench hang (333)
5549 - fix permission problem with file removal (286)
5550 - fix removal of OSCs from LOV when they fail
5551 - fix NULL deref during bulk timeout (214)
5552 - fix problems related to multiple filesystems on one MDS (241)
5553 - fixed serious subtle metadata locking bugs
5554 - free locks on clients when inodes are removed due to memory
5556 - fix inode pointer in lock data (285)
5557 - partial support for multiple MDS on a single host (241)
5558 - data locks weren't cancelled at clear_inode time (290, 311)
5559 - intent locks could lead to unbounded lock growth (205)
5560 - added a maximum lock count, an LRU list, and a flusher
5561 - fix multiple rename (365)
5562 - properly abstracted the echo client
5563 - OSC locked 1 byte too many; fixed
5564 - rewrote brw callback code:
5565 - fixed recovery bugs related to LOVs (306)
5566 - fixed too-many-pages-in-one-write crash (191)
5567 - fixed (again) crash in sync_io_timeout (214)
5568 - probably fixed callback-related race (385)
5570 - Add capability to MDS protocol
5571 - LDLM cancellations and callbacks on different portals
5573 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5576 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5577 - defer open object destruction to close time (601981, 138)
5578 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5579 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5580 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5581 - don't call obd_setattr in ll_file_release for destroyed objects
5582 * protocol change to lustre_msg: move |version| and add |flags|
5583 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5584 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5585 * added replay of create, unlink, link and rename operations during
5586 MDS failover; recovery should be much more robust now
5587 * remove failed OSCs from LOVs (only lov_create uses this so far)
5588 * the lustre-HOWTO was brought (more) up to date (582544)
5590 2002-10-23 Phil Schwan <phil@clusterfs.com>
5593 - in-use dentries weren't being reused properly (617851)
5594 - prevent multiple LDLM setup (599178)
5595 - fix LOV size calculations for truncate (617853)
5596 - fix client handling of MDS intent errors (POSIX)
5597 - fix permission bug in lovstripe.c test (624321)
5598 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5599 - truncate past end of file could corrupt data
5600 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5601 - a race in recovery could return ETIMEDOUT to apps (623947)
5602 - building outside the source directory was fixed
5603 * the lustre-HOWTO was brought (more) up to date (582544)
5604 * major progress was made on recovery functionality
5606 2002-10-10 Phil Schwan <phil@clusterfs.com>
5609 - recovery deadlock fix
5610 - rm -rf causes LBUG fix (617817)
5611 - file open by multiple tasks fix (618962)
5612 - directory permissions bugs (602707 and 620007)
5613 - journal_stop fixed with locking (611313)
5614 - O_APPEND failures resolved (618273, perhaps 614459)
5615 - lconf PATH fix (619770)
5616 - IA64 build fix (621450)
5617 - RPC buffer sizes scale with amount of memory
5619 2002-10-01 Phil Schwan <phil@clusterfs.com>
5622 - locks would be cancelled without throwing away data pages,
5623 resulting in inconsistent data (605627)
5624 - inode attributes were not always being refreshed (605627, 612449)
5625 - lconf now continues to cleanup after lctl reports an error
5626 - MDS now enforces user permissions (602707)
5627 - lprocfs cleanup fixed, but not yet enabled (614157)
5628 - fixed infinite server hang, should a client not respond to an AST
5629 - avoid going into recovery if user calls readlink() with a buffer
5630 that's too small (613941)
5631 - AST RPCs no longer require replies (614867) -- this may be changed
5632 - don't crash server if client sends an IOV that's too big (611336)
5633 - fixed lock conversion deadlock (611892)
5634 - fixed the following of symlinks (614622)
5635 * recovery: the server can remove locks from a client that dies, other
5636 clients can make progress
5637 * more extN patch fixes
5638 * compile-time configurable ptlrpc buffer allocations
5640 - collaborative read cache document
5641 - Lustre Lite Performance CDR document-in-progress
5643 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5646 - fix typo in patch-2.4.18
5648 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5651 - clear ptlrpc request each time in handle_incoming_request()
5652 - unlink of files now destroys the object on the OST
5654 2002-09-19 Peter Braam <braam@clusterfs.com>
5656 * add hard link support
5657 * change obdfile creation method
5658 * kernel patch changed
5660 2002-09-19 Peter Braam <braam@clusterfs.com>
5663 - stack overflow bug in extN fixed
5665 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5667 * documentation updates
5668 - add man pages for config tools
5669 - update tests/README to describe testing with new config tools
5670 - finish metadata API descriptions
5671 * bug fixes and cleanups
5672 - statfs workaround for 16TB limit
5673 - LOV stripe allocation improved, can stripe on subset of OSTs
5674 - LOV file size/IO offset was wrong for files > 4GB in size
5675 - object EA data was being dropped, caused files to be unreadable
5676 - memory overflow with non-LOV OST caused memory corruption
5677 - fixed regression tests to work with new config tools, obdfilter
5678 - fixed bug when directory size became larger than 1 block
5679 - fixed bug (for single client case) when PWD was deleted
5680 - invalidate local directory pages when doing intent-based ops
5681 - avoid LDLM oops when lock callback contained bad data
5683 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5685 * documentation updates
5686 * bug fixes and cleanups
5687 - configuration tools
5690 - 64-bit compile warnings
5691 - 64-bit internal statfs data
5693 * test_brw on persistent OST devices
5695 * lprocfs (disabled)
5697 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5699 * documentation updates
5700 * bug fixes and cleanups
5701 * configuration tools
5703 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5706 * many small fixes to 0_5_4
5707 * io/network handling
5708 * thinkos in MDS operations
5710 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5713 * crucial basic fixes to 0.5.3
5714 * IOR, Iozone work over Elan
5717 2002-08-07 Phil Schwan <phil@clusterfs.com>
5718 * version 0_5_3, our first alpha
5719 * we use the new Portals iovs
5720 * documentation updates
5721 * bug fixes and cleanups
5722 * small changes in the DLM wire protocol
5724 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5725 * version 0_5_1 with some initial stability,
5726 * locking on MD and file I/O.
5727 * documentation updates
5728 * several bug fixes since 0.5.0
5729 * small changes in wire protocol
5731 2002-07-18 Phil Schwan <phil@clusterfs.com>
5733 * delivered as Lustre Light Alpha
5734 * fixed a crash after handling invalid MDS requests
5735 * fixed directory pages for architectures with non-4k pages sizes
5737 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5738 * release version v0_4_4
5739 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5741 * Fixes a number of interruption problems with OST operations.
5742 * Update documentation for portals header changes
5743 * Move all wire protocol structs/defines to lustre_idl.h
5744 * Fixes symlink length bug.
5745 * Add tcpdump to repository.
5747 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5748 * release version v0_4_3
5749 * Fixes statfs for inodes on extN.
5750 * Fixes bug in runtests which would delete /etc/hosts.
5751 * Use 64-bit object IDs wherever possible (not into VFS though)
5752 Remove ost_get_info, which is unused by lustre, and out of date.
5754 2002-07-03 Peter Braam <braam@clusterfs.com>
5755 * release version v0_4_2 Fixes a lookup error (type not passed)
5756 * move forward to head of Portals
5757 * move forward to latest Lustre kernel
5759 2002-06-25 Peter Braam <braam@clusterfs.com>
5760 * release version v0_4_1. Hopefully stable on single node use.