1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.37 (SLES 10),
5 2.6.18-128.1.1.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.5.sun2
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
16 Severity : enhancement
18 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
22 Description: resolve race between obd_disconnect and class_disconnect_exports
23 Details : if obd_disconnect will be called to already disconnected export he
24 forget release one reference and osc module can't unloaded.
26 Severity : enhancement
28 Description: Allow tuning service thread via /proc
29 Details : For each service a new
30 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
31 created that can be used to set min/max thread counts, and get the
32 current number of running threads.
36 Descriptoin: don't return error if have particaly created objects for file.
37 Details : lov_update_create_set uses set->set_success as index for created objects,
38 so if some requests will be failed, they will have hole at end of
39 array and we can use qos_shrink_lsm for allocate correct lsm.
41 Severity : enhancement
43 Description: Update OFED support to 1.4.1
47 Description: Reduce small size read RPC
48 Details : Set read-ahead limite for every file and only do read-ahead when
49 available read-ahead pages are bigger than 1M to avoid small size
52 Severity : enhancement
54 Description: move AT tunable parameters for more consistent usage
55 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
57 Severity : enhancement
59 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
60 Details : allow skip disconnected ost for send statfs request and hide error
66 Description: fix racy locking of mballoc block bitmaps causing BUG
67 Details : The locking of the mballoc buddy bitmap and the in-memory
68 block bitmap was using two different spin locks in some
69 cases. This made it possible to incorrectly access the
70 mballoc bitmap while another process was modifying it,
71 causing a sanity assertion to fail. While no on-disk corruption
72 was reported, there was some risk of this happening.
75 Frequency : rare, on llog test 6
77 Descriptoin: don't allow connect to already connected import
78 Details : allowing connect to already connected import is hide connecting problem.
83 Descriptoin: allow kill process which wait statahead result
84 Details : for some reasons 'ls' can stick in waiting result from statahead,
85 in this case need way for kill this process.
87 Severity : enhancement
89 Description: Add state history info file, enhance import info file
90 Details : Track import connection state changes in a new osc/mdc proc file;
91 add overview-type data to the osc/mdc import proc file.
93 Severity : enhancement
95 Description: MDS create should not wait for statfs RPC while holding DLM lock.
97 Severity : enhancement
99 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
104 Descriptoin: OST grant too much space to client even there are not enough space.
105 Details : Client will shrink its grant cache to OST if there are no write
106 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
107 this grant cache if there are already not enough avaible space
108 (left_space < total_clients * 32M).
111 Frequency : start MDS on uncleanly shutdowned MDS device
113 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
114 Details : stay in waiting mds<>ost recovery finished produce random bugs
115 due race between two ll_sync thread for one lov target. send
116 ACTIVATE event only if connect realy finished and import have
120 Frequency : rare, connect and disconnect target at same time
122 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
123 Details : don't call obd_disconnect under lov_lock. this long time
124 operation and can block ptlrpcd which answer to connect request.
129 Descriptoin: don't lose wakeup for imp_recovery_waitq
130 Details : recover_import_no_retry or invalidate_import and import_close can
131 both sleep on imp_recovery_waitq, but we was send only one wakeup
135 Frequency : always with long access acl
137 Descriptoin: mds can't pack reply with long acl.
138 Details : mds don't control size of acl but they limited by reint/getattr
141 Severity : enhancement
143 Description: Update to SLES10 kernel-2.6.16.60-0.33.
145 Severity : enhancement
147 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
150 Frequency : start MDS on uncleanly shutdowned MDS device
152 Descriptoin: aborting recovery hang on MDS
153 Details : don't throttle destroy RPCs for the MDT.
156 Frequency : on remount
158 Description: external journal device not working after the remount
159 Details : clear dev_rdonly flag for external journal devices in
165 Description: shutdown vs evict race
166 Details : client_disconnect_export vs connect request race.
167 if client will evicted at this time - we start invalidate
168 thread without referece to import and import can be freed
172 Frequency : rare, need acl's on inode.
174 Description: client can't handle ost additional correctly
175 Details : if ost was added after client connected to mds client can have
176 hit lnet_try_match_md ... to big messages to wide striped files.
177 in this case need teach client to handle config events about add
178 lov target and update client max ea size at that event.
180 Severity : enhancement
182 Description: Changelogs
183 Details : Changelogs are a lightweight mechanism to track filesystem
184 metadata and namespace changes. The changelog is recorded
185 permanently on the MDTs, and is periodically "consumed" / purged
186 when records are no longer needed.
188 Severity : enhancement
190 Description: compact fld format with extents
191 Details : Store range of seq rather than every seq in FLD. Seq
192 controller update FLD rather than clients. In Case of CMD, mdt0
193 has FLD, all other metadata server act as non persistent proxy
194 for FLD queries and cache fld entries in fld cache.
199 Description: don't skip ost target if they assigned to file
200 Details : Drop slow OSCs if we can, but not for requested start idx.
201 This means "if OSC is slow and it is not the requested
202 start OST, then it can be skipped, otherwise skip it only
203 if it is inactive/recovering/out-of-space.
207 Description: more cleanup in mds_lov
208 Details : not send LOV EA under replay, we can't know about they size at this
209 time. Don't allow client connect to mds before any ost connected,
210 for avoid problems with LOV EA size and returning EIO to client.
212 Severity : enhancement
214 Description: Interoperability at server side (Disk interoperability)
216 Severity : enhancement
218 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
220 Severity : enhancement
222 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
224 Severity : enhancement
226 Description: New client IO stack (CLIO).
228 Severity : enhancement
230 Description: Commit on sharing. Eliminate inter-client dependencies between
231 uncommitted transactions by doing transaction commits.
232 Thereby clients may recovery independently.
235 Frequency : Create a symlink file with a very long name
237 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
238 Details : If there is no extra space in the request for early cancels,
239 ldlm_req_handles_avail() returns 0 instead of a negative value.
241 Severity : enhancement
243 Description: Add /proc entry for import status
244 Details : The mdc, osc, and mgc import directories now have
245 an import directory that contains useful import data for debugging
248 Severity : enhancement
250 Description: Re-disable certain /proc logging
251 Details : Enable and disable client's offset_stats, extents_stats and
252 extents_stats_per_process stats logging on the fly.
255 Frequency : Only on FC kernels 2.6.22+
257 Description: oops in statahead
258 Details : Do not drop reference count for the dentry from VFS when lookup,
259 VFS will do that by itself.
261 Severity : enhancement
263 Description: Generic /proc file permissions
264 Details : Set /Proc file permissions in a more generic way to enable non-
265 root users operate on some /proc files.
269 Description: Hitting mdc_commit_close() ASSERTION
270 Details : Properly handle request reference release in
271 ll_release_openhandle().
275 Description: quota recovery deadlock during mds failover
276 Details : This patch includes att18982, att18236, att18237 in bz14840.
278 1. fix osts hang when mds does failover with quotaon
279 2. prevent watchdog storm when osts threads wait for the
284 Frequency : only patchless client
285 Description: add workaround for race between add/remove dentry from hash
287 Severity : enhancement
289 Description: Allow OST glimpses to return PW locks
293 Description: LBUG when llog conf file is full
294 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
298 Description: Prevent import from entering FULL state when server in recovery
302 Description: service mount cannot take device name with ":"
303 Details : Only when device name contains ":/" will mount treat it as
309 Description: replace ptlrpcd with the statahead thread to interpret the async
310 statahead RPC callback
314 Frequency : on recovery
315 Description: I/O failures after umount during fail back
316 Details : if client reconnected to restarted server we need join to recovery
317 instead of find server handler is changed and process self
318 eviction with cancel all locks.
320 Severity : enhancement
322 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
324 Severity : enhancement
326 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
328 Severity : enhancement
330 Description: Upcall on Lustre log has been dumped
331 Details : Allow for a user mode script to be called once a Lustre log has
332 been dumped. It passes the filename of the dumped log to the
333 script, the location of the script can be specified via
334 /proc/sys/lnet/debug_log_upcall.
339 Description: avoid idr_remove called for id which is not allocated.
340 Details : Move assigment s_dev for clustered nfs to end of initialization,
341 to avoid problem with error handling.
346 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
347 Details : When connection is reused this not moved from CONN_UNUSED_HASH
348 into CONN_USED_HASH and this prodice warning when put connection
349 again in unused hash.
352 Severity : enhancement
354 Description: File striping can now be set to use an arbitrary pool of OSTs.
356 Severity : enhancement
358 Description: Export bytes_read/bytes_write count on OSC/OST.
362 Description: Early reply size mismatch, MGC loses connection
363 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
364 the connect flags are properly negotiated.
369 Description: quotas are not honored with O_DIRECT
370 Details : all writes with the flag O_DIRECT will use grants which leads to
371 this problem. Now using OBD_BRW_SYNC to guard this.
375 Description: add quota statistics
376 Details : 1. sort out quota proc entries and proc code.
377 2. add quota statistics
379 Severity : enhancement
381 Description: enable quota support for HEAD.
385 Description: Properly propagate oinfo flags from lov to osc for statfs
386 Details : restore missing copy oi_flags to lov requests.
388 Severity : enhancement
390 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
391 : l_getgroups(8), lst(8), routerstat(8)
392 Details : included man pages for llobdstat(8), llstat(8),
393 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
395 Severity : enhancement
397 Description: configure's --enable-quota should check the
398 : kernel .config for CONFIG_QUOTA
399 Details : configure is terminated if --enable-quota is passed but
400 : no quota support is in kernel
404 Description: Remove portals compatibility
405 Details : Remove portals compatibility, not interoperable with releases
410 Description: Resolve device initialization race
411 Details : Prevent proc handler from accessing devices added to the
412 obd_devs array but yet be intialized.
414 Severity : enhancement
416 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
418 Severity : enhancement
420 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
425 Description: Using wrong pointer in osc_brw_prep_request
426 Details : Access to array[-1] can produce panic if kernel compiled with
427 CONFIG_PAGE_ALLOC enabled
431 Description: Client runs out of low memory
432 Details : Consider only lowmem when counting initial number of llap pages
436 Description: *optional* service tags registration
437 Details : if the "service tags" package is installed on a Lustre node
438 When the filesystem is mounted, a local-node service tag will
439 be created. See http://inventory.sun.com/ for more information
440 about the Service Tags asset management system.
444 Description: Kernel BUG tries to release flock
445 Details : Lustre does not destroy flock lock before last reference goes
446 away. So always drop flock locks when client is evicted and
447 perform unlock regardless of successfulness of speaking to MDS.
451 Description: add refcount protection for osc callbacks, avoid panic on shutdown
455 Description: sanity test 65a fails if stripecount of -1 is set
456 Details : handle -1 striping on filesystem in ll_dirstripe_verify
461 Description: ASSERTION(CheckWriteback(page,cmd)) failed
462 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
465 Severity : enhancement
467 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
471 Description: do not process already freed flock
472 Details : flock can possibly be freed by another thread before it reaches
473 to ldlm_flock_completion_ast.
477 Description: LBUG during stress test
478 Details : Need properly lock accesses the flock deadlock detection list.
482 Description: oops in page fault handler
483 Details : kernel page fault handler can return two special 'pages' in error
484 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
488 Description: timeout with invalidate import.
489 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
490 should be handled by ptlrpcd. This produce long age waiting and
491 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
493 Severity : enhancement
495 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
499 Description: enable MGS and MDT services start separately
500 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
501 a co-located MGS without starting the MGS, which is a complement
502 to 'nosvc' mount option.
506 Frequency : after recovery
507 Description: precreate to many object's after del orphan.
508 Details : del orphan st in oscc last_id == next_id and this triger growing
509 count of precreated objects. Set flag LOW to skip increase count
510 of precreated objects.
514 Frequency : rare, on clear nid stats
515 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
516 Details : when clean nid stats sometimes try destroy live entry,
517 and this produce panic in free.
521 Description: Stack overflow during MDS log replay
522 ease stack pressure by using a thread dealing llog_process.
526 Description: wait until IO finished before start new when do lock cancel.
527 Details : VM protocol want old IO finished before start new, in this case
528 need wait until PG_writeback is cleared until check dirty flag and
529 call writepages in lock cancel callback.
531 Severity : enhancement
533 Description: using special macro for print time and cleanup in includes.
537 Description: mds_mfd_close() ASSERTION(rc == 0)
538 Details : In mds_mfd_close(), we need protect inode's writecount change
539 within its orphan write semaphore to prevent possible races.
543 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
547 Frequency : rare, on shutdown ost
548 Description: don't hit live lock with umount ost.
549 Details : shrink_dcache_parent can be in long loop with destroy dentries,
550 use shrink_dcache_sb instead.
554 Description: don't panic with use echo client
555 Details : echo client pass NULL as client nid pointer and this produce null
560 Description: fix build on ppc32
561 Details : compile code with -m64 flag produce wrong object file for ppc32.
565 Description: add message levels for liblustreapi
569 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
570 Details : Move the 'good_osts' check before the 'total_bavail' check. This
571 will result in an -EAGAIN and in the exit call path we call
572 alloc_rr() which will with increasing aggressiveness attempt to
573 aquire precreated objects on the minimum number of required OSCs.
577 Description: Use old size assignment to avoid deadlock
578 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
579 the scheduling while holding a spinlock. We do not need locking
580 for size in ll_update_inode() because size is only updated from
581 the MDS for directories or files without objects, so there is no
582 other place to do the update, and concurrent access to such inodes
583 are protected by the inode lock.
587 Description: resolve "_IOWR redefined" build error on SLES10
591 Description: dump the memory debugging after all modules are unloaded to
592 suppress false negative in conf_sanity test 39
594 Severity : enhancement
596 Description: build kernel-ib packages for OFED 1.3 in our release cycle
601 Description: fix SLES kernel versioning
602 Details : the kernel version for our SLES 10 kernel did not include a "-"
603 before the "smp" at the end. while this was not a problem in
604 general, it did mean that software trying to use the kernel
605 version to try to detect a vendor specific kernel would fail.
606 this was most evident by the OFED build scripts.
610 Description: Don't update lov_desc members until making sure they are valid
611 Details : When updating lov_desc members via proc fs, need fix their
612 validities before doing the real update.
616 Description: don't put request into delay list while invalidate in flight.
617 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
618 import in flight. this produce timeout for invalidate and sometimes
619 can cause stale data.
623 Frequency : on ppc only
624 Description: not convert ost objects for directory because it's not exist.
625 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
629 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
631 Severity : enhancement
633 Description: Update to support 2.6.22.14 vanilla kernel.
637 Frequency : rare, on recovery
638 Description: read procfs can produce deadlock in some situation
639 Details : Holding lprocfs lock which send rpc can produce block for destroy
640 obd objects and this also block reconnect with -EALREADY. This
641 isn't fix all lprocfs bugs - but make it rare.
643 Severity : enhancement
645 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
648 Frequency : frequent on X2 node
650 Description: mdc_set_open_replay_data LBUG
651 Details : Set replay data for requests that are eligible for replay.
655 Description: lustre_mgs: operation 101 on unconnected MGS
656 Details : When MGC is disconnected from MGS long enough, MGS will evict the
657 MGC, and late on MGC cannot successfully connect to MGS and a lot
658 of the error messages complaining that MGS is not connected.
662 Frequency : on network error
663 Description: panic with double free request if network error
664 Details : mdc_finish_enqueue is finish request if any network error ocuring,
665 but it's true only for synchronus enqueue, for async enqueue
666 (via ptlrpcd) this incorrect and ptlrpcd want finish request
669 Severity : enhancement
671 Description: client-side metadata stat-ahead during readdir(directory readahead)
672 Details : perform client-side metadata stat-ahead when the client detects
673 readdir and sequential stat of dir entries therein
676 Frequency : on start mds
678 Description: Implement get_info(last_id) in obdfilter.
681 Frequency : occasional
683 Description: Correctly check stale fid, not start epoch if ost not support SOM
684 Details : open with flag O_CREATE need set old fid in op_fid3 because
685 op_fid2 was overwritten with new generated fid, but mds can answer
686 with one of these two fids and both is not stale. Setattr
687 incorrectly started an epoch and assume will be called
688 done_writing, but without SOM done_writing ever being called.
691 Frequency : rare, depends on device drivers and load
693 Description: MDS or OSS nodes crash due to stack overflow
694 Details : Code changes in 1.8.0 increased the stack usage of some functions.
695 In some cases, in conjunction with device drivers that use a lot
696 of stack the MDS (or possibly OSS) service threads could overflow
697 the stack. One change which was identified to consume additional
698 stack has been reworked to avoid the extra stack usage.
701 Frequency : occasional
703 Description: Do not fail import if osc_interpret_create gets -EAGAIN
704 Details : If osc_interpret_create got -EAGAIN it immediately exits and
705 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
706 oscc_has_objects and see OSC has no objests and call
707 oscc_internal_create to resend create request.
709 Severity : enhancement
711 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
713 Severity : enhancement
715 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
718 Frequency : very rare
720 Description: Fix replay if there is an un-replied request and open
721 Details : In some cases, older replay request will revert the
722 mcd->mcd_last_xid on MDS which is used to record the client's
725 Severity : enhancement
727 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
729 Severity : enhancement
731 Description: Add rhel5 support to HEAD.
733 Serverity : enhancement
735 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
740 Description: Don't allow skipping OSTs if index has been specified.
741 Details : Don't allow skipping OSTs if index has been specified, make
742 locking in internal create lots better.
746 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
747 Details : access to bitfield in structure is always rounded to long
748 and this produce problem with not atomic change any bit.
752 Description: Lustre make rpms failed.
753 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
754 builds Lustre rpms from tarball.
757 Frequency : rare on shutdown ost
759 Description: If llog cancel was not send before clean_exports phase, this can
760 produce deadlock in llog code.
761 Details : If llog thread has last reference to obd and call class_import_put
762 this produce deadlock because llog_cleanup_commit_master wait when
763 last llog_commit_thread exited, but this never success because was
764 called from llog_commit_thread.
768 Description: allow userland application know is lost one of stripes.
769 Details : fill lvb_blocks with error code on ost and return it to
770 application if error flag found.
774 Description: NULL lov_tgts causing MDS oops
775 Details : more safe checks for NULL lov_tgts for avoid oops.
777 Severity : enhancement
779 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
783 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
784 Details : If an OST is down the MDS will hang indefinitely in
785 obd_statfs_rqset() waiting for the statfs data. While for
786 MDS QOS usage of statfs, it should not stuck in waiting.
788 Severity : enhancement
790 Description: remote_acl support
791 Details : Support ACL-based permission check for remote user.
792 Support setfacl/getfacl for remote user with the utils
793 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
794 the system "{s,g}etfacl" utils.
796 Severity : enhancement
798 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
800 Severity : enhancement
802 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
806 Description: make mgs_setparam() handle fsname containing dash
807 Details : fsname containing a dash does not work with lctl conf_param
809 Severity : enhancement
811 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
813 Severity : enhancement
815 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
819 Frequency : rare, at shutdown
820 Description: access already free / zero obd_namespace.
821 Details : if client_disconnect_export was called without force flag set,
822 and exist connect request in flight, this can produce access to
823 NULL pointer (or already free pointer) when connect_interpret
824 store ocd flags in obd_namespace.
828 Frequency : only at startup
829 Description: not alloc memory with spinlock held.
830 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
831 if any spinlock held.
833 Severity : enhancement
835 Description: make lustre randomly fail allocating memory
836 Details : Make lustre randomly failed allocating memory for testing purpose.
838 Severity : enhancement
840 Description: lost problems with lov objid file
841 Details : Fixes some scability and access to not inited memory problems
842 in work with lov objdid file.
847 Description: lfs find does not continue on file error
848 Details : Continue other files processing when a file/dir is absent.
852 Description: Inconsistent usage of lustre_pack_reply()
853 Details : Standardize the usage of lustre_pack_reply() such that it
854 always generate a CERROR on failure.
857 Frequency : occasional
859 Description: MDS hang or stay in waiting lock
860 Details : If client receive lock with CBPENDING flag ldlm need send lock
861 cancel as separate rpc, to avoid situation when cancel request
862 can't processed due all i/o threads stay in wait lock.
866 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
870 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
874 Description: Update to SLES9 kernel-2.6.5-7.287.3.
878 Description: Update to RHEL4 latest kernel.
880 Severity : enhancement
882 Description: Build SLES10 patchless client fails
883 Details : The configure was broken by run ./configure with
884 --with-linux-obj=.... argument for patchless client. When the
885 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
886 can't search header adequately. Use absolute path such as
887 -I($LINUX)/include instead.
891 Description: interrupt oig_wait produce painc on resend.
892 Details : brw_redo_request can be used for resend requests from ptlrpcd and
893 private set, and this produce situation when rq_ptlrpcd_data not
894 copyed to new allocated request and triggered LBUG on assert
895 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
896 ptlrpcd set if request is changed and can be safety changed to use
901 Description: LASSERT_{REQ,REP}SWAB macros are buggy
902 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
903 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
904 these macros and replace them with their difinitions instead.
908 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
909 Details : Update which_patch & target file for SLES10 latest kernel.
912 Description: Remove client patches from SLES 10 kernel.
913 Details : This causes SLES 10 clients to behave as patchless clients
914 even on a Lustre-patched (server) kernel.
916 Severity : enhancement
918 Description: self-adjustable client's lru lists
919 Details : use adaptive algorithm for managing client cached locks lru
920 lists according to current server load, other client's work
921 pattern, memory activities, etc. Both, server and client
922 side namespaces provide number of proc tunables for controlling
925 Severity : enhancement
927 Description: light-weight GSS support
928 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
929 and reduce performance overhead.
931 Severity : enhancement
933 Description: Linux keyring support
934 Details : Support using service of Linux keyring for Lustre GSS internal
935 context refresh/cache mechanism.
939 Description: Fix errors in lfs documentation
940 Details : Fixes man pages
944 Description: don't use GFP_* in generic Lustre code.
945 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
949 Description: obdclass is limited by single OBD_ALLOC(idarray)
950 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
954 Description: warnings with build patchless client with vanila 2.6.19 and up
955 Details : change old ctl_table style and replace ctl_table/ctl_table_header
956 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
960 Frequency : Only for SLES
961 Description: sanity_quota fail test_1
962 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
963 file, which makes TSTID[2] inunique.
967 Frequency : Only for SLES9
968 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
969 Details : Update target/ChangeLog/which_patch .
973 Frequency : Only for test_10 in sanity-quota.sh
974 Description: a bug in quota test code
975 Details : checking a test flag in a wrong place
979 Frequency : Only for test_13 in sanity-quota.sh
980 Description: a bug in quota test script
981 Details : shouldn't check overfull allocated quota
985 Description: jbd statistics
986 Details : Port older jbd statistics patch for sles10
988 Severity : enhancement
990 Description: Add printing inode info into message about error in writepage.
995 Description: reply_lock_interpret crash due to race with it and lock cancel.
996 Details : Do not replay locks that are being cancelled. Do not reference
997 locks by their address during replay, just by their handle.
1001 Frequency : When flocks are used.
1002 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1003 Details : Flock locks might destroy just granted lock if it could be merged
1004 with another existing flock, this is done in completion handler,
1005 so teach ldlm_cli_enquque_fini that this is a valid case for
1011 Description: Oops in read and write path when failing to allocate lock.
1012 Details : Check if lock allocation failed and return error back.
1016 Description: lstripe command fails for valid OST index
1017 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1018 instead of lov->desc.ld_active_tgt_count.
1022 Description: sanity.sh failed test 103
1023 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1028 Description: df doesn't work properly if diskfs blocksize != 4K
1029 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1033 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1034 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1036 Severity : enhancement
1038 Description: Adaptive timeouts
1039 Details : RPC timeouts adapt to changing server load and network
1040 conditions to reduce resend attempts and improve recovery time.
1044 Description: llapi_file_create() does not allow some changes
1045 Details : add llapi_file_open() that allows specifying the mode and
1046 open flags, and also returns an open file handle.
1050 Description: merge and cleanup kernel patches.
1051 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1055 Description: Add journal checksum support.(Kernel part)
1056 Details : The journal checksum feature adds two new flags i.e
1057 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1058 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1059 indicates that the commit block contains the checksum for
1060 the blocks described by the descriptor blocks. Now commit
1061 record can be sent to disk without waiting for descriptor
1062 blocks to be written to disk. This behavior is controlled
1063 using JBD2_FEATURE_ASYNC_COMMIT flag.
1067 Description: OSS needs mutliple precreate threads
1068 Details : Add ability to start more than one create thread per OSS.
1072 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1073 Details : execute lfs setstripe on client
1077 Description: mds_obd_create error creating tmp object
1078 Details : When the user sets quota on root, llog will be affected and can't
1079 create files and write files.
1082 Frequency : Always on ia64 patchless client, and possibly others.
1084 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1085 Details : This allows the patchless client to be loaded on architectures
1086 without this export.
1090 Description: RedHat Update kernel for RHEL5
1091 Details : Add the kernel config file for RHEL5.
1095 Description: RedHat Update kernel for RHEL5
1096 Details : Modify the kernel config file more closer RHEL5.
1100 Description: Build failure against Centos5 (RHEL5)
1101 Details : Define PAGE_SIZE when it isn't present.
1105 Description: test_7 on sanity-quota.sh can't be used on separate machines
1106 Details : add facet to handle it
1110 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1111 Details : with NFS, the anon dentry's parent was set to itself in
1112 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1113 corresponding dentry other than use rec->ur_name.
1115 Severity : enhancement
1117 Description: lfs setstripe enhancement
1118 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1122 Description: enable data checksumming by default
1123 Details : enable checksum by default, allow --disable-checksum
1124 configure option and "-o nochecksum" mount option. Checksums
1125 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1129 Description: lustre support for RHEL5
1130 Details : Add support for RHEL5.
1134 Description: Client eviction due to failover config
1135 Details : after a connection loss, the lustre client should attempt to
1136 reconnect to the last active server first before trying the
1137 other potential connections.
1141 Description: when mds and osts use different quota unit(32bit and 64bit),
1142 quota will be released repeatly.
1143 Details : void sending multiple quota reqs to mds, which will keep the
1144 status between the reqs.
1148 Description: osts not allocated evenly to files
1149 Details : change the condition to increase offset_idx
1152 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1154 Description: Data corruption for OSTs that are formatted larger than 2TB
1156 Details : When generating the bio request for lustre file writes the
1157 sector number would overflow a temporary variable before being
1158 used for the IO. The data reads correctly from Lustre (which
1159 will overflow in a similar manner) but other file data or
1160 filesystem metadata may be corrupted in some cases.
1164 Description: Tune the kernel for good SCSI performance.
1165 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1166 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1171 Description: rewrite ext2-derived code in obdclass/uuid.c
1172 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1173 from scratch preserving functionality.
1177 Description: "lfs find -obd UUID" prints directories
1178 Details : "lfs find -obd UUID" will return all directory names instead
1179 of just file names. It is incorrect because the directories
1180 do not reside on the OSTs.
1184 Description: use i_size_read and i_size_write in 2.6 port
1185 Details : replace inode->i_size access with i_size_read/write()
1188 Frequency : when removing large files
1190 Description: scheduling issue during removal of large Lustre files
1191 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1192 It causes scheduling issues when removing large files (17TB in the
1196 Frequency : only with liblustre clients on XT3
1198 Description: evictions taking too long
1199 Details : allow llrd to evict clients directly on OSTs
1202 Frequency : only on ppc
1204 Description: /proc/fs/lustre/devices broken on ppc
1205 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1206 We should make sure the type of 'index' is loff_t and then cast
1207 explicitly as needed below. Do not assign an explicitly cast
1211 Frequency : only for rhel5
1213 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1214 Details : Modify the target file & which_kernel.
1216 Severity : enhancement
1218 Description: omit set fsid for export NFS
1219 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1220 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1224 Frequency : Always, for kernels after 2.6.16
1225 Description: Fix warning idr_remove called for id=.. which is not allocated.
1226 Details : Last kernels save old s_dev before kill super and not allow
1227 to restore from callback - restore it before call kill_anon_super.
1231 Description: buffer overruns could theoretically occur
1232 Details : llapi_semantic_traverse() modifies the "path" argument by
1233 appending values to the end of the origin string, and a
1234 overrun may occur. Adding buffer overrun check in liblustreapi.
1238 Description: Fix error on 'ls .' at the top of the Lustre mount.
1239 Details : Don't revalidate dentry if it is a root dentry.
1243 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1244 Details : Modify vm-tunables-rhel4.patch.
1248 Description: Kernel config for 2.6.18-vanilla.
1249 Details : Modify targets/2.6-vanilla.target.in.
1250 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1251 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1252 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1253 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1257 Description: improve handling recoverable errors
1258 Details : if request processig with error which can be recoverable on server
1259 request should be resend, otherwise page released from cache and
1264 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1265 Details : Modify target file & which_patch.
1266 A flaw was found in the IA32 system call emulation provided
1267 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1268 value could be stored in the %RAX register, which could trigger an
1269 out-of-bounds system call table access. An untrusted local user
1270 could exploit this flaw to run code in the kernel
1271 (ie a root privilege escalation). (CVE-2007-4573).
1275 Description: O_DIRECT bypasses client statistics.
1276 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1277 read_ahead_stats were updated. copied the stats section from
1278 osc_send_oap_rpc() into async_internal().
1282 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1286 Description: change order of libsysio includes
1287 Details : '#include sysio.h' should always come before '#include xtio.h'
1291 Description: handle "serious error: objid * already exists" more gracefully
1292 Details : If LAST_ID value on disk is smaller than the objects existing in
1293 the O/0/d* directories, it indicates disk corruption and causes an
1294 LBUG(). If the object is 0-length, then we should use the existing
1295 object. This will help to avoid a full fsck in most cases.
1297 Severity : enhancement
1299 Description: adapt the lustre_config script to support the upgrade case
1300 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1301 server targets to 1.6 in parallel.
1305 Description: To avoid grant space > avaible space when the disk is almost
1306 full. Without this patch you might see the error "grant XXXX >
1307 available" or some LBUG about grant, when the disk is almost
1309 Details : In filter_check_grant, for non_grant cache write, we should
1310 check the left space by if (*left > ungranted + bytes), instead
1311 of (*left > ungranted), because only we are sure the left space
1312 is enough for another "bytes", then the ungrant space should be
1313 increase. In client, we should update cl_avail_grant only there
1314 is OBD_MD_FLGRANT in the reply.
1318 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1319 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1320 described in CVE-2007-4573. This problem could allow untrusted
1321 local users to gain root access.
1324 Frequency : when using O_DIRECT and quotas
1326 Description: Incorrect file ownership on O_DIRECT output files
1327 Details : block usage reported by 'lfs quota' does not take into account
1328 files that have been written with O_DIRECT.
1333 Description: touch file failed when fs is not full
1334 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1335 otherwise we can get ENOSP while fs is not full.
1339 Description: parallel lock callbacks
1340 Details : Instead of sending blocking and completion callbacks as separated
1341 requests, adding them to a set and sending in parallel.
1344 Frequency : only for Cray XT3
1345 Bugzilla : 12829/13455
1346 Description: Changing primary group doesn't change the group lustre assigns to
1348 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1349 overridden with the primary group provided by the group upcall,
1350 whereas the supplied fsgid can be trusted if it is in the list of
1351 supplementary groups returned by the group upcall.
1353 Severity : enhancement
1355 Description: Allow masking D_WARNING, D_ERROR messages from console
1356 Details : Console messages can now be disabled via lnet.printk.
1360 Description: User code with malformed file open parameter crashes client node
1361 Details : Before packing join_file req, all the related reference should be
1362 checked carefully in case some malformed flags cause fake
1363 join_file req on client.
1367 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1368 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1369 mode is equal to the granted mode. However, it should also check
1370 that LDLM_FL_CP_REQD is not set.
1374 Description: Heavy nfs access might result in deadlocks
1375 Details : After ELC code landed, it is now improper to enqueue any mds
1376 locks under och_sem, because enqueue might want to decide to
1377 cancel open locks for same inode we are holding och_sem for.
1381 Description: Client eviction while running blogbench
1382 Details : A lot of unlink operations with concurrent I/O can lead to a
1383 deadlock causing evictions. To address the problem, the number of
1384 oustanding OST_DESTROY requests is now throttled to
1385 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1386 callbacks are processed in priority.
1390 Description: enable ACLs on MDS by default
1391 Details : ACLs must be enabled on MDS by default.
1394 Frequency : PPC/PPC64 only
1396 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1397 Details : fix duplicated definitions between asm-ppc64/types.h and
1398 lustre_types.h on PPC.
1401 Frequency : PPC/PPC64 only
1403 Description: asm-ppc/segment.h does not exist
1404 Details : fix compile issue on PPC.
1408 Description: better handle error messages in extents code
1411 Frequency : RHEL4 only
1413 Description: mkfs is very slow on IA64/RHEL4
1414 Details : A performance regression has been discovered in the MPT Fusion
1415 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1416 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1417 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1419 Severity : enhancement
1421 Description: SNMP support enhancement
1422 Details : Adding total number of sampled request for an MDS node in snmp
1425 Severity : enhancement
1427 Description: Optimize ldlm waiting list processing for PR extent locks
1428 Details : When processing waiting list for read extent lock and meeting read
1429 lock that is same or wider to it that is not contended, skip
1430 processing rest of the list and immediatelly return current
1431 status of conflictness, since we are guaranteed there are no
1432 conflicting locks in the rest of the list.
1436 Description: Time out and refuse to reconnect
1437 Details : When the failover node is the primary node, it is possible
1438 to have two identical connections in imp_conn_list. We must
1439 compare not conn's pointers but NIDs, otherwise we can defeat
1440 connection throttling.
1444 Description: port llog fixes from b1_6 into HEAD
1445 Details : Port llog reference couting and some llog cleanups from b1_6
1446 (bug 10800) into HEAD, for protect from panic and access to already
1447 free llog structures.
1451 Description: Detect stride IO mode in read-ahead
1452 Details : When a client does stride read, read-ahead should detect that and
1453 read-ahead pages according to the detected stride pattern.
1457 Description: data checksumming impacts single node performance
1458 Details : add support for several checksum algorithm. Currently, only CRC32
1459 and Adler-32 are supported. The checksum type can be changed on
1460 the fly via /proc/fs/lustre/osc/*/checksum_type.
1464 Description: use adler32 for page checksums
1465 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1470 Description: build for x2 fails
1471 Details : fix compile issue on Cray systems.
1475 Description: Properly match for duplicate locks
1476 Details : Due to different lock order from skiplists code, we need to
1477 traverse entire list for now
1480 Frequency : only on PPC/SLES10
1482 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1483 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1484 be defined. Add a hack in mkfs_lustre.c to work around this compile
1489 Description: LASSERT on MDS when client holding flock lock dies
1490 Details : ldlm pool logic depends on number of granted locks equal to
1491 number of released locks which is not true for flock locks, so
1492 just exclude such locks from consideration.
1496 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1497 Details : Use fsfilt_sync() for both the whole filesystem sync and
1498 individual file sync to eliminate dangerous inode locking
1499 with I_LOCK that can lead to a deadlock.
1503 Description: performance in 1.6.3
1504 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1505 sure that 1MB requests are not fragmented by the block layer.
1507 Severity : enhancement
1509 Description: organize the server-side client stats on per-nid basis
1510 Details : Change the structure of stats under obdfilter and mds to
1518 The "uuid"s file would list the uuids of _active_ exports.
1519 And the clear entry is to clear all stats and stale nids.
1521 Severity : enhancement
1523 Description: eliminate client locks in face of contention
1524 Details : file contention detection and lockless i/o implementation
1525 for contended files.
1529 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1532 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1534 Description: MDS may lose file striping (and hence file data) in some cases
1535 Details : If there are additional extended attributes stored on the MDS,
1536 in particular ACLs, SELinux, or user attributes (if user_xattr
1537 is specified for the client mount options) then there is a risk
1538 of attribute loss. Additionally, the Lustre file striping
1539 needs to be larger than default (e.g. striped over all OSTs),
1540 and an additional attribute must be stored initially in the
1541 inode and then increase in size enough to be moved to the
1542 external attribute block (e.g. ACL growing in size) for file
1547 Description: skiplist implementation simplification
1548 Details : skiplists are used to group compatible locks on granted list
1549 that was implemented as tracking first and last lock of each
1550 lock group the patch changes that to using doubly linked lists
1554 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1555 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1556 because LASSERT on both the data supplied by a client, and the
1557 data on disk is dangerous and incorrect.
1559 Severity : enhancement
1561 Description: Slow truncate/writes to huge files at high offsets.
1562 Details : Directly associate cached pages to lock that protect those pages,
1563 this allows us to quickly find what pages to write and remove
1564 once lock callback is received.
1568 Description: more ldlm soft lockups
1569 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1570 starve other threads from the resource lock for a long time in
1571 case of long waiting queue, so change the debug level from
1572 D_OTHER to the less frequently used D_INFO.
1574 Severity : enhancement
1576 Description: add -gid, -group, -uid, -user options to lfs find
1580 Description: Hung threads in invalidate_inode_pages2_range
1581 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1582 once one is completed. As a result, some RPCs are stuck in the
1583 queue and are never sent.
1587 Description: filter threads hungs on waiting journal commit
1588 Details : Cleanup filter group llog code, then only filter group llog will
1589 be only created in the MDS/OST syncing process.
1593 Description: Procfs and llog threads access destoryed import sometimes.
1594 Details : Sync the import destoryed process with procfs and llog threads by
1595 the import refcount and semaphore.
1597 Severity : enhancement
1599 Description: openlock cache of b1_6 port to HEAD
1604 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1605 Details : If insertion of an extent fails, then discard the inode
1606 preallocation and free data blocks else it can lead to duplicate
1611 Description: don't always update ctime in ext3_xattr_set_handle()
1612 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1613 In some cases the ctime should not be updated, for example for
1614 2.0->1.8 compatibility it is necessary to delete an xattr and it
1615 should not update the ctime.
1619 Bugzilla : 15713/16362
1620 Description: Assertion in iopen_connect_dentry in 1.6.3
1621 Details : looking up an inode via iopen with the wrong generation number can
1622 populate the dcache with a disconneced dentry while the inode
1623 number is in the process of being reallocated. This causes an
1624 assertion failure in iopen since the inode's dentry list contains
1625 both a connected and disconnected dentry.
1629 Description: assertion failure in ldlm_handle2lock()
1630 Details : fix a race between class_handle_unhash() and class_handle2object()
1631 introduced in lustre 1.6.5 by bug 13622.
1636 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1637 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1642 Description: oops in sd_iostats_seq_show()
1643 Details : unloading/reloading the scsi low level driver triggers a kernel
1644 bug when trying to access the sd iostat file.
1649 Description: Kernel panics during QLogic driver reload
1650 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1651 patch, causing memory corruption.
1656 Description: journal_dev option does not work in b1_6
1657 Details : pass mount option during pre-mount.
1659 Severity : enhancement
1661 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1662 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1663 extent information of a file. It can be used to map logical blocks
1664 in a file to physical blocks in the block device.
1668 Description: LDLM soft lockups - improvement
1669 Details : It is be possible to send the lock handle along with each read
1670 or write request because the client is already doing a lock match
1671 itself so there isn't any reason the OST should have to re-do that
1675 Frequency : only on Cray X2
1677 Description: X2 build failures
1678 Details : fix build failures on Cray X2.
1682 Description: xid & resent requests
1683 Details : Initialize RPC XID from clock at startup (randomly if clock is
1686 Severity : enhancement
1688 Description: Add lustre_start utility to start or stop multiple Lustre servers
1693 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1694 Details : In case of memory pressure, list_del() can be called twice on
1695 req->rq_history_list, causing a kernel oops.
1699 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1700 Details : fix stack overflow in the distributed lock manager by defering
1701 export eviction after a failed AST to the elt thread instead of
1702 handling it in the dlm interpret routine.
1706 Description: Convert some comments to new format.
1707 Details : Update documenting comments to match doxygen conventions.
1711 Description: Grammar fixes.
1712 Details : A couple of trivial spelling fixes.
1716 Description: OSD_COUNTERS-mandatory
1717 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1718 to simplify the code.
1722 Description: simplify cmm_device freeing logic.
1723 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1727 Description: Add lockdep support to dt_object_operations locking interface.
1728 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1729 parameter indicating lock ordering. Update mdd code to use new
1734 Description: Introduce failloc constants for lockless IO tests.
1735 Details : Add two new failloc constants to test lockless IO. Only one of
1736 them in implemented---another is checked in yet to be landed
1741 Description: Add lockdep support for inode mutex.
1742 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1743 in the situations where Lustre has to lock more than one inode
1748 Description: Add optional invariants checking support.
1749 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1750 configure switch. This macro is to be used for consistency and
1751 sanity checks that are too expensive to be left in `production'
1756 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1757 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1762 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1763 Details : Use spin_lock_nested() in (the only) situation where more than
1764 one ldlm_lock is locked simultaneously. Also, fix possible
1765 dead-lock in ldlm_lock_change_resource() by enforcing particular
1770 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1771 Details : Instead of specifying each ldlm_lock call-back through separate
1772 parameter, wrap them into struct ldlm_callback_suite.
1776 Description: Kill join_lru obd method and its callers.
1777 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1778 in memory---a requirement implemented through ->o_join_lru() obd
1779 method in HEAD. Remove this method and its users.
1783 Description: Add asynchronous ldlm ENQUEUE completion handler.
1784 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1785 case that stock ldlm_completion_ast() cannot handle as it waits
1786 until lock is granted. Introduce new ldlm_completion_ast_async()
1787 for this. Also comment ldlm_completion_ast().
1791 Description: ldlm_error <-> errno conversion.
1792 Details : Add functions to map (rather arbitrary) between LDLM error codes
1793 and standard errno values. CLIO needs this to prevent LDLM specific
1794 constants from escaping ldlm and osc.
1798 Description: Kill unused ldlm_handle2lock_ns() function.
1799 Details : Kill unused ldlm_handle2lock_ns() function.
1803 Description: Add lu_ref support to ldlm_lock
1804 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1805 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1806 and ldlm_resource. LDLM interface has to be changed, because code that
1807 releases a reference on a lock, has to "know" what reference this is.
1808 In the most frequent case
1810 lock = ldlm_handle2lock(handle);
1812 LDLM_LOCK_PUT(lock);
1814 no changes are required. When any other reference (received _not_ from
1815 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1816 instead of LDLM_LOCK_PUT().
1818 Arguably, changes are pervasive, and interface requires some discipline
1819 for proper use. On the other hand, it was very instrumental in finding
1820 a few leaked lock references.
1824 Description: Add ldlm_lock_addref_try().
1825 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1826 attempts to addref a lock that might be being canceled
1831 Description: Add ldlm_weigh_callback().
1832 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1833 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1834 hard-coded `number of pages' logic.
1838 Description: Add lockdep annotations to llog code.
1839 Details : Use appropriately tagged _nested() locking calls in the places
1840 where llog takes more than one ->lgh_lock lock.
1844 Description: Add loi_kms_set().
1845 Details : Wrap kms updates into a helper function.
1849 Description: Constify instances of struct lsm_operations.
1850 Details : Constify instances of struct lsm_operations.
1854 Description: lu_conf support.
1855 Details : On a server, a file system object is uniquely identified
1856 by a fid, which is sufficient to locate and load all object
1857 state (inode). On a client, on the other hand, more data are
1858 necessary instantiate an object. Change lu_object_find() and
1859 friends to take additional `lu_conf' argument describing object.
1860 Typically this includes layout information.
1864 Description: lu_context fixes.
1865 Details : Introduce new lu_context functions that are needed on the client
1866 side, where some system threads (ptlrpcd) are shared by multiple
1867 modules, and so cannot be stopped during module shutdown.
1871 Description: Add start and stop methods to lu_device_type_operations.
1872 Details : Introduce two new methods in lu_device_type_operations, that are
1873 invoked when first instance of a given type is created and last one
1874 is destroyed respectively. This is need by CLIO.
1878 Description: Add lu_ref support to struct lu_device.
1879 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1880 track leaked references.
1884 Description: Introduce lu_kmem_descr.
1885 Details : lu_kmem_descr and its companion interface allow to create
1886 and destroy a number of kmem caches at once.
1890 Description: Fix lu_object finalization race.
1891 Details : Fix a race between lu_object_find() finding an object and its
1892 concurrent finalization. This race is (most likely) not possible
1893 on the server, but might happen on the client.
1897 Description: Introduce lu_ref interface.
1898 Details : lu_ref is a debugging module allowing to track references to
1899 a given object. It is quite cpu expensive, and has to be
1900 explicitly enabled with --enable-lu_ref. See usage description
1905 Description: Factor lu_site procfs stats into a separate function.
1906 Details : Separate lu_site stats printing code into a separate function
1907 that can be reused on a client.
1911 Description: Constify instances of struct {lu,dt,md}_device_operations.
1912 Details : Constify instances of struct {lu,dt,md}_device_operations.
1916 Description: Introduce struct md_site and move meta-data specific parts of
1917 struct lu_site here.
1918 Details : Move md-specific fields out of struct lu_site into special struct
1919 md_site, so that lu_site can be used on a client.
1923 Description: Kill mdd_lov_destroy().
1924 Details : Remove unused mdd code.
1928 Description: Add st_block checking to multistat.c.
1929 Details : Add st_block checking to multistat.c.
1933 Description: Add lu_ref support to struct obd_device.
1934 Details : Add lu_ref tracking to obd_device.
1938 Description: Kill obd_set_fail_loc().
1939 Details : Remove unused code.
1943 Description: Add special type for ptlrpc_request interpret functions.
1944 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1945 there. Actual usage will be in CLIO.
1949 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1950 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1951 doesn't work with lockdep.
1955 Description: Add rwv.c test program.
1956 Details : New testing program exercising readv(2) and writev(2) (Qian).
1960 Description: Add sendfile.c test program.
1961 Details : New testing program exercising sendfile(2) (Jay).
1965 Description: Ratelimit a message that can be very frequent.
1966 Details : Ratelimit a memory allocation failure message that can
1971 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1972 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1976 Description: Liblustre build fixes.
1977 Details : Liblustre build fixes.
1981 Description: libcfs: add cfs_{need,cond}_resched() interface.
1982 Details : libcfs: add cfs_{need,cond}_resched() definition and
1983 implementations for Linux, NT, and liblustre.
1985 Severity : enhancement
1987 Description: More exported tunables for mballoc
1988 Details : Add support for tunable preallocation window and new tunables for
1989 large/small requests
1993 Description: Detect corruption of block bitmap and checking for preallocations
1994 Details : Checks validity of on-disk block bitmap. Also it does better
1995 checking of number of applied preallocations. When corruption is
1996 found, it turns filesystem readonly to prevent further corruptions.
2000 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2001 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2002 window to 0. If it does hit cache miss, and read-pattern is still
2003 stride-io mode, does not reset the stride window, but also does not
2004 increase the stride window length in this case.
2008 Frequency : only for big-endian servers
2009 Description: Check if system is big-endian while mounting fs with extents feature
2010 Details : Mounting a filesystem with extents feature will fail on big-endian
2011 systems since ext3-based ldiskfs is not supported on big-endian
2012 systems. This can be over-riden with "bigendian_extents" mount option.
2014 Severity : enhancement
2016 Description: The root squash functionality
2017 Details : A security feature, which is to prevent users from being able
2018 to mount lustre on their desktop, run as root, and delete
2019 all of the files in the filesystem. The goal is accomplished by
2020 remapping user id (UID) and group id (GID) of the root user to
2021 a UID and GID specified by the system administartor via Lustre
2022 configuration management server (MGS). The functionality also
2023 allows to specify sets of clients for which the remapping does
2028 Description: Excessive recovery window
2029 Details : With AT enabled, the recovery window can be excessively long (6000+
2030 seconds). To address this problem, we no longer use
2031 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2032 timeout no longer depends on the service time, it is set to
2033 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2034 time via pb_service_time.
2038 Description: Watchdog triggered on MDS failover
2039 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2040 the OSTs know that the MDS "UUID" can be reused for the same export
2041 from a different NID, so we do not need to wait for the export to be
2045 Frequency : rare, only if using MMP with Linux RAID
2047 Description: MMP doesn't work with Linux RAID
2048 Details : While using HA for Lustre servers with Linux RAID, it is possible
2049 that MMP will not detect multiple mounts. To make this work we
2050 need to unplug the device queue in RAID when the MMP block is being
2051 written. Also while reading the MMP block, we should read it from
2052 disk and not the cached one.
2054 Severity : enhancement
2056 Description: open file using fid
2057 Details : A file can be opened using just its fid, like
2058 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2061 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2064 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2065 Details : add proper locking for creation of jbd_4k slab cache
2069 Description: MMP check in ext3_remount() fails without displaying any error
2070 Details : When multiple mount protection fails during remount, proper error
2073 --------------------------------------------------------------------------------
2075 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2077 * Support for kernels:
2078 2.6.5-7.283 (SLES 9),
2079 2.6.9-55.EL (RHEL 4),
2080 2.6.16.46-0.14 (SLES 10),
2081 2.6.18.8 vanilla (kernel.org)
2082 * Client support for unpatched kernels:
2083 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2084 2.6.16 - 2.6.22 vanilla (kernel.org)
2085 * Due to recently discovered recovery problems, we do not recommend
2086 using patchless RHEL 4 clients with this or any earlier release.
2087 * Recommended e2fsprogs version: 1.39.cfs8
2088 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2089 * Starting with this release, the ldiskfs backing filesystem required
2090 by Lustre is now in its own package, lustre-ldiskfs. This package
2091 should be installed. It is versioned separately from Lustre and
2092 may be released separately in future.
2096 Description: block reactivating mgc import until all deactivates complete
2097 Details : Fix race when failing back MDT/MGS to itself (testing)
2100 Frequency : at statup only
2102 Description: mds_lov_synchronize race leads to various problems
2103 Details : simultaneous MDT->OST connections at startup can cause the
2104 sync to abort, leaving the OSC in a bad state.
2106 Severity : enhancement
2108 Description: add optional extra BUILD_VERSION info
2109 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2110 to override the lustre version.
2113 Frequency : 2.6.18 servers only
2115 Description: ll_kern_mount() doesn't release the module reference
2116 Details : The ldiskfs module reference count never drops down to 0
2117 because ll_kern_mount() doesn't release the module reference.
2122 Description: server LBUG when using old ost_num_threads parameter
2123 Details : Accept the old ost_num_threads parameter but warn that it
2124 is deprecated, and fix an off-by-one error that caused an LBUG.
2129 Description: Transient SCSI error results in persistent IO issue
2130 Details : iobuf->dr_error is not reinitialized to 0 between two
2134 Frequency : sometimes when underlying device returns I/O errors
2136 Description: OSTs not going read-only during write failures
2137 Details : OSTs are not remounted read-only when the journal commit threads
2138 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2139 instead of the ext3 wrappers.
2143 Description: poor connect scaling with increasing client count
2144 Details : Don't run filter_grant_sanity_check for more than 100 exports
2145 to improve scaling for large numbers of clients.
2148 Frequency : SLES10 only
2150 Description: sanity-quota.sh quotacheck failed: rc = -22
2151 Details : Quotas cannot be enabled on SLES10.
2154 Frequency : liblustre clients only
2156 Description: getdirentries does not give error when run on compute nodes
2157 Details : getdirentries does not fail when the size specified as an argument
2158 is too small to contain at least one entry
2160 Severity : enhancement
2162 Description: Add LNET router traceability for debug purposes
2163 Details : If a checksum failure occurs with a router as part of the
2164 IO path, the NID of the last router that forwarded the bulk data
2165 is printed so it can be identified.
2170 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2171 Details : Due to a race condition, liblustre clients were occasionally
2172 evicted incorrectly.
2174 Severity : enhancement
2176 Description: lfs setstripe use optional parameters instead of postional
2179 Severity : enhancement
2181 Description: Nanosecond timestamp support for ldiskfs
2182 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2183 resolution timestamps. There is not yet support for this at
2184 the Lustre filesystem level.
2187 Frequency : during server recovery
2189 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2190 Details : request with rq_no_resend flag not awake l_wait_event if they get
2194 Frequency : nfs export on patchless client
2196 Description: connectathon hang when test nfs export over patchless client
2197 Details : Disconnected dentry cannot be found with lookup, so we do not need
2198 to unhash it or make it invalid
2201 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2202 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2203 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2208 Description: open req refcounting wrong on reconnect
2209 Details : If reconnect happened between getting open reply from server and
2210 call to mdc_set_replay_data in ll_file_open, we will schedule
2211 replay for unreferenced request that we are about to free.
2212 Subsequent close will crash in variety of ways.
2213 Check that request is still eligible for replay in
2214 mdc_set_replay_data().
2219 Description: disable writes to filesystem when reading health_check file
2220 Details : the default for reading the health_check proc file has changed
2221 to NOT do a journal transaction and write to disk, because this
2222 can cause reads of the /proc file to hang and block HA state
2223 checking on a healthy but otherwise heavily loaded system. It
2224 is possible to return to the previous behaviour during configure
2225 with --enable-health-write.
2227 Severity : enhancement
2229 Description: 64-bit inode version
2230 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2231 to the inode. This will be required for version-based recovery.
2236 Description: MDS fails to start if a duplicate client export is detected
2237 Details : in some rare cases it was possible for a client to connect to
2238 an MDS multiple times. Upon recovery the MDS would detect this
2239 and fail during startup. Handle this more gracefully.
2241 Severity : enhancement
2243 Description: Add -o localflock option to simulate old noflock
2245 Details : This will achieve local-only flock/fcntl locks
2251 Description: log_commit_thread vs filter_destroy race leads to crash
2252 Details : Take import reference before releasing llog record semaphore
2257 Description: Wrong request locking in request set processing
2258 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2259 imp_delayed_list, in this place should be used imp_lock.
2262 Frequency : when reconnection
2264 Description: Grant Leak when osc reconnect to OST
2265 Details : When osc reconnect ost, OST(filter) should check whether it
2266 should grant more space to client by comparing fed_grant and
2267 cl_avail_grant, and return the granted space to client instead
2268 of "new granted" space, because client will call osc_init_grant
2269 to update the client grant space info.
2272 Frequency : when client reconnect to OST
2274 Description: Grant Leak when osc do resend and replay bulk write
2275 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2276 bulk write request, because the grant info while be sync between
2277 OSC and OST when reconnect, and we should ignore the grant info
2278 these of resend/replay write req.
2283 Description: Grant space more than avaiable left space sometimes.
2284 Details : When then OST is about to be full, if two bulk writing from
2285 different clients came to OST. Accord the avaliable space of the
2286 OST, the first req should be permitted, and the second one
2287 should be denied by ENOSPC. But if the seconde arrived before
2288 the first one is commited. The OST might wrongly permit second
2289 writing, which will cause grant space > avaiable space.
2292 Frequency : when client is evicted
2294 Description: Grant might be wrongly erased when osc is evicted by OST
2295 Details : when the import is evicted by server, it will fork another
2296 thread ptlrpc_invalidate_import_thread to invalidate the
2297 import, where the grant will be set to 0. While the original
2298 thread will update the grant it got when connecting. So if
2299 the former happened latter, the grant will be wrongly errased
2300 because of this race.
2305 Description: Checking Stale with correct fid
2306 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2307 to check whether it is stale, because sometimes, we want the
2308 enqueue happened anyway, and op_data.fid2 will not be initialized.
2310 Severity : enhancement
2312 Description: update patchless client
2313 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2316 Frequency : only with 2.4 kernel
2318 Description: random memory corruption
2319 Details : size of struct ll_inode_info is to big for union inode.u and this
2320 can be cause of random memory corruption.
2325 Description: Memory leak in recovery
2326 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2327 It should also check obd_fail before fsfilt_start, otherwise if
2328 fsfilt_start return -EROFS,(failover mds during mds recovery).
2329 then the req will return with repmsg->transno = 0 and rc = EROFS.
2330 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2331 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2332 be freed no matter whether fsfilt_commit success or not.
2335 Frequency : only with huge count clients
2337 Description: Prevents from taking the superblock lock in llap_from_page for
2339 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2340 for llap_from_page call in ll_removepage prevents from taking the
2341 superblock lock for a soon died page.
2346 Description: Not check open intent error before release open handle
2347 Details : in some rare cases, the open intent error is not checked before
2348 release open handle, which may cause
2349 ASSERTION(open_req->rq_transno != 0), because it tries to release
2350 the failed open handle.
2355 Description: Set cat log bitmap only after create log success.
2356 Details : in some rare cases, the cat log bitmap is set too early. and it
2357 should be set only after create log success.
2361 Description: Accessing a block bevice can re-enable I/O when Lustre is
2362 tearing down a device.
2363 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2368 Description: service threads may hog cpus when there are a lot of requests
2369 Details : Insert cond_resched to give other threads a chance to use some CPU
2374 Description: the cat log was not initialized in recovery
2375 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2376 unlink log on mds will not be initialized until mds post
2377 recovery. And also in mds post recovery, the unlink log will
2378 initialization will be done asynchronausly, so there will be race
2379 between add unlink log and unlink log initialization.
2383 Description: brw_stats were being printed incorrectly
2384 Details : brw_stats were being printed as log2 but all of them were not
2385 recorded as log2. Also remove some code duplication arising from
2386 filter_tally_{read,write}.
2390 Frequency : rare, only in recovery.
2391 Description: ASSERTION(req->rq_type != LI_POISON) failed
2392 Details : imp_lock should be held while iterating over imp_sending_list for
2393 prevent destroy request after get timeout in ptlrpc_queue_wait.
2397 Description: replay-single.sh test 52 fails
2398 Details : A lock's skiplist need to be cleanup when it being unlinked
2399 from its resource list.
2403 Description: Short directio read returns full requested size rather than
2405 Details : Direct I/O operations should return actual amount of bytes
2406 transferred rather than requested size.
2408 Severity : enhancement
2410 Description: metadata RPC reduction (e.g. for rm performance)
2411 Details : decrease the amount of synchronous RPC between clients and servers
2412 by canceling conflicing lock before the operation on the client
2413 and packing thier handles into the main operation RPC to server.
2415 Severity : enhancement
2417 Description: Async OSC create to avoid the blocking unnecessarily.
2418 Details : If a OST has no remain object, system will block on the creating
2419 when need to create a new object on this OST. Now, ways use
2420 pre-created objects when available, instead of blocking on an
2421 empty osc while others are not empty. If we must block, we block
2422 for the shortest possible period of time.
2426 Description: improve handling recoverable errors
2427 Details : if request processig with error which can be recoverable on server
2428 request should be resend, otherwise page released from cache and
2431 Severity : enhancement
2433 Description: refine locking for avoid write wrong info into lov_objid file
2434 Details : fix possible races with add new target and write/update data in
2437 --------------------------------------------------------------------------------
2439 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2444 Frequency : on some architectures
2446 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2447 Details : Uninitialized flags sometimes cause configuration commands to
2451 Frequency : patchless clients only
2453 Description: missing __iget() symbol export
2454 Details : The __iget() symbol export is missing. To avoid the need for
2455 this on patchless clients the deathrow inode reaper is turned
2456 off, and we depend on the VM to clean up old inodes. This
2457 dependency was during via the fix for bug 12181.
2459 --------------------------------------------------------------------------------
2461 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2463 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2464 INTEROPERATE with older versions automatically. In many cases a
2465 special upgrade step is needed. Please read the
2466 user documentation before upgrading any part of a live system.
2467 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2468 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2469 * WARNING: Lustre configuration and startup changes are required with
2470 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2474 Severity : enhancement
2476 Description: MountConf
2477 Details : Lustre configuration is now managed via mkfs and mount
2478 commands instead of lmc and lconf. New obd types (MGS, MGC)
2479 are added for dynamic configuration management. See
2480 https://mail.clusterfs.com/wikis/lustre/MountConf for
2483 Severity : enhancement
2485 Description: dynamic OST addition
2486 Details : OSTs can now be added to a live filesystem
2488 Severity : enhancement
2490 Description: startup order invariance
2491 Details : MDTs and OSTs can be started in any order. Clients only
2492 require the MDT to complete startup.
2494 Severity : enhancement
2496 Description: parallel, asynchronous orphan cleanup
2497 Details : orphan cleanup is now performed in separate threads for each
2498 OST, allowing parallel non-blocking operation.
2500 Severity : enhancement
2502 Description: optimized stripe assignment
2503 Details : stripe assignments are now made based on ost space available,
2504 ost previous usage, and OSS previous usage, in order to try
2505 to optimize storage space and networking resources.
2507 Severity : enhancement
2509 Description: Permanently set tunables
2510 Details : All writable /proc/fs/lustre tunables can now be permanently
2511 set on a per-server basis, at mkfs time or on a live system.
2513 Severity : enhancement
2515 Description: Lustre message v2
2516 Details : Add lustre message format v2.
2518 Severity : enhancement
2520 Description: client OST exclusion list
2521 Details : Clients can be started with a list of OSTs that should be
2522 declared "inactive" for known non-responsive OSTs.
2526 Description: SPEC SFS validation failure on NFS v2 over lustre.
2527 Details : Changes the blocksize for regular files to be 2x RPC size,
2528 and not depend on stripe size.
2530 Severity : enhancement
2532 Description: Multiple MD RPCs in flight.
2533 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2534 To avoid overly-overloading MDS, introduce a limit on number of
2535 MDS RPCs in flight for a single client and add /proc controls
2536 to adjust this limit.
2538 Severity : enhancement
2540 Description: client read/write statistics
2541 Details : Add client read/write call usage stats for performance
2542 analysis of user processes.
2543 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2544 file access. extents_stats shows chunk size distribution.
2545 extents_stats_per_process show chunk size distribution per
2548 Severity : enhancement
2550 Description: mds statistics
2551 Details : Add detailed mds operations statistics in
2552 /proc/fs/lustre/mds/*/stats.
2556 Description: Failure of copying files with lustre special EAs.
2557 Details : Client side always return success for setxattr call for lustre
2558 special xattr (currently only "trusted.lov").
2563 Description: Refcount LNET uuids
2564 Details : The global LNET uuid list grew linearly with every startup;
2565 refcount repeated list entries instead of always adding to
2568 Severity : enhancement
2570 Description: Dynamic service threads
2571 Details : Within a small range, start extra service threads
2572 automatically when the request queue builds up.
2575 Frequency : mixed-endian client/server environments
2577 Description: mixed-endian crashes
2578 Details : The new msg_v2 system had some failures in mixed-endian
2581 Severity : enhancement
2583 Description: Easy OST removal
2584 Details : OSTs can be permanently deactivated with e.g. 'lctl
2585 conf_param lustre-OST0001.osc.active=0'
2587 Severity : enhancement
2589 Description: MGS proc entries
2590 Details : Added basic proc entries for the MGS showing what filesystems
2593 Severity : enhancement
2595 Description: provide MGS failover
2596 Details : Added config lock reacquisition after MGS server failover.
2598 Severity : enhancement
2600 Description: add Linux 2.4 support
2601 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2605 Description: a large application tries to do I/O to the same resource and dies
2606 in the middle of it.
2607 Details : Check the req->rq_arrival time after the call to
2608 ost_brw_lock_get(), but before we do anything about
2609 processing it & sending the BULK transfer request. This
2610 should help move old stale pending locks off the queue as
2611 quickly as obd_timeout.
2614 Frequency : when an incorrect nid is specified during startup
2616 Description: ptlrpc connect to non-existant node causes kernel crash
2617 Details : LNET can't be re-entered from an event callback, which
2618 happened when we expire a message after the export has been
2619 cleaned up. Instead, hand the zombie cleanup off to another
2622 Severity : enhancement
2624 Description: plain/inodebits lock performance improvement
2625 Details : Grouping plain/inodebits in granted list by their request modes
2626 and bits policy, thus improving the performance of search through
2630 Frequency : only if OST filesystem is corrupted
2632 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2633 Details : for a short time RPCs with bulk IO are in the replay list,
2634 but replay of bulk IOs is unimplemented. If the OST filesystem
2635 is corrupted due to disk cache incoherency and then replay is
2636 started it is possible to trip an assertion. Avoid putting
2637 committed RPCs into the replay list at all to avoid this issue.
2640 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2642 Description: System hang on startup
2643 Details : This bug allowed the liblustre (e.g. catamount) client to
2644 return to the app before handling all startup RPCs. This
2645 could leave the node unresponsive to lustre network traffic
2646 and manifested as a server ptllnd timeout.
2648 Severity : enhancement
2650 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2651 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2652 boolean to control whether to print peer debug info when a
2653 client's RPC times out.
2656 Frequency : only for kernels with patches from Lustre below 1.4.3
2658 Description: Remove old rdonly API
2659 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2662 Frequency : only for devices with external journals
2664 Description: Set external device read-only also
2665 Details : During a commanded failover stop, we set the disk device
2666 read-only while the server shuts down. We now also set any
2667 external journal device read-only at the same time.
2670 Frequency : when upgrading from 1.4 while trying to change parameters
2672 Description: The wrong (new) MDC name was used when setting parameters for
2673 upgraded MDT's. Also allows changing of OSC (and MDC)
2674 parameters if --writeconf is specified at tunefs upgrade time.
2677 Frequency : when setting specific ost indicies
2679 Description: QOS code breaks on skipped indicies
2680 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2681 created with --index need not be sequential.
2683 Severity : enhancement
2685 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2686 Details : The uninit_groups feature works in conjunction with the kernel
2687 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2688 up the pass1 processing of e2fsck. This is a read-only feature
2689 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2690 mount filesystems that have had this feature enabled.
2692 Severity : enhancement
2694 Description: Improve multi-block allocation algorithm to avoid fragmentation
2695 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2696 allocation locality and avoid filesystem fragmentation.
2698 ------------------------------------------------------------------------------
2700 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2702 * Support for kernels:
2703 2.4.21-47.0.1.EL (RHEL 3)
2704 2.6.5-7.283 (SLES 9)
2705 2.6.9-42.0.10.EL (RHEL 4)
2706 2.6.12.6 vanilla (kernel.org)
2707 2.6.16.27-0.9 (SLES 10)
2708 * Recommended e2fsprogs version: 1.39.cfs5
2710 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2714 Frequency : occasional, depends on client load and configuration
2715 Bugzilla : 12181, 12203
2716 Description: data loss for recently-modified files
2718 Details : In some cases it is possible that recently written or created
2719 files may not be written to disk in a timely manner (this should
2720 normally be within 30s unless client IO load is very high).
2721 The problem appears as zero-length files or files that are a
2722 multiple of 1MB in size after a client crash or client eviction
2723 that are missing data at the end of the file.
2725 This problem is more likely to be hit on clients where files are
2726 repeatedly created and unlinked in the same directory, clients
2727 have a large amount of RAM, have many CPUs, the filesystem has
2728 many OSTs, the clients are rebooted frequently, and/or the files
2729 are not accessed by other nodes after being written.
2731 The presence of the problem can be detected by looking at
2732 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2733 smaller than the second (nr_unused) then dirty files will not
2734 be flushed automatically to disk. "sync; sleep 10" should be
2735 run several times on the node before unmounting it to update
2736 Lustre (this is also safe to run on nodes without this problem).
2738 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2739 that can cause this same problem, so customers using that kernel
2740 also need to update the kernel in addition to Lustre. In order
2741 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2743 It is normal that files written just before a client crash (less
2744 than 30s) may not yet have been flushed to disk, even for local
2748 Frequency : frequent on thin XT3 nodes
2750 Description: UUID collision on thin XT3 Linux nodes
2751 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2752 randomly, since we relied on an insufficiently-seeded PRNG.
2757 Description: OSS hangs after "All ost request buffers busy"
2758 Details : A deadlock between quota and journal operations caused OSS
2759 hangs after printing "All ost request buffers busy."
2762 Frequency : always on liblustre builds
2764 Description: Cleanup compiler warnings on liblustre
2767 Frequency : always on liblustre builds on XT3
2769 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2773 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2774 > 32000 subdirectories
2775 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2776 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2777 subdirectory count crosses 32000. This will aid e2fsck to
2778 correctly handle more than 32000 subdirectories.
2781 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2783 Description: System hang on startup
2784 Details : This bug allowed the liblustre (e.g. catamount) client to
2785 return to the app before handling all startup RPCs. This
2786 could leave the node unresponsive to lustre network traffic
2787 and manifested as a server ptllnd timeout.
2789 Severity : enhancement
2791 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2792 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2793 boolean to control whether to print peer debug info when a
2794 client's RPC times out.
2799 Description: make O_SYNC working on 2.6 kernels
2800 Details : 2.6 kernels use different method for mark pages for write,
2801 so need add a code to lustre for O_SYNC work.
2806 Description: Failure to close file and release space on NFS
2807 Details : Put inode details into lock acquired in ll_intent_file_open.
2808 Use mdc_intent_lock in ll_intent_open to properly
2809 detect all kind of errors unhandled by mdc_enqueue.
2814 Description: proc file read during shutdown sometimes raced obd removal,
2816 Details : Add lock to prevent obd access after proc file removal.
2819 Frequency : Only for files larger than 4GB on 32-bit clients.
2821 Description: improperly doing page alignment of locks
2822 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2823 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2828 Description: RPCs being resent when they shouldn't be
2829 Details : Some RPCs that should not be resent are being resent. This
2830 can cause inconsistencies in the RPC state machine. Do not
2831 resend such requests.
2834 Frequency : rare, only with NFS export
2836 Description: Crash on NFS re-export node
2837 Details : under very unusual load conditions an assertion is hit in
2838 ll_intent_file_open()
2841 Frequency : only if OST filesystem is corrupted
2843 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2844 Details : for a short time RPCs with bulk IO are in the replay list,
2845 but replay of bulk IOs is unimplemented. If the OST filesystem
2846 is corrupted due to disk cache incoherency and then replay is
2847 started it is possible to trip an assertion. Avoid putting
2848 committed RPCs into the replay list at all to avoid this issue.
2853 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2854 Details : Large single O_DIRECT read and write calls can fail to allocate
2855 a sufficiently large buffer to process the request. In case of
2856 allocation failure the allocation is retried with a smaller
2857 buffer and broken into smaller requests.
2859 Severity : enhancement
2861 Description: Add -o localflock option to simulate old noflock behaviour.
2862 Details : This will achieve local-only flock/fcntl locks coherentness.
2867 Description: versioning check is incomplete
2868 Details : Checking the version difference of client vs. server, report
2869 error if the gap is too big.
2874 Description: add support PG_writeback bit
2875 Details : add support for PG_writeback bit for Lustre, for more carefull
2876 work with page cache in 2.6 kernel. This also fix some deadlocks
2877 and remove hack for work O_SYNC with 2.6 kernel.
2879 Severity : enhancement
2881 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2882 Details : The uninit_groups feature works in conjunction with the kernel
2883 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2884 up the pass1 processing of e2fsck. This is a read-only feature
2885 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2886 mount filesystems that have had this feature enabled.
2888 Severity : enhancement
2890 Description: Improve multi-block allocation algorithm to avoid fragmentation
2891 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2892 allocation locality and avoid filesystem fragmentation.
2894 ------------------------------------------------------------------------------
2896 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2898 * Support for kernels:
2899 2.6.9-42.0.3.EL (RHEL 4)
2900 2.6.5-7.276 (SLES 9)
2901 2.4.21-47.0.1.EL (RHEL 3)
2902 2.6.12.6 vanilla (kernel.org)
2903 2.6.16.21-0.8 (SLES10)
2904 * Recommended e2fsprogs version: 1.39.cfs2-0
2906 * The backwards-compatible /proc/sys/portals symlink has been removed
2907 in this release. Before upgrading, please ensure that you change
2908 any configuration scripts or /etc/sysctl.conf files that access
2909 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2910 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2911 in advance of the upgrade on any system running Lustre 1.4.6 or
2912 newer, since /proc/sys/lnet was added in that version.
2913 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2917 Frequency : only when quota is used
2919 Description: avoid scanning export list for quota master
2920 Details : Change the algorithms to avoid scanning export list in order
2921 to improve the efficiency.
2924 Frequency : MDS failover only, very rarely
2926 Description: "went back in time" messages on mds failover
2927 Details : The greatest transno may be lost when the current operation
2928 finishes with an error (transno==0) and the client's last_rcvd
2929 record is over-written. Save the greatest transno in the
2930 mds_last_transno for this case.
2933 Frequency : always for specific kernels and striping counts
2935 Description: client may get "Matching packet too big" without ACL support
2936 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2937 when trying to access files in certain configurations. The
2938 clients should in fact be denied when mounting because they do
2939 not understand ACLs.
2942 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2944 Description: many clients connecting with IO in progress causes connect timeouts
2945 Details : Avoid synchronous journal commits to avoid delays caused by many
2946 clients connecting/disconnecting when bulk IO is in progress.
2947 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2948 OST_IO_PORTAL to avoid delays behind potentially many pending
2952 Frequency : occasionally with multiple writers to a single file
2954 Description: shared writes to file may result in wrong size reported by stat()
2955 Details : Allow growing of kms when extent lock is cancelled
2958 Frequency : always with random mmap IO to multi-striped file
2960 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2961 Details : Only if the hole is at the end of OST object so that kms is too
2962 small. Fix is to increase kms accordingly in ll_nopage.
2965 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2967 Description: writes to a missing object would leak memory on the OST
2968 Details : If there is an inconsistency between the MDS and OST filesystems,
2969 such that the MDS references an object that doesn't exist, writes
2970 to that object will leak memory due to incorrect cleanup in the
2971 error handling path, eventually running out of memory on the OST.
2976 Description: Creating too long symlink causes lustre errors
2977 Details : Check symlink and name lengths before sending requests to MDS.
2980 Frequency : only if flock is enabled (not on by default)
2982 Description: posix locks not released on fd closure on 2.6.9+
2983 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2984 kernels, this caused such locks not to be released on fd close and
2985 then assertions on fs unmount about still used locks.
2988 Frequency : MDS failover only, very rarely
2990 Description: clients may get ASSERTION(granted_lock != NULL)
2991 Details : When request was taking a long time, and a client was resending
2992 a getattr by name lock request. The were multiple lock requests
2993 with the same client lock handle and
2994 mds_getattr_name->fixup_handle_for_resent_request found one of the
2995 lock handles but later failed with ASSERTION(granted_lock != NULL).
3000 Description: handle->h_buffer_credits > 0, assertion failure
3001 Details : h_buffer_credits is zero after truncate, causing assertion
3002 failure. This patch extends the transaction or creates a new
3006 Frequency : NFS re-export or patchless client
3007 Bugzilla : 11179, 10796
3008 Description: Crash on NFS re-export node (__d_move)
3009 Details : We do not want to hash the dentry if we don't have a lock.
3010 But if this dentry is later used in d_move, we'd hit uninitialised
3011 list head d_hash, so we just do this to init d_hash field but
3012 leave dentry unhashed.
3015 Frequency : NFS re-export or patchless client
3017 Description: NFS exports has problem with symbolic link
3018 Details : lustre client didn't properly install dentry when re-exported
3019 to NFS or running patchless client.
3022 Frequency : NFS re-export or patchless client
3024 Description: Various nfs/patchless fixes.
3025 Details : fixes reuse disconected alias for lookup process - this fixes
3026 warning "find_exported_dentry: npd != pd",
3027 fix permission error with open files at nfs.
3028 fix apply umask when do revalidate.
3031 Frequency : occasional
3033 Description: Crash on NFS re-export node
3034 Details : calling clear_page() on the wrong pointer triggered oops in
3035 generic_mapping_read().
3038 Frequency : rarely, using O_DIRECT IO
3040 Description: unaligned directio crashes client with LASSERT
3041 Details : check for unaligned buffers before trying any requests.
3044 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3046 Description: stale data returned from RAID cache
3047 Details : If only a small amount of IO is done to the RAID device before
3048 reading it again it is possible to get stale data from the RAID
3049 cache instead of reading it from disk.
3052 Frequency : always for sles10 kernel
3054 Description: sles10 support
3055 Details : ll_follow_link: compile fixes and using of nd_set_link
3056 under newer kernels.
3059 Frequency : depends on arch, kernel and compiler version, always on sles10
3062 Description: recursive or deep enough symlinks cause stack overflow
3063 Details : getting rid of large stack-allocated variable in
3067 Frequency : depends on hardware
3069 Description: lustre write performance loss in the SLES10 kernel
3070 Details : the performance loss is caused by using of write barriers in the
3071 ext3 code. The SLES10 kernel turns barrier support on by
3072 default. The fix is to undo that change for ldiskfs.
3074 ------------------------------------------------------------------------------
3076 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3078 * Support for kernels:
3079 2.6.9-42.0.3EL (RHEL 4)
3080 2.6.5-7.276 (SLES 9)
3081 2.4.21-47.0.1.EL (RHEL 3)
3082 2.6.12.6 vanilla (kernel.org)
3086 Frequency : quota enabled and large files being deleted
3088 Description: releasing more than 4GB of quota at once hangs OST
3089 Details : If a user deletes more than 4GB of files on a single OST it
3090 will cause the OST to spin in an infinite loop. Release
3091 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3096 Description: statfs data retrieved from /proc may be stale or zero
3097 Details : When reading per-device statfs data from /proc, in the
3098 {kbytes,files}_{total,free,avail} files, it may appear
3099 as zero or be out of date.
3102 Frequency : systems with MD RAID1 external journal devices
3104 Description: lconf's call to blkid is confused by RAID1 journal devices
3105 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3106 of returning all block devices that match the journal UUID.
3109 Frequency : always, for aggregate stripe size over 4GB
3111 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3112 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3113 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3116 Frequency : always if "lfs find" used on a local file/directory
3118 Description: "lfs find" segfaults if used on a local file/directory
3119 Details : The case where a directory component was not specified wasn't
3120 handled correctly. Handle this properly.
3123 Frequency : always on ppc64
3125 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3126 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3127 perform endianness conversion.
3130 Frequency : rarely (truncate to non-zero file size after write under load)
3131 Bugzilla : 10730, 10687
3132 Description: Files padded with zeros to next 4K multiple
3133 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3134 it is possible that files that are truncated to a non-zero size
3135 immediately after being written are filled with zero bytes beyond
3136 the truncated size. No file data is lost.
3138 Severity : enhancement
3140 Description: Allow recovery/failover for liblustre clients.
3141 Details : liblustre clients were unaware of failover configurations until
3144 Severity : enhancement
3146 Description: user file locks should fail when not mounting with flock option
3147 Details : Set up an error-returning stub in ll_file_operations.lock field
3148 to prevent incorrect behaviour when client is mounted without
3149 flock option. Also, set up properly f_op->flock field for
3153 Frequency : always on ia64
3155 Description: "lfs df" loops on printing out MDS statfs information
3156 Details : The obd_ioctl_data was not initialized and in some systems
3157 this caused a failure during the ioctl that did not return
3158 an error. Initialize the struct and return an error on failure.
3161 Frequency : SLES 9 only
3163 Description: Error of copying files with lustre special EAs as root
3164 Details : Client side always return success for setxattr call for lustre
3165 special xattr (currently only "trusted.lov").
3168 Frequency : rarely on clusters with both ia64+i386 clients
3170 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3171 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3172 clients are doing concurrent writes to the same file it is
3173 possible that the ia64 clients may LASSERT because the OST
3174 extent locks are not PAGE_SIZE aligned. Ensure that grown
3175 locks are always aligned on the request boundary.
3178 Frequency : specific use, occasional
3180 Description: Overwriting in use executable truncates on-disk binary image
3181 Details : If one node attempts to overwrite an executable in use by
3182 another node, we now correctly return ETXTBSY instead of
3183 truncating the file.
3185 Severity : enhancement
3187 Description: Async OSC create to avoid the blocking unnecessarily.
3188 Details : If a OST has no remain object, system will block on the creating
3189 when need to create a new object on this OST. Now, ways use
3190 pre-created objects when available, instead of blocking on an
3191 empty osc while others are not empty. If we must block, we block
3192 for the shortest possible period of time.
3197 Description: chmod on Lustre root is propagated to other clients
3198 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3199 invalid by the follow_mount time.
3202 Frequency : liblustre clients only
3204 Description: Race in 'instant cancel' lock handling could lead to such locks
3205 never to be granted in case of SMP MDS
3206 Details : Do not destroy not yet granted but cbpending locks in
3210 Frequency : replay/resend of open
3212 Description: non null lock assetion failure in mds_intent_policy
3213 Details : Trying to replay/resend lockless open requests resulted in
3214 mds_open() returning 0 with no lock. Now it sets a flag if
3215 a lock is going to be returned.
3217 Severity : enhancement
3219 Description: Checksum enhancements
3220 Details : New checksum enhancements allow for resending RPCs that failed
3223 Severity : enhancement
3225 Description: Tunables on number of dirty pages in cacche
3226 Details : Allow to set limit on number of dirty pages cached.
3231 Description: client crash on unmount - lock still has references
3232 Details : In some error handling cases it was possible to leak a lock
3233 reference on a client while accessing a file. This was not
3234 harmful to the client during operation, but would cause the
3235 client to crash when the filesystem is unmounted.
3238 Frequency : specific case, rare
3240 Description: ETXTBSY on mds though file not in use by client
3241 Details : ETXTBSY is no longer incorrectly returned when attempting to
3242 chmod or chown a directory that the user previously tried to
3243 execute or a currently-executing binary.
3246 Frequency : extremely rare except on liblustre-based clients
3248 Description: Lustre space not freed when files are deleted
3249 Details : Clean up open-unlinked files after client eviction. Previously
3250 the unlink was skipped and the files remained as orphans.
3255 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3256 Details : In some cases it was possible to send a blocking callback to
3257 a client doing a glimpse, even though that client didn't get
3258 a lock granted. When the glimpse lock is cancelled on the OST
3259 the freed lock is left on the waiting list and corrupted the list.
3262 Frequency : all core dumps
3264 Description: Broke core dumps to lustre
3265 Details : Negative dentry may be unhashed if parent does not have UPDATE
3266 lock, but some callers, e.g. do_coredump, expect dentry to be
3267 hashed after successful create, hash it in ll_create_it.
3269 ------------------------------------------------------------------------------
3271 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3273 * Support for kernels:
3274 2.6.9-42.0.2.EL (RHEL 4)
3275 2.6.5-7.276 (SLES 9)
3276 2.4.21-40.EL (RHEL 3)
3277 2.6.12.6 vanilla (kernel.org)
3281 Frequency : always on RHEL 3
3283 Description: Number of open files grows over time
3284 Details : The number of open files grows over time, whether or not
3285 Lustre is started. This was due to a filp leak introduced
3286 by one of our kernel patches.
3288 ------------------------------------------------------------------------------
3290 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3292 * Support for kernels:
3293 2.6.9-42.EL (RHEL 4)
3294 2.6.5-7.276 (SLES 9)
3295 2.4.21-40.EL (RHEL 3)
3296 2.6.12.6 vanilla (kernel.org)
3301 Bugzilla : 5719, 9635, 9792, 9684
3302 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3303 Details : If a server is under heavy load and cannot reply to new
3304 connection requests before the client resends the (re)connect,
3305 the connection handling code can behave badly if two service
3306 threads are concurrently handing separate (re)connections from
3307 the same client. Add better locking to the connection handling
3308 code, and ensure that only a single connection will be processed
3309 for a given client UUID, even if the lock is dropped.
3311 Severity : enhancement
3313 Description: add TCP zero-copy support to kernel
3314 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3315 sends if the hardware supports scatter-gather and checksumming.
3316 This allows socklnd to do client-write and server-read more
3317 efficiently and reduce CPU utilization from skbuf copying.
3320 Frequency : only if NFS exporting from client
3322 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3323 Details : If setting attributes on a file created under NFS that had
3324 never been opened it would be possible to oops the client
3325 if the file had no objects.
3328 Frequency : always for liblustre
3330 Description: liblustre client does MDS+OSTs setattr RPC for each write
3331 Details : When doing a write from a liblustre client, the client
3332 incorrectly issued an RPC to the MDS and each OST the file was
3333 striped over in order to update the timestamps. When writing
3334 with small chunks and many clients this could overwhelm the MDS
3335 with RPCs. In all cases it would slow down the write because
3336 these RPCs are unnecessary.
3338 Severity : enhancement
3340 Description: allow number of MDS service threads to be changed at module load
3341 Details : It is now possible to change the number of MDS service threads
3342 running. Adding "options mds mds_num_threads={N}" to the MDS's
3343 /etc/modprobe.conf will set the number of threads for the next
3344 time Lustre is restarted (assuming the "mds" module is also
3345 reloaded at that time). The default number of threads will
3346 stay the same, 32 for most systems.
3351 Description: OST crash if filesystem is unformatted or corrupt
3352 Details : If an OST is started on a device that has never been formatted
3353 or if the filesystem is corrupt and cannot even mount then the
3354 error handling cleanup routines would dereference a NULL pointer.
3359 Description: NULL pointer deref in llap_from_page.
3360 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3361 incorrect) mapping if the page was truncated/reclaimed while it was
3362 searched for. Check for this condition and skip such pages when
3363 doing readahead. Introduce extra check to llap_from_page() to
3364 verify page->mapping->host is non-NULL (so page is not anonymous).
3367 Frequency : Sometimes when using sys_sendfile
3369 Description: "page not covered by a lock" warnings from ll_readpage
3370 Details : sendfile called ll_readpage without right page locks present.
3371 Now we introduced ll_file_sendfile that does necessary locking
3372 around call to generic_file_sendfile() much like we do in
3376 Frequency : with certain MDS communication failures at client mount time
3378 Description: NULL pointer deref after failed client mount
3379 Details : a client connection request may delayed by the network layer
3380 and not be sent until after the PTLRPC layer has timed out the
3381 request. If the client fails the mount immediately it will try
3382 to clean up before the network times out the request. Add a
3383 reference from the request import to the obd device and delay
3384 the cleanup until the network drops the request.
3387 Frequency : occasionally during client (re)connect
3389 Description: assertion failure during client (re)connect
3390 Details : processing a client connection request may be delayed by the
3391 client or server longer than the client connect timeout. This
3392 causes the client to resend the connection request. If the
3393 original connection request is replied in this interval, the
3394 client may trip an assertion failure in ptlrpc_connect_interpret()
3395 which thought it would be the only running connect process.
3398 Frequency : only with obd_echo servers and clients that are rebooted
3400 Description: kernel BUG accessing uninitialized data structure
3401 Details : When running an obd_echo server it did not start the ping_evictor
3402 thread, and when a client was evicted an uninitialized data
3403 structure was accessed. Start the ping_evictor in the RPC
3404 service startup instead of the OBD startup.
3406 Severity : enhancement
3407 Bugzilla : 10193 (patchless)
3408 Description: Remove dependency on various unexported kernel interfaces.
3409 Details : No longer need reparent_to_init, exit_mm, exit_files,
3410 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3413 Frequency : rare (only users of deprecated and unsupported LDAP config)
3415 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3416 Details : LDAP apparently contains 'lustreName' attributes instead of
3417 'name'. A simple remapping of the name is sufficient.
3420 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3422 Description: waiting_locks_callback trips kernel BUG if client is evicted
3423 Details : Running with the dump_on_timeout debug flag turned on makes
3424 it possible that the waiting_locks_callback() can try to dump
3425 the Lustre kernel debug logs from an interrupt handler. Defer
3426 this log dumping to the expired_lock_main() thread.
3428 Severity : enhancement
3430 Description: Support NFS exporting on 2.6 kernels.
3431 Details : Implement non-rawops metadata methods for NFS server to use without
3432 changing NFS server code.
3435 Frequency : very rare (synthetic metadata workload only)
3437 Description: two racing renames might cause an MDS thread to deadlock
3438 Details : Running the "racer" program may cause one MDS thread to rename
3439 a file from being the source of a rename to being the target of
3440 a rename at exactly the same time that another thread is doing
3441 so, and the second thread has already enqueued these locks after
3442 doing a lookup of the target and is trying to relock them in
3443 order. Ensure that we don't try to re-lock the same resource.
3446 Frequency : only very large systems with liblustre clients
3448 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3449 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3450 all OSTs in parallel. This allows the eviction of stale liblustre
3451 clients to proceed much faster than if they were done in series,
3452 and also offers similar improvements for other set_info RPCs.
3457 Description: excessive CPU usage during initial read phase on client
3458 Details : During the initial read phase on a client, it would agressively
3459 retry readahead on the file, consuming too much CPU and impacting
3460 performance (since 1.4.5.8). Improve the readahead algorithm
3461 to avoid this, and also improve some other common cases (read
3462 of small files in particular, where "small" is files smaller than
3463 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3468 Description: MDS crash when receiving packet with unknown intent.
3469 Details : Do not LBUG in unknown intent case, just return -EFAULT
3471 Severity : enhancement
3472 Bugzilla : 9293, 9385
3473 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3474 Details : Do not serialize getattr (non-intent version) and statfs.
3477 Frequency : occasional, when OST network is overloaded/intermittent
3479 Description: client evicted by OST after bulk IO timeout
3480 Details : If a client sends a bulk IO request (read or write) the OST
3481 may evict the client if it is unresposive to its data GET/PUT
3482 request. This is incorrect if the network is overloaded (takes
3483 too long to transfer the RPC data) or dropped the OST GET/PUT
3484 request. There is no need to evict the client at all, since
3485 the pinger and/or lock callbacks will handle this, and the
3486 client can restart the bulk request.
3489 Frequency : Always when mmapping file with no objects
3491 Description: client crashes when mmapping file with no objects
3492 Details : Check that we actually have objects in a file before doing any
3493 operations on objects in ll_vm_open, ll_vm_close and
3499 Description: Request leak when working with deleted CWD
3500 Details : Introduce advanced request refcount tracking for requests
3501 referenced from lustre intent.
3503 Severity : Enhancement
3505 Description: Cache open file handles on client.
3506 Details : MDS now will return special lock along with openhandle, if
3507 requested and client is allowed to hold openhandle, even if unused,
3508 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3509 closing files for every read/write openration.
3511 Severity : Enhancement
3513 Description: Cache open negative dentries on client when possible.
3514 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3515 negative dentries on lock revocation.
3520 Description: Remounting a client read-only wasn't possible with a zconf mount
3521 Details : It wasn't possible to remount a client read-only with llmount.
3523 Severity : enhancement
3524 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3525 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3526 MPICH is included to add Lustre-specific ADIO interfaces.
3527 This is based closely on the UFS ADIO layer and only differs
3528 in file creation, in order to allow the OST striping to be set.
3529 This is user-contributed code and not supported by CFS.
3534 Description: extended inode attributes (immutable, append-only) work improperly
3535 when 2.4 and 2.6 kernels are used on client/server or vice versa
3536 Details : Introduce kernel-independent values for these flags.
3538 Severity : enhancement
3541 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3542 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3543 granularity. Now, Fractional MB granularity is be supported,
3544 this is very useful for low memory system.
3546 Severity : enhancement
3548 Description: Getattr by fid
3549 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3550 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3551 have locks and avoids getattr rpc requests for it on every stat.
3554 Frequency : Always, for filesystems larger than 2TB
3556 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3557 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3558 filesystems larger than 2TB to be mounted with mballoc enabled.
3561 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3563 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3564 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3565 filesystems larger than 2TB is mounted then the kernel will
3566 silently corrupt the start of the filesystem. CONFIG_LBD is
3567 enabled for all CFS-supported kernels, but the possibility of
3568 this happening with a modified kernel config exists.
3570 Severity : enhancement
3572 Description: add client O_DIRECT support for 2.6 kernels
3573 Details : It is now possible to do O_DIRECT reads and writes to files
3574 in the Lustre client mountpoint on 2.6 kernel clients.
3576 Severity : enhancement
3578 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3579 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3580 parallel, not waiting for response from every OST before sending
3581 a rpc to the next OST.
3586 Description: setattr vs write race when updating file timestamps
3587 Details : Client processes that update a file timestamp into the past
3588 right after writing to the file (e.g. tar) it is possible that
3589 the updated file modification time can be reset to the current
3590 time due to a race between processing the setattr and write RPC.
3592 Severity : enhancement
3594 Description: Bring 'lfs find' closer in line with regular Linux find.
3595 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3596 -print0 options and obtains all the needed info through the lustre
3599 Severity : enhancement
3601 Description: support up to 1024 configured devices on one node
3602 Details : change obd_dev array from statically allocated to dynamically
3603 allocated structs as they are first used to reduce memory usage
3608 Description: Flush dirty partially truncated pages during truncate
3609 Details : Immediatelly flush partially truncated pages in filter_setattr,
3610 this way we completely avoid having any pages in page cache on OST
3611 and can retire ugly workarounds during writes to flush such pages.
3616 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3617 Details : protect inode from truncation within vfs_unlink() context
3618 just take a reference before calling vfs_unlink() and release it
3619 when parent's i_sem is free.
3624 Description: last_id value checked outside lock on OST caused LASSERT failure
3625 Details : If there were multiple MDS->OST object precreate requests in
3626 flight, it was possible that the OST's last object id was checked
3627 outside a lock and incorrectly tripped an assertion. Move checks
3628 inside locks, and discard old precreate requests.
3631 Frequency : always, if extents are used on OSTs
3633 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3634 Details : OSTs using the extents format would not zero the high 16 bits of
3635 the index physical block number. This is not a problem for any
3636 OST filesystems smaller than 16TB, and no kernels support ext3
3637 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3638 new/modified files) and can be fixed for existing filesystems
3639 with e2fsprogs-1.39-cfs1.
3644 Description: import connection selection may be incorrect if timer wraps
3645 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3646 import connections to be ignored if the 32-bit jiffies counter
3647 wraps. Use a 64-bit jiffies counter.
3650 Frequency : very large clusters immediately after boot
3652 Description: LNET request buffers exhausted under heavy short-term load
3653 Details : If a large number of client requests are generated on a service
3654 that has previously never seen so many requests it is possible
3655 that the request buffer growth cannot keep up with the spike in
3656 demand. Instead of dropping incoming requests, they are held in
3657 the LND until the RPC service can accept more requests.
3660 Frequency : Sometimes during replay
3662 Description: Assertion failure in ll_local_open after replay.
3663 Details : If replay happened on an open request reply before we were able
3664 to set replay handler, reply will become not swabbed tripping the
3665 assertion in ll_local_open. Now we set the handler right after
3666 recognising of open request
3669 Frequency : very rare
3671 Description: kernel reports "badness in vsnprintf"
3672 Details : Reading from the "recovery_status" /proc file in small chunks
3673 may cause a negative length in lprocfs_obd_rd_recovery_status()
3674 call to vsnprintf() (which is otherwise harmless). Exit early
3675 if there is no more space in the output buffer.
3677 Severity : enhancement
3679 Description: clear OBD RPC statistics by writing to them
3680 Details : It is now possible to clear the OBD RPC statistics by writing
3681 to the "stats" file.
3686 Description: Client mtime is not the same on different clients after utimes
3687 Details : In some cases, the client was using the utimes() syscall on
3688 a file cached on another node. The clients now validate the
3689 ctime from the MDS + OSTs to determine which one is right.
3694 Description: Inability to activate failout mode
3695 Details : lconf script incorrectly assumed that in pythong string's numeric
3696 value is used in comparisons.
3699 Frequency : always with multiple stripes per file
3701 Description: Inefficient object allocation for mutli-stripe files
3702 Details : When selecting which OSTs to stripe files over, for files with
3703 a stripe count that divides evenly into the number of OSTs,
3704 the MDS is always picking the same starting OST for each file.
3705 Return the OST selection heuristic to the original design.
3710 Description: mount failures may take full timeout to return an error
3711 Details : Under some heavy load conditions it is possible that a
3712 failed mount can wait for the full obd_timeout interval,
3713 possibly several minutes, before reporting an error.
3714 Instead return an error as soon as the status is known.
3716 Frequency : quota enabled and large files being deleted
3718 Description: releasing more than 4GB of quota at once hangs OST
3719 Details : If a user deletes more than 4GB of files on a single OST it
3720 will cause the OST to spin in an infinite loop. Release
3721 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3726 Description: statfs data retrieved from /proc may be stale or zero
3727 Details : When reading per-device statfs data from /proc, in the
3728 {kbytes,files}_{total,free,avail} files, it may appear
3729 as zero or be out of date.
3732 Frequency : systems with MD RAID1 external journal devices
3734 Description: lconf's call to blkid is confused by RAID1 journal devices
3735 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3736 of returning all block devices that match the journal UUID.
3739 Frequency : always, for aggregate stripe size over 4GB
3741 Description: assertion fails when trying to use 4GB stripe size
3742 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3743 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3746 Frequency : always on ppc64
3748 Description: the first write on an ext3 filesystem with mballoc got stuck
3749 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3750 perform endianness conversion.
3752 ------------------------------------------------------------------------------
3754 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3756 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3757 INTEROPERATE with older versions automatically. Please read the
3758 user documentation before upgrading any part of a live system.
3759 * WARNING: Lustre networking configuration changes are required with
3760 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3763 * Support for kernels:
3764 2.6.9-22.0.2.EL (RHEL 4)
3765 2.6.5-7.244 (SLES 9)
3766 2.6.12.6 vanilla (kernel.org)
3769 Severity : enhancement
3770 Bugzilla : 7981/8208
3771 Description: Introduced Lustre Networking (LNET)
3772 Details : LNET is new networking infrastructure for Lustre, it includes
3773 a reorganized network configuration mode (see the user
3774 documentation for full details) as well as support for routing
3775 between different network fabrics. Lustre Networking Devices
3776 (LNDs) for the supported network fabrics have also been
3777 created for this new infrastructure.
3779 Severity : enhancement
3780 Description: Introduced Access control lists
3781 Details : clients can set ACLs on files and directories in order to have
3782 more fine-grained permissions than the standard Unix UGO+RWX.
3783 The MDS must be started with the "-o acl" mount option.
3785 Severity : enhancement
3786 Description: Introduced filesystem quotas
3787 Details : Administrators may now establish per-user quotas on the
3790 Severity : enhancement
3792 Description: Configuration change for the XT3
3793 The PTLLND is now used to run Lustre over Portals on the XT3
3794 The configure option(s) --with-cray-portals are no longer used.
3795 Rather --with-portals=<path-to-portals-includes> is used to
3796 enable building on the XT3. In addition to enable XT3 specific
3797 features the option --enable-cray-xt3 must be used.
3802 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3803 Details : A race between allocating a new llcd and re-getting the llcd_lock
3804 allowed another thread to grab newly-allocated llcd.
3806 Severity : enhancement
3808 Description: 2.6 OST async journal commit and locking fix to improve performance
3809 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3810 2.6 kernels are now async as they already were in 2.4 kernels so
3811 that they can commit concurrently with the network bulk transfer.
3812 For block-allocated files the filter allocation semaphore is held
3813 to avoid filesystem fragmentation during allocation. BKL lock
3814 removed for 2.6 xattr operations where it is no longer needed.
3819 Description: lconf incorrectly determined whether two IP networks could talk
3820 Details : In some more complicated routing and multiple-network
3821 configurations, lconf will avoid trying to make a network
3822 connection to a disjoint part of the IP space. It was doing the
3823 math incorrectly for one set of cases.
3828 Description: Fix for potential infinite loop processing records in an llog.
3829 Details : If an llog record is corrupted/zeroed, it is possible to loop
3830 forever in llog_process(). Validate the llog record length
3831 and skip the remainder of the block on error.
3834 Frequency : occasional (liblustre only)
3836 Description: liblustre could not open files whose last component is a symlink
3837 Details : sysio_path_walk() would incorrectly pass the open intent to
3838 intermediate path components.
3841 Frequency : rare (liblustre only with non-standard tuning)
3842 Bugzilla : 7201 (7350)
3843 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3844 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3845 after it has been released, only do this for referenced locks
3847 Severity : enhancement
3849 Description: specifying an (invalid) directory default stripe_size of -1
3850 would reset the directory default striping
3851 Details : stripe_size -1 was used internally to signal directory stripe
3852 removal, now use "all default" to signal dir stripe removal
3853 as a directory striping of "all default" is not useful
3856 Frequency : common for large clusters running liblustre clients
3858 Description: doing an ls when liblustre clients are running is slow
3859 Details : sending a glimpse AST to a liblustre client waits for every AST
3860 to time out, as liblustre clients will not respond. Since they
3861 cannot cache data we refresh the OST lock LVB from disk instead.
3863 Severity : enhancement
3865 Description: doing an ls at the same time as file IO can be slow
3866 Details : enqueue and other "small" requests can be blocked behind many
3867 large IO requests. Create a new OST IO portal for non-IO
3868 requests so they can be processed faster.
3871 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3873 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3874 created file with mode 000
3875 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3876 allow an ftruncate() as a non-root user to a file with mode 000.
3877 Lustre now respects this flag to disable mode checks when
3878 truncating a file owned by the user
3881 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3884 Description: Revoking locks from clients that went dead or catatonic might take
3886 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3887 cancellation of such locks instant on servers without waiting for
3888 any reply from clients. Clients drops these locks when cancel
3889 notification from server is received without replying.
3892 Frequency : liblustre-only, when liblustre client dies or becomes busy
3894 Description: Doing ls on Linux clients can take a long time with active
3896 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3897 granting such locks to it in the first place if possible. Locks
3898 are taken by proxy on the OST during the read or write and
3899 dropped immediately afterward. Add connect flags handling, do
3900 not grant locks to liblustre clients for glimpse ASTs.
3902 Severity : enhancement
3904 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3905 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3906 sequential reads and behaves badly with applications doing random
3907 reads. Improve it by reading ahead at least read region, and
3908 avoiding excessive large RPC for small reads.
3910 Severity : enhancement
3912 Description: Creating more than 1000 files for a single job may cause a load
3913 imbalance on the OSTs if there are also a large number of OSTs.
3914 Details : qos_prep_create() uses an OST index reseed value that is an
3915 even multiple of the number of available OSTs so that if the
3916 reseed happens in the middle of the object allocation it will
3917 still utilize the OSTs as uniformly as possible.
3922 Description: OST or MDS may oops in ping_evictor_main()
3923 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3924 but doesn't restart at beginning of obd_exports_timed list
3927 Severity : enhancement
3929 Description: improve by-nid export eviction on the MDS and OST
3930 Details : allow multiple exports with the same NID to be evicted at one
3931 time without re-searching the exports list.
3934 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3936 Description: MDS may oops in groups_free()
3937 Details : in rare race conditions a newly allocated group_info struct is
3938 freed again, and this can be NULL. The 2.4 compatibility code
3939 for groups_free() checked for a NULL pointer, but 2.6 did not.
3942 Frequency : common for liblustre clients doing little filesystem IO
3943 Bugzilla : 9352, 7313
3944 Description: server may evict liblustre clients accessing contended locks
3945 Details : if a client is granted a lock or receives a completion AST
3946 with a blocking AST already set it would not reply to the AST
3947 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3950 Frequency : lfs setstripe, only systems with more than 160 OSTs
3952 Description: unable to set striping with a starting offset beyond OST 160
3953 Details : llapi_create_file() incorrectly limited the starting stripe
3954 index to the maximum single-file stripe count.
3957 Frequency : LDAP users only
3959 Description: lconf did not handle in-kernel recovery with LDAP properly
3960 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3962 Severity : enhancement
3964 Description: bind OST threads to NUMA nodes to improve performance
3965 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3966 node and do their allocations there to localize memory access
3968 Severity : enhancement
3970 Description: llmount can determine client NID directly from Myrinet (GM)
3971 Details : the client NID code from gmnalnid was moved directly into
3972 llmount, removing the need to use this or specifying the
3973 client NID explicitly when mounting GM clients with zeroconf
3976 Frequency : if client is started with down MDS
3978 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3979 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3980 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3981 ptlrpc_import_delayed_req() because we didn't check if the
3982 request was interrupted, and we also didn't break out of the
3983 event loop if there was no timeout
3988 Description: data loss during non-page-aligned writes to a single file from
3989 both multiple nodes and multiple threads on one node at same time
3990 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3991 inconsistency led to false short-reads, that were cached and later
3992 used by ->prepare_write() to fill in partially written page,
3993 leading to data loss.
3996 Frequency : always, if lconf --abort_recovery used
3998 Description: lconf --abort_recovery fails with 'Operation not supported'
3999 Details : lconf was attempting to abort recovery on the MDT device and not
4002 Severity : enhancement
4004 Description: remove cleanup logs
4005 Details : replace lconf-generated cleanup logs with lustre internal
4006 cleanup routines. Eliminates the need for client-cleanup and
4009 Severity : enhancement
4011 Description: add support for EAs (user and system) on lustre filesystems
4012 Details : it is now possible to store extended attributes in the Lustre
4013 client filesystem, and with the user_xattr mount option it
4014 is possible to allow users to store EAs on their files also
4016 Severity : enhancement
4018 Description: Add possibility (config option) to show minimal available OST free
4020 Details : When compiled with --enable-mindf configure option, statfs(2)
4021 (and so, df) will return least minimal free space available from
4022 all OSTs as amount of free space on FS, instead of summary of
4023 free spaces of all OSTs.
4025 Severity : enhancement
4027 Description: do not expand extent locks acquired on OST-side
4028 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4029 by server: they are not cached anyway.
4032 Frequency : when mmap is used/binaries executed from Lustre
4034 Description: Unmmap pages before throwing them away from read cache.
4035 Details : llap_shrink cache now attempts to unmap pages before discarding
4036 them (if unmapping failed - do not discard). SLES9 kernel has
4037 extra checks that trigger if this unmapping is not done first.
4042 Description: lconf didn't resolve symlinks before checking to see whether a
4043 given mountpoint was already in use
4046 Frequency : when migrating failover services
4047 Bugzilla : 6395, 9514
4048 Description: When migrating a subset of services from a node (e.g. failback
4049 from a failover service node) the remaining services would
4050 time out and evict clients.
4051 Details : lconf --force (implied by --failover) sets the global obd_timeout
4052 to 5 seconds in order to quickly disconnect, but this caused
4053 other RPCs to time out too quickly. Do not change the global
4054 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4056 Severity : enhancement
4057 Frequency : if MDS is started with down OST
4058 Bugzilla : 9439,5706
4059 Description: Allow startup/shutdown of an MDS without depending on the
4060 availability of the OSTs.
4061 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4062 Add appropriate locking and lov-osc refcounts for safe
4063 cleaning. Add osc abort_inflight calls in case the
4064 synchronize never started.
4067 Frequency : occasional (Cray XT3 only)
4069 Description: root not authorized to access files in CRAY_PORTALS environment
4070 Details : The client process capabilities were not honoured on the MDS in
4071 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4072 been accessed by an authorized user then root was able to access
4073 the file on the local client also. The root user capabilities
4074 are now allowed on the MDS, as this environment has secure UID.
4077 Frequency : occasional
4079 Description: ldiskfs "too long searching" message happens too often
4080 Details : A debugging message (otherwise harmless) prints too often on
4081 the OST console. This has been reduced to only happen when
4082 there are fragmentation problems on the filesystem.
4087 Description: Division by zero in statfs when all OSCs are inactive
4088 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4089 lov_statfs divides by value returned by lov_get_stripecnt().
4093 Bugzilla : 9489, 3273
4094 Description: First write from each client to each OST was only 4kB in size,
4095 to initialize client writeback cache, which caused sub-optimal
4096 RPCs and poor layout on disk for the first writen file.
4097 Details : Clients now request an initial cache grant at (re)connect time
4098 and so that they can start streaming writes to the cache right
4099 away and always do full-sized RPCs if there is enough data.
4100 If the OST is rebooted the client also re-establishes its grant
4101 so that client cached writes will be honoured under the grant.
4106 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4107 Details : Now I/O RPCs go to different portal number and (presumably) fast
4108 lock requests (and glimses) and other RPCs get their own service
4109 threads pool that should be able to service those RPCs
4112 Severity : enhancement
4114 Description: Ability to exchange lustre version between client and servers and
4115 issue warnings at client side if client is too old. Also for
4116 liblustre clients there is ability to refuse connection of too old
4118 Details : New 'version' field is added to connect data structure that is
4119 filled with version info. That info is later checked by server and
4123 Frequency : rare, liblustre only.
4124 Bugzilla : 9296, 9581
4125 Description: Two simultaneous writes from liblustre at offset within same page
4126 might proceed at the same time overwriting eachother with stale
4128 Details : I/O lock withing llu_file_prwv was released too early, before data
4129 actually was hitting the wire. Extended lock-holding time until
4130 server acknowledges receiving data.
4133 Frequency : extremely rare. Never observed in practice.
4135 Description: avoid generating lustre_handle cookie of 0.
4136 Details : class_handle_hash() generates handle cookies by incrementing
4137 global counter, and can hit 0 occasionaly (this is unlikely, but
4138 not impossible, because initial value of cookie counter is
4139 selected randonly). Value of 0 is used as a sentinel meaning
4140 "unassigned handle" --- avoid it. Also coalesce two critical
4141 sections in this function into one.
4143 Severity : enhancement
4145 Description: allow liblustre clients to delegate truncate locking to OST
4146 Details : To avoid overhead of locking, liblustre client instructs OST to
4147 take extent lock in ost_punch() on client's behalf. New connection
4148 flag is added to handle backward compatibility.
4150 Severity : enhancement
4151 Bugzilla : 4928, 7341, 9758
4152 Description: allow number of OST service threads to be specified
4153 Details : a module parameter allows the number of OST service threads
4154 to be specified via "options ost ost_num_threads={N}" in the
4155 OSS's /etc/modules.conf or /etc/modprobe.conf.
4159 Bugzilla : 6146, 9635, 9895
4160 Description: servers crash with bad pointer in target_handle_connect()
4161 Details : In rare cases when a client is reconnecting it was possible that
4162 the connection request was the last reference for that export.
4163 We would temporarily drop the export reference and get a new
4164 one, but this may have been the last reference and the export
4165 was just destroyed. Get new reference before dropping old one.
4167 Severity : enhancement
4168 Frequency : if client is started with failover MDS
4170 Description: Allow multiple MDS hostnames in the mount command
4171 Details : Try to read the configuration from all specified MDS
4172 hostnames during a client mount in case the "primary"
4175 Severity : enhancement
4177 Description: Stop sending data to evicted clients as soon as possible.
4178 Details : Check if the client we are about to send or are sending data to
4179 was evicted already. (Check is done every second of waiting,
4180 for which l_wait_event interface was extended to allow checking
4181 of exit condition at specified intervals).
4184 Frequency : rare, normally only when NFS exporting is done from client
4186 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4188 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4191 Frequency : always, if srand() is called before liblustre initialization
4193 Description: Liblustre uses system PRNG disturbing its usage by user application
4194 Details : Introduce internal to lustre fast and high-quality PRNG for
4195 lustre usage and make liblustre and some other places in generic
4196 lustre code to use it.
4198 Severity : enhancement
4199 Bugzilla : 9477, 9557, 9870
4200 Description: Verify that the MDS configuration logs are updated when xml is
4201 Details : Check if the .xml configuration logs are newer than the config
4202 logs stored on the MDS and report an error if this is the case.
4203 Request --write-conf, or allow starting with --old_conf.
4205 Severity : enhancement
4207 Description: Handle symlinks in the path when checking if Lustre is mounted.
4208 Details : Resolve intermediate symlinks when checking if a client has
4209 mounted a filesystem to avoid duplicate client mounts.
4214 Description: lconf can hit an error exception but still return success.
4215 Details : The lconf command catches the Command error exception at the top
4216 level script context and will exit with the associated exit
4217 status, but doesn't ensure that this exit status is non-zero.
4222 Description: failure of ptlrpc thread startup can cause oops
4223 Details : Starting a ptlrpc service thread can fail if there are a large
4224 number of threads or the server memory is very fragmented.
4225 Handle this without oopsing.
4228 Frequency : always, only if liblustre and non-default acceptor port was used
4230 Description: liblustre cannot connect to servers with non-default acceptor port
4231 Details : tcpnal_set_default_params() was not called and was therefore
4232 ignoring the environment varaible TCPNAL_PORT, as well as other
4233 TCPNAL_ environment variables
4238 Description: two objects could be created on the same OST for a single file
4239 Details : If an OST is down, in some cases it was possible to create two
4240 objects on a single OST for a single file. No problems other
4241 than potential performance impact and spurious error messages.
4245 Bugzilla : 5681, 9562
4246 Description: Client may oops in ll_unhash_aliases
4247 Details : Client dcache may become inconsistent in race condition.
4248 In some cases "getcwd" can fail if the current directory is
4254 Description: Inode refcounting problems in NFS export code
4255 Details : link_raw functions used to call d_instantiate without obtaining
4256 extra inode reference first.
4260 Bugzilla : 9942, 9903
4261 Description: Referencing freed requests leading to crash, memleaks with NFS.
4262 Details : We used to require that call to ll_revalidate_it was always
4263 followed by ll_lookup_it. Also with revalidate_special() it is
4264 possible to call ll_revalidate_it() twice for the same dentry
4265 even if first occurence returned success. This fix changes semantic
4266 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4267 reference on a request referred from the intent.
4268 ll_intent_release() then releases such a request.
4271 Frequency : rare, normally benchmark loads only
4273 Description: unlinked inodes were kept in memory on the client
4274 Details : If a client is repeatedly creating and unlinking files it
4275 can accumulate a lot of stale inodes in the inode slab cache.
4276 If there is no other client load running this can cause the
4277 client node to run out of memory. Instead flush old inodes
4278 from client cache that have the same inode number as a new inode.
4281 Frequency : SLES9 2.6.5 kernel and long filenames only
4282 Bugzilla : 9969, 10379
4283 Description: utime reports stale NFS file handle
4284 Details : SLES9 uses out-of-dentry names in some cases, which confused
4285 the lustre dentry revalidation. Change it to always use the
4289 Frequency : rare, unless heavy write-truncate concurrency is continuous
4290 Bugzilla : 4180, 6984, 7171, 9963, 9331
4291 Description: OST becomes very slow and/or deadlocked during object unlink
4292 Details : filter_destroy() was holding onto the parent directory lock
4293 while truncating+unlinking objects. For very large objects this
4294 may block other threads for a long time and slow overall OST
4295 responsiveness. It may also be possible to get a lock ordering
4296 deadlock in this case, or run out of journal credits because of
4297 the combined truncate+unlink. Solution is to do object truncate
4298 first in one transaction without parent lock, and then do the
4299 final unlink in a new transaction with the parent lock. This
4300 reduces the lock hold time dramatically.
4303 Frequency : rare, 2.4 kernels only
4305 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4306 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4307 before unmounting their filesystem in order to avoid deadlock.
4308 The kernel_locked() function in 2.4 kernels only checks whether
4309 the kernel lock is held, not whether it is this process that is
4310 holding it as 2.6 kernels do.
4315 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4316 Details : The client ptlrpc code may be trying to reconnect to a down
4317 server before a previous connection attempt has timed out.
4318 Increase the reconnect interval to be longer than the connection
4319 timeout interval to avoid sending duplicate connections to
4323 Frequency : echo_client brw_test command
4325 Description: fix echo_client to work with OST preallocated code
4326 Details : OST preallocation code (5137) didn't take echo_client IO path
4327 into account: echo_client calls filter methods outside of any
4328 OST thread and, hence, there is no per-thread preallocated
4329 pages and buffers to use. Solution: hijack pga pages for IO. As
4330 a byproduct, this avoids unnecessary data copying.
4334 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4335 Description: Client can oops in mdc_commit_close() after open replay
4336 Details : It was possible for the MDS to return an open request with no
4337 transaction number in mds_finish_transno() if the client was
4338 evicted, but without actually returning an error. Clients
4339 would later try to replay that open and may trip an assertion
4340 Simplify the client close codepath, and always return an error
4341 from the MDS in case the open is not successful.
4344 Frequency : rare, 2.6 OSTs only
4346 Description: OST may deadlock under high load on fragmented files
4347 Details : If there was a heavy load and highly-fragmented OST filesystems
4348 it was possible to have all the OST threads deadlock waiting on
4349 allocation of biovecs, because the biovecs were not released
4350 until the entire RPC IO was completed. Instead, release biovecs
4351 as soon as they are complete to ensure forward IO progress.
4353 Severity : enhancement
4355 Description: Support for specifying external journal device at mount
4356 Details : If an OST or MDS device is formatted with an external journal
4357 device, this device major/minor is stored in the ext3 superblock
4358 and may not be valid for failover. Allow detecting and
4359 specifying the external journal at mount time.
4364 Description: Mounting an MDS with pending unlinked files may cause oops
4365 Details : target_finish_recovery() calls mds_postrecov() which returned
4366 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4367 considers this an error and immediately begins cleaning up the
4368 lov, just after starting the mds_lov process
4370 Severity : enhancement
4372 Description: Implement 'lfs df' to report actual free space on per-OST basis
4373 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4374 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4375 usage of objects; '-h' to report in human readable format.
4377 ------------------------------------------------------------------------------
4379 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4386 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4387 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4388 at startup during orphan unlinking. Instead, skip buddy update
4389 before setup, as it will be regenerated after recovery is complete.
4394 Description: If an OST is inactive, its locks might reference stale inodes.
4395 Details : lov_change_cbdata() must iterate over all namespaces, even if
4396 they are inactive to clear inode references from the lock.
4398 Severity : enhancement
4399 Frequency : occasional, if non-standard max_dirty_mb used
4401 Description: Client will block write RPCs if not enough grant
4402 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4403 then the client will block writes while waiting for another RPC
4404 to complete instead of consuming its dirty limit. With change
4405 we get improved performance when max_dirty_mb is small.
4407 Severity : enhancement
4408 Bugzilla : 3389, 6253
4409 Description: Add support for supplementary groups on the MDS.
4410 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4411 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4412 lookups for user supplementary groups into a cache.
4416 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4417 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4418 for liblustre O_EXCL is enforced by MDS.
4422 Description: readdir never returns NULL in liblustre.
4423 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4424 dentry in struct dirent.
4428 Frequency : liblustre only, depends on application IO pattern
4429 Description: liblustre clients evicted if not contacting servers
4430 Details : Don't put liblustre clients into the ping_evictor list, so
4431 they will not be evicted by the pinger ever.
4433 Severity : enhancement
4435 Description: Add ability to evict clients by NID from MDS.
4436 Details : By echoing "nid:$NID" string into
4437 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4438 $NID would be instantly evicted from this MDS and from all active
4439 OSTs connected to it.
4443 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4444 Details : lookup_it_finish() used to query file size from OSTs that was not
4449 Description: service threads change working directory to that of init
4450 Details : Starting lustre service threads may pin the working directory
4451 of the parent thread, making that filesystem busy. Threads
4452 now change to the working directory of init to avoid this.
4456 Frequency : during shutdown only
4457 Description: shutdown with a failed MDS or OST can cause unmount to hang
4458 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4463 Frequency : 2.6 only
4464 Description: chmod/chown may include an extra supplementary group
4465 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4466 supplementary group and if none is specified this is used.
4469 Bugzilla : 5479 (6816)
4471 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4472 Details : If lookup is in progress on a file that is unlinked we might try
4473 to revalidate the inode and fail in revalidate after lookup is
4474 complete and ll_file_open() enqueues the open again but
4475 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4478 Frequency : always, if lconf --abort_recovery used
4480 Description: lconf --abort_recovery fails with 'Operation not supported'
4481 Details : lconf was attempting to abort recovery on the MDT device and not
4484 ------------------------------------------------------------------------------
4486 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4491 Frequency : rare (only unsupported configurations with a node running as an
4493 Bugzilla : 6514, 5137
4494 Description: Mounting a Lustre file system on a node running as an OST could
4496 Details : OSTs now preallocates memory needed to write out data at
4497 startup, instead of when needed, to avoid having to
4498 allocate memory in possibly low memory situations.
4499 Specifically, if the file system is mounted on on OST,
4500 memory pressure could force it to try to write out data,
4501 which it needed to allocate memory to do. Due to the low
4502 memory, it would be unable to do so and the node would
4503 become unresponsive.
4505 Severity : enhancement
4507 Description: Addition of lconf --service command line option
4508 Details : lconf now accepts a '--service <arg>' option, which is
4509 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4511 Severity : enhancement
4513 Description: Failover mode is now the default for OSTs.
4514 Details : By default, OSTs will now run in failover mode. To return to
4515 the old behaviour, add '--failout' to the lmc line for OSTs.
4517 Severity : enhancement
4519 Description: Health checks are now provided for MDS and OSTs
4520 Details : Additional detailed health check information on MSD and OSTs
4521 is now provided through the procfs health_check value.
4524 Frequency : occasional, depends on IO load
4526 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4527 after numerous create/delete cycles
4528 Details : The ext3 inode allocation policy would not allocate new inodes
4529 very well on the OSTs because there are no new directories
4530 being created. Instead we look for groups with free space if
4531 the parent directories are nearly full.
4536 Description: Network or server problems during mount may cause partially
4537 mounted clients instead of returning an error.
4538 Details : The config llog parsing code may overwrite the error return
4539 code during mount error handling, returning success instead
4545 Description: MDS can fail to allocate large reply buffers
4546 Details : After long uptimes the MDS can fail to allocate large reply
4547 buffers (e.g. zconf client mount config records) due to memory
4548 fragmentation or consumption by the buffer cache. Preallocate
4549 some large reply buffers so that these replies can be sent even
4550 under memory pressure.
4554 Frequency : rare (liblustre)
4555 Description: fsx running with liblustre complained that using truncate() to
4556 extend the file doesn't work. This patch corrects that issue.
4557 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4558 ATTR_SIZE and lsm use in llu_setattr_raw.
4562 Frequency : rare, only 2.6 kernels
4563 Description: Unusual file access patterns on the MDS may result in inode
4564 data being lost in very rare circumstances.
4565 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4566 code in the 2.6 kernel caused the inode and/or EA data not to
4567 be read from disk, causing single-file corruption.
4571 Frequency : rare, only 2.6 filesystems using extents
4572 Description: Heavy concurrent write and delete load may cause data corruption.
4573 Details : It was possible under high-load situations to have an extent
4574 metadata block in the block device cache from a just-unlinked
4575 file overwrite a newly-allocated data block. We now unmap any
4576 metadata buffers that alias just-allocated data blocks.
4580 Frequency : filesystems with default stripe_count larger than 77
4581 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4582 Details : lconf specifies an inode size of 4096 bytes when the default
4583 stripe_count is larger than 77. This conflicts with the default
4584 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4585 this case to avoid pinning too much memory for large EAs.
4587 ------------------------------------------------------------------------------
4589 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4594 Frequency : rare (extremely heavy IO load with hundreds of clients)
4596 Description: Client is evicted, gets IO error writing to file
4597 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4598 caused clients to be evicted for AST timeouts. The fixes in
4599 bug 5192 mean we no longer need to have such short AST timeouts
4600 so ldlm_timeout has been increased.
4603 Frequency : occasional during --force or --failover shutdown under load
4604 Bugzilla : 5949, 4834
4605 Description: Server oops/LBUG if stopped with --force or --failover under load
4606 Details : a collection of import/export refcount and cleanup ordering
4607 issues fixed for safer force cleanup
4610 Frequency : only filesystems larger than 120 OSTs
4611 Bugzilla : 5990, 6223
4612 Description: lfs getstripe would oops on a very large filesystem
4613 Details : lov_getconfig used kfree on vmalloc'd memory
4616 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4617 Bugzilla : 6242, 6243
4618 Description: reading from files that had been truncated to a non-zero size
4619 but never opened returned no data
4620 Details : ll_file_read() reads zeros from no-object files to EOF
4625 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4626 Details : There's little harm in aborting MDS/OSS recovery and letting it
4627 try again, so I removed the LASSERT and return an error instead.
4629 Severity : enhancement
4631 Description: New debugging infrastructure for tracking down data corruption
4632 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4633 (b) cover more of the client-side code path, and (c) try to narrow
4634 down where problems occurred
4638 Bugzilla : 3819, 4364, 4397, 6313
4639 Description: Racing close and eviction MDS could cause assertion in mds_close
4640 Details : It was possible to get multiple mfd references during close and
4641 client eviction, leading to one thread referencing a freed mfd.
4643 Severity: : enhancement
4644 Bugzilla : 3262, 6359
4645 Description: Attempts to reconnect to servers are now more aggressive.
4646 Details : This builds on the enhanced upcall-less recovery that was added
4647 in 1.4.2. When trying to reconnect to servers, clients will
4648 now try each server in the failover group every 10 seconds. By
4649 default, clients would previously try one server every 25 seconds.
4654 Description: After recovery, certain operations trigger a failed
4655 assertion on a client.
4656 Details : Failing over an mds, using lconf -d --failover, while a
4657 client was doing a readdir() call would cause the client to
4658 LBUG after recovery completed and the readdir() was resent.
4660 Severity : enhancement
4662 Description: Default groups are now added by lconf
4663 Details : You can now run lconf --group <servicename> without having to
4664 manually add groups with lmc.
4667 Frequency : occasional
4669 Description: Nodes with an elan id of 0 trigger a failed assertion
4672 Frequency : always when accessing e.g. tty/console device nodes
4674 Description: tty and some other devices nodes cannot be used on lustre
4675 Details : file's private_data field is used by device data and lustre
4676 values in there got lost. New field was added to struct file to
4677 store fs-specific private data.
4680 Frequency : when exporting Lustre via NFS
4682 Description: NFSD failed occasionally when looking up a path component
4683 Details : NFSD is looking up ".." which was broken in ext3 directories
4684 that had grown large enough to become hashed.
4687 Frequency : Clusters with multiple interfaces not on the same subnet
4689 Description: Nodes will repeatedly try to reconnect to an interface which it
4690 cannot reach and report an error to the log.
4691 Details : Extra peer list entries will be created by lconf with some peers
4692 unreachable. lconf now validates the peer before adding it.
4695 Frequency : Only if a default stripe is set on the filesystem root.
4697 Description: Setting a default stripe on the filesystem root prevented the
4698 filesystem from being remounted.
4699 Details : The client was sending extra request flags in the root getattr
4700 request and did not allocate a reply buffer for the dir EA.
4703 Frequency : occasional, higher if lots of files are accessed by one client
4704 Bugzilla : 6159, 6097
4705 Description: Client trips assertion regarding lsm mismatch/magic
4706 Details : While revalidating inodes the VFS looks up inodes with ifind()
4707 and in rare cases can find an inode that is being freed.
4708 The ll_test_inode() code will free the lsm during ifind()
4709 when it finds an existing inode and then the VFS later attaches
4710 this free lsm to a new inode.
4714 Bugzilla : 6422, 7030
4715 Description: MDS deadlock between mkdir and client eviction
4716 Details : Creating a new file via mkdir or mknod (starting a transaction
4717 and getting the ns lock) can deadlock with client eviction
4718 (gets ns lock and trying to finish a synchronous transaction).
4721 Frequency : occasional
4722 Description: While starting a server, the fsfilt_ext3 module could not be
4724 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4725 kernels. Previously, lconf would still use the ext3 name
4726 when trying to load modules. Now, it will correctly use
4727 ext3 on 2.4 and ldiskfs on 2.6.
4729 Severity : enhancement
4730 Description: The default stripe count has been changed to 1
4731 Details : The interpretation of the default stripe count (0, to lfs
4732 or lmc) has been changed to mean striping across a single
4733 OST, rather than all available. For general usage we have
4734 found a stripe count of 1 or 2 works best.
4736 Severity : enhancement
4737 Description: Add support for compiling against Cray portals.
4738 Details : Conditional compiling for some areas that are different
4742 Frequency : occasional
4743 Bugzilla : 6409, 6834
4744 Description: Creating files with an explicit stripe count may lead to
4745 a failed assertion on the MDS
4746 Details : If some OSTs are full or unavailable, creating files may
4747 trigger a failed assertion on the MDS. Now, Lustre will
4748 try to use other servers or return an error to the
4752 Frequency : occasional
4754 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4755 caused lock timeout during readahead (since 1.4.2).
4756 Details : Processes doing readahead might match a lock that hasn't been
4757 granted yet if there are overlapping and conflicting lock
4758 requests. The readahead process waits on ungranted lock
4759 (original lock is CBPENDING), while OST waits for that process
4760 to cancel CBPENDING read lock and eventually evicts client.
4762 Severity : enhancement
4764 Description: Initial enabling of flock support for clients
4765 Details : Implements fcntl advisory locking and file status functions.
4766 This feature is provided as an optional mount flag (default
4767 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4768 locking are implemented yet, and those that are are not guaranteed
4769 to be completely correct in production environments.
4770 mount -t lustre -o [flock|noflock] ...
4773 Frequency : occasional
4775 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4776 assertion in the ext3 JBD (journaling) layer.
4777 Details : The b_committed_data struct is protected by the big kernel lock
4778 in 2.4 kernels, serializing journal_commit_transaction() and
4779 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4780 access to this struct. In 2.6 kernels there is finer grained
4781 locking to improve SMP performance of the JBD layer.
4785 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4790 Description: merge and cleanup kernel patches.
4793 -----------------------------------------------------------------------------
4795 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4797 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4798 versions of Lustre. Please update all servers and clients to
4799 version 1.4.2 or later at the same time. You must also run
4800 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4801 to update the configuration logs.
4803 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4804 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4805 - swab llog records of type '0' so we get proper header size/idx (5861)
4806 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4807 - fix rename of one directory over another leaking an inode (5953)
4808 - avoid SetPageDirty on 2.6 (5981)
4809 - don't re-add just-being-destroyed locks to the waiting list (5653)
4810 - when creating new directories, inherit the parent's custom
4811 striping settings if present parent (3048)
4812 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4813 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4814 - don't hold client locks on temporary worklist from l_lru (5666)
4815 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4816 - automatically evict dead clients (5921)
4817 - Update file size properly in create+truncate+fstat case (6196)
4818 - Do not unhash mountpoint dentries, do not allow removal of
4820 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4821 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4822 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4823 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4824 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4825 - Correctly handle reads of files with no objects (6243)
4826 - lctl recover will also mark a device active if deactivate used (5933)
4828 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4829 - allow --write-conf on an MDS with different nettype than client (5619)
4830 - don't write config llogs to MDS for mounts not from that MDS (5617)
4831 - lconf should create multiple TCP connections from a client (5201)
4832 - init scripts are now turned off by default; run chkconfig --on
4833 lustre and chkconfig --on lustrefs to use them
4834 - upcalls are no longer needed for clients to recover to failover
4836 - add --abort-recovery option to lconf to abort recovery on device
4838 - add support for an arbitrary number of OSTs (3026)
4839 - Quota support protocol changes.
4840 - forward compatibility changes to wire structs (6007)
4841 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4842 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4843 - improved "lustre status" script
4844 - initialize blocksize for non-regular files (6062)
4845 - added --disable-server and --disable-client configure options (5782)
4846 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4847 - Vanilla 2.4.29 support
4848 - increase maximum number of obd devices to 520 (6242)
4849 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4850 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4851 SLES 9 2.6.5 kernels
4852 - limit stripes per file to 160 (the maximum EA size) (6093)
4854 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4857 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4858 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4859 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4860 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4861 - small ext3 extents cleanups and fixes (5733)
4862 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4863 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4864 - avoid needless client->OST connect, fix handle mismatch (5317)
4865 - fix DLM error path that led to out-of-sync client, long delays (5779)
4866 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4867 - fix several locking issues related to i_size (5492,5624,5654,5672)
4868 - don't move pending lock onto export if it is already evicted (5683)
4869 - fix kernel oops when creating .foo in unlinked directory (5548)
4870 - fix deadlock in obdfilter statistics vs. object create (5811)
4871 - use time_{before,after} to avoid timer jiffies wrap (5882)
4872 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4873 - Do not leak request if server was not able to process it (5154)
4874 - If mds_open unable to find parent dir, make that negative lookup(5154)
4875 - don't create new directories with extent-mapping (5909, 5936)
4877 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4878 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4879 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4880 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4881 - support for RHEL4 kernel on i686 (5773)
4882 - provide error messages when incompatible logs are encountered (5898)
4884 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4885 * version 1.4.0.10 (1.4.1 release candidate 1)
4887 - don't keep a lock reference when lock is not granted (4238)
4888 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4889 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4890 - drop import inflight refcount on signal_completed_replay error (5255)
4891 - unlock page after async write error during send (3677)
4892 - handle missing objects in filter_preprw_read properly (5265)
4893 - no transno return for symlink open, don't save no-trasno open (3440)
4894 - don't try to complete elan receive that already failed (4012)
4895 - free RPC server reply state on error (5406)
4896 - clean up thread from ptlrpc_start_thread() on error (5160)
4897 - readahead could read extra page into cache that wasn't ejected (5388)
4898 - prevent races in class_attach/setup/cleanup/detach (5260)
4899 - don't dereference de->d_inode after l_dput of de (5458)
4900 - use "int" for stripe value returned from lock_to_stripe (5544)
4901 - mballoc allocation and error-checking fixes in 2.6 (5504)
4902 - block device patches to fix I/O request sizes in 2.6 (5482)
4903 - look up hostnames for IB nals (5602)
4904 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4905 - don't start multiple acceptors for the same port (5277)
4906 - fix incorrect LASSERT in mds_getattr_name (5635)
4907 - export a proc file for general "ping" checking (5628)
4908 - fix "lfs check" to not block when the MDS is down (5628)
4910 - service request history (4965)
4911 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4912 - create an "evict_client" /proc entry on OSTs, like the MDS has
4913 - fix mount usage message, return errors per mount(8) (5168)
4914 - change grep [] to grep "[]" in tests so they work in more UMLs
4915 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4916 - remove ancient LOV_MAGIC_V0 EA support (5047)
4917 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4918 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4919 - allow readahead allocations to fail when low on memory (5383)
4920 - mmap locking landed again, after considerable improvement (2828)
4921 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4923 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4926 - send OST transaction number in read/write reply to free req (4966)
4927 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4928 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4929 - the watchdog thread now runs as interruptible (5246)
4930 - flock/lockf fixes (but it's still disabled, pending 5135)
4931 - don't use EXT3 constants in llite code (5094)
4932 - memory shortage at startup could cause assertion (5176)
4934 - reorganization of lov code
4935 - single portals codebase
4937 - add extents/mballoc support (5025)
4938 - direct I/O reads in the obdfilter (4048)
4939 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4941 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4944 - send OST transaction number in read/write reply to free req (4966)
4945 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4946 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4947 - the watchdog thread now runs as interruptible (5246)
4948 - handle missing objects in filter_preprw_read properly (5265)
4949 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4950 - drop import inflight refcount on signal_completed_replay error (5255)
4951 - unlock page after async write error during send (3677)
4952 - return original error code on reconstructed replies (3761)
4953 - no transno return for symlink open, don't save no-trasno open (3440)
4955 - add pid to ldlm debugging output (4922)
4956 - bump the watchdog timeouts -- we can't handle 30sec yet
4957 - extra debugging for orphan dentry/inode bug (5259)
4959 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4962 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4963 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4964 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4965 - fix NULL dereference in /proc/sys/portals/routes (4827)
4966 - allow failed mdc_close() operations to be interrupted (4561)
4967 - stop precreate on OST before MDS would time out on it (4778)
4968 - don't send partial-page writes before EOF from client (4410)
4969 - discard client grant for sub-page writes on large-page clients (4520)
4970 - don't free dentries not owned by NFS code, check generation (4806)
4971 - fix lsm leak if mds_create_objects() fails (4801)
4972 - limit debug_daemon file size, always print CERROR messages (4789)
4973 - use transno after validating reply (3892)
4974 - process timed out requests if import state changes (3754)
4975 - update mtime on OST during writes, return in glimpse (4829)
4976 - add mkfsoptions to LDAP (4679)
4977 - use ->max_readahead method instead of zapping global ra (5039)
4978 - don't interrupt __l_wait_event() during strace
4980 - add software watchdogs to catch hung threads quickly (4941)
4981 - make lustrefs init script start after nfs is mounted
4982 - fix CWARN/ERROR duplication (4930)
4983 - return async write errors to application if possible (2248)
4984 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4985 - print NAL number in %x format (4645)
4986 - update barely-supported suse-2.4.21-171 series (4842)
4987 - support for sles 9 %post scripts
4988 - support for building 2.6 kernel-source packages
4989 - support for sles km_* packages
4991 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4994 - ignore -ENOENT errors in osc_destroy (3639)
4995 - notify osc create thread that OSC is being cleaned up (4600)
4996 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4997 - reconstruct ost_handle() like mds_handle() (4657)
4998 - create a new thread to do import eviction to avoid deadlock (3969)
4999 - let lconf resolve symlinked-to devices (4629)
5000 - don't unlink "objects" from directory with default EA (4554)
5001 - hold socknal file ref over connect in case target is down (4394)
5002 - allow more than 32000 subdirectories in a single directory (3244)
5003 - fix blocks count for O_DIRECT writes (3751)
5004 - OST returns ENOSPC from object create when no space left (4539)
5005 - don't send truncate RPC if file size isn't changing (4410)
5006 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5007 - bind to privileged port in socknal and tcpnal (3689)
5009 - rate limit CERROR/CWARN console message to avoid overload (4519)
5010 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5011 - basic mmap support (3918)
5012 - kernel patch series update from b1_4 (4711)
5014 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5017 - avoid crash during MDS cleanup with OST shut down (2775)
5018 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5019 - don't use bad inodes on the MDS (3744)
5020 - dynamic object preallocation to improve recovery speed (4236)
5021 - don't hold spinlock over lock dumping or change debug flags (4401)
5022 - don't zero obd_dev when it is force cleaned (3651)
5023 - print grants to console if they go negative (4431)
5024 - "lctl deactivate" will stop automatic recovery attempts (3406)
5025 - look for existing locks in ldlm_handle_enqueue() (3764)
5026 - don't resolve lock handle twice in recovery avoiding race (4401)
5027 - revalidate should check working dir is a directory (4134)
5029 - don't always mark "slow" obdfilter messages as errors (4418)
5031 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5034 - don't close LustreDB during write_conf until it is done (3860)
5035 - fix typo in lconf for_each_profile (3821)
5036 - allow dumping logs from multiple threads at one time (3820)
5037 - don't allow multiple threads in OSC recovery (3812)
5038 - fix debug_size parameters (3864)
5039 - fix mds_postrecov to initialize import for llog ctxt (3121)
5040 - replace config semaphore with spinlock (3306)
5041 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5042 - don't allow enqueue to complete on a destroyed export (3822)
5043 - down write_lock before checking llog header bitmap (3825)
5044 - recover from lock replay timeout (3764)
5045 - up llog sem before sending rpc (3652)
5046 - reduce ns lock hold times when setting kms (3267)
5047 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5048 - fix NULL deref and obd_dev leak on setup error (3312)
5049 - replace some LBUG about llog ops with error handling (3841)
5050 - don't match INVALID dentries from d_lookup and spin (3784)
5051 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5052 - fix invalid assertion in ptlrpc_set_wait (3880)
5054 - add libwrap support for the TCP acceptor (3996)
5055 - add /proc/sys/portals/routes for non-root route listing (3994)
5056 - allow setting MDS UUID in .xml (2580)
5057 - print the stack of a process that LBUGs (4228)
5059 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5062 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5063 - make sure to unset replay flag from failed open requests (3440)
5064 - if default stripe count is 0, use OST count for inode size (3636)
5065 - update parent mtime/ctime on client for create/unlink (2611)
5066 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5067 - free recovery state on server during a forced cleanup (3571)
5068 - unregister_reply for resent reqs (3063)
5069 - loop back devices mounting and status check on 2.6 (3563)
5070 - fix resource-creation race that can provoke i_size == 0 (3513)
5071 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5072 - more debugging for page-accounting assertion (3746)
5073 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5074 - avoid deadlock after precreation failure (3758)
5075 - fix race and lock order deadlock in orphan handling (3450, 3750)
5076 - add validity checks when grabbing inodes from l_ast_data (3599)
5078 - add /proc/.../recovery_status to obdfilter (3428)
5079 - lightweight CDEBUG infrastructure, debug daemon (3668)
5080 - change default OSC RPC parameters to be better on small clusters
5081 - turn off OST read cache for files smaller than 32MB
5082 - install man pages and include them in rpms (3100)
5083 - add new init script for (un)mounting lustre filesystems (2593)
5084 - run chkconfig in %post for init scripts (3701)
5085 - drop scimac NAL (unmaintained)
5087 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5090 - clean kiobufs before and after use (3485)
5091 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5092 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5093 - change init script to fail more clearly if not run as root (1528)
5094 - allow clients to reconnect during replay (1742)
5095 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5096 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5097 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5098 - fix cond_resched (3554)
5099 - only evict unfinished clients after recovery (3515)
5100 - allow bulk resend, prevent data loss (3570)
5101 - dynamic ptlrpc request buffer allocation (2102)
5102 - don't allow unlinking open directory if it isn't empty (2904)
5103 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5104 - remove extraneous obd dereference causing LASSERT failure (3334)
5105 - don't use get_cycles() when creating temp. files on the mds (3156)
5106 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5107 - handle EEXIST for set-stripe, set proper directory name (3336)
5109 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5110 - fix ksocknal_fmb_callback() error messages (2918)
5112 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5115 - don't copy lvb into (possibly NULL) reply on error (2983)
5116 - don't deref dentry after dput, don't free lvb on error (2922)
5117 - use the kms to determine writeback rpc length (2947)
5118 - increment oti_logcookies when osc is inactive (2948)
5119 - update client's i_blocks count via lvb messages (2543)
5120 - handle intent open/close of special files properly (1557)
5121 - mount MDS with errors=remount-ro, like obdfilter (2009)
5122 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5123 - don't use cancelling-locks' kms values (2947)
5124 - use highest lock extent for kms, not last one (2925)
5125 - don't dereference ERR_PTR() dentry in error handling path (3107)
5126 - fix thread race in portals_debug_dumplog() (3122)
5127 - create lprocfs device entries at setup instead of at attach (1519)
5128 - common AST error handler, don't evict client on completion race (3145)
5129 - zero nameidata in detach_mnt in 2.6 (3118)
5130 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5131 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5132 - initialize RPC timeout timer earlier for 2.6 (3219)
5133 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5134 - print nal/nid for unknown nid (3258)
5135 - additional checks for oscc recovery before doing precreate (3284)
5136 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5137 - don't crash in mdc_close for bad permissions on open (3285)
5138 - zero i_rdev for non-device files (3147)
5139 - clear page->private before handing to FS, better assertion (3119)
5140 - tune the read pipeline (3236)
5141 - fix incorrect decref of invalidated dentry (2350)
5142 - provide read-ahead stats and refine rpc in flight stats (3328)
5143 - don't hold journal transaction open across create RPC (3313)
5144 - update atime on MDS at close time (3265)
5145 - close LDAP connection when recovering to avoid server load (3315)
5146 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5147 - don't leak open file on MDS after open resend (3325)
5148 - serialize filter_precreate and filter_destroy_precreated (3329)
5149 - loop device shouldn't call sync_dev() for nul device (3092)
5150 - clear page cache after eviction (2766)
5151 - resynchronize MDS->OST in background (2824)
5152 - refuse to mount the same filesystem twice on same mountpoint (3394)
5153 - allow llmount to create routes for mounting behind routers (3320)
5154 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5155 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5156 - fix rare problem with rename on htree directories (3417)
5158 - allow default OST striping configuration per directory (1414)
5159 - fix compilation for qswnal for 2.6 kernels (3125)
5160 - increase maximum number of MDS request buffers for large systems
5161 - change liblustreapi to be useful for external progs like lfsck (3098)
5162 - increase local configuration timeout for slow disks (3353)
5163 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5165 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5168 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5169 - don't overwrite extent policy data in reply if lock was blocked (2901)
5170 - drop filter export grants atomically with removal from device (2663)
5171 - del obd_self_export from work_list in class_disconnect_exports (2908)
5172 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5173 - swab reply message in mdc_close, other PPC fixes (2464)
5174 - fix destroying of named logs (2325)
5175 - overwrite old logs when running lconf --write_conf (2264)
5176 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5177 - fix race in target_handle_connect (2898)
5178 - mds_reint_create() should take same inode create lock (2926)
5179 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5180 - don't close files for self_export to avoid uninitialized obd (2936)
5181 - allow MDS with the same name as client node (2939)
5182 - hold dentry reference for closed log files for unlink (2325)
5183 - reserve space for all logs during transactions (2059)
5184 - don't evict page beyond end of stripe extent (2925)
5185 - don't oops on a deleted current working directory (2399)
5186 - handle hard links to targets without a parent properly (2517)
5187 - don't dereference NULL lock when racing during eviction (2867)
5188 - don't grow lock extents when lots of conflicting locks (2919)
5190 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5193 - account for cache space usage on clients to avoid data loss (974)
5194 - lfsck support in lustre kernel code (2349)
5195 - reduce journal credits needed for BRW writes (2370)
5196 - orphan handling to avoid losing space on client/server crashes
5197 - ptlrpcd can be blocked, stopping ALL progress (2477)
5198 - use lock value blocks to assist in proper KMS, faster stat (1021)
5199 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5200 - recovery for initial connections (2355)
5201 - fixes for mds_cleanup_orphans (1934)
5202 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5203 - block all file creations until orphan recovery completes (1901)
5204 - client remove rq_connection from request struct (2423)
5205 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5206 - recovery timer race (2670)
5207 - mdc_close recovey bug (2532)
5208 - ptlrpc cleanup bug (2710)
5209 - mds timeout on local locks (2588)
5210 - namespace lock held during RPCs (2431)
5211 - handle interrupted sync write properly (2503)
5212 - don't try to handle a message that hasn't been replied to (2699)
5213 - client assert failure during cleanup after abort recovery (2701)
5214 - leak mdc device after failed mount (2712)
5215 - ptlrpc_check_set allows timedout requests to complete (2714)
5216 - wait for inflight reqs when ptlrpcd finishes (2710)
5217 - make sure unregistered services are removed from the srv_list
5218 - reset bulk XID's when resending them (caught by 1138 test)
5219 - unregister_bulk after timeout
5220 - fix lconf error (2694)
5221 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5222 - readahead locks pages, leaves pending causing memory pressure (2673)
5223 - increase OST request buffers to 4096 on large machines (2729)
5224 - fix up permission of existing directories in simple_mkdir (2661)
5225 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5226 - don't assign transno to errored transactions (2742)
5227 - don't delete objects on OST if given a bogus objid from MDS (2751)
5228 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5229 - if rq_no_resend, then timeout request after recovery (2432)
5230 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5231 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5232 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5233 - MDS recovery shouldn't delete objects that it has given out (2730)
5234 - if enqueue arrives after completion, don't clobber LVB (2819)
5235 - don't unlock pages twice when trigger_group_io returns error (2814)
5236 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5237 - don't write pages to disk if there was an error (1450)
5238 - don't ping imports that have recovery disabled (2676)
5239 - take buffered bytes into account when balancing socknal conn (2817)
5240 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5241 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5242 - remove little-endian swabbing of llog records (1987)
5243 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5244 - retry reposting request buffers if they fail (1191)
5245 - grow extent at grant time to avoid granting a revoked lock (2809)
5246 - lock revoke doesn't evict page if covered by a second lock (2765)
5247 - disable VM readahead to avoid reading outside lock extents (2805)
5249 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5250 - updated kernel patches for hp-2.4.20 kernel (2681)
5252 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5255 - fix truncated write corruption (2366)
5256 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5258 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5259 - ensure len(uuid) < 37 in lmc (1171)
5260 - fix ia64 OOPS in llog_test (2255)
5261 - zero end of page at obdfilter for partial page writes (2648)
5262 - don't leave stale dentries around after renames (bug 2428)
5263 - fix timeouts when evicting a client with a single lock held (2642)
5264 - set deadline for the initial HELLO message to drain (2634)
5265 - print out dotted-quad IP addresses in the socknal (2302)
5267 - additional debugging for MDS client eviction problem (2443)
5268 - fix mkfsoptions support for osts (2603, 2604)
5270 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5273 - add series for the vanilla 2.6.0 kernel
5274 - add series for the vanilla 2.4.24 kernel
5275 - add series for a cray x86/64 UL kernel drop
5276 - fix xattr patches for the vanilla 2.4.19 series
5278 - generate true UUIDs in lmc (1171)
5279 - have portals stack dumping break in UML (2466)
5280 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5281 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5282 - wake callback waiting threads on client eviction (2460)
5283 - Add --ptldebug and --subsystem to lmc (1719)
5284 - update assertion to allow safe interrupt allocation
5285 - set rq_no_resend for cancel requests (2432)
5286 - recalculate ptlrpcd timeout after resend (2494)
5287 - call vfs_rmdir when removing pending directories (2368)
5288 - fix renaming a file to itself (2429)
5289 - lmc creates a default one-stripe lov (2454)
5290 - expand procfs space to handle large clusters (2326)
5291 - increase UML stack to avoid overflow
5292 - update lconf's list of debug and subsystem masks
5293 - fix lfs find --obd (2510)
5294 - /proc tunable for disabling filter read caching (2591)
5295 - stop rpm packages from altering slapd.conf (2301)
5296 - disable nagle in the socknal under 0conf (2578)
5297 - choose mds inode size based on stripe count (2572)
5298 - fix kernel-source rpm problems (2516)
5300 - add --disable-doc to avoid pdf generation (2421)
5301 - update documentation, tests, type-os, comments
5302 - avoid format warnings on ia64
5303 - remove the TOE NAL
5304 - tiny code cleanups by removing unused fields
5306 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5309 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5310 - remove the most blatant lies from BUILDING file (2371)
5311 - change default debug level to reasonable production setting
5312 - reduce client side cache size to reduce cache flush time
5313 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5314 - make TCP zerocopy and pinger support enabled by default (2476)
5315 - sync writes completed after process exits caused crashes (2319)
5316 - maintain correct mount count on the MDS (2356)
5317 - backout 1557, because 2316 wasn't really fixed
5318 - better file I/O statistics gathering in /proc
5319 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5320 - another kernel patch to fix zero-copy TCP function export
5321 - don't take duplicate lock when processing re-sent getattr (2420)
5322 - lctl uses obd_self_export instead of creating new conn (2353)
5323 - MDS/OST recovery case which requires object creation asserted (2425)
5324 - move lfs from /usr/sbin to /usr/bin in packages
5325 - fix race between mds_client_add and mds_client_free (2417)
5326 - use kmalloc instead of slabs in portals (2430)
5327 - don't create duplicate records when a failover MDS is present (2442)
5328 - remove unnecessary mount age check (2332)
5329 - don't remove directory inodes from locks prematurely (2451)
5330 - don't break if MDS service name is the same as hostname (2103)
5331 - fix races in client write RPC generation when cache full (2482)
5333 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5336 - remove now-unused request->rq_obd (278)
5337 - if an allocation fails, print out how much memory we've used (1933)
5338 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5339 - add the "configurable stack size" patch to most series files (1256)
5340 - ability to write large log records, for 100+ OST configs (2306)
5341 - fix NULL deref when filter_prep fails (2314)
5342 - fix operator precedence error in filter_sync
5343 - dynamic allocation of socknal TX descriptors (2315)
5344 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5345 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5346 - ensure that CWARN messages in Portals always get to the syslog
5347 - __init/__exit are not for prototype decls (ldlm_init/exit)
5348 - x86-64 compile warning fixes
5349 - fix gateway LMC keyword conflict (2318)
5350 - fix MDS lock inversions in getattr/reint paths (1844)
5351 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5352 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5353 - DLM race condition prevented some lock evictions (2328)
5354 - ENOMEM detection and retry on socknal sends (2230)
5355 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5356 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5357 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5358 - better error messages when a client is rejected during recovery (1505)
5359 - avoid cancelling locks which were never granted, after failure (2330)
5360 - fix i_sem/journal inversion in mds_client_add (2333)
5361 - fix truncate/getattr lock cycle deadlock (2334)
5362 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5363 - fix two rare exit paths which could leak an l_lock() ref (2321)
5364 - fencepost error in MDS/OST orphan recovery (2226)
5365 - make log record alignment 8 bytes (1988)
5366 - lstripe now fails when requested offset > ost_count (2237)
5367 - ensure that all kernel series have a complete list.h (1607)
5368 - fix crashes in special-file operations (2316)
5369 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5371 - allow configurable automake binary, for testing new versions
5372 - small update to the lfs documentation
5374 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5376 * fix negative export reference count in fsfilt_sync (2312)
5378 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5379 * release candidate 0.9.1
5381 - orphans are moved into the PENDING directory for possible recovery
5382 - replayed opens now open by fid for orphan/rename safety (1042)
5383 - last close of an orphan inode generates a transno (683)
5384 - chdir() and mount() now pin the directory entry (1020)
5385 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5386 - discard very old requests without processing them (1502)
5387 - remove some common, well-understood CERRORs (1505)
5388 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5389 - clear "grant" flags in OST replies until OST grant code lands (1644)
5390 - fix read performance by not clobbering i_blksize on client (1598)
5391 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5392 - make LRU size a /proc tunable, clears locks when reduced (707)
5393 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5394 - server locks take references on exports now (1558)
5395 - build fixes for 2.4.20-rh trees (1663)
5396 - return an error from lov_create if all OSCs are inactive (1751)
5397 - fix import levels when a reconnect happens without a timeout (1597)
5398 - exit early from mds_open if we get a lookup error (1749)
5399 - partial page read at EOF wouldn't wait for disk before sending (1642)
5400 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5401 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5402 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5403 - write/truncate lock inversion (1639)
5404 - Don't auto-load obdclass, portals modules during cleanup (1495)
5405 - fix timestamps from jumping to "now" (1763)
5406 - extra journal assertions (1648)
5407 - add an extra multiunlink test (1771)
5408 - fix read_record/write_record API (1776)
5409 - fix leak of offset_extent, possible incorrect i_size later (1772)
5410 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5411 - Debugging for the kqswnal_get_idle_tx problems (1820)
5412 - Allow recovery to be attempted multiple times (1536)
5413 - Write out MDS last_rcvd file after it is first created (1600)
5414 - Fix tx_descriptor leak in failed transmit situations (1827)
5415 - ext3 journaling fixes for assertion failure after IO error (1871)
5416 - class_export_put() on freed export after completion AST error (1896)
5417 - Fix revalidate looping in VFS (1322)
5418 - Don't access a freed export during MDS_REINT timeout (1521)
5419 - Add open-unlink recovery support on the MDS (1673,1764)
5420 - Return an error if no MDS data was read from last_rcvd (1946)
5421 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5422 - Don't setup a disk device that doesn't match exported UUID (317)
5423 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5424 - avoid committing NULL handle in force close
5425 - local.sh is now a one-stripe LOV configuration
5426 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5427 - fix ext3 htree duplicate directory entry corruption (1516)
5428 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5429 - update to new LOV EA format (2097)
5430 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5432 2003-06-15 Phil Schwan <phil@clusterfs.com>
5435 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5436 - per-import/export recovery handling (958, 931, 959)
5437 - multiple last-rcvd slots, for serving multiple FSes (949)
5438 - connections are again shared between multiple imp/exports (963, 964)
5439 - "umount -f" would hang if any requests needed to be sent (393, 978)
5440 - avoid pinning large req buffer by copying for queued messages (989)
5441 - add "uuid" to "lctl device" command to help upcalls (991)
5442 - "open" RPCs with transnos would confuse recovery counters (1037)
5443 - do proper endian conversion of all wire messages (288, 340, 891)
5444 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5445 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5446 - fix ptlrpc_connection leak in target_handle_connect (1174)
5447 - fix import refcounting bug in OST and MDS cleanup (1134)
5448 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5449 - fix very unlikely obd_types race condition (501)
5450 - remove osc_open hack for echo_client (1187)
5451 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5452 - a failure in read_inode2 leads to deadlock (1139)
5453 - cancel ack-locks as soon as transaction is committed (1072)
5454 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5455 - make sure to commitrw after any preprw to avoid deadlock (1162)
5456 - failing to execute a file in a lustre FS would lock inode (1203)
5457 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5458 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5459 - obd_uuid2tgt would walk off the end of the list (1255)
5460 - on IA64 the osc would give portals incorrect bulk size (1258)
5461 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5462 - fix lock inversion caused by new llite matching code (1282)
5463 - limit the number of dirty pages on a client to 10MB (1286)
5464 - timed out locks were not being corrected cancelled (1289)
5465 - fix O_DIRECT above 4GB on IA-32 (1292)
5466 * major user-visible changes
5467 - fail out/fail over policy now controlled by the upcall (993)
5469 - add OBD_PING to check server availability and failure (954)
5470 - lustre messages are now sent in sending host order (288, 340, 891)
5471 - add eadatalen to MDS getattr reply (340)
5472 - OST read replies may contain second buffer, with per-page status (593)
5474 2003-03-11 Phil Schwan <phil@clusterfs.com>
5477 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5478 - failed open()s could cause deadlock; fixed (867, 869)
5479 - stop cancelling OST locks when files are closed (481)
5480 - overlapping XID spaces caused network corruption (851, 853)
5481 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5482 - setattr_raw added, to do single-RPC, server-side setattrs
5483 - lmc/lconf syntax change for OST UUIDs
5484 - fix crashy race condition between ptlrpc_free_req and osc_close
5485 - don't use request in mdc_enqueue if we hit a timeout (889)
5486 - don't set the inode i_size for regular files from the MDS (896)
5487 - handle out of order completion AST (842)
5488 - don't LBUG if a lock request times out after receiving AST (913)
5489 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5490 - if a bad lock AST arrives, send an error instead of dropping entirely
5491 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5492 - fix leak in bulk IO when only partially completed (899, 900, 926)
5493 - fix O_DIRECT for ia64 (55)
5494 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5495 - C-z after timeout could hang a process forever; fixed (977)
5497 - client-side I/O cache (678, 924, 929, 941, 970)
5499 - READPAGE and SETATTRs which don't take server-side locks get
5502 2003-02-11 Phil Schwan <phil@clusterfs.com>
5505 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5506 - Fix fsfilt_extN_readpage() to read a full page of directory
5507 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5508 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5509 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5510 - Most intent operations were removed, in favour of a new RPC mode
5511 that does a single RPC to the server and bypasses most of the VFS
5512 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5513 - Aggressively cancel local locks on DLM servers
5514 - mds_reint_unlink sends EA to the client if it's the last nlink.
5515 client uses that EA to unlink OST objects.
5516 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5517 - recursive symlinks were fixed (439)
5518 - fixed NULL deref in DEBUG_REQ
5519 - filter_update_lastobjid no longer calls sync, which annoyed extN
5520 - fixed multi-client small-writes to a single file problem (445)
5521 - fixed mtime updates during file writes (607)
5522 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5523 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5524 - fixed Linux OST size reporting problem (444, 656)
5525 - OST now updates object mtime with writes or setattr (607, 619)
5526 - client verifies file size before zeroing page past EOF (445)
5527 - OST now writes last allocated objid to disk with allocation (108)
5528 - LOV on echo now works (409)
5530 - mds_reint_unlink sends a new buffer, with the EA included. this
5531 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5532 set if a regular file was being unlinked, and it was the last link
5533 - use PtlGet from the target for bulk writes (315)
5534 - OST now updates object mtime with writes or setattr (607, 619)
5535 - LDLM now has a grant-time callback to revalidate locked items, if
5537 - Many MDS operations were reorganized to combat race conditions
5539 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5540 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5542 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5545 - Fully reactivate OST imports after reconnection (512, others)
5546 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5547 - More graceful error handling for truncating on dead OST (515)
5548 - Don't error out unless we're actually accessing dead stripes (474)
5549 - Fix garbage sizes when stripes are missing (410)
5550 - LRU counters were broken, causing constant lock purge (433, 432)
5551 - garbage on read from stripes with failed OSTs (441)
5552 - mark OSCs as active before reconnecting during recovery (438)
5553 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5554 - lfind did not preserve OST order in output (443)
5555 - symlinks cause hung clients, incorrect data (439)
5556 - stop dereferencing request after dropping refcount (457)
5557 - don't LASSERT(spin_is_locked) on non-SMP (455)
5558 - fixes for many rename() bugs
5559 - fstat didn't correctly synchronize attributes (399)
5560 - server must handle lock cancellation during blocking AST prep (487)
5561 - bulk descriptors were free()d too soon (511)
5562 - fix paths in lconf, which would load incorrect modules (451, 507)
5563 - fix confusing lconf 'host not found' error message (386)
5564 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5565 - fix race condition in mdc_blocking_ast() for inode access (526)
5566 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5567 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5568 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5569 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5570 - fix lstripe in several ways (kernel side) (527)
5571 - fix request leak in ldlm_cli_enqueue (262)
5572 - incorrect OSC was marked inactive after OST failure
5573 - call mds_fs_cleanup before unmounting filesystem (524)
5574 - fix races between taking ns_lock and ldlm_lock_change_resource
5575 - fix races updating LOV export open file list
5576 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5577 - fix recovery NULL deref in ldlm_cli_cancel_unused
5578 - fix some DLM races by using new hash table for lock handles (419)
5579 - permit the client to specify desired inodes, at replay
5580 - duplicate requests when we queue them for replay reintegration
5581 - fix last_rcvd offset calculation
5582 - sync after each recovered transaction, so we always make progress
5583 - never, not always, ERESTART requests without transnos
5584 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5586 - skip replay if the MDS says that the client is already connected
5587 - don't check for a recovery-enabled export to match lctl's UUID
5588 - don't INC_USE_COUNT for phantom exports
5589 - don't crash when cleaning up phantom exports (567)
5590 - don't double-finish or set replay data for errored mdc_open requests
5591 - abort requests when they time out, so we don't get old replies
5592 - send/receive replies for AST messages again
5593 - if the client says that it doesn't have the lock, cancel it on the
5595 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5596 mark it as destroyed, it will all work out when decref is called
5597 - fix module use counts (22, 581)
5599 - ASTs now expect a reply (server cancels lock on error reply)
5601 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5604 - fix many simultaneous client startup (392)
5605 - fix dentry->d_it clobbering
5606 - credentials weren't being shipped for readdir/getattr operations
5607 - remove invalid assertions triggered during some concurrent MD
5609 - proper Lustre versions added (336, 389)
5610 - fix memory leak for create error case (398)
5611 - fix LOV locking bug that would get cli/srv out of sync
5612 - fix echo client over LOV (409)
5613 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5614 - fix double-O_EXCL intent crash (424)
5615 - avoid sending multiple lock CANCELs (352)
5617 - MDS can do multi-client recovery (modulo bugs in new code)
5619 - many updates, edits, cleanups
5621 2002-11-18 Phil Schwan <phil@clusterfs.com>
5624 - fix null d_it dereference (346)
5625 - fix full OST/dbench hang (333)
5626 - fix permission problem with file removal (286)
5627 - fix removal of OSCs from LOV when they fail
5628 - fix NULL deref during bulk timeout (214)
5629 - fix problems related to multiple filesystems on one MDS (241)
5630 - fixed serious subtle metadata locking bugs
5631 - free locks on clients when inodes are removed due to memory
5633 - fix inode pointer in lock data (285)
5634 - partial support for multiple MDS on a single host (241)
5635 - data locks weren't cancelled at clear_inode time (290, 311)
5636 - intent locks could lead to unbounded lock growth (205)
5637 - added a maximum lock count, an LRU list, and a flusher
5638 - fix multiple rename (365)
5639 - properly abstracted the echo client
5640 - OSC locked 1 byte too many; fixed
5641 - rewrote brw callback code:
5642 - fixed recovery bugs related to LOVs (306)
5643 - fixed too-many-pages-in-one-write crash (191)
5644 - fixed (again) crash in sync_io_timeout (214)
5645 - probably fixed callback-related race (385)
5647 - Add capability to MDS protocol
5648 - LDLM cancellations and callbacks on different portals
5650 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5653 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5654 - defer open object destruction to close time (601981, 138)
5655 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5656 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5657 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5658 - don't call obd_setattr in ll_file_release for destroyed objects
5659 * protocol change to lustre_msg: move |version| and add |flags|
5660 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5661 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5662 * added replay of create, unlink, link and rename operations during
5663 MDS failover; recovery should be much more robust now
5664 * remove failed OSCs from LOVs (only lov_create uses this so far)
5665 * the lustre-HOWTO was brought (more) up to date (582544)
5667 2002-10-23 Phil Schwan <phil@clusterfs.com>
5670 - in-use dentries weren't being reused properly (617851)
5671 - prevent multiple LDLM setup (599178)
5672 - fix LOV size calculations for truncate (617853)
5673 - fix client handling of MDS intent errors (POSIX)
5674 - fix permission bug in lovstripe.c test (624321)
5675 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5676 - truncate past end of file could corrupt data
5677 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5678 - a race in recovery could return ETIMEDOUT to apps (623947)
5679 - building outside the source directory was fixed
5680 * the lustre-HOWTO was brought (more) up to date (582544)
5681 * major progress was made on recovery functionality
5683 2002-10-10 Phil Schwan <phil@clusterfs.com>
5686 - recovery deadlock fix
5687 - rm -rf causes LBUG fix (617817)
5688 - file open by multiple tasks fix (618962)
5689 - directory permissions bugs (602707 and 620007)
5690 - journal_stop fixed with locking (611313)
5691 - O_APPEND failures resolved (618273, perhaps 614459)
5692 - lconf PATH fix (619770)
5693 - IA64 build fix (621450)
5694 - RPC buffer sizes scale with amount of memory
5696 2002-10-01 Phil Schwan <phil@clusterfs.com>
5699 - locks would be cancelled without throwing away data pages,
5700 resulting in inconsistent data (605627)
5701 - inode attributes were not always being refreshed (605627, 612449)
5702 - lconf now continues to cleanup after lctl reports an error
5703 - MDS now enforces user permissions (602707)
5704 - lprocfs cleanup fixed, but not yet enabled (614157)
5705 - fixed infinite server hang, should a client not respond to an AST
5706 - avoid going into recovery if user calls readlink() with a buffer
5707 that's too small (613941)
5708 - AST RPCs no longer require replies (614867) -- this may be changed
5709 - don't crash server if client sends an IOV that's too big (611336)
5710 - fixed lock conversion deadlock (611892)
5711 - fixed the following of symlinks (614622)
5712 * recovery: the server can remove locks from a client that dies, other
5713 clients can make progress
5714 * more extN patch fixes
5715 * compile-time configurable ptlrpc buffer allocations
5717 - collaborative read cache document
5718 - Lustre Lite Performance CDR document-in-progress
5720 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5723 - fix typo in patch-2.4.18
5725 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5728 - clear ptlrpc request each time in handle_incoming_request()
5729 - unlink of files now destroys the object on the OST
5731 2002-09-19 Peter Braam <braam@clusterfs.com>
5733 * add hard link support
5734 * change obdfile creation method
5735 * kernel patch changed
5737 2002-09-19 Peter Braam <braam@clusterfs.com>
5740 - stack overflow bug in extN fixed
5742 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5744 * documentation updates
5745 - add man pages for config tools
5746 - update tests/README to describe testing with new config tools
5747 - finish metadata API descriptions
5748 * bug fixes and cleanups
5749 - statfs workaround for 16TB limit
5750 - LOV stripe allocation improved, can stripe on subset of OSTs
5751 - LOV file size/IO offset was wrong for files > 4GB in size
5752 - object EA data was being dropped, caused files to be unreadable
5753 - memory overflow with non-LOV OST caused memory corruption
5754 - fixed regression tests to work with new config tools, obdfilter
5755 - fixed bug when directory size became larger than 1 block
5756 - fixed bug (for single client case) when PWD was deleted
5757 - invalidate local directory pages when doing intent-based ops
5758 - avoid LDLM oops when lock callback contained bad data
5760 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5762 * documentation updates
5763 * bug fixes and cleanups
5764 - configuration tools
5767 - 64-bit compile warnings
5768 - 64-bit internal statfs data
5770 * test_brw on persistent OST devices
5772 * lprocfs (disabled)
5774 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5776 * documentation updates
5777 * bug fixes and cleanups
5778 * configuration tools
5780 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5783 * many small fixes to 0_5_4
5784 * io/network handling
5785 * thinkos in MDS operations
5787 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5790 * crucial basic fixes to 0.5.3
5791 * IOR, Iozone work over Elan
5794 2002-08-07 Phil Schwan <phil@clusterfs.com>
5795 * version 0_5_3, our first alpha
5796 * we use the new Portals iovs
5797 * documentation updates
5798 * bug fixes and cleanups
5799 * small changes in the DLM wire protocol
5801 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5802 * version 0_5_1 with some initial stability,
5803 * locking on MD and file I/O.
5804 * documentation updates
5805 * several bug fixes since 0.5.0
5806 * small changes in wire protocol
5808 2002-07-18 Phil Schwan <phil@clusterfs.com>
5810 * delivered as Lustre Light Alpha
5811 * fixed a crash after handling invalid MDS requests
5812 * fixed directory pages for architectures with non-4k pages sizes
5814 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5815 * release version v0_4_4
5816 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5818 * Fixes a number of interruption problems with OST operations.
5819 * Update documentation for portals header changes
5820 * Move all wire protocol structs/defines to lustre_idl.h
5821 * Fixes symlink length bug.
5822 * Add tcpdump to repository.
5824 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5825 * release version v0_4_3
5826 * Fixes statfs for inodes on extN.
5827 * Fixes bug in runtests which would delete /etc/hosts.
5828 * Use 64-bit object IDs wherever possible (not into VFS though)
5829 Remove ost_get_info, which is unused by lustre, and out of date.
5831 2002-07-03 Peter Braam <braam@clusterfs.com>
5832 * release version v0_4_2 Fixes a lookup error (type not passed)
5833 * move forward to head of Portals
5834 * move forward to latest Lustre kernel
5836 2002-06-25 Peter Braam <braam@clusterfs.com>
5837 * release version v0_4_1. Hopefully stable on single node use.