1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (SLES 10),
5 2.6.18-128.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.6.sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
18 Description: Don't allow make backward step on assignin osc next id.
19 Details : race between allocation next id and ll_sync thread can be cause
20 of set wrong osc next id and can be kill valid ost objects.
24 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
28 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
32 Description: Error handling in osc_statfs_interpret() has been improved.
33 Details : Check in osc_statfs_interpret() for EBADR.
37 Description: Do not send statfs() requests to OSTs disabled by administrator.
38 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
40 Severity : enhancement
42 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
44 Severity : enhancement
46 Description: Enhancement for lfs(1) command to use numeric uid/gid.
50 Description: Improved cleanup procedure for conf-sanity test_46a.
54 Description: Do not start bulk transfer, if client tried to reconnect.
55 Details : Do not start bulk transfer, if client tried to reconnect,
56 but failed due to active rpcs. Check for exp_abort_active_req
61 Description: Prevent inconsistences between linux and lustre mount structures.
62 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
63 Make the sleep interruptible.
65 Severity : enhancement
67 Description: provide server to client comms path
68 Details : server to client communications path via new LDLM_SET_INFO rpc
72 Description: truncate starts GFP_FS allocation under transaction causing deadlock
73 Details : ldiskfs_truncate calls grab_cache_page which may start page
74 allocation under an open transaction. This may lead to
75 calling prune_icache with consequent lustre reentrance.
79 Description: send correctly lsm on open replay
80 Details : MDS is trust to LSM size on replay open, but client can set wrong size
85 Description: enable client interface failover
86 Details : When a child reconnects from another NID, properly update export
87 nid hash position and ldlm reverse import.
89 Severity : enhancement
91 Description: Communicate OST degraded/readonly state via statfs to MDS
92 Details : Flags in the statfs returned from OSTs indicate whether the
93 OST is in a degraded RAID state, or if the filesystem has
94 turned read-only after a filesystem error is detected.
99 Description: don't panic if EPROTO was hit when reading symlink
100 Details : correctly handling request reference in error cases.
102 Severity : enhancement
104 Description: Add LustreNetLink, a kernel-userspace communcation path.
106 Severity : enhancement
108 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
113 Description: access to llog context before init.
114 Details : move handling CATALOGS file at osc layer and forbid access to llog
118 Frequency : rare, only when client is writing to OST on local node
120 Description: Avoid deadlock for local client writes
121 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
122 memory freeing context. This allows OST threads to set the
123 PF_MEMALLOC flag on task structures in order to allocate memory
124 from reserved pools and complete IO.
125 Use GFP_HIGHUSER for OST allocations for non-local client writes,
126 so that the OST threads generate memory pressure and allow
127 inactive pages to be reclaimed.
129 Severity : enhancement
131 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
136 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
137 Details : move ldlm namespace creation in setup phase to avoid grab
138 _lprocfs_lock with cli_sem held.
142 Description: Temporarily disable grant shrink.
143 Details : Disable the feature for debugging.
146 Frequency : only when formatting test filesystems
148 Description: Unable to run several mkfs.lustre on loop devices at the same time
149 Details : mkfs.lustre returns error 256 on the concurrent loop devices
150 formatting. The solution is to proper handle the error.
152 Severity : enhancement
154 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
156 Severity : enhancement
158 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
163 Description: resolve race between obd_disconnect and class_disconnect_exports
164 Details : if obd_disconnect will be called to already disconnected export he
165 forget release one reference and osc module can't unloaded.
167 Severity : enhancement
169 Description: Allow tuning service thread via /proc
170 Details : For each service a new
171 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
172 created that can be used to set min/max thread counts, and get the
173 current number of running threads.
178 Descriptoin: don't return error if have partially created objects for file.
179 Details : lov_update_create_set uses set->set_success as index for created
180 objects, so if some requests will be failed, they will have hole
181 at end of array and qos_shrink_lsm for allocate correct lsm.
183 Severity : enhancement
185 Description: Update OFED support to 1.4.1
190 Description: Reduce small size read RPC
191 Details : Set read-ahead limit for every file and only do read-ahead when
192 available read-ahead pages are bigger than 1M to avoid small size
195 Severity : enhancement
197 Description: move AT tunable parameters for more consistent usage
198 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
200 Severity : enhancement
202 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
203 Details : allow skip disconnected ost for send statfs request and hide error
209 Description: fix racy locking of mballoc block bitmaps causing BUG
210 Details : The locking of the mballoc buddy bitmap and the in-memory
211 block bitmap was using two different spin locks in some
212 cases. This made it possible to incorrectly access the
213 mballoc bitmap while another process was modifying it,
214 causing a sanity assertion to fail. While no on-disk corruption
215 was reported, there was some risk of this happening.
218 Frequency : rare, on llog test 6
220 Descriptoin: don't allow connect to already connected import
221 Details : allowing connect to already connected import is hide connecting problem.
226 Descriptoin: allow kill process which wait statahead result
227 Details : for some reasons 'ls' can stick in waiting result from statahead,
228 in this case need way for kill this process.
230 Severity : enhancement
232 Description: Add state history info file, enhance import info file
233 Details : Track import connection state changes in a new osc/mdc proc file;
234 add overview-type data to the osc/mdc import proc file.
236 Severity : enhancement
238 Description: MDS create should not wait for statfs RPC while holding DLM lock.
240 Severity : enhancement
242 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
247 Descriptoin: OST grant too much space to client even there are not enough space.
248 Details : Client will shrink its grant cache to OST if there are no write
249 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
250 this grant cache if there are already not enough avaible space
251 (left_space < total_clients * 32M).
254 Frequency : start MDS on uncleanly shutdowned MDS device
256 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
257 Details : stay in waiting mds<>ost recovery finished produce random bugs
258 due race between two ll_sync thread for one lov target. send
259 ACTIVATE event only if connect realy finished and import have
263 Frequency : rare, connect and disconnect target at same time
265 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
266 Details : don't call obd_disconnect under lov_lock. this long time
267 operation and can block ptlrpcd which answer to connect request.
272 Descriptoin: don't lose wakeup for imp_recovery_waitq
273 Details : recover_import_no_retry or invalidate_import and import_close can
274 both sleep on imp_recovery_waitq, but we was send only one wakeup
278 Frequency : always with long access acl
280 Descriptoin: mds can't pack reply with long acl.
281 Details : mds don't control size of acl but they limited by reint/getattr
284 Severity : enhancement
286 Description: Update to SLES10 kernel-2.6.16.60-0.33.
288 Severity : enhancement
290 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
293 Frequency : start MDS on uncleanly shutdowned MDS device
295 Descriptoin: aborting recovery hang on MDS
296 Details : don't throttle destroy RPCs for the MDT.
299 Frequency : on remount
301 Description: external journal device not working after the remount
302 Details : clear dev_rdonly flag for external journal devices in
308 Description: shutdown vs evict race
309 Details : client_disconnect_export vs connect request race.
310 if client will evicted at this time - we start invalidate
311 thread without referece to import and import can be freed
315 Frequency : rare, need acl's on inode.
317 Description: client can't handle ost additional correctly
318 Details : if ost was added after client connected to mds client can have
319 hit lnet_try_match_md ... to big messages to wide striped files.
320 in this case need teach client to handle config events about add
321 lov target and update client max ea size at that event.
323 Severity : enhancement
325 Description: Changelogs
326 Details : Changelogs are a lightweight mechanism to track filesystem
327 metadata and namespace changes. The changelog is recorded
328 permanently on the MDTs, and is periodically "consumed" / purged
329 when records are no longer needed.
331 Severity : enhancement
333 Description: compact fld format with extents
334 Details : Store range of seq rather than every seq in FLD. Seq
335 controller update FLD rather than clients. In Case of CMD, mdt0
336 has FLD, all other metadata server act as non persistent proxy
337 for FLD queries and cache fld entries in fld cache.
342 Description: don't skip ost target if they assigned to file
343 Details : Drop slow OSCs if we can, but not for requested start idx.
344 This means "if OSC is slow and it is not the requested
345 start OST, then it can be skipped, otherwise skip it only
346 if it is inactive/recovering/out-of-space.
350 Description: more cleanup in mds_lov
351 Details : not send LOV EA under replay, we can't know about they size at this
352 time. Don't allow client connect to mds before any ost connected,
353 for avoid problems with LOV EA size and returning EIO to client.
355 Severity : enhancement
357 Description: Interoperability at server side (Disk interoperability)
359 Severity : enhancement
361 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
363 Severity : enhancement
365 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
367 Severity : enhancement
369 Description: New client IO stack (CLIO).
371 Severity : enhancement
373 Description: Commit on sharing. Eliminate inter-client dependencies between
374 uncommitted transactions by doing transaction commits.
375 Thereby clients may recovery independently.
378 Frequency : Create a symlink file with a very long name
380 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
381 Details : If there is no extra space in the request for early cancels,
382 ldlm_req_handles_avail() returns 0 instead of a negative value.
384 Severity : enhancement
386 Description: Add /proc entry for import status
387 Details : The mdc, osc, and mgc import directories now have
388 an import directory that contains useful import data for debugging
391 Severity : enhancement
393 Description: Re-disable certain /proc logging
394 Details : Enable and disable client's offset_stats, extents_stats and
395 extents_stats_per_process stats logging on the fly.
398 Frequency : Only on FC kernels 2.6.22+
400 Description: oops in statahead
401 Details : Do not drop reference count for the dentry from VFS when lookup,
402 VFS will do that by itself.
404 Severity : enhancement
406 Description: Generic /proc file permissions
407 Details : Set /Proc file permissions in a more generic way to enable non-
408 root users operate on some /proc files.
412 Description: Hitting mdc_commit_close() ASSERTION
413 Details : Properly handle request reference release in
414 ll_release_openhandle().
418 Description: quota recovery deadlock during mds failover
419 Details : This patch includes att18982, att18236, att18237 in bz14840.
421 1. fix osts hang when mds does failover with quotaon
422 2. prevent watchdog storm when osts threads wait for the
427 Frequency : only patchless client
428 Description: add workaround for race between add/remove dentry from hash
430 Severity : enhancement
432 Description: Allow OST glimpses to return PW locks
436 Description: LBUG when llog conf file is full
437 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
441 Description: Prevent import from entering FULL state when server in recovery
445 Description: service mount cannot take device name with ":"
446 Details : Only when device name contains ":/" will mount treat it as
452 Description: replace ptlrpcd with the statahead thread to interpret the async
453 statahead RPC callback
457 Frequency : on recovery
458 Description: I/O failures after umount during fail back
459 Details : if client reconnected to restarted server we need join to recovery
460 instead of find server handler is changed and process self
461 eviction with cancel all locks.
463 Severity : enhancement
465 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
467 Severity : enhancement
469 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
471 Severity : enhancement
473 Description: Upcall on Lustre log has been dumped
474 Details : Allow for a user mode script to be called once a Lustre log has
475 been dumped. It passes the filename of the dumped log to the
476 script, the location of the script can be specified via
477 /proc/sys/lnet/debug_log_upcall.
482 Description: avoid idr_remove called for id which is not allocated.
483 Details : Move assigment s_dev for clustered nfs to end of initialization,
484 to avoid problem with error handling.
489 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
490 Details : When connection is reused this not moved from CONN_UNUSED_HASH
491 into CONN_USED_HASH and this prodice warning when put connection
492 again in unused hash.
495 Severity : enhancement
497 Description: File striping can now be set to use an arbitrary pool of OSTs.
499 Severity : enhancement
501 Description: Export bytes_read/bytes_write count on OSC/OST.
505 Description: Early reply size mismatch, MGC loses connection
506 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
507 the connect flags are properly negotiated.
512 Description: quotas are not honored with O_DIRECT
513 Details : all writes with the flag O_DIRECT will use grants which leads to
514 this problem. Now using OBD_BRW_SYNC to guard this.
518 Description: add quota statistics
519 Details : 1. sort out quota proc entries and proc code.
520 2. add quota statistics
522 Severity : enhancement
524 Description: enable quota support for HEAD.
528 Description: Properly propagate oinfo flags from lov to osc for statfs
529 Details : restore missing copy oi_flags to lov requests.
531 Severity : enhancement
533 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
534 : l_getgroups(8), lst(8), routerstat(8)
535 Details : included man pages for llobdstat(8), llstat(8),
536 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
538 Severity : enhancement
540 Description: configure's --enable-quota should check the
541 : kernel .config for CONFIG_QUOTA
542 Details : configure is terminated if --enable-quota is passed but
543 : no quota support is in kernel
547 Description: Remove portals compatibility
548 Details : Remove portals compatibility, not interoperable with releases
553 Description: Resolve device initialization race
554 Details : Prevent proc handler from accessing devices added to the
555 obd_devs array but yet be intialized.
557 Severity : enhancement
559 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
561 Severity : enhancement
563 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
568 Description: Using wrong pointer in osc_brw_prep_request
569 Details : Access to array[-1] can produce panic if kernel compiled with
570 CONFIG_PAGE_ALLOC enabled
574 Description: Client runs out of low memory
575 Details : Consider only lowmem when counting initial number of llap pages
579 Description: *optional* service tags registration
580 Details : if the "service tags" package is installed on a Lustre node
581 When the filesystem is mounted, a local-node service tag will
582 be created. See http://inventory.sun.com/ for more information
583 about the Service Tags asset management system.
587 Description: Kernel BUG tries to release flock
588 Details : Lustre does not destroy flock lock before last reference goes
589 away. So always drop flock locks when client is evicted and
590 perform unlock regardless of successfulness of speaking to MDS.
594 Description: add refcount protection for osc callbacks, avoid panic on shutdown
598 Description: sanity test 65a fails if stripecount of -1 is set
599 Details : handle -1 striping on filesystem in ll_dirstripe_verify
604 Description: ASSERTION(CheckWriteback(page,cmd)) failed
605 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
608 Severity : enhancement
610 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
614 Description: do not process already freed flock
615 Details : flock can possibly be freed by another thread before it reaches
616 to ldlm_flock_completion_ast.
620 Description: LBUG during stress test
621 Details : Need properly lock accesses the flock deadlock detection list.
625 Description: oops in page fault handler
626 Details : kernel page fault handler can return two special 'pages' in error
627 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
631 Description: timeout with invalidate import.
632 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
633 should be handled by ptlrpcd. This produce long age waiting and
634 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
636 Severity : enhancement
638 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
642 Description: enable MGS and MDT services start separately
643 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
644 a co-located MGS without starting the MGS, which is a complement
645 to 'nosvc' mount option.
649 Frequency : after recovery
650 Description: precreate to many object's after del orphan.
651 Details : del orphan st in oscc last_id == next_id and this triger growing
652 count of precreated objects. Set flag LOW to skip increase count
653 of precreated objects.
657 Frequency : rare, on clear nid stats
658 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
659 Details : when clean nid stats sometimes try destroy live entry,
660 and this produce panic in free.
664 Description: Stack overflow during MDS log replay
665 ease stack pressure by using a thread dealing llog_process.
669 Description: wait until IO finished before start new when do lock cancel.
670 Details : VM protocol want old IO finished before start new, in this case
671 need wait until PG_writeback is cleared until check dirty flag and
672 call writepages in lock cancel callback.
674 Severity : enhancement
676 Description: using special macro for print time and cleanup in includes.
680 Description: mds_mfd_close() ASSERTION(rc == 0)
681 Details : In mds_mfd_close(), we need protect inode's writecount change
682 within its orphan write semaphore to prevent possible races.
686 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
690 Frequency : rare, on shutdown ost
691 Description: don't hit live lock with umount ost.
692 Details : shrink_dcache_parent can be in long loop with destroy dentries,
693 use shrink_dcache_sb instead.
697 Description: don't panic with use echo client
698 Details : echo client pass NULL as client nid pointer and this produce null
703 Description: fix build on ppc32
704 Details : compile code with -m64 flag produce wrong object file for ppc32.
708 Description: add message levels for liblustreapi
712 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
713 Details : Move the 'good_osts' check before the 'total_bavail' check. This
714 will result in an -EAGAIN and in the exit call path we call
715 alloc_rr() which will with increasing aggressiveness attempt to
716 aquire precreated objects on the minimum number of required OSCs.
720 Description: Use old size assignment to avoid deadlock
721 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
722 the scheduling while holding a spinlock. We do not need locking
723 for size in ll_update_inode() because size is only updated from
724 the MDS for directories or files without objects, so there is no
725 other place to do the update, and concurrent access to such inodes
726 are protected by the inode lock.
730 Description: resolve "_IOWR redefined" build error on SLES10
734 Description: dump the memory debugging after all modules are unloaded to
735 suppress false negative in conf_sanity test 39
737 Severity : enhancement
739 Description: build kernel-ib packages for OFED 1.3 in our release cycle
744 Description: fix SLES kernel versioning
745 Details : the kernel version for our SLES 10 kernel did not include a "-"
746 before the "smp" at the end. while this was not a problem in
747 general, it did mean that software trying to use the kernel
748 version to try to detect a vendor specific kernel would fail.
749 this was most evident by the OFED build scripts.
753 Description: Don't update lov_desc members until making sure they are valid
754 Details : When updating lov_desc members via proc fs, need fix their
755 validities before doing the real update.
759 Description: don't put request into delay list while invalidate in flight.
760 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
761 import in flight. this produce timeout for invalidate and sometimes
762 can cause stale data.
766 Frequency : on ppc only
767 Description: not convert ost objects for directory because it's not exist.
768 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
772 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
774 Severity : enhancement
776 Description: Update to support 2.6.22.14 vanilla kernel.
780 Frequency : rare, on recovery
781 Description: read procfs can produce deadlock in some situation
782 Details : Holding lprocfs lock which send rpc can produce block for destroy
783 obd objects and this also block reconnect with -EALREADY. This
784 isn't fix all lprocfs bugs - but make it rare.
786 Severity : enhancement
788 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
791 Frequency : frequent on X2 node
793 Description: mdc_set_open_replay_data LBUG
794 Details : Set replay data for requests that are eligible for replay.
798 Description: lustre_mgs: operation 101 on unconnected MGS
799 Details : When MGC is disconnected from MGS long enough, MGS will evict the
800 MGC, and late on MGC cannot successfully connect to MGS and a lot
801 of the error messages complaining that MGS is not connected.
805 Frequency : on network error
806 Description: panic with double free request if network error
807 Details : mdc_finish_enqueue is finish request if any network error ocuring,
808 but it's true only for synchronus enqueue, for async enqueue
809 (via ptlrpcd) this incorrect and ptlrpcd want finish request
812 Severity : enhancement
814 Description: client-side metadata stat-ahead during readdir(directory readahead)
815 Details : perform client-side metadata stat-ahead when the client detects
816 readdir and sequential stat of dir entries therein
819 Frequency : on start mds
821 Description: Implement get_info(last_id) in obdfilter.
824 Frequency : occasional
826 Description: Correctly check stale fid, not start epoch if ost not support SOM
827 Details : open with flag O_CREATE need set old fid in op_fid3 because
828 op_fid2 was overwritten with new generated fid, but mds can answer
829 with one of these two fids and both is not stale. Setattr
830 incorrectly started an epoch and assume will be called
831 done_writing, but without SOM done_writing ever being called.
834 Frequency : rare, depends on device drivers and load
836 Description: MDS or OSS nodes crash due to stack overflow
837 Details : Code changes in 1.8.0 increased the stack usage of some functions.
838 In some cases, in conjunction with device drivers that use a lot
839 of stack the MDS (or possibly OSS) service threads could overflow
840 the stack. One change which was identified to consume additional
841 stack has been reworked to avoid the extra stack usage.
844 Frequency : occasional
846 Description: Do not fail import if osc_interpret_create gets -EAGAIN
847 Details : If osc_interpret_create got -EAGAIN it immediately exits and
848 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
849 oscc_has_objects and see OSC has no objests and call
850 oscc_internal_create to resend create request.
852 Severity : enhancement
854 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
856 Severity : enhancement
858 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
861 Frequency : very rare
863 Description: Fix replay if there is an un-replied request and open
864 Details : In some cases, older replay request will revert the
865 mcd->mcd_last_xid on MDS which is used to record the client's
868 Severity : enhancement
870 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
872 Severity : enhancement
874 Description: Add rhel5 support to HEAD.
876 Serverity : enhancement
878 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
883 Description: Don't allow skipping OSTs if index has been specified.
884 Details : Don't allow skipping OSTs if index has been specified, make
885 locking in internal create lots better.
889 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
890 Details : access to bitfield in structure is always rounded to long
891 and this produce problem with not atomic change any bit.
895 Description: Lustre make rpms failed.
896 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
897 builds Lustre rpms from tarball.
900 Frequency : rare on shutdown ost
902 Description: If llog cancel was not send before clean_exports phase, this can
903 produce deadlock in llog code.
904 Details : If llog thread has last reference to obd and call class_import_put
905 this produce deadlock because llog_cleanup_commit_master wait when
906 last llog_commit_thread exited, but this never success because was
907 called from llog_commit_thread.
911 Description: allow userland application know is lost one of stripes.
912 Details : fill lvb_blocks with error code on ost and return it to
913 application if error flag found.
917 Description: NULL lov_tgts causing MDS oops
918 Details : more safe checks for NULL lov_tgts for avoid oops.
920 Severity : enhancement
922 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
926 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
927 Details : If an OST is down the MDS will hang indefinitely in
928 obd_statfs_rqset() waiting for the statfs data. While for
929 MDS QOS usage of statfs, it should not stuck in waiting.
931 Severity : enhancement
933 Description: remote_acl support
934 Details : Support ACL-based permission check for remote user.
935 Support setfacl/getfacl for remote user with the utils
936 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
937 the system "{s,g}etfacl" utils.
939 Severity : enhancement
941 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
943 Severity : enhancement
945 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
949 Description: make mgs_setparam() handle fsname containing dash
950 Details : fsname containing a dash does not work with lctl conf_param
952 Severity : enhancement
954 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
956 Severity : enhancement
958 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
962 Frequency : rare, at shutdown
963 Description: access already free / zero obd_namespace.
964 Details : if client_disconnect_export was called without force flag set,
965 and exist connect request in flight, this can produce access to
966 NULL pointer (or already free pointer) when connect_interpret
967 store ocd flags in obd_namespace.
971 Frequency : only at startup
972 Description: not alloc memory with spinlock held.
973 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
974 if any spinlock held.
976 Severity : enhancement
978 Description: make lustre randomly fail allocating memory
979 Details : Make lustre randomly failed allocating memory for testing purpose.
981 Severity : enhancement
983 Description: lost problems with lov objid file
984 Details : Fixes some scability and access to not inited memory problems
985 in work with lov objdid file.
990 Description: lfs find does not continue on file error
991 Details : Continue other files processing when a file/dir is absent.
995 Description: Inconsistent usage of lustre_pack_reply()
996 Details : Standardize the usage of lustre_pack_reply() such that it
997 always generate a CERROR on failure.
1000 Frequency : occasional
1002 Description: MDS hang or stay in waiting lock
1003 Details : If client receive lock with CBPENDING flag ldlm need send lock
1004 cancel as separate rpc, to avoid situation when cancel request
1005 can't processed due all i/o threads stay in wait lock.
1009 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1013 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1017 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1021 Description: Update to RHEL4 latest kernel.
1023 Severity : enhancement
1025 Description: Build SLES10 patchless client fails
1026 Details : The configure was broken by run ./configure with
1027 --with-linux-obj=.... argument for patchless client. When the
1028 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1029 can't search header adequately. Use absolute path such as
1030 -I($LINUX)/include instead.
1034 Description: interrupt oig_wait produce painc on resend.
1035 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1036 private set, and this produce situation when rq_ptlrpcd_data not
1037 copyed to new allocated request and triggered LBUG on assert
1038 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1039 ptlrpcd set if request is changed and can be safety changed to use
1044 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1045 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1046 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1047 these macros and replace them with their difinitions instead.
1051 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1052 Details : Update which_patch & target file for SLES10 latest kernel.
1055 Description: Remove client patches from SLES 10 kernel.
1056 Details : This causes SLES 10 clients to behave as patchless clients
1057 even on a Lustre-patched (server) kernel.
1059 Severity : enhancement
1061 Description: self-adjustable client's lru lists
1062 Details : use adaptive algorithm for managing client cached locks lru
1063 lists according to current server load, other client's work
1064 pattern, memory activities, etc. Both, server and client
1065 side namespaces provide number of proc tunables for controlling
1068 Severity : enhancement
1070 Description: light-weight GSS support
1071 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1072 and reduce performance overhead.
1074 Severity : enhancement
1076 Description: Linux keyring support
1077 Details : Support using service of Linux keyring for Lustre GSS internal
1078 context refresh/cache mechanism.
1082 Description: Fix errors in lfs documentation
1083 Details : Fixes man pages
1087 Description: don't use GFP_* in generic Lustre code.
1088 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1092 Description: obdclass is limited by single OBD_ALLOC(idarray)
1093 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1097 Description: warnings with build patchless client with vanila 2.6.19 and up
1098 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1099 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1103 Frequency : Only for SLES
1104 Description: sanity_quota fail test_1
1105 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1106 file, which makes TSTID[2] inunique.
1110 Frequency : Only for SLES9
1111 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1112 Details : Update target/ChangeLog/which_patch .
1116 Frequency : Only for test_10 in sanity-quota.sh
1117 Description: a bug in quota test code
1118 Details : checking a test flag in a wrong place
1122 Frequency : Only for test_13 in sanity-quota.sh
1123 Description: a bug in quota test script
1124 Details : shouldn't check overfull allocated quota
1128 Description: jbd statistics
1129 Details : Port older jbd statistics patch for sles10
1131 Severity : enhancement
1133 Description: Add printing inode info into message about error in writepage.
1138 Description: reply_lock_interpret crash due to race with it and lock cancel.
1139 Details : Do not replay locks that are being cancelled. Do not reference
1140 locks by their address during replay, just by their handle.
1144 Frequency : When flocks are used.
1145 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1146 Details : Flock locks might destroy just granted lock if it could be merged
1147 with another existing flock, this is done in completion handler,
1148 so teach ldlm_cli_enquque_fini that this is a valid case for
1154 Description: Oops in read and write path when failing to allocate lock.
1155 Details : Check if lock allocation failed and return error back.
1159 Description: lstripe command fails for valid OST index
1160 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1161 instead of lov->desc.ld_active_tgt_count.
1165 Description: sanity.sh failed test 103
1166 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1171 Description: df doesn't work properly if diskfs blocksize != 4K
1172 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1176 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1177 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1179 Severity : enhancement
1181 Description: Adaptive timeouts
1182 Details : RPC timeouts adapt to changing server load and network
1183 conditions to reduce resend attempts and improve recovery time.
1187 Description: llapi_file_create() does not allow some changes
1188 Details : add llapi_file_open() that allows specifying the mode and
1189 open flags, and also returns an open file handle.
1193 Description: merge and cleanup kernel patches.
1194 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1198 Description: Add journal checksum support.(Kernel part)
1199 Details : The journal checksum feature adds two new flags i.e
1200 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1201 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1202 indicates that the commit block contains the checksum for
1203 the blocks described by the descriptor blocks. Now commit
1204 record can be sent to disk without waiting for descriptor
1205 blocks to be written to disk. This behavior is controlled
1206 using JBD2_FEATURE_ASYNC_COMMIT flag.
1210 Description: OSS needs mutliple precreate threads
1211 Details : Add ability to start more than one create thread per OSS.
1215 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1216 Details : execute lfs setstripe on client
1220 Description: mds_obd_create error creating tmp object
1221 Details : When the user sets quota on root, llog will be affected and can't
1222 create files and write files.
1225 Frequency : Always on ia64 patchless client, and possibly others.
1227 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1228 Details : This allows the patchless client to be loaded on architectures
1229 without this export.
1233 Description: RedHat Update kernel for RHEL5
1234 Details : Add the kernel config file for RHEL5.
1238 Description: RedHat Update kernel for RHEL5
1239 Details : Modify the kernel config file more closer RHEL5.
1243 Description: Build failure against Centos5 (RHEL5)
1244 Details : Define PAGE_SIZE when it isn't present.
1248 Description: test_7 on sanity-quota.sh can't be used on separate machines
1249 Details : add facet to handle it
1253 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1254 Details : with NFS, the anon dentry's parent was set to itself in
1255 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1256 corresponding dentry other than use rec->ur_name.
1258 Severity : enhancement
1260 Description: lfs setstripe enhancement
1261 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1265 Description: enable data checksumming by default
1266 Details : enable checksum by default, allow --disable-checksum
1267 configure option and "-o nochecksum" mount option. Checksums
1268 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1272 Description: lustre support for RHEL5
1273 Details : Add support for RHEL5.
1277 Description: Client eviction due to failover config
1278 Details : after a connection loss, the lustre client should attempt to
1279 reconnect to the last active server first before trying the
1280 other potential connections.
1284 Description: when mds and osts use different quota unit(32bit and 64bit),
1285 quota will be released repeatly.
1286 Details : void sending multiple quota reqs to mds, which will keep the
1287 status between the reqs.
1291 Description: osts not allocated evenly to files
1292 Details : change the condition to increase offset_idx
1295 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1297 Description: Data corruption for OSTs that are formatted larger than 2TB
1299 Details : When generating the bio request for lustre file writes the
1300 sector number would overflow a temporary variable before being
1301 used for the IO. The data reads correctly from Lustre (which
1302 will overflow in a similar manner) but other file data or
1303 filesystem metadata may be corrupted in some cases.
1307 Description: Tune the kernel for good SCSI performance.
1308 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1309 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1314 Description: rewrite ext2-derived code in obdclass/uuid.c
1315 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1316 from scratch preserving functionality.
1320 Description: "lfs find -obd UUID" prints directories
1321 Details : "lfs find -obd UUID" will return all directory names instead
1322 of just file names. It is incorrect because the directories
1323 do not reside on the OSTs.
1327 Description: use i_size_read and i_size_write in 2.6 port
1328 Details : replace inode->i_size access with i_size_read/write()
1331 Frequency : when removing large files
1333 Description: scheduling issue during removal of large Lustre files
1334 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1335 It causes scheduling issues when removing large files (17TB in the
1339 Frequency : only with liblustre clients on XT3
1341 Description: evictions taking too long
1342 Details : allow llrd to evict clients directly on OSTs
1345 Frequency : only on ppc
1347 Description: /proc/fs/lustre/devices broken on ppc
1348 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1349 We should make sure the type of 'index' is loff_t and then cast
1350 explicitly as needed below. Do not assign an explicitly cast
1354 Frequency : only for rhel5
1356 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1357 Details : Modify the target file & which_kernel.
1359 Severity : enhancement
1361 Description: omit set fsid for export NFS
1362 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1363 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1367 Frequency : Always, for kernels after 2.6.16
1368 Description: Fix warning idr_remove called for id=.. which is not allocated.
1369 Details : Last kernels save old s_dev before kill super and not allow
1370 to restore from callback - restore it before call kill_anon_super.
1374 Description: buffer overruns could theoretically occur
1375 Details : llapi_semantic_traverse() modifies the "path" argument by
1376 appending values to the end of the origin string, and a
1377 overrun may occur. Adding buffer overrun check in liblustreapi.
1381 Description: Fix error on 'ls .' at the top of the Lustre mount.
1382 Details : Don't revalidate dentry if it is a root dentry.
1386 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1387 Details : Modify vm-tunables-rhel4.patch.
1391 Description: Kernel config for 2.6.18-vanilla.
1392 Details : Modify targets/2.6-vanilla.target.in.
1393 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1394 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1395 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1396 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1400 Description: improve handling recoverable errors
1401 Details : if request processig with error which can be recoverable on server
1402 request should be resend, otherwise page released from cache and
1407 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1408 Details : Modify target file & which_patch.
1409 A flaw was found in the IA32 system call emulation provided
1410 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1411 value could be stored in the %RAX register, which could trigger an
1412 out-of-bounds system call table access. An untrusted local user
1413 could exploit this flaw to run code in the kernel
1414 (ie a root privilege escalation). (CVE-2007-4573).
1418 Description: O_DIRECT bypasses client statistics.
1419 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1420 read_ahead_stats were updated. copied the stats section from
1421 osc_send_oap_rpc() into async_internal().
1425 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1429 Description: change order of libsysio includes
1430 Details : '#include sysio.h' should always come before '#include xtio.h'
1434 Description: handle "serious error: objid * already exists" more gracefully
1435 Details : If LAST_ID value on disk is smaller than the objects existing in
1436 the O/0/d* directories, it indicates disk corruption and causes an
1437 LBUG(). If the object is 0-length, then we should use the existing
1438 object. This will help to avoid a full fsck in most cases.
1440 Severity : enhancement
1442 Description: adapt the lustre_config script to support the upgrade case
1443 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1444 server targets to 1.6 in parallel.
1448 Description: To avoid grant space > avaible space when the disk is almost
1449 full. Without this patch you might see the error "grant XXXX >
1450 available" or some LBUG about grant, when the disk is almost
1452 Details : In filter_check_grant, for non_grant cache write, we should
1453 check the left space by if (*left > ungranted + bytes), instead
1454 of (*left > ungranted), because only we are sure the left space
1455 is enough for another "bytes", then the ungrant space should be
1456 increase. In client, we should update cl_avail_grant only there
1457 is OBD_MD_FLGRANT in the reply.
1461 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1462 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1463 described in CVE-2007-4573. This problem could allow untrusted
1464 local users to gain root access.
1467 Frequency : when using O_DIRECT and quotas
1469 Description: Incorrect file ownership on O_DIRECT output files
1470 Details : block usage reported by 'lfs quota' does not take into account
1471 files that have been written with O_DIRECT.
1476 Description: touch file failed when fs is not full
1477 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1478 otherwise we can get ENOSP while fs is not full.
1482 Description: parallel lock callbacks
1483 Details : Instead of sending blocking and completion callbacks as separated
1484 requests, adding them to a set and sending in parallel.
1487 Frequency : only for Cray XT3
1488 Bugzilla : 12829/13455
1489 Description: Changing primary group doesn't change the group lustre assigns to
1491 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1492 overridden with the primary group provided by the group upcall,
1493 whereas the supplied fsgid can be trusted if it is in the list of
1494 supplementary groups returned by the group upcall.
1496 Severity : enhancement
1498 Description: Allow masking D_WARNING, D_ERROR messages from console
1499 Details : Console messages can now be disabled via lnet.printk.
1503 Description: User code with malformed file open parameter crashes client node
1504 Details : Before packing join_file req, all the related reference should be
1505 checked carefully in case some malformed flags cause fake
1506 join_file req on client.
1510 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1511 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1512 mode is equal to the granted mode. However, it should also check
1513 that LDLM_FL_CP_REQD is not set.
1517 Description: Heavy nfs access might result in deadlocks
1518 Details : After ELC code landed, it is now improper to enqueue any mds
1519 locks under och_sem, because enqueue might want to decide to
1520 cancel open locks for same inode we are holding och_sem for.
1524 Description: Client eviction while running blogbench
1525 Details : A lot of unlink operations with concurrent I/O can lead to a
1526 deadlock causing evictions. To address the problem, the number of
1527 oustanding OST_DESTROY requests is now throttled to
1528 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1529 callbacks are processed in priority.
1533 Description: enable ACLs on MDS by default
1534 Details : ACLs must be enabled on MDS by default.
1537 Frequency : PPC/PPC64 only
1539 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1540 Details : fix duplicated definitions between asm-ppc64/types.h and
1541 lustre_types.h on PPC.
1544 Frequency : PPC/PPC64 only
1546 Description: asm-ppc/segment.h does not exist
1547 Details : fix compile issue on PPC.
1551 Description: better handle error messages in extents code
1554 Frequency : RHEL4 only
1556 Description: mkfs is very slow on IA64/RHEL4
1557 Details : A performance regression has been discovered in the MPT Fusion
1558 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1559 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1560 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1562 Severity : enhancement
1564 Description: SNMP support enhancement
1565 Details : Adding total number of sampled request for an MDS node in snmp
1568 Severity : enhancement
1570 Description: Optimize ldlm waiting list processing for PR extent locks
1571 Details : When processing waiting list for read extent lock and meeting read
1572 lock that is same or wider to it that is not contended, skip
1573 processing rest of the list and immediatelly return current
1574 status of conflictness, since we are guaranteed there are no
1575 conflicting locks in the rest of the list.
1579 Description: Time out and refuse to reconnect
1580 Details : When the failover node is the primary node, it is possible
1581 to have two identical connections in imp_conn_list. We must
1582 compare not conn's pointers but NIDs, otherwise we can defeat
1583 connection throttling.
1587 Description: port llog fixes from b1_6 into HEAD
1588 Details : Port llog reference couting and some llog cleanups from b1_6
1589 (bug 10800) into HEAD, for protect from panic and access to already
1590 free llog structures.
1594 Description: Detect stride IO mode in read-ahead
1595 Details : When a client does stride read, read-ahead should detect that and
1596 read-ahead pages according to the detected stride pattern.
1600 Description: data checksumming impacts single node performance
1601 Details : add support for several checksum algorithm. Currently, only CRC32
1602 and Adler-32 are supported. The checksum type can be changed on
1603 the fly via /proc/fs/lustre/osc/*/checksum_type.
1607 Description: use adler32 for page checksums
1608 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1613 Description: build for x2 fails
1614 Details : fix compile issue on Cray systems.
1618 Description: Properly match for duplicate locks
1619 Details : Due to different lock order from skiplists code, we need to
1620 traverse entire list for now
1623 Frequency : only on PPC/SLES10
1625 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1626 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1627 be defined. Add a hack in mkfs_lustre.c to work around this compile
1632 Description: LASSERT on MDS when client holding flock lock dies
1633 Details : ldlm pool logic depends on number of granted locks equal to
1634 number of released locks which is not true for flock locks, so
1635 just exclude such locks from consideration.
1639 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1640 Details : Use fsfilt_sync() for both the whole filesystem sync and
1641 individual file sync to eliminate dangerous inode locking
1642 with I_LOCK that can lead to a deadlock.
1646 Description: performance in 1.6.3
1647 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1648 sure that 1MB requests are not fragmented by the block layer.
1650 Severity : enhancement
1652 Description: organize the server-side client stats on per-nid basis
1653 Details : Change the structure of stats under obdfilter and mds to
1661 The "uuid"s file would list the uuids of _active_ exports.
1662 And the clear entry is to clear all stats and stale nids.
1664 Severity : enhancement
1666 Description: eliminate client locks in face of contention
1667 Details : file contention detection and lockless i/o implementation
1668 for contended files.
1672 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1675 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1677 Description: MDS may lose file striping (and hence file data) in some cases
1678 Details : If there are additional extended attributes stored on the MDS,
1679 in particular ACLs, SELinux, or user attributes (if user_xattr
1680 is specified for the client mount options) then there is a risk
1681 of attribute loss. Additionally, the Lustre file striping
1682 needs to be larger than default (e.g. striped over all OSTs),
1683 and an additional attribute must be stored initially in the
1684 inode and then increase in size enough to be moved to the
1685 external attribute block (e.g. ACL growing in size) for file
1690 Description: skiplist implementation simplification
1691 Details : skiplists are used to group compatible locks on granted list
1692 that was implemented as tracking first and last lock of each
1693 lock group the patch changes that to using doubly linked lists
1697 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1698 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1699 because LASSERT on both the data supplied by a client, and the
1700 data on disk is dangerous and incorrect.
1702 Severity : enhancement
1704 Description: Slow truncate/writes to huge files at high offsets.
1705 Details : Directly associate cached pages to lock that protect those pages,
1706 this allows us to quickly find what pages to write and remove
1707 once lock callback is received.
1711 Description: more ldlm soft lockups
1712 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1713 starve other threads from the resource lock for a long time in
1714 case of long waiting queue, so change the debug level from
1715 D_OTHER to the less frequently used D_INFO.
1717 Severity : enhancement
1719 Description: add -gid, -group, -uid, -user options to lfs find
1723 Description: Hung threads in invalidate_inode_pages2_range
1724 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1725 once one is completed. As a result, some RPCs are stuck in the
1726 queue and are never sent.
1730 Description: filter threads hungs on waiting journal commit
1731 Details : Cleanup filter group llog code, then only filter group llog will
1732 be only created in the MDS/OST syncing process.
1736 Description: Procfs and llog threads access destoryed import sometimes.
1737 Details : Sync the import destoryed process with procfs and llog threads by
1738 the import refcount and semaphore.
1740 Severity : enhancement
1742 Description: openlock cache of b1_6 port to HEAD
1747 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1748 Details : If insertion of an extent fails, then discard the inode
1749 preallocation and free data blocks else it can lead to duplicate
1754 Description: don't always update ctime in ext3_xattr_set_handle()
1755 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1756 In some cases the ctime should not be updated, for example for
1757 2.0->1.8 compatibility it is necessary to delete an xattr and it
1758 should not update the ctime.
1762 Bugzilla : 15713/16362
1763 Description: Assertion in iopen_connect_dentry in 1.6.3
1764 Details : looking up an inode via iopen with the wrong generation number can
1765 populate the dcache with a disconneced dentry while the inode
1766 number is in the process of being reallocated. This causes an
1767 assertion failure in iopen since the inode's dentry list contains
1768 both a connected and disconnected dentry.
1772 Description: assertion failure in ldlm_handle2lock()
1773 Details : fix a race between class_handle_unhash() and class_handle2object()
1774 introduced in lustre 1.6.5 by bug 13622.
1779 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1780 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1785 Description: oops in sd_iostats_seq_show()
1786 Details : unloading/reloading the scsi low level driver triggers a kernel
1787 bug when trying to access the sd iostat file.
1792 Description: Kernel panics during QLogic driver reload
1793 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1794 patch, causing memory corruption.
1799 Description: journal_dev option does not work in b1_6
1800 Details : pass mount option during pre-mount.
1802 Severity : enhancement
1804 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1805 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1806 extent information of a file. It can be used to map logical blocks
1807 in a file to physical blocks in the block device.
1811 Description: LDLM soft lockups - improvement
1812 Details : It is be possible to send the lock handle along with each read
1813 or write request because the client is already doing a lock match
1814 itself so there isn't any reason the OST should have to re-do that
1818 Frequency : only on Cray X2
1820 Description: X2 build failures
1821 Details : fix build failures on Cray X2.
1825 Description: xid & resent requests
1826 Details : Initialize RPC XID from clock at startup (randomly if clock is
1829 Severity : enhancement
1831 Description: Add lustre_start utility to start or stop multiple Lustre servers
1836 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1837 Details : In case of memory pressure, list_del() can be called twice on
1838 req->rq_history_list, causing a kernel oops.
1842 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1843 Details : fix stack overflow in the distributed lock manager by defering
1844 export eviction after a failed AST to the elt thread instead of
1845 handling it in the dlm interpret routine.
1849 Description: Convert some comments to new format.
1850 Details : Update documenting comments to match doxygen conventions.
1854 Description: Grammar fixes.
1855 Details : A couple of trivial spelling fixes.
1859 Description: OSD_COUNTERS-mandatory
1860 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1861 to simplify the code.
1865 Description: simplify cmm_device freeing logic.
1866 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1870 Description: Add lockdep support to dt_object_operations locking interface.
1871 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1872 parameter indicating lock ordering. Update mdd code to use new
1877 Description: Introduce failloc constants for lockless IO tests.
1878 Details : Add two new failloc constants to test lockless IO. Only one of
1879 them in implemented---another is checked in yet to be landed
1884 Description: Add lockdep support for inode mutex.
1885 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1886 in the situations where Lustre has to lock more than one inode
1891 Description: Add optional invariants checking support.
1892 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1893 configure switch. This macro is to be used for consistency and
1894 sanity checks that are too expensive to be left in `production'
1899 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1900 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1905 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1906 Details : Use spin_lock_nested() in (the only) situation where more than
1907 one ldlm_lock is locked simultaneously. Also, fix possible
1908 dead-lock in ldlm_lock_change_resource() by enforcing particular
1913 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1914 Details : Instead of specifying each ldlm_lock call-back through separate
1915 parameter, wrap them into struct ldlm_callback_suite.
1919 Description: Kill join_lru obd method and its callers.
1920 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1921 in memory---a requirement implemented through ->o_join_lru() obd
1922 method in HEAD. Remove this method and its users.
1926 Description: Add asynchronous ldlm ENQUEUE completion handler.
1927 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1928 case that stock ldlm_completion_ast() cannot handle as it waits
1929 until lock is granted. Introduce new ldlm_completion_ast_async()
1930 for this. Also comment ldlm_completion_ast().
1934 Description: ldlm_error <-> errno conversion.
1935 Details : Add functions to map (rather arbitrary) between LDLM error codes
1936 and standard errno values. CLIO needs this to prevent LDLM specific
1937 constants from escaping ldlm and osc.
1941 Description: Kill unused ldlm_handle2lock_ns() function.
1942 Details : Kill unused ldlm_handle2lock_ns() function.
1946 Description: Add lu_ref support to ldlm_lock
1947 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1948 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1949 and ldlm_resource. LDLM interface has to be changed, because code that
1950 releases a reference on a lock, has to "know" what reference this is.
1951 In the most frequent case
1953 lock = ldlm_handle2lock(handle);
1955 LDLM_LOCK_PUT(lock);
1957 no changes are required. When any other reference (received _not_ from
1958 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1959 instead of LDLM_LOCK_PUT().
1961 Arguably, changes are pervasive, and interface requires some discipline
1962 for proper use. On the other hand, it was very instrumental in finding
1963 a few leaked lock references.
1967 Description: Add ldlm_lock_addref_try().
1968 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1969 attempts to addref a lock that might be being canceled
1974 Description: Add ldlm_weigh_callback().
1975 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1976 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1977 hard-coded `number of pages' logic.
1981 Description: Add lockdep annotations to llog code.
1982 Details : Use appropriately tagged _nested() locking calls in the places
1983 where llog takes more than one ->lgh_lock lock.
1987 Description: Add loi_kms_set().
1988 Details : Wrap kms updates into a helper function.
1992 Description: Constify instances of struct lsm_operations.
1993 Details : Constify instances of struct lsm_operations.
1997 Description: lu_conf support.
1998 Details : On a server, a file system object is uniquely identified
1999 by a fid, which is sufficient to locate and load all object
2000 state (inode). On a client, on the other hand, more data are
2001 necessary instantiate an object. Change lu_object_find() and
2002 friends to take additional `lu_conf' argument describing object.
2003 Typically this includes layout information.
2007 Description: lu_context fixes.
2008 Details : Introduce new lu_context functions that are needed on the client
2009 side, where some system threads (ptlrpcd) are shared by multiple
2010 modules, and so cannot be stopped during module shutdown.
2014 Description: Add start and stop methods to lu_device_type_operations.
2015 Details : Introduce two new methods in lu_device_type_operations, that are
2016 invoked when first instance of a given type is created and last one
2017 is destroyed respectively. This is need by CLIO.
2021 Description: Add lu_ref support to struct lu_device.
2022 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2023 track leaked references.
2027 Description: Introduce lu_kmem_descr.
2028 Details : lu_kmem_descr and its companion interface allow to create
2029 and destroy a number of kmem caches at once.
2033 Description: Fix lu_object finalization race.
2034 Details : Fix a race between lu_object_find() finding an object and its
2035 concurrent finalization. This race is (most likely) not possible
2036 on the server, but might happen on the client.
2040 Description: Introduce lu_ref interface.
2041 Details : lu_ref is a debugging module allowing to track references to
2042 a given object. It is quite cpu expensive, and has to be
2043 explicitly enabled with --enable-lu_ref. See usage description
2048 Description: Factor lu_site procfs stats into a separate function.
2049 Details : Separate lu_site stats printing code into a separate function
2050 that can be reused on a client.
2054 Description: Constify instances of struct {lu,dt,md}_device_operations.
2055 Details : Constify instances of struct {lu,dt,md}_device_operations.
2059 Description: Introduce struct md_site and move meta-data specific parts of
2060 struct lu_site here.
2061 Details : Move md-specific fields out of struct lu_site into special struct
2062 md_site, so that lu_site can be used on a client.
2066 Description: Kill mdd_lov_destroy().
2067 Details : Remove unused mdd code.
2071 Description: Add st_block checking to multistat.c.
2072 Details : Add st_block checking to multistat.c.
2076 Description: Add lu_ref support to struct obd_device.
2077 Details : Add lu_ref tracking to obd_device.
2081 Description: Kill obd_set_fail_loc().
2082 Details : Remove unused code.
2086 Description: Add special type for ptlrpc_request interpret functions.
2087 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2088 there. Actual usage will be in CLIO.
2092 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2093 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2094 doesn't work with lockdep.
2098 Description: Add rwv.c test program.
2099 Details : New testing program exercising readv(2) and writev(2) (Qian).
2103 Description: Add sendfile.c test program.
2104 Details : New testing program exercising sendfile(2) (Jay).
2108 Description: Ratelimit a message that can be very frequent.
2109 Details : Ratelimit a memory allocation failure message that can
2114 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2115 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2119 Description: Liblustre build fixes.
2120 Details : Liblustre build fixes.
2124 Description: libcfs: add cfs_{need,cond}_resched() interface.
2125 Details : libcfs: add cfs_{need,cond}_resched() definition and
2126 implementations for Linux, NT, and liblustre.
2128 Severity : enhancement
2130 Description: More exported tunables for mballoc
2131 Details : Add support for tunable preallocation window and new tunables for
2132 large/small requests
2136 Description: Detect corruption of block bitmap and checking for preallocations
2137 Details : Checks validity of on-disk block bitmap. Also it does better
2138 checking of number of applied preallocations. When corruption is
2139 found, it turns filesystem readonly to prevent further corruptions.
2143 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2144 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2145 window to 0. If it does hit cache miss, and read-pattern is still
2146 stride-io mode, does not reset the stride window, but also does not
2147 increase the stride window length in this case.
2151 Frequency : only for big-endian servers
2152 Description: Check if system is big-endian while mounting fs with extents feature
2153 Details : Mounting a filesystem with extents feature will fail on big-endian
2154 systems since ext3-based ldiskfs is not supported on big-endian
2155 systems. This can be over-riden with "bigendian_extents" mount option.
2157 Severity : enhancement
2159 Description: The root squash functionality
2160 Details : A security feature, which is to prevent users from being able
2161 to mount lustre on their desktop, run as root, and delete
2162 all of the files in the filesystem. The goal is accomplished by
2163 remapping user id (UID) and group id (GID) of the root user to
2164 a UID and GID specified by the system administartor via Lustre
2165 configuration management server (MGS). The functionality also
2166 allows to specify sets of clients for which the remapping does
2171 Description: Excessive recovery window
2172 Details : With AT enabled, the recovery window can be excessively long (6000+
2173 seconds). To address this problem, we no longer use
2174 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2175 timeout no longer depends on the service time, it is set to
2176 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2177 time via pb_service_time.
2181 Description: Watchdog triggered on MDS failover
2182 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2183 the OSTs know that the MDS "UUID" can be reused for the same export
2184 from a different NID, so we do not need to wait for the export to be
2188 Frequency : rare, only if using MMP with Linux RAID
2190 Description: MMP doesn't work with Linux RAID
2191 Details : While using HA for Lustre servers with Linux RAID, it is possible
2192 that MMP will not detect multiple mounts. To make this work we
2193 need to unplug the device queue in RAID when the MMP block is being
2194 written. Also while reading the MMP block, we should read it from
2195 disk and not the cached one.
2197 Severity : enhancement
2199 Description: open file using fid
2200 Details : A file can be opened using just its fid, like
2201 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2204 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2207 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2208 Details : add proper locking for creation of jbd_4k slab cache
2212 Description: MMP check in ext3_remount() fails without displaying any error
2213 Details : When multiple mount protection fails during remount, proper error
2216 Severity : enhancement
2218 Description: Allow stripe size to be up to 4G-64k
2219 Details : Fix math logic to allow large stripe sizes.
2223 Description: add check for >8TB ldiskfs filesystems
2224 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2225 Don't allow >8TB ldiskfs filesystems to be mounted without
2226 force_over_8tb mount option
2231 Descriptoin: OSS complains "too many blocks" and crashes
2232 Details : In filter_commitrw_write():
2233 1. lquota_chkquota();
2234 2. obdo_from_inode();
2235 3. lquota_pending_commit();
2236 "2." has changed uid and gid in this case, then caused
2237 lquota_chkquota() and lquota_pending_commit() to use
2238 different uid and gid. That is the root of the bug.
2242 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2243 Details : Detect and print out the 'stripe count = -1' case properly.
2245 --------------------------------------------------------------------------------
2247 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2249 * Support for kernels:
2250 2.6.5-7.283 (SLES 9),
2251 2.6.9-55.EL (RHEL 4),
2252 2.6.16.46-0.14 (SLES 10),
2253 2.6.18.8 vanilla (kernel.org)
2254 * Client support for unpatched kernels:
2255 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2256 2.6.16 - 2.6.22 vanilla (kernel.org)
2257 * Due to recently discovered recovery problems, we do not recommend
2258 using patchless RHEL 4 clients with this or any earlier release.
2259 * Recommended e2fsprogs version: 1.39.cfs8
2260 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2261 * Starting with this release, the ldiskfs backing filesystem required
2262 by Lustre is now in its own package, lustre-ldiskfs. This package
2263 should be installed. It is versioned separately from Lustre and
2264 may be released separately in future.
2268 Description: block reactivating mgc import until all deactivates complete
2269 Details : Fix race when failing back MDT/MGS to itself (testing)
2272 Frequency : at statup only
2274 Description: mds_lov_synchronize race leads to various problems
2275 Details : simultaneous MDT->OST connections at startup can cause the
2276 sync to abort, leaving the OSC in a bad state.
2278 Severity : enhancement
2280 Description: add optional extra BUILD_VERSION info
2281 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2282 to override the lustre version.
2285 Frequency : 2.6.18 servers only
2287 Description: ll_kern_mount() doesn't release the module reference
2288 Details : The ldiskfs module reference count never drops down to 0
2289 because ll_kern_mount() doesn't release the module reference.
2294 Description: server LBUG when using old ost_num_threads parameter
2295 Details : Accept the old ost_num_threads parameter but warn that it
2296 is deprecated, and fix an off-by-one error that caused an LBUG.
2301 Description: Transient SCSI error results in persistent IO issue
2302 Details : iobuf->dr_error is not reinitialized to 0 between two
2306 Frequency : sometimes when underlying device returns I/O errors
2308 Description: OSTs not going read-only during write failures
2309 Details : OSTs are not remounted read-only when the journal commit threads
2310 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2311 instead of the ext3 wrappers.
2315 Description: poor connect scaling with increasing client count
2316 Details : Don't run filter_grant_sanity_check for more than 100 exports
2317 to improve scaling for large numbers of clients.
2320 Frequency : SLES10 only
2322 Description: sanity-quota.sh quotacheck failed: rc = -22
2323 Details : Quotas cannot be enabled on SLES10.
2326 Frequency : liblustre clients only
2328 Description: getdirentries does not give error when run on compute nodes
2329 Details : getdirentries does not fail when the size specified as an argument
2330 is too small to contain at least one entry
2332 Severity : enhancement
2334 Description: Add LNET router traceability for debug purposes
2335 Details : If a checksum failure occurs with a router as part of the
2336 IO path, the NID of the last router that forwarded the bulk data
2337 is printed so it can be identified.
2342 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2343 Details : Due to a race condition, liblustre clients were occasionally
2344 evicted incorrectly.
2346 Severity : enhancement
2348 Description: lfs setstripe use optional parameters instead of postional
2351 Severity : enhancement
2353 Description: Nanosecond timestamp support for ldiskfs
2354 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2355 resolution timestamps. There is not yet support for this at
2356 the Lustre filesystem level.
2359 Frequency : during server recovery
2361 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2362 Details : request with rq_no_resend flag not awake l_wait_event if they get
2366 Frequency : nfs export on patchless client
2368 Description: connectathon hang when test nfs export over patchless client
2369 Details : Disconnected dentry cannot be found with lookup, so we do not need
2370 to unhash it or make it invalid
2373 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2374 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2375 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2380 Description: open req refcounting wrong on reconnect
2381 Details : If reconnect happened between getting open reply from server and
2382 call to mdc_set_replay_data in ll_file_open, we will schedule
2383 replay for unreferenced request that we are about to free.
2384 Subsequent close will crash in variety of ways.
2385 Check that request is still eligible for replay in
2386 mdc_set_replay_data().
2391 Description: disable writes to filesystem when reading health_check file
2392 Details : the default for reading the health_check proc file has changed
2393 to NOT do a journal transaction and write to disk, because this
2394 can cause reads of the /proc file to hang and block HA state
2395 checking on a healthy but otherwise heavily loaded system. It
2396 is possible to return to the previous behaviour during configure
2397 with --enable-health-write.
2399 Severity : enhancement
2401 Description: 64-bit inode version
2402 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2403 to the inode. This will be required for version-based recovery.
2408 Description: MDS fails to start if a duplicate client export is detected
2409 Details : in some rare cases it was possible for a client to connect to
2410 an MDS multiple times. Upon recovery the MDS would detect this
2411 and fail during startup. Handle this more gracefully.
2413 Severity : enhancement
2415 Description: Add -o localflock option to simulate old noflock
2417 Details : This will achieve local-only flock/fcntl locks
2423 Description: log_commit_thread vs filter_destroy race leads to crash
2424 Details : Take import reference before releasing llog record semaphore
2429 Description: Wrong request locking in request set processing
2430 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2431 imp_delayed_list, in this place should be used imp_lock.
2434 Frequency : when reconnection
2436 Description: Grant Leak when osc reconnect to OST
2437 Details : When osc reconnect ost, OST(filter) should check whether it
2438 should grant more space to client by comparing fed_grant and
2439 cl_avail_grant, and return the granted space to client instead
2440 of "new granted" space, because client will call osc_init_grant
2441 to update the client grant space info.
2444 Frequency : when client reconnect to OST
2446 Description: Grant Leak when osc do resend and replay bulk write
2447 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2448 bulk write request, because the grant info while be sync between
2449 OSC and OST when reconnect, and we should ignore the grant info
2450 these of resend/replay write req.
2455 Description: Grant space more than avaiable left space sometimes.
2456 Details : When then OST is about to be full, if two bulk writing from
2457 different clients came to OST. Accord the avaliable space of the
2458 OST, the first req should be permitted, and the second one
2459 should be denied by ENOSPC. But if the seconde arrived before
2460 the first one is commited. The OST might wrongly permit second
2461 writing, which will cause grant space > avaiable space.
2464 Frequency : when client is evicted
2466 Description: Grant might be wrongly erased when osc is evicted by OST
2467 Details : when the import is evicted by server, it will fork another
2468 thread ptlrpc_invalidate_import_thread to invalidate the
2469 import, where the grant will be set to 0. While the original
2470 thread will update the grant it got when connecting. So if
2471 the former happened latter, the grant will be wrongly errased
2472 because of this race.
2477 Description: Checking Stale with correct fid
2478 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2479 to check whether it is stale, because sometimes, we want the
2480 enqueue happened anyway, and op_data.fid2 will not be initialized.
2482 Severity : enhancement
2484 Description: update patchless client
2485 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2488 Frequency : only with 2.4 kernel
2490 Description: random memory corruption
2491 Details : size of struct ll_inode_info is to big for union inode.u and this
2492 can be cause of random memory corruption.
2497 Description: Memory leak in recovery
2498 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2499 It should also check obd_fail before fsfilt_start, otherwise if
2500 fsfilt_start return -EROFS,(failover mds during mds recovery).
2501 then the req will return with repmsg->transno = 0 and rc = EROFS.
2502 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2503 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2504 be freed no matter whether fsfilt_commit success or not.
2507 Frequency : only with huge count clients
2509 Description: Prevents from taking the superblock lock in llap_from_page for
2511 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2512 for llap_from_page call in ll_removepage prevents from taking the
2513 superblock lock for a soon died page.
2518 Description: Not check open intent error before release open handle
2519 Details : in some rare cases, the open intent error is not checked before
2520 release open handle, which may cause
2521 ASSERTION(open_req->rq_transno != 0), because it tries to release
2522 the failed open handle.
2527 Description: Set cat log bitmap only after create log success.
2528 Details : in some rare cases, the cat log bitmap is set too early. and it
2529 should be set only after create log success.
2533 Description: Accessing a block bevice can re-enable I/O when Lustre is
2534 tearing down a device.
2535 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2540 Description: service threads may hog cpus when there are a lot of requests
2541 Details : Insert cond_resched to give other threads a chance to use some CPU
2546 Description: the cat log was not initialized in recovery
2547 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2548 unlink log on mds will not be initialized until mds post
2549 recovery. And also in mds post recovery, the unlink log will
2550 initialization will be done asynchronausly, so there will be race
2551 between add unlink log and unlink log initialization.
2555 Description: brw_stats were being printed incorrectly
2556 Details : brw_stats were being printed as log2 but all of them were not
2557 recorded as log2. Also remove some code duplication arising from
2558 filter_tally_{read,write}.
2562 Frequency : rare, only in recovery.
2563 Description: ASSERTION(req->rq_type != LI_POISON) failed
2564 Details : imp_lock should be held while iterating over imp_sending_list for
2565 prevent destroy request after get timeout in ptlrpc_queue_wait.
2569 Description: replay-single.sh test 52 fails
2570 Details : A lock's skiplist need to be cleanup when it being unlinked
2571 from its resource list.
2575 Description: Short directio read returns full requested size rather than
2577 Details : Direct I/O operations should return actual amount of bytes
2578 transferred rather than requested size.
2580 Severity : enhancement
2582 Description: metadata RPC reduction (e.g. for rm performance)
2583 Details : decrease the amount of synchronous RPC between clients and servers
2584 by canceling conflicing lock before the operation on the client
2585 and packing thier handles into the main operation RPC to server.
2587 Severity : enhancement
2589 Description: Async OSC create to avoid the blocking unnecessarily.
2590 Details : If a OST has no remain object, system will block on the creating
2591 when need to create a new object on this OST. Now, ways use
2592 pre-created objects when available, instead of blocking on an
2593 empty osc while others are not empty. If we must block, we block
2594 for the shortest possible period of time.
2598 Description: improve handling recoverable errors
2599 Details : if request processig with error which can be recoverable on server
2600 request should be resend, otherwise page released from cache and
2603 Severity : enhancement
2605 Description: refine locking for avoid write wrong info into lov_objid file
2606 Details : fix possible races with add new target and write/update data in
2609 --------------------------------------------------------------------------------
2611 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2616 Frequency : on some architectures
2618 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2619 Details : Uninitialized flags sometimes cause configuration commands to
2623 Frequency : patchless clients only
2625 Description: missing __iget() symbol export
2626 Details : The __iget() symbol export is missing. To avoid the need for
2627 this on patchless clients the deathrow inode reaper is turned
2628 off, and we depend on the VM to clean up old inodes. This
2629 dependency was during via the fix for bug 12181.
2631 --------------------------------------------------------------------------------
2633 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2635 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2636 INTEROPERATE with older versions automatically. In many cases a
2637 special upgrade step is needed. Please read the
2638 user documentation before upgrading any part of a live system.
2639 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2640 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2641 * WARNING: Lustre configuration and startup changes are required with
2642 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2646 Severity : enhancement
2648 Description: MountConf
2649 Details : Lustre configuration is now managed via mkfs and mount
2650 commands instead of lmc and lconf. New obd types (MGS, MGC)
2651 are added for dynamic configuration management. See
2652 https://mail.clusterfs.com/wikis/lustre/MountConf for
2655 Severity : enhancement
2657 Description: dynamic OST addition
2658 Details : OSTs can now be added to a live filesystem
2660 Severity : enhancement
2662 Description: startup order invariance
2663 Details : MDTs and OSTs can be started in any order. Clients only
2664 require the MDT to complete startup.
2666 Severity : enhancement
2668 Description: parallel, asynchronous orphan cleanup
2669 Details : orphan cleanup is now performed in separate threads for each
2670 OST, allowing parallel non-blocking operation.
2672 Severity : enhancement
2674 Description: optimized stripe assignment
2675 Details : stripe assignments are now made based on ost space available,
2676 ost previous usage, and OSS previous usage, in order to try
2677 to optimize storage space and networking resources.
2679 Severity : enhancement
2681 Description: Permanently set tunables
2682 Details : All writable /proc/fs/lustre tunables can now be permanently
2683 set on a per-server basis, at mkfs time or on a live system.
2685 Severity : enhancement
2687 Description: Lustre message v2
2688 Details : Add lustre message format v2.
2690 Severity : enhancement
2692 Description: client OST exclusion list
2693 Details : Clients can be started with a list of OSTs that should be
2694 declared "inactive" for known non-responsive OSTs.
2698 Description: SPEC SFS validation failure on NFS v2 over lustre.
2699 Details : Changes the blocksize for regular files to be 2x RPC size,
2700 and not depend on stripe size.
2702 Severity : enhancement
2704 Description: Multiple MD RPCs in flight.
2705 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2706 To avoid overly-overloading MDS, introduce a limit on number of
2707 MDS RPCs in flight for a single client and add /proc controls
2708 to adjust this limit.
2710 Severity : enhancement
2712 Description: client read/write statistics
2713 Details : Add client read/write call usage stats for performance
2714 analysis of user processes.
2715 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2716 file access. extents_stats shows chunk size distribution.
2717 extents_stats_per_process show chunk size distribution per
2720 Severity : enhancement
2722 Description: mds statistics
2723 Details : Add detailed mds operations statistics in
2724 /proc/fs/lustre/mds/*/stats.
2728 Description: Failure of copying files with lustre special EAs.
2729 Details : Client side always return success for setxattr call for lustre
2730 special xattr (currently only "trusted.lov").
2735 Description: Refcount LNET uuids
2736 Details : The global LNET uuid list grew linearly with every startup;
2737 refcount repeated list entries instead of always adding to
2740 Severity : enhancement
2742 Description: Dynamic service threads
2743 Details : Within a small range, start extra service threads
2744 automatically when the request queue builds up.
2747 Frequency : mixed-endian client/server environments
2749 Description: mixed-endian crashes
2750 Details : The new msg_v2 system had some failures in mixed-endian
2753 Severity : enhancement
2755 Description: Easy OST removal
2756 Details : OSTs can be permanently deactivated with e.g. 'lctl
2757 conf_param lustre-OST0001.osc.active=0'
2759 Severity : enhancement
2761 Description: MGS proc entries
2762 Details : Added basic proc entries for the MGS showing what filesystems
2765 Severity : enhancement
2767 Description: provide MGS failover
2768 Details : Added config lock reacquisition after MGS server failover.
2770 Severity : enhancement
2772 Description: add Linux 2.4 support
2773 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2777 Description: a large application tries to do I/O to the same resource and dies
2778 in the middle of it.
2779 Details : Check the req->rq_arrival time after the call to
2780 ost_brw_lock_get(), but before we do anything about
2781 processing it & sending the BULK transfer request. This
2782 should help move old stale pending locks off the queue as
2783 quickly as obd_timeout.
2786 Frequency : when an incorrect nid is specified during startup
2788 Description: ptlrpc connect to non-existant node causes kernel crash
2789 Details : LNET can't be re-entered from an event callback, which
2790 happened when we expire a message after the export has been
2791 cleaned up. Instead, hand the zombie cleanup off to another
2794 Severity : enhancement
2796 Description: plain/inodebits lock performance improvement
2797 Details : Grouping plain/inodebits in granted list by their request modes
2798 and bits policy, thus improving the performance of search through
2802 Frequency : only if OST filesystem is corrupted
2804 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2805 Details : for a short time RPCs with bulk IO are in the replay list,
2806 but replay of bulk IOs is unimplemented. If the OST filesystem
2807 is corrupted due to disk cache incoherency and then replay is
2808 started it is possible to trip an assertion. Avoid putting
2809 committed RPCs into the replay list at all to avoid this issue.
2812 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2814 Description: System hang on startup
2815 Details : This bug allowed the liblustre (e.g. catamount) client to
2816 return to the app before handling all startup RPCs. This
2817 could leave the node unresponsive to lustre network traffic
2818 and manifested as a server ptllnd timeout.
2820 Severity : enhancement
2822 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2823 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2824 boolean to control whether to print peer debug info when a
2825 client's RPC times out.
2828 Frequency : only for kernels with patches from Lustre below 1.4.3
2830 Description: Remove old rdonly API
2831 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2834 Frequency : only for devices with external journals
2836 Description: Set external device read-only also
2837 Details : During a commanded failover stop, we set the disk device
2838 read-only while the server shuts down. We now also set any
2839 external journal device read-only at the same time.
2842 Frequency : when upgrading from 1.4 while trying to change parameters
2844 Description: The wrong (new) MDC name was used when setting parameters for
2845 upgraded MDT's. Also allows changing of OSC (and MDC)
2846 parameters if --writeconf is specified at tunefs upgrade time.
2849 Frequency : when setting specific ost indicies
2851 Description: QOS code breaks on skipped indicies
2852 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2853 created with --index need not be sequential.
2855 Severity : enhancement
2857 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2858 Details : The uninit_groups feature works in conjunction with the kernel
2859 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2860 up the pass1 processing of e2fsck. This is a read-only feature
2861 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2862 mount filesystems that have had this feature enabled.
2864 Severity : enhancement
2866 Description: Improve multi-block allocation algorithm to avoid fragmentation
2867 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2868 allocation locality and avoid filesystem fragmentation.
2870 ------------------------------------------------------------------------------
2872 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2874 * Support for kernels:
2875 2.4.21-47.0.1.EL (RHEL 3)
2876 2.6.5-7.283 (SLES 9)
2877 2.6.9-42.0.10.EL (RHEL 4)
2878 2.6.12.6 vanilla (kernel.org)
2879 2.6.16.27-0.9 (SLES 10)
2880 * Recommended e2fsprogs version: 1.39.cfs5
2882 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2886 Frequency : occasional, depends on client load and configuration
2887 Bugzilla : 12181, 12203
2888 Description: data loss for recently-modified files
2890 Details : In some cases it is possible that recently written or created
2891 files may not be written to disk in a timely manner (this should
2892 normally be within 30s unless client IO load is very high).
2893 The problem appears as zero-length files or files that are a
2894 multiple of 1MB in size after a client crash or client eviction
2895 that are missing data at the end of the file.
2897 This problem is more likely to be hit on clients where files are
2898 repeatedly created and unlinked in the same directory, clients
2899 have a large amount of RAM, have many CPUs, the filesystem has
2900 many OSTs, the clients are rebooted frequently, and/or the files
2901 are not accessed by other nodes after being written.
2903 The presence of the problem can be detected by looking at
2904 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2905 smaller than the second (nr_unused) then dirty files will not
2906 be flushed automatically to disk. "sync; sleep 10" should be
2907 run several times on the node before unmounting it to update
2908 Lustre (this is also safe to run on nodes without this problem).
2910 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2911 that can cause this same problem, so customers using that kernel
2912 also need to update the kernel in addition to Lustre. In order
2913 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2915 It is normal that files written just before a client crash (less
2916 than 30s) may not yet have been flushed to disk, even for local
2920 Frequency : frequent on thin XT3 nodes
2922 Description: UUID collision on thin XT3 Linux nodes
2923 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2924 randomly, since we relied on an insufficiently-seeded PRNG.
2929 Description: OSS hangs after "All ost request buffers busy"
2930 Details : A deadlock between quota and journal operations caused OSS
2931 hangs after printing "All ost request buffers busy."
2934 Frequency : always on liblustre builds
2936 Description: Cleanup compiler warnings on liblustre
2939 Frequency : always on liblustre builds on XT3
2941 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2945 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2946 > 32000 subdirectories
2947 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2948 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2949 subdirectory count crosses 32000. This will aid e2fsck to
2950 correctly handle more than 32000 subdirectories.
2953 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2955 Description: System hang on startup
2956 Details : This bug allowed the liblustre (e.g. catamount) client to
2957 return to the app before handling all startup RPCs. This
2958 could leave the node unresponsive to lustre network traffic
2959 and manifested as a server ptllnd timeout.
2961 Severity : enhancement
2963 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2964 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2965 boolean to control whether to print peer debug info when a
2966 client's RPC times out.
2971 Description: make O_SYNC working on 2.6 kernels
2972 Details : 2.6 kernels use different method for mark pages for write,
2973 so need add a code to lustre for O_SYNC work.
2978 Description: Failure to close file and release space on NFS
2979 Details : Put inode details into lock acquired in ll_intent_file_open.
2980 Use mdc_intent_lock in ll_intent_open to properly
2981 detect all kind of errors unhandled by mdc_enqueue.
2986 Description: proc file read during shutdown sometimes raced obd removal,
2988 Details : Add lock to prevent obd access after proc file removal.
2991 Frequency : Only for files larger than 4GB on 32-bit clients.
2993 Description: improperly doing page alignment of locks
2994 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2995 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3000 Description: RPCs being resent when they shouldn't be
3001 Details : Some RPCs that should not be resent are being resent. This
3002 can cause inconsistencies in the RPC state machine. Do not
3003 resend such requests.
3006 Frequency : rare, only with NFS export
3008 Description: Crash on NFS re-export node
3009 Details : under very unusual load conditions an assertion is hit in
3010 ll_intent_file_open()
3013 Frequency : only if OST filesystem is corrupted
3015 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3016 Details : for a short time RPCs with bulk IO are in the replay list,
3017 but replay of bulk IOs is unimplemented. If the OST filesystem
3018 is corrupted due to disk cache incoherency and then replay is
3019 started it is possible to trip an assertion. Avoid putting
3020 committed RPCs into the replay list at all to avoid this issue.
3025 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3026 Details : Large single O_DIRECT read and write calls can fail to allocate
3027 a sufficiently large buffer to process the request. In case of
3028 allocation failure the allocation is retried with a smaller
3029 buffer and broken into smaller requests.
3031 Severity : enhancement
3033 Description: Add -o localflock option to simulate old noflock behaviour.
3034 Details : This will achieve local-only flock/fcntl locks coherentness.
3039 Description: versioning check is incomplete
3040 Details : Checking the version difference of client vs. server, report
3041 error if the gap is too big.
3046 Description: add support PG_writeback bit
3047 Details : add support for PG_writeback bit for Lustre, for more carefull
3048 work with page cache in 2.6 kernel. This also fix some deadlocks
3049 and remove hack for work O_SYNC with 2.6 kernel.
3051 Severity : enhancement
3053 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3054 Details : The uninit_groups feature works in conjunction with the kernel
3055 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3056 up the pass1 processing of e2fsck. This is a read-only feature
3057 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3058 mount filesystems that have had this feature enabled.
3060 Severity : enhancement
3062 Description: Improve multi-block allocation algorithm to avoid fragmentation
3063 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3064 allocation locality and avoid filesystem fragmentation.
3066 ------------------------------------------------------------------------------
3068 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3070 * Support for kernels:
3071 2.6.9-42.0.3.EL (RHEL 4)
3072 2.6.5-7.276 (SLES 9)
3073 2.4.21-47.0.1.EL (RHEL 3)
3074 2.6.12.6 vanilla (kernel.org)
3075 2.6.16.21-0.8 (SLES10)
3076 * Recommended e2fsprogs version: 1.39.cfs2-0
3078 * The backwards-compatible /proc/sys/portals symlink has been removed
3079 in this release. Before upgrading, please ensure that you change
3080 any configuration scripts or /etc/sysctl.conf files that access
3081 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3082 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3083 in advance of the upgrade on any system running Lustre 1.4.6 or
3084 newer, since /proc/sys/lnet was added in that version.
3085 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3089 Frequency : only when quota is used
3091 Description: avoid scanning export list for quota master
3092 Details : Change the algorithms to avoid scanning export list in order
3093 to improve the efficiency.
3096 Frequency : MDS failover only, very rarely
3098 Description: "went back in time" messages on mds failover
3099 Details : The greatest transno may be lost when the current operation
3100 finishes with an error (transno==0) and the client's last_rcvd
3101 record is over-written. Save the greatest transno in the
3102 mds_last_transno for this case.
3105 Frequency : always for specific kernels and striping counts
3107 Description: client may get "Matching packet too big" without ACL support
3108 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3109 when trying to access files in certain configurations. The
3110 clients should in fact be denied when mounting because they do
3111 not understand ACLs.
3114 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3116 Description: many clients connecting with IO in progress causes connect timeouts
3117 Details : Avoid synchronous journal commits to avoid delays caused by many
3118 clients connecting/disconnecting when bulk IO is in progress.
3119 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3120 OST_IO_PORTAL to avoid delays behind potentially many pending
3124 Frequency : occasionally with multiple writers to a single file
3126 Description: shared writes to file may result in wrong size reported by stat()
3127 Details : Allow growing of kms when extent lock is cancelled
3130 Frequency : always with random mmap IO to multi-striped file
3132 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3133 Details : Only if the hole is at the end of OST object so that kms is too
3134 small. Fix is to increase kms accordingly in ll_nopage.
3137 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3139 Description: writes to a missing object would leak memory on the OST
3140 Details : If there is an inconsistency between the MDS and OST filesystems,
3141 such that the MDS references an object that doesn't exist, writes
3142 to that object will leak memory due to incorrect cleanup in the
3143 error handling path, eventually running out of memory on the OST.
3148 Description: Creating too long symlink causes lustre errors
3149 Details : Check symlink and name lengths before sending requests to MDS.
3152 Frequency : only if flock is enabled (not on by default)
3154 Description: posix locks not released on fd closure on 2.6.9+
3155 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3156 kernels, this caused such locks not to be released on fd close and
3157 then assertions on fs unmount about still used locks.
3160 Frequency : MDS failover only, very rarely
3162 Description: clients may get ASSERTION(granted_lock != NULL)
3163 Details : When request was taking a long time, and a client was resending
3164 a getattr by name lock request. The were multiple lock requests
3165 with the same client lock handle and
3166 mds_getattr_name->fixup_handle_for_resent_request found one of the
3167 lock handles but later failed with ASSERTION(granted_lock != NULL).
3172 Description: handle->h_buffer_credits > 0, assertion failure
3173 Details : h_buffer_credits is zero after truncate, causing assertion
3174 failure. This patch extends the transaction or creates a new
3178 Frequency : NFS re-export or patchless client
3179 Bugzilla : 11179, 10796
3180 Description: Crash on NFS re-export node (__d_move)
3181 Details : We do not want to hash the dentry if we don't have a lock.
3182 But if this dentry is later used in d_move, we'd hit uninitialised
3183 list head d_hash, so we just do this to init d_hash field but
3184 leave dentry unhashed.
3187 Frequency : NFS re-export or patchless client
3189 Description: NFS exports has problem with symbolic link
3190 Details : lustre client didn't properly install dentry when re-exported
3191 to NFS or running patchless client.
3194 Frequency : NFS re-export or patchless client
3196 Description: Various nfs/patchless fixes.
3197 Details : fixes reuse disconected alias for lookup process - this fixes
3198 warning "find_exported_dentry: npd != pd",
3199 fix permission error with open files at nfs.
3200 fix apply umask when do revalidate.
3203 Frequency : occasional
3205 Description: Crash on NFS re-export node
3206 Details : calling clear_page() on the wrong pointer triggered oops in
3207 generic_mapping_read().
3210 Frequency : rarely, using O_DIRECT IO
3212 Description: unaligned directio crashes client with LASSERT
3213 Details : check for unaligned buffers before trying any requests.
3216 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3218 Description: stale data returned from RAID cache
3219 Details : If only a small amount of IO is done to the RAID device before
3220 reading it again it is possible to get stale data from the RAID
3221 cache instead of reading it from disk.
3224 Frequency : always for sles10 kernel
3226 Description: sles10 support
3227 Details : ll_follow_link: compile fixes and using of nd_set_link
3228 under newer kernels.
3231 Frequency : depends on arch, kernel and compiler version, always on sles10
3234 Description: recursive or deep enough symlinks cause stack overflow
3235 Details : getting rid of large stack-allocated variable in
3239 Frequency : depends on hardware
3241 Description: lustre write performance loss in the SLES10 kernel
3242 Details : the performance loss is caused by using of write barriers in the
3243 ext3 code. The SLES10 kernel turns barrier support on by
3244 default. The fix is to undo that change for ldiskfs.
3246 ------------------------------------------------------------------------------
3248 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3250 * Support for kernels:
3251 2.6.9-42.0.3EL (RHEL 4)
3252 2.6.5-7.276 (SLES 9)
3253 2.4.21-47.0.1.EL (RHEL 3)
3254 2.6.12.6 vanilla (kernel.org)
3258 Frequency : quota enabled and large files being deleted
3260 Description: releasing more than 4GB of quota at once hangs OST
3261 Details : If a user deletes more than 4GB of files on a single OST it
3262 will cause the OST to spin in an infinite loop. Release
3263 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3268 Description: statfs data retrieved from /proc may be stale or zero
3269 Details : When reading per-device statfs data from /proc, in the
3270 {kbytes,files}_{total,free,avail} files, it may appear
3271 as zero or be out of date.
3274 Frequency : systems with MD RAID1 external journal devices
3276 Description: lconf's call to blkid is confused by RAID1 journal devices
3277 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3278 of returning all block devices that match the journal UUID.
3281 Frequency : always, for aggregate stripe size over 4GB
3283 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3284 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3285 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3288 Frequency : always if "lfs find" used on a local file/directory
3290 Description: "lfs find" segfaults if used on a local file/directory
3291 Details : The case where a directory component was not specified wasn't
3292 handled correctly. Handle this properly.
3295 Frequency : always on ppc64
3297 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3298 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3299 perform endianness conversion.
3302 Frequency : rarely (truncate to non-zero file size after write under load)
3303 Bugzilla : 10730, 10687
3304 Description: Files padded with zeros to next 4K multiple
3305 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3306 it is possible that files that are truncated to a non-zero size
3307 immediately after being written are filled with zero bytes beyond
3308 the truncated size. No file data is lost.
3310 Severity : enhancement
3312 Description: Allow recovery/failover for liblustre clients.
3313 Details : liblustre clients were unaware of failover configurations until
3316 Severity : enhancement
3318 Description: user file locks should fail when not mounting with flock option
3319 Details : Set up an error-returning stub in ll_file_operations.lock field
3320 to prevent incorrect behaviour when client is mounted without
3321 flock option. Also, set up properly f_op->flock field for
3325 Frequency : always on ia64
3327 Description: "lfs df" loops on printing out MDS statfs information
3328 Details : The obd_ioctl_data was not initialized and in some systems
3329 this caused a failure during the ioctl that did not return
3330 an error. Initialize the struct and return an error on failure.
3333 Frequency : SLES 9 only
3335 Description: Error of copying files with lustre special EAs as root
3336 Details : Client side always return success for setxattr call for lustre
3337 special xattr (currently only "trusted.lov").
3340 Frequency : rarely on clusters with both ia64+i386 clients
3342 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3343 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3344 clients are doing concurrent writes to the same file it is
3345 possible that the ia64 clients may LASSERT because the OST
3346 extent locks are not PAGE_SIZE aligned. Ensure that grown
3347 locks are always aligned on the request boundary.
3350 Frequency : specific use, occasional
3352 Description: Overwriting in use executable truncates on-disk binary image
3353 Details : If one node attempts to overwrite an executable in use by
3354 another node, we now correctly return ETXTBSY instead of
3355 truncating the file.
3357 Severity : enhancement
3359 Description: Async OSC create to avoid the blocking unnecessarily.
3360 Details : If a OST has no remain object, system will block on the creating
3361 when need to create a new object on this OST. Now, ways use
3362 pre-created objects when available, instead of blocking on an
3363 empty osc while others are not empty. If we must block, we block
3364 for the shortest possible period of time.
3369 Description: chmod on Lustre root is propagated to other clients
3370 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3371 invalid by the follow_mount time.
3374 Frequency : liblustre clients only
3376 Description: Race in 'instant cancel' lock handling could lead to such locks
3377 never to be granted in case of SMP MDS
3378 Details : Do not destroy not yet granted but cbpending locks in
3382 Frequency : replay/resend of open
3384 Description: non null lock assetion failure in mds_intent_policy
3385 Details : Trying to replay/resend lockless open requests resulted in
3386 mds_open() returning 0 with no lock. Now it sets a flag if
3387 a lock is going to be returned.
3389 Severity : enhancement
3391 Description: Checksum enhancements
3392 Details : New checksum enhancements allow for resending RPCs that failed
3395 Severity : enhancement
3397 Description: Tunables on number of dirty pages in cacche
3398 Details : Allow to set limit on number of dirty pages cached.
3403 Description: client crash on unmount - lock still has references
3404 Details : In some error handling cases it was possible to leak a lock
3405 reference on a client while accessing a file. This was not
3406 harmful to the client during operation, but would cause the
3407 client to crash when the filesystem is unmounted.
3410 Frequency : specific case, rare
3412 Description: ETXTBSY on mds though file not in use by client
3413 Details : ETXTBSY is no longer incorrectly returned when attempting to
3414 chmod or chown a directory that the user previously tried to
3415 execute or a currently-executing binary.
3418 Frequency : extremely rare except on liblustre-based clients
3420 Description: Lustre space not freed when files are deleted
3421 Details : Clean up open-unlinked files after client eviction. Previously
3422 the unlink was skipped and the files remained as orphans.
3427 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3428 Details : In some cases it was possible to send a blocking callback to
3429 a client doing a glimpse, even though that client didn't get
3430 a lock granted. When the glimpse lock is cancelled on the OST
3431 the freed lock is left on the waiting list and corrupted the list.
3434 Frequency : all core dumps
3436 Description: Broke core dumps to lustre
3437 Details : Negative dentry may be unhashed if parent does not have UPDATE
3438 lock, but some callers, e.g. do_coredump, expect dentry to be
3439 hashed after successful create, hash it in ll_create_it.
3441 ------------------------------------------------------------------------------
3443 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3445 * Support for kernels:
3446 2.6.9-42.0.2.EL (RHEL 4)
3447 2.6.5-7.276 (SLES 9)
3448 2.4.21-40.EL (RHEL 3)
3449 2.6.12.6 vanilla (kernel.org)
3453 Frequency : always on RHEL 3
3455 Description: Number of open files grows over time
3456 Details : The number of open files grows over time, whether or not
3457 Lustre is started. This was due to a filp leak introduced
3458 by one of our kernel patches.
3460 ------------------------------------------------------------------------------
3462 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3464 * Support for kernels:
3465 2.6.9-42.EL (RHEL 4)
3466 2.6.5-7.276 (SLES 9)
3467 2.4.21-40.EL (RHEL 3)
3468 2.6.12.6 vanilla (kernel.org)
3473 Bugzilla : 5719, 9635, 9792, 9684
3474 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3475 Details : If a server is under heavy load and cannot reply to new
3476 connection requests before the client resends the (re)connect,
3477 the connection handling code can behave badly if two service
3478 threads are concurrently handing separate (re)connections from
3479 the same client. Add better locking to the connection handling
3480 code, and ensure that only a single connection will be processed
3481 for a given client UUID, even if the lock is dropped.
3483 Severity : enhancement
3485 Description: add TCP zero-copy support to kernel
3486 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3487 sends if the hardware supports scatter-gather and checksumming.
3488 This allows socklnd to do client-write and server-read more
3489 efficiently and reduce CPU utilization from skbuf copying.
3492 Frequency : only if NFS exporting from client
3494 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3495 Details : If setting attributes on a file created under NFS that had
3496 never been opened it would be possible to oops the client
3497 if the file had no objects.
3500 Frequency : always for liblustre
3502 Description: liblustre client does MDS+OSTs setattr RPC for each write
3503 Details : When doing a write from a liblustre client, the client
3504 incorrectly issued an RPC to the MDS and each OST the file was
3505 striped over in order to update the timestamps. When writing
3506 with small chunks and many clients this could overwhelm the MDS
3507 with RPCs. In all cases it would slow down the write because
3508 these RPCs are unnecessary.
3510 Severity : enhancement
3512 Description: allow number of MDS service threads to be changed at module load
3513 Details : It is now possible to change the number of MDS service threads
3514 running. Adding "options mds mds_num_threads={N}" to the MDS's
3515 /etc/modprobe.conf will set the number of threads for the next
3516 time Lustre is restarted (assuming the "mds" module is also
3517 reloaded at that time). The default number of threads will
3518 stay the same, 32 for most systems.
3523 Description: OST crash if filesystem is unformatted or corrupt
3524 Details : If an OST is started on a device that has never been formatted
3525 or if the filesystem is corrupt and cannot even mount then the
3526 error handling cleanup routines would dereference a NULL pointer.
3531 Description: NULL pointer deref in llap_from_page.
3532 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3533 incorrect) mapping if the page was truncated/reclaimed while it was
3534 searched for. Check for this condition and skip such pages when
3535 doing readahead. Introduce extra check to llap_from_page() to
3536 verify page->mapping->host is non-NULL (so page is not anonymous).
3539 Frequency : Sometimes when using sys_sendfile
3541 Description: "page not covered by a lock" warnings from ll_readpage
3542 Details : sendfile called ll_readpage without right page locks present.
3543 Now we introduced ll_file_sendfile that does necessary locking
3544 around call to generic_file_sendfile() much like we do in
3548 Frequency : with certain MDS communication failures at client mount time
3550 Description: NULL pointer deref after failed client mount
3551 Details : a client connection request may delayed by the network layer
3552 and not be sent until after the PTLRPC layer has timed out the
3553 request. If the client fails the mount immediately it will try
3554 to clean up before the network times out the request. Add a
3555 reference from the request import to the obd device and delay
3556 the cleanup until the network drops the request.
3559 Frequency : occasionally during client (re)connect
3561 Description: assertion failure during client (re)connect
3562 Details : processing a client connection request may be delayed by the
3563 client or server longer than the client connect timeout. This
3564 causes the client to resend the connection request. If the
3565 original connection request is replied in this interval, the
3566 client may trip an assertion failure in ptlrpc_connect_interpret()
3567 which thought it would be the only running connect process.
3570 Frequency : only with obd_echo servers and clients that are rebooted
3572 Description: kernel BUG accessing uninitialized data structure
3573 Details : When running an obd_echo server it did not start the ping_evictor
3574 thread, and when a client was evicted an uninitialized data
3575 structure was accessed. Start the ping_evictor in the RPC
3576 service startup instead of the OBD startup.
3578 Severity : enhancement
3579 Bugzilla : 10193 (patchless)
3580 Description: Remove dependency on various unexported kernel interfaces.
3581 Details : No longer need reparent_to_init, exit_mm, exit_files,
3582 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3585 Frequency : rare (only users of deprecated and unsupported LDAP config)
3587 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3588 Details : LDAP apparently contains 'lustreName' attributes instead of
3589 'name'. A simple remapping of the name is sufficient.
3592 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3594 Description: waiting_locks_callback trips kernel BUG if client is evicted
3595 Details : Running with the dump_on_timeout debug flag turned on makes
3596 it possible that the waiting_locks_callback() can try to dump
3597 the Lustre kernel debug logs from an interrupt handler. Defer
3598 this log dumping to the expired_lock_main() thread.
3600 Severity : enhancement
3602 Description: Support NFS exporting on 2.6 kernels.
3603 Details : Implement non-rawops metadata methods for NFS server to use without
3604 changing NFS server code.
3607 Frequency : very rare (synthetic metadata workload only)
3609 Description: two racing renames might cause an MDS thread to deadlock
3610 Details : Running the "racer" program may cause one MDS thread to rename
3611 a file from being the source of a rename to being the target of
3612 a rename at exactly the same time that another thread is doing
3613 so, and the second thread has already enqueued these locks after
3614 doing a lookup of the target and is trying to relock them in
3615 order. Ensure that we don't try to re-lock the same resource.
3618 Frequency : only very large systems with liblustre clients
3620 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3621 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3622 all OSTs in parallel. This allows the eviction of stale liblustre
3623 clients to proceed much faster than if they were done in series,
3624 and also offers similar improvements for other set_info RPCs.
3629 Description: excessive CPU usage during initial read phase on client
3630 Details : During the initial read phase on a client, it would agressively
3631 retry readahead on the file, consuming too much CPU and impacting
3632 performance (since 1.4.5.8). Improve the readahead algorithm
3633 to avoid this, and also improve some other common cases (read
3634 of small files in particular, where "small" is files smaller than
3635 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3640 Description: MDS crash when receiving packet with unknown intent.
3641 Details : Do not LBUG in unknown intent case, just return -EFAULT
3643 Severity : enhancement
3644 Bugzilla : 9293, 9385
3645 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3646 Details : Do not serialize getattr (non-intent version) and statfs.
3649 Frequency : occasional, when OST network is overloaded/intermittent
3651 Description: client evicted by OST after bulk IO timeout
3652 Details : If a client sends a bulk IO request (read or write) the OST
3653 may evict the client if it is unresposive to its data GET/PUT
3654 request. This is incorrect if the network is overloaded (takes
3655 too long to transfer the RPC data) or dropped the OST GET/PUT
3656 request. There is no need to evict the client at all, since
3657 the pinger and/or lock callbacks will handle this, and the
3658 client can restart the bulk request.
3661 Frequency : Always when mmapping file with no objects
3663 Description: client crashes when mmapping file with no objects
3664 Details : Check that we actually have objects in a file before doing any
3665 operations on objects in ll_vm_open, ll_vm_close and
3671 Description: Request leak when working with deleted CWD
3672 Details : Introduce advanced request refcount tracking for requests
3673 referenced from lustre intent.
3675 Severity : Enhancement
3677 Description: Cache open file handles on client.
3678 Details : MDS now will return special lock along with openhandle, if
3679 requested and client is allowed to hold openhandle, even if unused,
3680 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3681 closing files for every read/write openration.
3683 Severity : Enhancement
3685 Description: Cache open negative dentries on client when possible.
3686 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3687 negative dentries on lock revocation.
3692 Description: Remounting a client read-only wasn't possible with a zconf mount
3693 Details : It wasn't possible to remount a client read-only with llmount.
3695 Severity : enhancement
3696 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3697 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3698 MPICH is included to add Lustre-specific ADIO interfaces.
3699 This is based closely on the UFS ADIO layer and only differs
3700 in file creation, in order to allow the OST striping to be set.
3701 This is user-contributed code and not supported by CFS.
3706 Description: extended inode attributes (immutable, append-only) work improperly
3707 when 2.4 and 2.6 kernels are used on client/server or vice versa
3708 Details : Introduce kernel-independent values for these flags.
3710 Severity : enhancement
3713 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3714 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3715 granularity. Now, Fractional MB granularity is be supported,
3716 this is very useful for low memory system.
3718 Severity : enhancement
3720 Description: Getattr by fid
3721 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3722 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3723 have locks and avoids getattr rpc requests for it on every stat.
3726 Frequency : Always, for filesystems larger than 2TB
3728 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3729 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3730 filesystems larger than 2TB to be mounted with mballoc enabled.
3733 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3735 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3736 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3737 filesystems larger than 2TB is mounted then the kernel will
3738 silently corrupt the start of the filesystem. CONFIG_LBD is
3739 enabled for all CFS-supported kernels, but the possibility of
3740 this happening with a modified kernel config exists.
3742 Severity : enhancement
3744 Description: add client O_DIRECT support for 2.6 kernels
3745 Details : It is now possible to do O_DIRECT reads and writes to files
3746 in the Lustre client mountpoint on 2.6 kernel clients.
3748 Severity : enhancement
3750 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3751 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3752 parallel, not waiting for response from every OST before sending
3753 a rpc to the next OST.
3758 Description: setattr vs write race when updating file timestamps
3759 Details : Client processes that update a file timestamp into the past
3760 right after writing to the file (e.g. tar) it is possible that
3761 the updated file modification time can be reset to the current
3762 time due to a race between processing the setattr and write RPC.
3764 Severity : enhancement
3766 Description: Bring 'lfs find' closer in line with regular Linux find.
3767 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3768 -print0 options and obtains all the needed info through the lustre
3771 Severity : enhancement
3773 Description: support up to 1024 configured devices on one node
3774 Details : change obd_dev array from statically allocated to dynamically
3775 allocated structs as they are first used to reduce memory usage
3780 Description: Flush dirty partially truncated pages during truncate
3781 Details : Immediatelly flush partially truncated pages in filter_setattr,
3782 this way we completely avoid having any pages in page cache on OST
3783 and can retire ugly workarounds during writes to flush such pages.
3788 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3789 Details : protect inode from truncation within vfs_unlink() context
3790 just take a reference before calling vfs_unlink() and release it
3791 when parent's i_sem is free.
3796 Description: last_id value checked outside lock on OST caused LASSERT failure
3797 Details : If there were multiple MDS->OST object precreate requests in
3798 flight, it was possible that the OST's last object id was checked
3799 outside a lock and incorrectly tripped an assertion. Move checks
3800 inside locks, and discard old precreate requests.
3803 Frequency : always, if extents are used on OSTs
3805 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3806 Details : OSTs using the extents format would not zero the high 16 bits of
3807 the index physical block number. This is not a problem for any
3808 OST filesystems smaller than 16TB, and no kernels support ext3
3809 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3810 new/modified files) and can be fixed for existing filesystems
3811 with e2fsprogs-1.39-cfs1.
3816 Description: import connection selection may be incorrect if timer wraps
3817 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3818 import connections to be ignored if the 32-bit jiffies counter
3819 wraps. Use a 64-bit jiffies counter.
3822 Frequency : very large clusters immediately after boot
3824 Description: LNET request buffers exhausted under heavy short-term load
3825 Details : If a large number of client requests are generated on a service
3826 that has previously never seen so many requests it is possible
3827 that the request buffer growth cannot keep up with the spike in
3828 demand. Instead of dropping incoming requests, they are held in
3829 the LND until the RPC service can accept more requests.
3832 Frequency : Sometimes during replay
3834 Description: Assertion failure in ll_local_open after replay.
3835 Details : If replay happened on an open request reply before we were able
3836 to set replay handler, reply will become not swabbed tripping the
3837 assertion in ll_local_open. Now we set the handler right after
3838 recognising of open request
3841 Frequency : very rare
3843 Description: kernel reports "badness in vsnprintf"
3844 Details : Reading from the "recovery_status" /proc file in small chunks
3845 may cause a negative length in lprocfs_obd_rd_recovery_status()
3846 call to vsnprintf() (which is otherwise harmless). Exit early
3847 if there is no more space in the output buffer.
3849 Severity : enhancement
3851 Description: clear OBD RPC statistics by writing to them
3852 Details : It is now possible to clear the OBD RPC statistics by writing
3853 to the "stats" file.
3858 Description: Client mtime is not the same on different clients after utimes
3859 Details : In some cases, the client was using the utimes() syscall on
3860 a file cached on another node. The clients now validate the
3861 ctime from the MDS + OSTs to determine which one is right.
3866 Description: Inability to activate failout mode
3867 Details : lconf script incorrectly assumed that in pythong string's numeric
3868 value is used in comparisons.
3871 Frequency : always with multiple stripes per file
3873 Description: Inefficient object allocation for mutli-stripe files
3874 Details : When selecting which OSTs to stripe files over, for files with
3875 a stripe count that divides evenly into the number of OSTs,
3876 the MDS is always picking the same starting OST for each file.
3877 Return the OST selection heuristic to the original design.
3882 Description: mount failures may take full timeout to return an error
3883 Details : Under some heavy load conditions it is possible that a
3884 failed mount can wait for the full obd_timeout interval,
3885 possibly several minutes, before reporting an error.
3886 Instead return an error as soon as the status is known.
3888 Frequency : quota enabled and large files being deleted
3890 Description: releasing more than 4GB of quota at once hangs OST
3891 Details : If a user deletes more than 4GB of files on a single OST it
3892 will cause the OST to spin in an infinite loop. Release
3893 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3898 Description: statfs data retrieved from /proc may be stale or zero
3899 Details : When reading per-device statfs data from /proc, in the
3900 {kbytes,files}_{total,free,avail} files, it may appear
3901 as zero or be out of date.
3904 Frequency : systems with MD RAID1 external journal devices
3906 Description: lconf's call to blkid is confused by RAID1 journal devices
3907 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3908 of returning all block devices that match the journal UUID.
3911 Frequency : always, for aggregate stripe size over 4GB
3913 Description: assertion fails when trying to use 4GB stripe size
3914 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3915 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3918 Frequency : always on ppc64
3920 Description: the first write on an ext3 filesystem with mballoc got stuck
3921 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3922 perform endianness conversion.
3924 ------------------------------------------------------------------------------
3926 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3928 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3929 INTEROPERATE with older versions automatically. Please read the
3930 user documentation before upgrading any part of a live system.
3931 * WARNING: Lustre networking configuration changes are required with
3932 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3935 * Support for kernels:
3936 2.6.9-22.0.2.EL (RHEL 4)
3937 2.6.5-7.244 (SLES 9)
3938 2.6.12.6 vanilla (kernel.org)
3941 Severity : enhancement
3942 Bugzilla : 7981/8208
3943 Description: Introduced Lustre Networking (LNET)
3944 Details : LNET is new networking infrastructure for Lustre, it includes
3945 a reorganized network configuration mode (see the user
3946 documentation for full details) as well as support for routing
3947 between different network fabrics. Lustre Networking Devices
3948 (LNDs) for the supported network fabrics have also been
3949 created for this new infrastructure.
3951 Severity : enhancement
3952 Description: Introduced Access control lists
3953 Details : clients can set ACLs on files and directories in order to have
3954 more fine-grained permissions than the standard Unix UGO+RWX.
3955 The MDS must be started with the "-o acl" mount option.
3957 Severity : enhancement
3958 Description: Introduced filesystem quotas
3959 Details : Administrators may now establish per-user quotas on the
3962 Severity : enhancement
3964 Description: Configuration change for the XT3
3965 The PTLLND is now used to run Lustre over Portals on the XT3
3966 The configure option(s) --with-cray-portals are no longer used.
3967 Rather --with-portals=<path-to-portals-includes> is used to
3968 enable building on the XT3. In addition to enable XT3 specific
3969 features the option --enable-cray-xt3 must be used.
3974 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3975 Details : A race between allocating a new llcd and re-getting the llcd_lock
3976 allowed another thread to grab newly-allocated llcd.
3978 Severity : enhancement
3980 Description: 2.6 OST async journal commit and locking fix to improve performance
3981 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3982 2.6 kernels are now async as they already were in 2.4 kernels so
3983 that they can commit concurrently with the network bulk transfer.
3984 For block-allocated files the filter allocation semaphore is held
3985 to avoid filesystem fragmentation during allocation. BKL lock
3986 removed for 2.6 xattr operations where it is no longer needed.
3991 Description: lconf incorrectly determined whether two IP networks could talk
3992 Details : In some more complicated routing and multiple-network
3993 configurations, lconf will avoid trying to make a network
3994 connection to a disjoint part of the IP space. It was doing the
3995 math incorrectly for one set of cases.
4000 Description: Fix for potential infinite loop processing records in an llog.
4001 Details : If an llog record is corrupted/zeroed, it is possible to loop
4002 forever in llog_process(). Validate the llog record length
4003 and skip the remainder of the block on error.
4006 Frequency : occasional (liblustre only)
4008 Description: liblustre could not open files whose last component is a symlink
4009 Details : sysio_path_walk() would incorrectly pass the open intent to
4010 intermediate path components.
4013 Frequency : rare (liblustre only with non-standard tuning)
4014 Bugzilla : 7201 (7350)
4015 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4016 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4017 after it has been released, only do this for referenced locks
4019 Severity : enhancement
4021 Description: specifying an (invalid) directory default stripe_size of -1
4022 would reset the directory default striping
4023 Details : stripe_size -1 was used internally to signal directory stripe
4024 removal, now use "all default" to signal dir stripe removal
4025 as a directory striping of "all default" is not useful
4028 Frequency : common for large clusters running liblustre clients
4030 Description: doing an ls when liblustre clients are running is slow
4031 Details : sending a glimpse AST to a liblustre client waits for every AST
4032 to time out, as liblustre clients will not respond. Since they
4033 cannot cache data we refresh the OST lock LVB from disk instead.
4035 Severity : enhancement
4037 Description: doing an ls at the same time as file IO can be slow
4038 Details : enqueue and other "small" requests can be blocked behind many
4039 large IO requests. Create a new OST IO portal for non-IO
4040 requests so they can be processed faster.
4043 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4045 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4046 created file with mode 000
4047 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4048 allow an ftruncate() as a non-root user to a file with mode 000.
4049 Lustre now respects this flag to disable mode checks when
4050 truncating a file owned by the user
4053 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4056 Description: Revoking locks from clients that went dead or catatonic might take
4058 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4059 cancellation of such locks instant on servers without waiting for
4060 any reply from clients. Clients drops these locks when cancel
4061 notification from server is received without replying.
4064 Frequency : liblustre-only, when liblustre client dies or becomes busy
4066 Description: Doing ls on Linux clients can take a long time with active
4068 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4069 granting such locks to it in the first place if possible. Locks
4070 are taken by proxy on the OST during the read or write and
4071 dropped immediately afterward. Add connect flags handling, do
4072 not grant locks to liblustre clients for glimpse ASTs.
4074 Severity : enhancement
4076 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4077 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4078 sequential reads and behaves badly with applications doing random
4079 reads. Improve it by reading ahead at least read region, and
4080 avoiding excessive large RPC for small reads.
4082 Severity : enhancement
4084 Description: Creating more than 1000 files for a single job may cause a load
4085 imbalance on the OSTs if there are also a large number of OSTs.
4086 Details : qos_prep_create() uses an OST index reseed value that is an
4087 even multiple of the number of available OSTs so that if the
4088 reseed happens in the middle of the object allocation it will
4089 still utilize the OSTs as uniformly as possible.
4094 Description: OST or MDS may oops in ping_evictor_main()
4095 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4096 but doesn't restart at beginning of obd_exports_timed list
4099 Severity : enhancement
4101 Description: improve by-nid export eviction on the MDS and OST
4102 Details : allow multiple exports with the same NID to be evicted at one
4103 time without re-searching the exports list.
4106 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4108 Description: MDS may oops in groups_free()
4109 Details : in rare race conditions a newly allocated group_info struct is
4110 freed again, and this can be NULL. The 2.4 compatibility code
4111 for groups_free() checked for a NULL pointer, but 2.6 did not.
4114 Frequency : common for liblustre clients doing little filesystem IO
4115 Bugzilla : 9352, 7313
4116 Description: server may evict liblustre clients accessing contended locks
4117 Details : if a client is granted a lock or receives a completion AST
4118 with a blocking AST already set it would not reply to the AST
4119 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4122 Frequency : lfs setstripe, only systems with more than 160 OSTs
4124 Description: unable to set striping with a starting offset beyond OST 160
4125 Details : llapi_create_file() incorrectly limited the starting stripe
4126 index to the maximum single-file stripe count.
4129 Frequency : LDAP users only
4131 Description: lconf did not handle in-kernel recovery with LDAP properly
4132 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4134 Severity : enhancement
4136 Description: bind OST threads to NUMA nodes to improve performance
4137 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4138 node and do their allocations there to localize memory access
4140 Severity : enhancement
4142 Description: llmount can determine client NID directly from Myrinet (GM)
4143 Details : the client NID code from gmnalnid was moved directly into
4144 llmount, removing the need to use this or specifying the
4145 client NID explicitly when mounting GM clients with zeroconf
4148 Frequency : if client is started with down MDS
4150 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4151 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4152 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4153 ptlrpc_import_delayed_req() because we didn't check if the
4154 request was interrupted, and we also didn't break out of the
4155 event loop if there was no timeout
4160 Description: data loss during non-page-aligned writes to a single file from
4161 both multiple nodes and multiple threads on one node at same time
4162 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4163 inconsistency led to false short-reads, that were cached and later
4164 used by ->prepare_write() to fill in partially written page,
4165 leading to data loss.
4168 Frequency : always, if lconf --abort_recovery used
4170 Description: lconf --abort_recovery fails with 'Operation not supported'
4171 Details : lconf was attempting to abort recovery on the MDT device and not
4174 Severity : enhancement
4176 Description: remove cleanup logs
4177 Details : replace lconf-generated cleanup logs with lustre internal
4178 cleanup routines. Eliminates the need for client-cleanup and
4181 Severity : enhancement
4183 Description: add support for EAs (user and system) on lustre filesystems
4184 Details : it is now possible to store extended attributes in the Lustre
4185 client filesystem, and with the user_xattr mount option it
4186 is possible to allow users to store EAs on their files also
4188 Severity : enhancement
4190 Description: Add possibility (config option) to show minimal available OST free
4192 Details : When compiled with --enable-mindf configure option, statfs(2)
4193 (and so, df) will return least minimal free space available from
4194 all OSTs as amount of free space on FS, instead of summary of
4195 free spaces of all OSTs.
4197 Severity : enhancement
4199 Description: do not expand extent locks acquired on OST-side
4200 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4201 by server: they are not cached anyway.
4204 Frequency : when mmap is used/binaries executed from Lustre
4206 Description: Unmmap pages before throwing them away from read cache.
4207 Details : llap_shrink cache now attempts to unmap pages before discarding
4208 them (if unmapping failed - do not discard). SLES9 kernel has
4209 extra checks that trigger if this unmapping is not done first.
4214 Description: lconf didn't resolve symlinks before checking to see whether a
4215 given mountpoint was already in use
4218 Frequency : when migrating failover services
4219 Bugzilla : 6395, 9514
4220 Description: When migrating a subset of services from a node (e.g. failback
4221 from a failover service node) the remaining services would
4222 time out and evict clients.
4223 Details : lconf --force (implied by --failover) sets the global obd_timeout
4224 to 5 seconds in order to quickly disconnect, but this caused
4225 other RPCs to time out too quickly. Do not change the global
4226 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4228 Severity : enhancement
4229 Frequency : if MDS is started with down OST
4230 Bugzilla : 9439,5706
4231 Description: Allow startup/shutdown of an MDS without depending on the
4232 availability of the OSTs.
4233 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4234 Add appropriate locking and lov-osc refcounts for safe
4235 cleaning. Add osc abort_inflight calls in case the
4236 synchronize never started.
4239 Frequency : occasional (Cray XT3 only)
4241 Description: root not authorized to access files in CRAY_PORTALS environment
4242 Details : The client process capabilities were not honoured on the MDS in
4243 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4244 been accessed by an authorized user then root was able to access
4245 the file on the local client also. The root user capabilities
4246 are now allowed on the MDS, as this environment has secure UID.
4249 Frequency : occasional
4251 Description: ldiskfs "too long searching" message happens too often
4252 Details : A debugging message (otherwise harmless) prints too often on
4253 the OST console. This has been reduced to only happen when
4254 there are fragmentation problems on the filesystem.
4259 Description: Division by zero in statfs when all OSCs are inactive
4260 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4261 lov_statfs divides by value returned by lov_get_stripecnt().
4265 Bugzilla : 9489, 3273
4266 Description: First write from each client to each OST was only 4kB in size,
4267 to initialize client writeback cache, which caused sub-optimal
4268 RPCs and poor layout on disk for the first writen file.
4269 Details : Clients now request an initial cache grant at (re)connect time
4270 and so that they can start streaming writes to the cache right
4271 away and always do full-sized RPCs if there is enough data.
4272 If the OST is rebooted the client also re-establishes its grant
4273 so that client cached writes will be honoured under the grant.
4278 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4279 Details : Now I/O RPCs go to different portal number and (presumably) fast
4280 lock requests (and glimses) and other RPCs get their own service
4281 threads pool that should be able to service those RPCs
4284 Severity : enhancement
4286 Description: Ability to exchange lustre version between client and servers and
4287 issue warnings at client side if client is too old. Also for
4288 liblustre clients there is ability to refuse connection of too old
4290 Details : New 'version' field is added to connect data structure that is
4291 filled with version info. That info is later checked by server and
4295 Frequency : rare, liblustre only.
4296 Bugzilla : 9296, 9581
4297 Description: Two simultaneous writes from liblustre at offset within same page
4298 might proceed at the same time overwriting eachother with stale
4300 Details : I/O lock withing llu_file_prwv was released too early, before data
4301 actually was hitting the wire. Extended lock-holding time until
4302 server acknowledges receiving data.
4305 Frequency : extremely rare. Never observed in practice.
4307 Description: avoid generating lustre_handle cookie of 0.
4308 Details : class_handle_hash() generates handle cookies by incrementing
4309 global counter, and can hit 0 occasionaly (this is unlikely, but
4310 not impossible, because initial value of cookie counter is
4311 selected randonly). Value of 0 is used as a sentinel meaning
4312 "unassigned handle" --- avoid it. Also coalesce two critical
4313 sections in this function into one.
4315 Severity : enhancement
4317 Description: allow liblustre clients to delegate truncate locking to OST
4318 Details : To avoid overhead of locking, liblustre client instructs OST to
4319 take extent lock in ost_punch() on client's behalf. New connection
4320 flag is added to handle backward compatibility.
4322 Severity : enhancement
4323 Bugzilla : 4928, 7341, 9758
4324 Description: allow number of OST service threads to be specified
4325 Details : a module parameter allows the number of OST service threads
4326 to be specified via "options ost ost_num_threads={N}" in the
4327 OSS's /etc/modules.conf or /etc/modprobe.conf.
4331 Bugzilla : 6146, 9635, 9895
4332 Description: servers crash with bad pointer in target_handle_connect()
4333 Details : In rare cases when a client is reconnecting it was possible that
4334 the connection request was the last reference for that export.
4335 We would temporarily drop the export reference and get a new
4336 one, but this may have been the last reference and the export
4337 was just destroyed. Get new reference before dropping old one.
4339 Severity : enhancement
4340 Frequency : if client is started with failover MDS
4342 Description: Allow multiple MDS hostnames in the mount command
4343 Details : Try to read the configuration from all specified MDS
4344 hostnames during a client mount in case the "primary"
4347 Severity : enhancement
4349 Description: Stop sending data to evicted clients as soon as possible.
4350 Details : Check if the client we are about to send or are sending data to
4351 was evicted already. (Check is done every second of waiting,
4352 for which l_wait_event interface was extended to allow checking
4353 of exit condition at specified intervals).
4356 Frequency : rare, normally only when NFS exporting is done from client
4358 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4360 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4363 Frequency : always, if srand() is called before liblustre initialization
4365 Description: Liblustre uses system PRNG disturbing its usage by user application
4366 Details : Introduce internal to lustre fast and high-quality PRNG for
4367 lustre usage and make liblustre and some other places in generic
4368 lustre code to use it.
4370 Severity : enhancement
4371 Bugzilla : 9477, 9557, 9870
4372 Description: Verify that the MDS configuration logs are updated when xml is
4373 Details : Check if the .xml configuration logs are newer than the config
4374 logs stored on the MDS and report an error if this is the case.
4375 Request --write-conf, or allow starting with --old_conf.
4377 Severity : enhancement
4379 Description: Handle symlinks in the path when checking if Lustre is mounted.
4380 Details : Resolve intermediate symlinks when checking if a client has
4381 mounted a filesystem to avoid duplicate client mounts.
4386 Description: lconf can hit an error exception but still return success.
4387 Details : The lconf command catches the Command error exception at the top
4388 level script context and will exit with the associated exit
4389 status, but doesn't ensure that this exit status is non-zero.
4394 Description: failure of ptlrpc thread startup can cause oops
4395 Details : Starting a ptlrpc service thread can fail if there are a large
4396 number of threads or the server memory is very fragmented.
4397 Handle this without oopsing.
4400 Frequency : always, only if liblustre and non-default acceptor port was used
4402 Description: liblustre cannot connect to servers with non-default acceptor port
4403 Details : tcpnal_set_default_params() was not called and was therefore
4404 ignoring the environment varaible TCPNAL_PORT, as well as other
4405 TCPNAL_ environment variables
4410 Description: two objects could be created on the same OST for a single file
4411 Details : If an OST is down, in some cases it was possible to create two
4412 objects on a single OST for a single file. No problems other
4413 than potential performance impact and spurious error messages.
4417 Bugzilla : 5681, 9562
4418 Description: Client may oops in ll_unhash_aliases
4419 Details : Client dcache may become inconsistent in race condition.
4420 In some cases "getcwd" can fail if the current directory is
4426 Description: Inode refcounting problems in NFS export code
4427 Details : link_raw functions used to call d_instantiate without obtaining
4428 extra inode reference first.
4432 Bugzilla : 9942, 9903
4433 Description: Referencing freed requests leading to crash, memleaks with NFS.
4434 Details : We used to require that call to ll_revalidate_it was always
4435 followed by ll_lookup_it. Also with revalidate_special() it is
4436 possible to call ll_revalidate_it() twice for the same dentry
4437 even if first occurence returned success. This fix changes semantic
4438 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4439 reference on a request referred from the intent.
4440 ll_intent_release() then releases such a request.
4443 Frequency : rare, normally benchmark loads only
4445 Description: unlinked inodes were kept in memory on the client
4446 Details : If a client is repeatedly creating and unlinking files it
4447 can accumulate a lot of stale inodes in the inode slab cache.
4448 If there is no other client load running this can cause the
4449 client node to run out of memory. Instead flush old inodes
4450 from client cache that have the same inode number as a new inode.
4453 Frequency : SLES9 2.6.5 kernel and long filenames only
4454 Bugzilla : 9969, 10379
4455 Description: utime reports stale NFS file handle
4456 Details : SLES9 uses out-of-dentry names in some cases, which confused
4457 the lustre dentry revalidation. Change it to always use the
4461 Frequency : rare, unless heavy write-truncate concurrency is continuous
4462 Bugzilla : 4180, 6984, 7171, 9963, 9331
4463 Description: OST becomes very slow and/or deadlocked during object unlink
4464 Details : filter_destroy() was holding onto the parent directory lock
4465 while truncating+unlinking objects. For very large objects this
4466 may block other threads for a long time and slow overall OST
4467 responsiveness. It may also be possible to get a lock ordering
4468 deadlock in this case, or run out of journal credits because of
4469 the combined truncate+unlink. Solution is to do object truncate
4470 first in one transaction without parent lock, and then do the
4471 final unlink in a new transaction with the parent lock. This
4472 reduces the lock hold time dramatically.
4475 Frequency : rare, 2.4 kernels only
4477 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4478 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4479 before unmounting their filesystem in order to avoid deadlock.
4480 The kernel_locked() function in 2.4 kernels only checks whether
4481 the kernel lock is held, not whether it is this process that is
4482 holding it as 2.6 kernels do.
4487 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4488 Details : The client ptlrpc code may be trying to reconnect to a down
4489 server before a previous connection attempt has timed out.
4490 Increase the reconnect interval to be longer than the connection
4491 timeout interval to avoid sending duplicate connections to
4495 Frequency : echo_client brw_test command
4497 Description: fix echo_client to work with OST preallocated code
4498 Details : OST preallocation code (5137) didn't take echo_client IO path
4499 into account: echo_client calls filter methods outside of any
4500 OST thread and, hence, there is no per-thread preallocated
4501 pages and buffers to use. Solution: hijack pga pages for IO. As
4502 a byproduct, this avoids unnecessary data copying.
4506 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4507 Description: Client can oops in mdc_commit_close() after open replay
4508 Details : It was possible for the MDS to return an open request with no
4509 transaction number in mds_finish_transno() if the client was
4510 evicted, but without actually returning an error. Clients
4511 would later try to replay that open and may trip an assertion
4512 Simplify the client close codepath, and always return an error
4513 from the MDS in case the open is not successful.
4516 Frequency : rare, 2.6 OSTs only
4518 Description: OST may deadlock under high load on fragmented files
4519 Details : If there was a heavy load and highly-fragmented OST filesystems
4520 it was possible to have all the OST threads deadlock waiting on
4521 allocation of biovecs, because the biovecs were not released
4522 until the entire RPC IO was completed. Instead, release biovecs
4523 as soon as they are complete to ensure forward IO progress.
4525 Severity : enhancement
4527 Description: Support for specifying external journal device at mount
4528 Details : If an OST or MDS device is formatted with an external journal
4529 device, this device major/minor is stored in the ext3 superblock
4530 and may not be valid for failover. Allow detecting and
4531 specifying the external journal at mount time.
4536 Description: Mounting an MDS with pending unlinked files may cause oops
4537 Details : target_finish_recovery() calls mds_postrecov() which returned
4538 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4539 considers this an error and immediately begins cleaning up the
4540 lov, just after starting the mds_lov process
4542 Severity : enhancement
4544 Description: Implement 'lfs df' to report actual free space on per-OST basis
4545 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4546 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4547 usage of objects; '-h' to report in human readable format.
4549 ------------------------------------------------------------------------------
4551 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4558 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4559 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4560 at startup during orphan unlinking. Instead, skip buddy update
4561 before setup, as it will be regenerated after recovery is complete.
4566 Description: If an OST is inactive, its locks might reference stale inodes.
4567 Details : lov_change_cbdata() must iterate over all namespaces, even if
4568 they are inactive to clear inode references from the lock.
4570 Severity : enhancement
4571 Frequency : occasional, if non-standard max_dirty_mb used
4573 Description: Client will block write RPCs if not enough grant
4574 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4575 then the client will block writes while waiting for another RPC
4576 to complete instead of consuming its dirty limit. With change
4577 we get improved performance when max_dirty_mb is small.
4579 Severity : enhancement
4580 Bugzilla : 3389, 6253
4581 Description: Add support for supplementary groups on the MDS.
4582 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4583 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4584 lookups for user supplementary groups into a cache.
4588 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4589 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4590 for liblustre O_EXCL is enforced by MDS.
4594 Description: readdir never returns NULL in liblustre.
4595 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4596 dentry in struct dirent.
4600 Frequency : liblustre only, depends on application IO pattern
4601 Description: liblustre clients evicted if not contacting servers
4602 Details : Don't put liblustre clients into the ping_evictor list, so
4603 they will not be evicted by the pinger ever.
4605 Severity : enhancement
4607 Description: Add ability to evict clients by NID from MDS.
4608 Details : By echoing "nid:$NID" string into
4609 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4610 $NID would be instantly evicted from this MDS and from all active
4611 OSTs connected to it.
4615 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4616 Details : lookup_it_finish() used to query file size from OSTs that was not
4621 Description: service threads change working directory to that of init
4622 Details : Starting lustre service threads may pin the working directory
4623 of the parent thread, making that filesystem busy. Threads
4624 now change to the working directory of init to avoid this.
4628 Frequency : during shutdown only
4629 Description: shutdown with a failed MDS or OST can cause unmount to hang
4630 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4635 Frequency : 2.6 only
4636 Description: chmod/chown may include an extra supplementary group
4637 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4638 supplementary group and if none is specified this is used.
4641 Bugzilla : 5479 (6816)
4643 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4644 Details : If lookup is in progress on a file that is unlinked we might try
4645 to revalidate the inode and fail in revalidate after lookup is
4646 complete and ll_file_open() enqueues the open again but
4647 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4650 Frequency : always, if lconf --abort_recovery used
4652 Description: lconf --abort_recovery fails with 'Operation not supported'
4653 Details : lconf was attempting to abort recovery on the MDT device and not
4656 ------------------------------------------------------------------------------
4658 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4663 Frequency : rare (only unsupported configurations with a node running as an
4665 Bugzilla : 6514, 5137
4666 Description: Mounting a Lustre file system on a node running as an OST could
4668 Details : OSTs now preallocates memory needed to write out data at
4669 startup, instead of when needed, to avoid having to
4670 allocate memory in possibly low memory situations.
4671 Specifically, if the file system is mounted on on OST,
4672 memory pressure could force it to try to write out data,
4673 which it needed to allocate memory to do. Due to the low
4674 memory, it would be unable to do so and the node would
4675 become unresponsive.
4677 Severity : enhancement
4679 Description: Addition of lconf --service command line option
4680 Details : lconf now accepts a '--service <arg>' option, which is
4681 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4683 Severity : enhancement
4685 Description: Failover mode is now the default for OSTs.
4686 Details : By default, OSTs will now run in failover mode. To return to
4687 the old behaviour, add '--failout' to the lmc line for OSTs.
4689 Severity : enhancement
4691 Description: Health checks are now provided for MDS and OSTs
4692 Details : Additional detailed health check information on MSD and OSTs
4693 is now provided through the procfs health_check value.
4696 Frequency : occasional, depends on IO load
4698 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4699 after numerous create/delete cycles
4700 Details : The ext3 inode allocation policy would not allocate new inodes
4701 very well on the OSTs because there are no new directories
4702 being created. Instead we look for groups with free space if
4703 the parent directories are nearly full.
4708 Description: Network or server problems during mount may cause partially
4709 mounted clients instead of returning an error.
4710 Details : The config llog parsing code may overwrite the error return
4711 code during mount error handling, returning success instead
4717 Description: MDS can fail to allocate large reply buffers
4718 Details : After long uptimes the MDS can fail to allocate large reply
4719 buffers (e.g. zconf client mount config records) due to memory
4720 fragmentation or consumption by the buffer cache. Preallocate
4721 some large reply buffers so that these replies can be sent even
4722 under memory pressure.
4726 Frequency : rare (liblustre)
4727 Description: fsx running with liblustre complained that using truncate() to
4728 extend the file doesn't work. This patch corrects that issue.
4729 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4730 ATTR_SIZE and lsm use in llu_setattr_raw.
4734 Frequency : rare, only 2.6 kernels
4735 Description: Unusual file access patterns on the MDS may result in inode
4736 data being lost in very rare circumstances.
4737 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4738 code in the 2.6 kernel caused the inode and/or EA data not to
4739 be read from disk, causing single-file corruption.
4743 Frequency : rare, only 2.6 filesystems using extents
4744 Description: Heavy concurrent write and delete load may cause data corruption.
4745 Details : It was possible under high-load situations to have an extent
4746 metadata block in the block device cache from a just-unlinked
4747 file overwrite a newly-allocated data block. We now unmap any
4748 metadata buffers that alias just-allocated data blocks.
4752 Frequency : filesystems with default stripe_count larger than 77
4753 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4754 Details : lconf specifies an inode size of 4096 bytes when the default
4755 stripe_count is larger than 77. This conflicts with the default
4756 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4757 this case to avoid pinning too much memory for large EAs.
4759 ------------------------------------------------------------------------------
4761 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4766 Frequency : rare (extremely heavy IO load with hundreds of clients)
4768 Description: Client is evicted, gets IO error writing to file
4769 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4770 caused clients to be evicted for AST timeouts. The fixes in
4771 bug 5192 mean we no longer need to have such short AST timeouts
4772 so ldlm_timeout has been increased.
4775 Frequency : occasional during --force or --failover shutdown under load
4776 Bugzilla : 5949, 4834
4777 Description: Server oops/LBUG if stopped with --force or --failover under load
4778 Details : a collection of import/export refcount and cleanup ordering
4779 issues fixed for safer force cleanup
4782 Frequency : only filesystems larger than 120 OSTs
4783 Bugzilla : 5990, 6223
4784 Description: lfs getstripe would oops on a very large filesystem
4785 Details : lov_getconfig used kfree on vmalloc'd memory
4788 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4789 Bugzilla : 6242, 6243
4790 Description: reading from files that had been truncated to a non-zero size
4791 but never opened returned no data
4792 Details : ll_file_read() reads zeros from no-object files to EOF
4797 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4798 Details : There's little harm in aborting MDS/OSS recovery and letting it
4799 try again, so I removed the LASSERT and return an error instead.
4801 Severity : enhancement
4803 Description: New debugging infrastructure for tracking down data corruption
4804 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4805 (b) cover more of the client-side code path, and (c) try to narrow
4806 down where problems occurred
4810 Bugzilla : 3819, 4364, 4397, 6313
4811 Description: Racing close and eviction MDS could cause assertion in mds_close
4812 Details : It was possible to get multiple mfd references during close and
4813 client eviction, leading to one thread referencing a freed mfd.
4815 Severity: : enhancement
4816 Bugzilla : 3262, 6359
4817 Description: Attempts to reconnect to servers are now more aggressive.
4818 Details : This builds on the enhanced upcall-less recovery that was added
4819 in 1.4.2. When trying to reconnect to servers, clients will
4820 now try each server in the failover group every 10 seconds. By
4821 default, clients would previously try one server every 25 seconds.
4826 Description: After recovery, certain operations trigger a failed
4827 assertion on a client.
4828 Details : Failing over an mds, using lconf -d --failover, while a
4829 client was doing a readdir() call would cause the client to
4830 LBUG after recovery completed and the readdir() was resent.
4832 Severity : enhancement
4834 Description: Default groups are now added by lconf
4835 Details : You can now run lconf --group <servicename> without having to
4836 manually add groups with lmc.
4839 Frequency : occasional
4841 Description: Nodes with an elan id of 0 trigger a failed assertion
4844 Frequency : always when accessing e.g. tty/console device nodes
4846 Description: tty and some other devices nodes cannot be used on lustre
4847 Details : file's private_data field is used by device data and lustre
4848 values in there got lost. New field was added to struct file to
4849 store fs-specific private data.
4852 Frequency : when exporting Lustre via NFS
4854 Description: NFSD failed occasionally when looking up a path component
4855 Details : NFSD is looking up ".." which was broken in ext3 directories
4856 that had grown large enough to become hashed.
4859 Frequency : Clusters with multiple interfaces not on the same subnet
4861 Description: Nodes will repeatedly try to reconnect to an interface which it
4862 cannot reach and report an error to the log.
4863 Details : Extra peer list entries will be created by lconf with some peers
4864 unreachable. lconf now validates the peer before adding it.
4867 Frequency : Only if a default stripe is set on the filesystem root.
4869 Description: Setting a default stripe on the filesystem root prevented the
4870 filesystem from being remounted.
4871 Details : The client was sending extra request flags in the root getattr
4872 request and did not allocate a reply buffer for the dir EA.
4875 Frequency : occasional, higher if lots of files are accessed by one client
4876 Bugzilla : 6159, 6097
4877 Description: Client trips assertion regarding lsm mismatch/magic
4878 Details : While revalidating inodes the VFS looks up inodes with ifind()
4879 and in rare cases can find an inode that is being freed.
4880 The ll_test_inode() code will free the lsm during ifind()
4881 when it finds an existing inode and then the VFS later attaches
4882 this free lsm to a new inode.
4886 Bugzilla : 6422, 7030
4887 Description: MDS deadlock between mkdir and client eviction
4888 Details : Creating a new file via mkdir or mknod (starting a transaction
4889 and getting the ns lock) can deadlock with client eviction
4890 (gets ns lock and trying to finish a synchronous transaction).
4893 Frequency : occasional
4894 Description: While starting a server, the fsfilt_ext3 module could not be
4896 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4897 kernels. Previously, lconf would still use the ext3 name
4898 when trying to load modules. Now, it will correctly use
4899 ext3 on 2.4 and ldiskfs on 2.6.
4901 Severity : enhancement
4902 Description: The default stripe count has been changed to 1
4903 Details : The interpretation of the default stripe count (0, to lfs
4904 or lmc) has been changed to mean striping across a single
4905 OST, rather than all available. For general usage we have
4906 found a stripe count of 1 or 2 works best.
4908 Severity : enhancement
4909 Description: Add support for compiling against Cray portals.
4910 Details : Conditional compiling for some areas that are different
4914 Frequency : occasional
4915 Bugzilla : 6409, 6834
4916 Description: Creating files with an explicit stripe count may lead to
4917 a failed assertion on the MDS
4918 Details : If some OSTs are full or unavailable, creating files may
4919 trigger a failed assertion on the MDS. Now, Lustre will
4920 try to use other servers or return an error to the
4924 Frequency : occasional
4926 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4927 caused lock timeout during readahead (since 1.4.2).
4928 Details : Processes doing readahead might match a lock that hasn't been
4929 granted yet if there are overlapping and conflicting lock
4930 requests. The readahead process waits on ungranted lock
4931 (original lock is CBPENDING), while OST waits for that process
4932 to cancel CBPENDING read lock and eventually evicts client.
4934 Severity : enhancement
4936 Description: Initial enabling of flock support for clients
4937 Details : Implements fcntl advisory locking and file status functions.
4938 This feature is provided as an optional mount flag (default
4939 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4940 locking are implemented yet, and those that are are not guaranteed
4941 to be completely correct in production environments.
4942 mount -t lustre -o [flock|noflock] ...
4945 Frequency : occasional
4947 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4948 assertion in the ext3 JBD (journaling) layer.
4949 Details : The b_committed_data struct is protected by the big kernel lock
4950 in 2.4 kernels, serializing journal_commit_transaction() and
4951 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4952 access to this struct. In 2.6 kernels there is finer grained
4953 locking to improve SMP performance of the JBD layer.
4957 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4962 Description: merge and cleanup kernel patches.
4965 -----------------------------------------------------------------------------
4967 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4969 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4970 versions of Lustre. Please update all servers and clients to
4971 version 1.4.2 or later at the same time. You must also run
4972 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4973 to update the configuration logs.
4975 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4976 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4977 - swab llog records of type '0' so we get proper header size/idx (5861)
4978 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4979 - fix rename of one directory over another leaking an inode (5953)
4980 - avoid SetPageDirty on 2.6 (5981)
4981 - don't re-add just-being-destroyed locks to the waiting list (5653)
4982 - when creating new directories, inherit the parent's custom
4983 striping settings if present parent (3048)
4984 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4985 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4986 - don't hold client locks on temporary worklist from l_lru (5666)
4987 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4988 - automatically evict dead clients (5921)
4989 - Update file size properly in create+truncate+fstat case (6196)
4990 - Do not unhash mountpoint dentries, do not allow removal of
4992 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4993 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4994 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4995 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4996 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4997 - Correctly handle reads of files with no objects (6243)
4998 - lctl recover will also mark a device active if deactivate used (5933)
5000 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5001 - allow --write-conf on an MDS with different nettype than client (5619)
5002 - don't write config llogs to MDS for mounts not from that MDS (5617)
5003 - lconf should create multiple TCP connections from a client (5201)
5004 - init scripts are now turned off by default; run chkconfig --on
5005 lustre and chkconfig --on lustrefs to use them
5006 - upcalls are no longer needed for clients to recover to failover
5008 - add --abort-recovery option to lconf to abort recovery on device
5010 - add support for an arbitrary number of OSTs (3026)
5011 - Quota support protocol changes.
5012 - forward compatibility changes to wire structs (6007)
5013 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5014 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5015 - improved "lustre status" script
5016 - initialize blocksize for non-regular files (6062)
5017 - added --disable-server and --disable-client configure options (5782)
5018 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5019 - Vanilla 2.4.29 support
5020 - increase maximum number of obd devices to 520 (6242)
5021 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5022 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5023 SLES 9 2.6.5 kernels
5024 - limit stripes per file to 160 (the maximum EA size) (6093)
5026 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5029 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5030 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5031 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5032 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5033 - small ext3 extents cleanups and fixes (5733)
5034 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5035 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5036 - avoid needless client->OST connect, fix handle mismatch (5317)
5037 - fix DLM error path that led to out-of-sync client, long delays (5779)
5038 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5039 - fix several locking issues related to i_size (5492,5624,5654,5672)
5040 - don't move pending lock onto export if it is already evicted (5683)
5041 - fix kernel oops when creating .foo in unlinked directory (5548)
5042 - fix deadlock in obdfilter statistics vs. object create (5811)
5043 - use time_{before,after} to avoid timer jiffies wrap (5882)
5044 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5045 - Do not leak request if server was not able to process it (5154)
5046 - If mds_open unable to find parent dir, make that negative lookup(5154)
5047 - don't create new directories with extent-mapping (5909, 5936)
5049 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5050 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5051 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5052 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5053 - support for RHEL4 kernel on i686 (5773)
5054 - provide error messages when incompatible logs are encountered (5898)
5056 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5057 * version 1.4.0.10 (1.4.1 release candidate 1)
5059 - don't keep a lock reference when lock is not granted (4238)
5060 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5061 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5062 - drop import inflight refcount on signal_completed_replay error (5255)
5063 - unlock page after async write error during send (3677)
5064 - handle missing objects in filter_preprw_read properly (5265)
5065 - no transno return for symlink open, don't save no-trasno open (3440)
5066 - don't try to complete elan receive that already failed (4012)
5067 - free RPC server reply state on error (5406)
5068 - clean up thread from ptlrpc_start_thread() on error (5160)
5069 - readahead could read extra page into cache that wasn't ejected (5388)
5070 - prevent races in class_attach/setup/cleanup/detach (5260)
5071 - don't dereference de->d_inode after l_dput of de (5458)
5072 - use "int" for stripe value returned from lock_to_stripe (5544)
5073 - mballoc allocation and error-checking fixes in 2.6 (5504)
5074 - block device patches to fix I/O request sizes in 2.6 (5482)
5075 - look up hostnames for IB nals (5602)
5076 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5077 - don't start multiple acceptors for the same port (5277)
5078 - fix incorrect LASSERT in mds_getattr_name (5635)
5079 - export a proc file for general "ping" checking (5628)
5080 - fix "lfs check" to not block when the MDS is down (5628)
5082 - service request history (4965)
5083 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5084 - create an "evict_client" /proc entry on OSTs, like the MDS has
5085 - fix mount usage message, return errors per mount(8) (5168)
5086 - change grep [] to grep "[]" in tests so they work in more UMLs
5087 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5088 - remove ancient LOV_MAGIC_V0 EA support (5047)
5089 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5090 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5091 - allow readahead allocations to fail when low on memory (5383)
5092 - mmap locking landed again, after considerable improvement (2828)
5093 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5095 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5098 - send OST transaction number in read/write reply to free req (4966)
5099 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5100 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5101 - the watchdog thread now runs as interruptible (5246)
5102 - flock/lockf fixes (but it's still disabled, pending 5135)
5103 - don't use EXT3 constants in llite code (5094)
5104 - memory shortage at startup could cause assertion (5176)
5106 - reorganization of lov code
5107 - single portals codebase
5109 - add extents/mballoc support (5025)
5110 - direct I/O reads in the obdfilter (4048)
5111 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5113 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5116 - send OST transaction number in read/write reply to free req (4966)
5117 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5118 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5119 - the watchdog thread now runs as interruptible (5246)
5120 - handle missing objects in filter_preprw_read properly (5265)
5121 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5122 - drop import inflight refcount on signal_completed_replay error (5255)
5123 - unlock page after async write error during send (3677)
5124 - return original error code on reconstructed replies (3761)
5125 - no transno return for symlink open, don't save no-trasno open (3440)
5127 - add pid to ldlm debugging output (4922)
5128 - bump the watchdog timeouts -- we can't handle 30sec yet
5129 - extra debugging for orphan dentry/inode bug (5259)
5131 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5134 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5135 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5136 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5137 - fix NULL dereference in /proc/sys/portals/routes (4827)
5138 - allow failed mdc_close() operations to be interrupted (4561)
5139 - stop precreate on OST before MDS would time out on it (4778)
5140 - don't send partial-page writes before EOF from client (4410)
5141 - discard client grant for sub-page writes on large-page clients (4520)
5142 - don't free dentries not owned by NFS code, check generation (4806)
5143 - fix lsm leak if mds_create_objects() fails (4801)
5144 - limit debug_daemon file size, always print CERROR messages (4789)
5145 - use transno after validating reply (3892)
5146 - process timed out requests if import state changes (3754)
5147 - update mtime on OST during writes, return in glimpse (4829)
5148 - add mkfsoptions to LDAP (4679)
5149 - use ->max_readahead method instead of zapping global ra (5039)
5150 - don't interrupt __l_wait_event() during strace
5152 - add software watchdogs to catch hung threads quickly (4941)
5153 - make lustrefs init script start after nfs is mounted
5154 - fix CWARN/ERROR duplication (4930)
5155 - return async write errors to application if possible (2248)
5156 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5157 - print NAL number in %x format (4645)
5158 - update barely-supported suse-2.4.21-171 series (4842)
5159 - support for sles 9 %post scripts
5160 - support for building 2.6 kernel-source packages
5161 - support for sles km_* packages
5163 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5166 - ignore -ENOENT errors in osc_destroy (3639)
5167 - notify osc create thread that OSC is being cleaned up (4600)
5168 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5169 - reconstruct ost_handle() like mds_handle() (4657)
5170 - create a new thread to do import eviction to avoid deadlock (3969)
5171 - let lconf resolve symlinked-to devices (4629)
5172 - don't unlink "objects" from directory with default EA (4554)
5173 - hold socknal file ref over connect in case target is down (4394)
5174 - allow more than 32000 subdirectories in a single directory (3244)
5175 - fix blocks count for O_DIRECT writes (3751)
5176 - OST returns ENOSPC from object create when no space left (4539)
5177 - don't send truncate RPC if file size isn't changing (4410)
5178 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5179 - bind to privileged port in socknal and tcpnal (3689)
5181 - rate limit CERROR/CWARN console message to avoid overload (4519)
5182 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5183 - basic mmap support (3918)
5184 - kernel patch series update from b1_4 (4711)
5186 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5189 - avoid crash during MDS cleanup with OST shut down (2775)
5190 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5191 - don't use bad inodes on the MDS (3744)
5192 - dynamic object preallocation to improve recovery speed (4236)
5193 - don't hold spinlock over lock dumping or change debug flags (4401)
5194 - don't zero obd_dev when it is force cleaned (3651)
5195 - print grants to console if they go negative (4431)
5196 - "lctl deactivate" will stop automatic recovery attempts (3406)
5197 - look for existing locks in ldlm_handle_enqueue() (3764)
5198 - don't resolve lock handle twice in recovery avoiding race (4401)
5199 - revalidate should check working dir is a directory (4134)
5201 - don't always mark "slow" obdfilter messages as errors (4418)
5203 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5206 - don't close LustreDB during write_conf until it is done (3860)
5207 - fix typo in lconf for_each_profile (3821)
5208 - allow dumping logs from multiple threads at one time (3820)
5209 - don't allow multiple threads in OSC recovery (3812)
5210 - fix debug_size parameters (3864)
5211 - fix mds_postrecov to initialize import for llog ctxt (3121)
5212 - replace config semaphore with spinlock (3306)
5213 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5214 - don't allow enqueue to complete on a destroyed export (3822)
5215 - down write_lock before checking llog header bitmap (3825)
5216 - recover from lock replay timeout (3764)
5217 - up llog sem before sending rpc (3652)
5218 - reduce ns lock hold times when setting kms (3267)
5219 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5220 - fix NULL deref and obd_dev leak on setup error (3312)
5221 - replace some LBUG about llog ops with error handling (3841)
5222 - don't match INVALID dentries from d_lookup and spin (3784)
5223 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5224 - fix invalid assertion in ptlrpc_set_wait (3880)
5226 - add libwrap support for the TCP acceptor (3996)
5227 - add /proc/sys/portals/routes for non-root route listing (3994)
5228 - allow setting MDS UUID in .xml (2580)
5229 - print the stack of a process that LBUGs (4228)
5231 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5234 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5235 - make sure to unset replay flag from failed open requests (3440)
5236 - if default stripe count is 0, use OST count for inode size (3636)
5237 - update parent mtime/ctime on client for create/unlink (2611)
5238 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5239 - free recovery state on server during a forced cleanup (3571)
5240 - unregister_reply for resent reqs (3063)
5241 - loop back devices mounting and status check on 2.6 (3563)
5242 - fix resource-creation race that can provoke i_size == 0 (3513)
5243 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5244 - more debugging for page-accounting assertion (3746)
5245 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5246 - avoid deadlock after precreation failure (3758)
5247 - fix race and lock order deadlock in orphan handling (3450, 3750)
5248 - add validity checks when grabbing inodes from l_ast_data (3599)
5250 - add /proc/.../recovery_status to obdfilter (3428)
5251 - lightweight CDEBUG infrastructure, debug daemon (3668)
5252 - change default OSC RPC parameters to be better on small clusters
5253 - turn off OST read cache for files smaller than 32MB
5254 - install man pages and include them in rpms (3100)
5255 - add new init script for (un)mounting lustre filesystems (2593)
5256 - run chkconfig in %post for init scripts (3701)
5257 - drop scimac NAL (unmaintained)
5259 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5262 - clean kiobufs before and after use (3485)
5263 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5264 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5265 - change init script to fail more clearly if not run as root (1528)
5266 - allow clients to reconnect during replay (1742)
5267 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5268 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5269 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5270 - fix cond_resched (3554)
5271 - only evict unfinished clients after recovery (3515)
5272 - allow bulk resend, prevent data loss (3570)
5273 - dynamic ptlrpc request buffer allocation (2102)
5274 - don't allow unlinking open directory if it isn't empty (2904)
5275 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5276 - remove extraneous obd dereference causing LASSERT failure (3334)
5277 - don't use get_cycles() when creating temp. files on the mds (3156)
5278 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5279 - handle EEXIST for set-stripe, set proper directory name (3336)
5281 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5282 - fix ksocknal_fmb_callback() error messages (2918)
5284 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5287 - don't copy lvb into (possibly NULL) reply on error (2983)
5288 - don't deref dentry after dput, don't free lvb on error (2922)
5289 - use the kms to determine writeback rpc length (2947)
5290 - increment oti_logcookies when osc is inactive (2948)
5291 - update client's i_blocks count via lvb messages (2543)
5292 - handle intent open/close of special files properly (1557)
5293 - mount MDS with errors=remount-ro, like obdfilter (2009)
5294 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5295 - don't use cancelling-locks' kms values (2947)
5296 - use highest lock extent for kms, not last one (2925)
5297 - don't dereference ERR_PTR() dentry in error handling path (3107)
5298 - fix thread race in portals_debug_dumplog() (3122)
5299 - create lprocfs device entries at setup instead of at attach (1519)
5300 - common AST error handler, don't evict client on completion race (3145)
5301 - zero nameidata in detach_mnt in 2.6 (3118)
5302 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5303 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5304 - initialize RPC timeout timer earlier for 2.6 (3219)
5305 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5306 - print nal/nid for unknown nid (3258)
5307 - additional checks for oscc recovery before doing precreate (3284)
5308 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5309 - don't crash in mdc_close for bad permissions on open (3285)
5310 - zero i_rdev for non-device files (3147)
5311 - clear page->private before handing to FS, better assertion (3119)
5312 - tune the read pipeline (3236)
5313 - fix incorrect decref of invalidated dentry (2350)
5314 - provide read-ahead stats and refine rpc in flight stats (3328)
5315 - don't hold journal transaction open across create RPC (3313)
5316 - update atime on MDS at close time (3265)
5317 - close LDAP connection when recovering to avoid server load (3315)
5318 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5319 - don't leak open file on MDS after open resend (3325)
5320 - serialize filter_precreate and filter_destroy_precreated (3329)
5321 - loop device shouldn't call sync_dev() for nul device (3092)
5322 - clear page cache after eviction (2766)
5323 - resynchronize MDS->OST in background (2824)
5324 - refuse to mount the same filesystem twice on same mountpoint (3394)
5325 - allow llmount to create routes for mounting behind routers (3320)
5326 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5327 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5328 - fix rare problem with rename on htree directories (3417)
5330 - allow default OST striping configuration per directory (1414)
5331 - fix compilation for qswnal for 2.6 kernels (3125)
5332 - increase maximum number of MDS request buffers for large systems
5333 - change liblustreapi to be useful for external progs like lfsck (3098)
5334 - increase local configuration timeout for slow disks (3353)
5335 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5337 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5340 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5341 - don't overwrite extent policy data in reply if lock was blocked (2901)
5342 - drop filter export grants atomically with removal from device (2663)
5343 - del obd_self_export from work_list in class_disconnect_exports (2908)
5344 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5345 - swab reply message in mdc_close, other PPC fixes (2464)
5346 - fix destroying of named logs (2325)
5347 - overwrite old logs when running lconf --write_conf (2264)
5348 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5349 - fix race in target_handle_connect (2898)
5350 - mds_reint_create() should take same inode create lock (2926)
5351 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5352 - don't close files for self_export to avoid uninitialized obd (2936)
5353 - allow MDS with the same name as client node (2939)
5354 - hold dentry reference for closed log files for unlink (2325)
5355 - reserve space for all logs during transactions (2059)
5356 - don't evict page beyond end of stripe extent (2925)
5357 - don't oops on a deleted current working directory (2399)
5358 - handle hard links to targets without a parent properly (2517)
5359 - don't dereference NULL lock when racing during eviction (2867)
5360 - don't grow lock extents when lots of conflicting locks (2919)
5362 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5365 - account for cache space usage on clients to avoid data loss (974)
5366 - lfsck support in lustre kernel code (2349)
5367 - reduce journal credits needed for BRW writes (2370)
5368 - orphan handling to avoid losing space on client/server crashes
5369 - ptlrpcd can be blocked, stopping ALL progress (2477)
5370 - use lock value blocks to assist in proper KMS, faster stat (1021)
5371 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5372 - recovery for initial connections (2355)
5373 - fixes for mds_cleanup_orphans (1934)
5374 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5375 - block all file creations until orphan recovery completes (1901)
5376 - client remove rq_connection from request struct (2423)
5377 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5378 - recovery timer race (2670)
5379 - mdc_close recovey bug (2532)
5380 - ptlrpc cleanup bug (2710)
5381 - mds timeout on local locks (2588)
5382 - namespace lock held during RPCs (2431)
5383 - handle interrupted sync write properly (2503)
5384 - don't try to handle a message that hasn't been replied to (2699)
5385 - client assert failure during cleanup after abort recovery (2701)
5386 - leak mdc device after failed mount (2712)
5387 - ptlrpc_check_set allows timedout requests to complete (2714)
5388 - wait for inflight reqs when ptlrpcd finishes (2710)
5389 - make sure unregistered services are removed from the srv_list
5390 - reset bulk XID's when resending them (caught by 1138 test)
5391 - unregister_bulk after timeout
5392 - fix lconf error (2694)
5393 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5394 - readahead locks pages, leaves pending causing memory pressure (2673)
5395 - increase OST request buffers to 4096 on large machines (2729)
5396 - fix up permission of existing directories in simple_mkdir (2661)
5397 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5398 - don't assign transno to errored transactions (2742)
5399 - don't delete objects on OST if given a bogus objid from MDS (2751)
5400 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5401 - if rq_no_resend, then timeout request after recovery (2432)
5402 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5403 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5404 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5405 - MDS recovery shouldn't delete objects that it has given out (2730)
5406 - if enqueue arrives after completion, don't clobber LVB (2819)
5407 - don't unlock pages twice when trigger_group_io returns error (2814)
5408 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5409 - don't write pages to disk if there was an error (1450)
5410 - don't ping imports that have recovery disabled (2676)
5411 - take buffered bytes into account when balancing socknal conn (2817)
5412 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5413 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5414 - remove little-endian swabbing of llog records (1987)
5415 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5416 - retry reposting request buffers if they fail (1191)
5417 - grow extent at grant time to avoid granting a revoked lock (2809)
5418 - lock revoke doesn't evict page if covered by a second lock (2765)
5419 - disable VM readahead to avoid reading outside lock extents (2805)
5421 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5422 - updated kernel patches for hp-2.4.20 kernel (2681)
5424 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5427 - fix truncated write corruption (2366)
5428 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5430 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5431 - ensure len(uuid) < 37 in lmc (1171)
5432 - fix ia64 OOPS in llog_test (2255)
5433 - zero end of page at obdfilter for partial page writes (2648)
5434 - don't leave stale dentries around after renames (bug 2428)
5435 - fix timeouts when evicting a client with a single lock held (2642)
5436 - set deadline for the initial HELLO message to drain (2634)
5437 - print out dotted-quad IP addresses in the socknal (2302)
5439 - additional debugging for MDS client eviction problem (2443)
5440 - fix mkfsoptions support for osts (2603, 2604)
5442 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5445 - add series for the vanilla 2.6.0 kernel
5446 - add series for the vanilla 2.4.24 kernel
5447 - add series for a cray x86/64 UL kernel drop
5448 - fix xattr patches for the vanilla 2.4.19 series
5450 - generate true UUIDs in lmc (1171)
5451 - have portals stack dumping break in UML (2466)
5452 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5453 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5454 - wake callback waiting threads on client eviction (2460)
5455 - Add --ptldebug and --subsystem to lmc (1719)
5456 - update assertion to allow safe interrupt allocation
5457 - set rq_no_resend for cancel requests (2432)
5458 - recalculate ptlrpcd timeout after resend (2494)
5459 - call vfs_rmdir when removing pending directories (2368)
5460 - fix renaming a file to itself (2429)
5461 - lmc creates a default one-stripe lov (2454)
5462 - expand procfs space to handle large clusters (2326)
5463 - increase UML stack to avoid overflow
5464 - update lconf's list of debug and subsystem masks
5465 - fix lfs find --obd (2510)
5466 - /proc tunable for disabling filter read caching (2591)
5467 - stop rpm packages from altering slapd.conf (2301)
5468 - disable nagle in the socknal under 0conf (2578)
5469 - choose mds inode size based on stripe count (2572)
5470 - fix kernel-source rpm problems (2516)
5472 - add --disable-doc to avoid pdf generation (2421)
5473 - update documentation, tests, type-os, comments
5474 - avoid format warnings on ia64
5475 - remove the TOE NAL
5476 - tiny code cleanups by removing unused fields
5478 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5481 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5482 - remove the most blatant lies from BUILDING file (2371)
5483 - change default debug level to reasonable production setting
5484 - reduce client side cache size to reduce cache flush time
5485 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5486 - make TCP zerocopy and pinger support enabled by default (2476)
5487 - sync writes completed after process exits caused crashes (2319)
5488 - maintain correct mount count on the MDS (2356)
5489 - backout 1557, because 2316 wasn't really fixed
5490 - better file I/O statistics gathering in /proc
5491 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5492 - another kernel patch to fix zero-copy TCP function export
5493 - don't take duplicate lock when processing re-sent getattr (2420)
5494 - lctl uses obd_self_export instead of creating new conn (2353)
5495 - MDS/OST recovery case which requires object creation asserted (2425)
5496 - move lfs from /usr/sbin to /usr/bin in packages
5497 - fix race between mds_client_add and mds_client_free (2417)
5498 - use kmalloc instead of slabs in portals (2430)
5499 - don't create duplicate records when a failover MDS is present (2442)
5500 - remove unnecessary mount age check (2332)
5501 - don't remove directory inodes from locks prematurely (2451)
5502 - don't break if MDS service name is the same as hostname (2103)
5503 - fix races in client write RPC generation when cache full (2482)
5505 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5508 - remove now-unused request->rq_obd (278)
5509 - if an allocation fails, print out how much memory we've used (1933)
5510 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5511 - add the "configurable stack size" patch to most series files (1256)
5512 - ability to write large log records, for 100+ OST configs (2306)
5513 - fix NULL deref when filter_prep fails (2314)
5514 - fix operator precedence error in filter_sync
5515 - dynamic allocation of socknal TX descriptors (2315)
5516 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5517 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5518 - ensure that CWARN messages in Portals always get to the syslog
5519 - __init/__exit are not for prototype decls (ldlm_init/exit)
5520 - x86-64 compile warning fixes
5521 - fix gateway LMC keyword conflict (2318)
5522 - fix MDS lock inversions in getattr/reint paths (1844)
5523 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5524 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5525 - DLM race condition prevented some lock evictions (2328)
5526 - ENOMEM detection and retry on socknal sends (2230)
5527 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5528 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5529 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5530 - better error messages when a client is rejected during recovery (1505)
5531 - avoid cancelling locks which were never granted, after failure (2330)
5532 - fix i_sem/journal inversion in mds_client_add (2333)
5533 - fix truncate/getattr lock cycle deadlock (2334)
5534 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5535 - fix two rare exit paths which could leak an l_lock() ref (2321)
5536 - fencepost error in MDS/OST orphan recovery (2226)
5537 - make log record alignment 8 bytes (1988)
5538 - lstripe now fails when requested offset > ost_count (2237)
5539 - ensure that all kernel series have a complete list.h (1607)
5540 - fix crashes in special-file operations (2316)
5541 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5543 - allow configurable automake binary, for testing new versions
5544 - small update to the lfs documentation
5546 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5548 * fix negative export reference count in fsfilt_sync (2312)
5550 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5551 * release candidate 0.9.1
5553 - orphans are moved into the PENDING directory for possible recovery
5554 - replayed opens now open by fid for orphan/rename safety (1042)
5555 - last close of an orphan inode generates a transno (683)
5556 - chdir() and mount() now pin the directory entry (1020)
5557 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5558 - discard very old requests without processing them (1502)
5559 - remove some common, well-understood CERRORs (1505)
5560 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5561 - clear "grant" flags in OST replies until OST grant code lands (1644)
5562 - fix read performance by not clobbering i_blksize on client (1598)
5563 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5564 - make LRU size a /proc tunable, clears locks when reduced (707)
5565 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5566 - server locks take references on exports now (1558)
5567 - build fixes for 2.4.20-rh trees (1663)
5568 - return an error from lov_create if all OSCs are inactive (1751)
5569 - fix import levels when a reconnect happens without a timeout (1597)
5570 - exit early from mds_open if we get a lookup error (1749)
5571 - partial page read at EOF wouldn't wait for disk before sending (1642)
5572 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5573 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5574 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5575 - write/truncate lock inversion (1639)
5576 - Don't auto-load obdclass, portals modules during cleanup (1495)
5577 - fix timestamps from jumping to "now" (1763)
5578 - extra journal assertions (1648)
5579 - add an extra multiunlink test (1771)
5580 - fix read_record/write_record API (1776)
5581 - fix leak of offset_extent, possible incorrect i_size later (1772)
5582 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5583 - Debugging for the kqswnal_get_idle_tx problems (1820)
5584 - Allow recovery to be attempted multiple times (1536)
5585 - Write out MDS last_rcvd file after it is first created (1600)
5586 - Fix tx_descriptor leak in failed transmit situations (1827)
5587 - ext3 journaling fixes for assertion failure after IO error (1871)
5588 - class_export_put() on freed export after completion AST error (1896)
5589 - Fix revalidate looping in VFS (1322)
5590 - Don't access a freed export during MDS_REINT timeout (1521)
5591 - Add open-unlink recovery support on the MDS (1673,1764)
5592 - Return an error if no MDS data was read from last_rcvd (1946)
5593 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5594 - Don't setup a disk device that doesn't match exported UUID (317)
5595 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5596 - avoid committing NULL handle in force close
5597 - local.sh is now a one-stripe LOV configuration
5598 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5599 - fix ext3 htree duplicate directory entry corruption (1516)
5600 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5601 - update to new LOV EA format (2097)
5602 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5604 2003-06-15 Phil Schwan <phil@clusterfs.com>
5607 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5608 - per-import/export recovery handling (958, 931, 959)
5609 - multiple last-rcvd slots, for serving multiple FSes (949)
5610 - connections are again shared between multiple imp/exports (963, 964)
5611 - "umount -f" would hang if any requests needed to be sent (393, 978)
5612 - avoid pinning large req buffer by copying for queued messages (989)
5613 - add "uuid" to "lctl device" command to help upcalls (991)
5614 - "open" RPCs with transnos would confuse recovery counters (1037)
5615 - do proper endian conversion of all wire messages (288, 340, 891)
5616 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5617 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5618 - fix ptlrpc_connection leak in target_handle_connect (1174)
5619 - fix import refcounting bug in OST and MDS cleanup (1134)
5620 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5621 - fix very unlikely obd_types race condition (501)
5622 - remove osc_open hack for echo_client (1187)
5623 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5624 - a failure in read_inode2 leads to deadlock (1139)
5625 - cancel ack-locks as soon as transaction is committed (1072)
5626 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5627 - make sure to commitrw after any preprw to avoid deadlock (1162)
5628 - failing to execute a file in a lustre FS would lock inode (1203)
5629 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5630 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5631 - obd_uuid2tgt would walk off the end of the list (1255)
5632 - on IA64 the osc would give portals incorrect bulk size (1258)
5633 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5634 - fix lock inversion caused by new llite matching code (1282)
5635 - limit the number of dirty pages on a client to 10MB (1286)
5636 - timed out locks were not being corrected cancelled (1289)
5637 - fix O_DIRECT above 4GB on IA-32 (1292)
5638 * major user-visible changes
5639 - fail out/fail over policy now controlled by the upcall (993)
5641 - add OBD_PING to check server availability and failure (954)
5642 - lustre messages are now sent in sending host order (288, 340, 891)
5643 - add eadatalen to MDS getattr reply (340)
5644 - OST read replies may contain second buffer, with per-page status (593)
5646 2003-03-11 Phil Schwan <phil@clusterfs.com>
5649 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5650 - failed open()s could cause deadlock; fixed (867, 869)
5651 - stop cancelling OST locks when files are closed (481)
5652 - overlapping XID spaces caused network corruption (851, 853)
5653 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5654 - setattr_raw added, to do single-RPC, server-side setattrs
5655 - lmc/lconf syntax change for OST UUIDs
5656 - fix crashy race condition between ptlrpc_free_req and osc_close
5657 - don't use request in mdc_enqueue if we hit a timeout (889)
5658 - don't set the inode i_size for regular files from the MDS (896)
5659 - handle out of order completion AST (842)
5660 - don't LBUG if a lock request times out after receiving AST (913)
5661 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5662 - if a bad lock AST arrives, send an error instead of dropping entirely
5663 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5664 - fix leak in bulk IO when only partially completed (899, 900, 926)
5665 - fix O_DIRECT for ia64 (55)
5666 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5667 - C-z after timeout could hang a process forever; fixed (977)
5669 - client-side I/O cache (678, 924, 929, 941, 970)
5671 - READPAGE and SETATTRs which don't take server-side locks get
5674 2003-02-11 Phil Schwan <phil@clusterfs.com>
5677 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5678 - Fix fsfilt_extN_readpage() to read a full page of directory
5679 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5680 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5681 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5682 - Most intent operations were removed, in favour of a new RPC mode
5683 that does a single RPC to the server and bypasses most of the VFS
5684 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5685 - Aggressively cancel local locks on DLM servers
5686 - mds_reint_unlink sends EA to the client if it's the last nlink.
5687 client uses that EA to unlink OST objects.
5688 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5689 - recursive symlinks were fixed (439)
5690 - fixed NULL deref in DEBUG_REQ
5691 - filter_update_lastobjid no longer calls sync, which annoyed extN
5692 - fixed multi-client small-writes to a single file problem (445)
5693 - fixed mtime updates during file writes (607)
5694 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5695 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5696 - fixed Linux OST size reporting problem (444, 656)
5697 - OST now updates object mtime with writes or setattr (607, 619)
5698 - client verifies file size before zeroing page past EOF (445)
5699 - OST now writes last allocated objid to disk with allocation (108)
5700 - LOV on echo now works (409)
5702 - mds_reint_unlink sends a new buffer, with the EA included. this
5703 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5704 set if a regular file was being unlinked, and it was the last link
5705 - use PtlGet from the target for bulk writes (315)
5706 - OST now updates object mtime with writes or setattr (607, 619)
5707 - LDLM now has a grant-time callback to revalidate locked items, if
5709 - Many MDS operations were reorganized to combat race conditions
5711 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5712 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5714 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5717 - Fully reactivate OST imports after reconnection (512, others)
5718 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5719 - More graceful error handling for truncating on dead OST (515)
5720 - Don't error out unless we're actually accessing dead stripes (474)
5721 - Fix garbage sizes when stripes are missing (410)
5722 - LRU counters were broken, causing constant lock purge (433, 432)
5723 - garbage on read from stripes with failed OSTs (441)
5724 - mark OSCs as active before reconnecting during recovery (438)
5725 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5726 - lfind did not preserve OST order in output (443)
5727 - symlinks cause hung clients, incorrect data (439)
5728 - stop dereferencing request after dropping refcount (457)
5729 - don't LASSERT(spin_is_locked) on non-SMP (455)
5730 - fixes for many rename() bugs
5731 - fstat didn't correctly synchronize attributes (399)
5732 - server must handle lock cancellation during blocking AST prep (487)
5733 - bulk descriptors were free()d too soon (511)
5734 - fix paths in lconf, which would load incorrect modules (451, 507)
5735 - fix confusing lconf 'host not found' error message (386)
5736 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5737 - fix race condition in mdc_blocking_ast() for inode access (526)
5738 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5739 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5740 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5741 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5742 - fix lstripe in several ways (kernel side) (527)
5743 - fix request leak in ldlm_cli_enqueue (262)
5744 - incorrect OSC was marked inactive after OST failure
5745 - call mds_fs_cleanup before unmounting filesystem (524)
5746 - fix races between taking ns_lock and ldlm_lock_change_resource
5747 - fix races updating LOV export open file list
5748 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5749 - fix recovery NULL deref in ldlm_cli_cancel_unused
5750 - fix some DLM races by using new hash table for lock handles (419)
5751 - permit the client to specify desired inodes, at replay
5752 - duplicate requests when we queue them for replay reintegration
5753 - fix last_rcvd offset calculation
5754 - sync after each recovered transaction, so we always make progress
5755 - never, not always, ERESTART requests without transnos
5756 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5758 - skip replay if the MDS says that the client is already connected
5759 - don't check for a recovery-enabled export to match lctl's UUID
5760 - don't INC_USE_COUNT for phantom exports
5761 - don't crash when cleaning up phantom exports (567)
5762 - don't double-finish or set replay data for errored mdc_open requests
5763 - abort requests when they time out, so we don't get old replies
5764 - send/receive replies for AST messages again
5765 - if the client says that it doesn't have the lock, cancel it on the
5767 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5768 mark it as destroyed, it will all work out when decref is called
5769 - fix module use counts (22, 581)
5771 - ASTs now expect a reply (server cancels lock on error reply)
5773 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5776 - fix many simultaneous client startup (392)
5777 - fix dentry->d_it clobbering
5778 - credentials weren't being shipped for readdir/getattr operations
5779 - remove invalid assertions triggered during some concurrent MD
5781 - proper Lustre versions added (336, 389)
5782 - fix memory leak for create error case (398)
5783 - fix LOV locking bug that would get cli/srv out of sync
5784 - fix echo client over LOV (409)
5785 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5786 - fix double-O_EXCL intent crash (424)
5787 - avoid sending multiple lock CANCELs (352)
5789 - MDS can do multi-client recovery (modulo bugs in new code)
5791 - many updates, edits, cleanups
5793 2002-11-18 Phil Schwan <phil@clusterfs.com>
5796 - fix null d_it dereference (346)
5797 - fix full OST/dbench hang (333)
5798 - fix permission problem with file removal (286)
5799 - fix removal of OSCs from LOV when they fail
5800 - fix NULL deref during bulk timeout (214)
5801 - fix problems related to multiple filesystems on one MDS (241)
5802 - fixed serious subtle metadata locking bugs
5803 - free locks on clients when inodes are removed due to memory
5805 - fix inode pointer in lock data (285)
5806 - partial support for multiple MDS on a single host (241)
5807 - data locks weren't cancelled at clear_inode time (290, 311)
5808 - intent locks could lead to unbounded lock growth (205)
5809 - added a maximum lock count, an LRU list, and a flusher
5810 - fix multiple rename (365)
5811 - properly abstracted the echo client
5812 - OSC locked 1 byte too many; fixed
5813 - rewrote brw callback code:
5814 - fixed recovery bugs related to LOVs (306)
5815 - fixed too-many-pages-in-one-write crash (191)
5816 - fixed (again) crash in sync_io_timeout (214)
5817 - probably fixed callback-related race (385)
5819 - Add capability to MDS protocol
5820 - LDLM cancellations and callbacks on different portals
5822 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5825 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5826 - defer open object destruction to close time (601981, 138)
5827 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5828 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5829 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5830 - don't call obd_setattr in ll_file_release for destroyed objects
5831 * protocol change to lustre_msg: move |version| and add |flags|
5832 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5833 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5834 * added replay of create, unlink, link and rename operations during
5835 MDS failover; recovery should be much more robust now
5836 * remove failed OSCs from LOVs (only lov_create uses this so far)
5837 * the lustre-HOWTO was brought (more) up to date (582544)
5839 2002-10-23 Phil Schwan <phil@clusterfs.com>
5842 - in-use dentries weren't being reused properly (617851)
5843 - prevent multiple LDLM setup (599178)
5844 - fix LOV size calculations for truncate (617853)
5845 - fix client handling of MDS intent errors (POSIX)
5846 - fix permission bug in lovstripe.c test (624321)
5847 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5848 - truncate past end of file could corrupt data
5849 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5850 - a race in recovery could return ETIMEDOUT to apps (623947)
5851 - building outside the source directory was fixed
5852 * the lustre-HOWTO was brought (more) up to date (582544)
5853 * major progress was made on recovery functionality
5855 2002-10-10 Phil Schwan <phil@clusterfs.com>
5858 - recovery deadlock fix
5859 - rm -rf causes LBUG fix (617817)
5860 - file open by multiple tasks fix (618962)
5861 - directory permissions bugs (602707 and 620007)
5862 - journal_stop fixed with locking (611313)
5863 - O_APPEND failures resolved (618273, perhaps 614459)
5864 - lconf PATH fix (619770)
5865 - IA64 build fix (621450)
5866 - RPC buffer sizes scale with amount of memory
5868 2002-10-01 Phil Schwan <phil@clusterfs.com>
5871 - locks would be cancelled without throwing away data pages,
5872 resulting in inconsistent data (605627)
5873 - inode attributes were not always being refreshed (605627, 612449)
5874 - lconf now continues to cleanup after lctl reports an error
5875 - MDS now enforces user permissions (602707)
5876 - lprocfs cleanup fixed, but not yet enabled (614157)
5877 - fixed infinite server hang, should a client not respond to an AST
5878 - avoid going into recovery if user calls readlink() with a buffer
5879 that's too small (613941)
5880 - AST RPCs no longer require replies (614867) -- this may be changed
5881 - don't crash server if client sends an IOV that's too big (611336)
5882 - fixed lock conversion deadlock (611892)
5883 - fixed the following of symlinks (614622)
5884 * recovery: the server can remove locks from a client that dies, other
5885 clients can make progress
5886 * more extN patch fixes
5887 * compile-time configurable ptlrpc buffer allocations
5889 - collaborative read cache document
5890 - Lustre Lite Performance CDR document-in-progress
5892 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5895 - fix typo in patch-2.4.18
5897 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5900 - clear ptlrpc request each time in handle_incoming_request()
5901 - unlink of files now destroys the object on the OST
5903 2002-09-19 Peter Braam <braam@clusterfs.com>
5905 * add hard link support
5906 * change obdfile creation method
5907 * kernel patch changed
5909 2002-09-19 Peter Braam <braam@clusterfs.com>
5912 - stack overflow bug in extN fixed
5914 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5916 * documentation updates
5917 - add man pages for config tools
5918 - update tests/README to describe testing with new config tools
5919 - finish metadata API descriptions
5920 * bug fixes and cleanups
5921 - statfs workaround for 16TB limit
5922 - LOV stripe allocation improved, can stripe on subset of OSTs
5923 - LOV file size/IO offset was wrong for files > 4GB in size
5924 - object EA data was being dropped, caused files to be unreadable
5925 - memory overflow with non-LOV OST caused memory corruption
5926 - fixed regression tests to work with new config tools, obdfilter
5927 - fixed bug when directory size became larger than 1 block
5928 - fixed bug (for single client case) when PWD was deleted
5929 - invalidate local directory pages when doing intent-based ops
5930 - avoid LDLM oops when lock callback contained bad data
5932 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5934 * documentation updates
5935 * bug fixes and cleanups
5936 - configuration tools
5939 - 64-bit compile warnings
5940 - 64-bit internal statfs data
5942 * test_brw on persistent OST devices
5944 * lprocfs (disabled)
5946 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5948 * documentation updates
5949 * bug fixes and cleanups
5950 * configuration tools
5952 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5955 * many small fixes to 0_5_4
5956 * io/network handling
5957 * thinkos in MDS operations
5959 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5962 * crucial basic fixes to 0.5.3
5963 * IOR, Iozone work over Elan
5966 2002-08-07 Phil Schwan <phil@clusterfs.com>
5967 * version 0_5_3, our first alpha
5968 * we use the new Portals iovs
5969 * documentation updates
5970 * bug fixes and cleanups
5971 * small changes in the DLM wire protocol
5973 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5974 * version 0_5_1 with some initial stability,
5975 * locking on MD and file I/O.
5976 * documentation updates
5977 * several bug fixes since 0.5.0
5978 * small changes in wire protocol
5980 2002-07-18 Phil Schwan <phil@clusterfs.com>
5982 * delivered as Lustre Light Alpha
5983 * fixed a crash after handling invalid MDS requests
5984 * fixed directory pages for architectures with non-4k pages sizes
5986 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5987 * release version v0_4_4
5988 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5990 * Fixes a number of interruption problems with OST operations.
5991 * Update documentation for portals header changes
5992 * Move all wire protocol structs/defines to lustre_idl.h
5993 * Fixes symlink length bug.
5994 * Add tcpdump to repository.
5996 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5997 * release version v0_4_3
5998 * Fixes statfs for inodes on extN.
5999 * Fixes bug in runtests which would delete /etc/hosts.
6000 * Use 64-bit object IDs wherever possible (not into VFS though)
6001 Remove ost_get_info, which is unused by lustre, and out of date.
6003 2002-07-03 Peter Braam <braam@clusterfs.com>
6004 * release version v0_4_2 Fixes a lookup error (type not passed)
6005 * move forward to head of Portals
6006 * move forward to latest Lustre kernel
6008 2002-06-25 Peter Braam <braam@clusterfs.com>
6009 * release version v0_4_1. Hopefully stable on single node use.