1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (SLES 10),
5 2.6.18-128.7.1.el5 (RHEL 5),
6 2.6.18-128.7.1.el5 (OEL 5),
7 2.6.22.14 vanilla (kernel.org).
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.16 - 2.6.21 vanilla (kernel.org)
11 * Recommended e2fsprogs version: 1.41.6.sun1
12 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
13 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
14 removed cwd "./" (refer to Bugzilla 14399).
15 * File join has been disabled in this release, refer to Bugzilla 16929.
17 Severity : enhancement
19 Description: correctly handle big reply message.
20 Details : send LNet event if reply is bigger then buffer and adjust this buffer
25 Description: Drop unnecessary __GFP_NOMEMALLOC flag from filter_get_page()
27 Severity : enhancement
29 Description: Add support for OEL5.
33 Description: Conf-sanity.sh 50g test - deactivated OST should not cause a panic.
37 Description: client could not reconnect to OST because of an active request.
38 Details : The client now retries to reconnect to the same server, if a
39 connect request failed with EBUSY or -EAGAIN. Test case 35b
40 conf-sanity.sh has been created.
44 Description: Don't allow make backward step on assignin osc next id.
45 Details : race between allocation next id and ll_sync thread can be cause
46 of set wrong osc next id and can be kill valid ost objects.
50 Description: Changes in raid5-large-io-rhel5.patch to calculate sectors properly
54 Description: Increase the default BLK_DEF_MAX_SECTORS value for RHEL5 and SLES11
58 Description: Error handling in osc_statfs_interpret() has been improved.
59 Details : Check in osc_statfs_interpret() for EBADR.
63 Description: Do not send statfs() requests to OSTs disabled by administrator.
64 Details : Check in lov_prep_statfs_set() for non-NULL ltd_exp.
66 Severity : enhancement
68 Description: Added sanity-quota test to for lfs(1) to work with numeric uid/gid.
70 Severity : enhancement
72 Description: Update kernel to RHEL5 2.6.18-128.7.1.el5.
74 Severity : enhancement
76 Description: Enhancement for lfs(1) command to use numeric uid/gid.
80 Description: Improved cleanup procedure for conf-sanity test_46a.
84 Description: Do not start bulk transfer, if client tried to reconnect.
85 Details : Do not start bulk transfer, if client tried to reconnect,
86 but failed due to active rpcs. Check for exp_abort_active_req
91 Description: Prevent inconsistences between linux and lustre mount structures.
92 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
93 Make the sleep interruptible.
95 Severity : enhancement
97 Description: provide server to client comms path
98 Details : server to client communications path via new LDLM_SET_INFO rpc
102 Description: truncate starts GFP_FS allocation under transaction causing deadlock
103 Details : ldiskfs_truncate calls grab_cache_page which may start page
104 allocation under an open transaction. This may lead to
105 calling prune_icache with consequent lustre reentrance.
109 Description: send correctly lsm on open replay
110 Details : MDS is trust to LSM size on replay open, but client can set wrong size
115 Description: enable client interface failover
116 Details : When a child reconnects from another NID, properly update export
117 nid hash position and ldlm reverse import.
119 Severity : enhancement
121 Description: Communicate OST degraded/readonly state via statfs to MDS
122 Details : Flags in the statfs returned from OSTs indicate whether the
123 OST is in a degraded RAID state, or if the filesystem has
124 turned read-only after a filesystem error is detected.
129 Description: don't panic if EPROTO was hit when reading symlink
130 Details : correctly handling request reference in error cases.
132 Severity : enhancement
134 Description: Add LustreNetLink, a kernel-userspace communcation path.
136 Severity : enhancement
138 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
143 Description: access to llog context before init.
144 Details : move handling CATALOGS file at osc layer and forbid access to llog
148 Frequency : rare, only when client is writing to OST on local node
150 Description: Avoid deadlock for local client writes
151 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
152 memory freeing context. This allows OST threads to set the
153 PF_MEMALLOC flag on task structures in order to allocate memory
154 from reserved pools and complete IO.
155 Use GFP_HIGHUSER for OST allocations for non-local client writes,
156 so that the OST threads generate memory pressure and allow
157 inactive pages to be reclaimed.
159 Severity : enhancement
161 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
166 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
167 Details : move ldlm namespace creation in setup phase to avoid grab
168 _lprocfs_lock with cli_sem held.
172 Description: Temporarily disable grant shrink.
173 Details : Disable the feature for debugging.
176 Frequency : only when formatting test filesystems
178 Description: Unable to run several mkfs.lustre on loop devices at the same time
179 Details : mkfs.lustre returns error 256 on the concurrent loop devices
180 formatting. The solution is to proper handle the error.
182 Severity : enhancement
184 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
186 Severity : enhancement
188 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
193 Description: resolve race between obd_disconnect and class_disconnect_exports
194 Details : if obd_disconnect will be called to already disconnected export he
195 forget release one reference and osc module can't unloaded.
197 Severity : enhancement
199 Description: Allow tuning service thread via /proc
200 Details : For each service a new
201 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
202 created that can be used to set min/max thread counts, and get the
203 current number of running threads.
208 Descriptoin: don't return error if have partially created objects for file.
209 Details : lov_update_create_set uses set->set_success as index for created
210 objects, so if some requests will be failed, they will have hole
211 at end of array and qos_shrink_lsm for allocate correct lsm.
213 Severity : enhancement
215 Description: Update OFED support to 1.4.1
220 Description: Reduce small size read RPC
221 Details : Set read-ahead limit for every file and only do read-ahead when
222 available read-ahead pages are bigger than 1M to avoid small size
225 Severity : enhancement
227 Description: move AT tunable parameters for more consistent usage
228 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
230 Severity : enhancement
232 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
233 Details : allow skip disconnected ost for send statfs request and hide error
239 Description: fix racy locking of mballoc block bitmaps causing BUG
240 Details : The locking of the mballoc buddy bitmap and the in-memory
241 block bitmap was using two different spin locks in some
242 cases. This made it possible to incorrectly access the
243 mballoc bitmap while another process was modifying it,
244 causing a sanity assertion to fail. While no on-disk corruption
245 was reported, there was some risk of this happening.
248 Frequency : rare, on llog test 6
250 Descriptoin: don't allow connect to already connected import
251 Details : allowing connect to already connected import is hide connecting problem.
256 Descriptoin: allow kill process which wait statahead result
257 Details : for some reasons 'ls' can stick in waiting result from statahead,
258 in this case need way for kill this process.
260 Severity : enhancement
262 Description: Add state history info file, enhance import info file
263 Details : Track import connection state changes in a new osc/mdc proc file;
264 add overview-type data to the osc/mdc import proc file.
266 Severity : enhancement
268 Description: MDS create should not wait for statfs RPC while holding DLM lock.
270 Severity : enhancement
272 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
277 Descriptoin: OST grant too much space to client even there are not enough space.
278 Details : Client will shrink its grant cache to OST if there are no write
279 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
280 this grant cache if there are already not enough avaible space
281 (left_space < total_clients * 32M).
284 Frequency : start MDS on uncleanly shutdowned MDS device
286 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
287 Details : stay in waiting mds<>ost recovery finished produce random bugs
288 due race between two ll_sync thread for one lov target. send
289 ACTIVATE event only if connect realy finished and import have
293 Frequency : rare, connect and disconnect target at same time
295 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
296 Details : don't call obd_disconnect under lov_lock. this long time
297 operation and can block ptlrpcd which answer to connect request.
302 Descriptoin: don't lose wakeup for imp_recovery_waitq
303 Details : recover_import_no_retry or invalidate_import and import_close can
304 both sleep on imp_recovery_waitq, but we was send only one wakeup
308 Frequency : always with long access acl
310 Descriptoin: mds can't pack reply with long acl.
311 Details : mds don't control size of acl but they limited by reint/getattr
314 Severity : enhancement
316 Description: Update to SLES10 kernel-2.6.16.60-0.33.
318 Severity : enhancement
320 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
323 Frequency : start MDS on uncleanly shutdowned MDS device
325 Descriptoin: aborting recovery hang on MDS
326 Details : don't throttle destroy RPCs for the MDT.
329 Frequency : on remount
331 Description: external journal device not working after the remount
332 Details : clear dev_rdonly flag for external journal devices in
338 Description: shutdown vs evict race
339 Details : client_disconnect_export vs connect request race.
340 if client will evicted at this time - we start invalidate
341 thread without referece to import and import can be freed
345 Frequency : rare, need acl's on inode.
347 Description: client can't handle ost additional correctly
348 Details : if ost was added after client connected to mds client can have
349 hit lnet_try_match_md ... to big messages to wide striped files.
350 in this case need teach client to handle config events about add
351 lov target and update client max ea size at that event.
353 Severity : enhancement
355 Description: Changelogs
356 Details : Changelogs are a lightweight mechanism to track filesystem
357 metadata and namespace changes. The changelog is recorded
358 permanently on the MDTs, and is periodically "consumed" / purged
359 when records are no longer needed.
361 Severity : enhancement
363 Description: compact fld format with extents
364 Details : Store range of seq rather than every seq in FLD. Seq
365 controller update FLD rather than clients. In Case of CMD, mdt0
366 has FLD, all other metadata server act as non persistent proxy
367 for FLD queries and cache fld entries in fld cache.
372 Description: don't skip ost target if they assigned to file
373 Details : Drop slow OSCs if we can, but not for requested start idx.
374 This means "if OSC is slow and it is not the requested
375 start OST, then it can be skipped, otherwise skip it only
376 if it is inactive/recovering/out-of-space.
380 Description: more cleanup in mds_lov
381 Details : not send LOV EA under replay, we can't know about they size at this
382 time. Don't allow client connect to mds before any ost connected,
383 for avoid problems with LOV EA size and returning EIO to client.
385 Severity : enhancement
387 Description: Interoperability at server side (Disk interoperability)
389 Severity : enhancement
391 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
393 Severity : enhancement
395 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
397 Severity : enhancement
399 Description: New client IO stack (CLIO).
401 Severity : enhancement
403 Description: Commit on sharing. Eliminate inter-client dependencies between
404 uncommitted transactions by doing transaction commits.
405 Thereby clients may recovery independently.
408 Frequency : Create a symlink file with a very long name
410 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
411 Details : If there is no extra space in the request for early cancels,
412 ldlm_req_handles_avail() returns 0 instead of a negative value.
414 Severity : enhancement
416 Description: Add /proc entry for import status
417 Details : The mdc, osc, and mgc import directories now have
418 an import directory that contains useful import data for debugging
421 Severity : enhancement
423 Description: Re-disable certain /proc logging
424 Details : Enable and disable client's offset_stats, extents_stats and
425 extents_stats_per_process stats logging on the fly.
428 Frequency : Only on FC kernels 2.6.22+
430 Description: oops in statahead
431 Details : Do not drop reference count for the dentry from VFS when lookup,
432 VFS will do that by itself.
434 Severity : enhancement
436 Description: Generic /proc file permissions
437 Details : Set /Proc file permissions in a more generic way to enable non-
438 root users operate on some /proc files.
442 Description: Hitting mdc_commit_close() ASSERTION
443 Details : Properly handle request reference release in
444 ll_release_openhandle().
448 Description: quota recovery deadlock during mds failover
449 Details : This patch includes att18982, att18236, att18237 in bz14840.
451 1. fix osts hang when mds does failover with quotaon
452 2. prevent watchdog storm when osts threads wait for the
457 Frequency : only patchless client
458 Description: add workaround for race between add/remove dentry from hash
460 Severity : enhancement
462 Description: Allow OST glimpses to return PW locks
466 Description: LBUG when llog conf file is full
467 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
471 Description: Prevent import from entering FULL state when server in recovery
475 Description: service mount cannot take device name with ":"
476 Details : Only when device name contains ":/" will mount treat it as
482 Description: replace ptlrpcd with the statahead thread to interpret the async
483 statahead RPC callback
487 Frequency : on recovery
488 Description: I/O failures after umount during fail back
489 Details : if client reconnected to restarted server we need join to recovery
490 instead of find server handler is changed and process self
491 eviction with cancel all locks.
493 Severity : enhancement
495 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
497 Severity : enhancement
499 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
501 Severity : enhancement
503 Description: Upcall on Lustre log has been dumped
504 Details : Allow for a user mode script to be called once a Lustre log has
505 been dumped. It passes the filename of the dumped log to the
506 script, the location of the script can be specified via
507 /proc/sys/lnet/debug_log_upcall.
512 Description: avoid idr_remove called for id which is not allocated.
513 Details : Move assigment s_dev for clustered nfs to end of initialization,
514 to avoid problem with error handling.
519 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
520 Details : When connection is reused this not moved from CONN_UNUSED_HASH
521 into CONN_USED_HASH and this prodice warning when put connection
522 again in unused hash.
525 Severity : enhancement
527 Description: File striping can now be set to use an arbitrary pool of OSTs.
529 Severity : enhancement
531 Description: Export bytes_read/bytes_write count on OSC/OST.
535 Description: Early reply size mismatch, MGC loses connection
536 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
537 the connect flags are properly negotiated.
542 Description: quotas are not honored with O_DIRECT
543 Details : all writes with the flag O_DIRECT will use grants which leads to
544 this problem. Now using OBD_BRW_SYNC to guard this.
548 Description: add quota statistics
549 Details : 1. sort out quota proc entries and proc code.
550 2. add quota statistics
552 Severity : enhancement
554 Description: enable quota support for HEAD.
558 Description: Properly propagate oinfo flags from lov to osc for statfs
559 Details : restore missing copy oi_flags to lov requests.
561 Severity : enhancement
563 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
564 : l_getgroups(8), lst(8), routerstat(8)
565 Details : included man pages for llobdstat(8), llstat(8),
566 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
568 Severity : enhancement
570 Description: configure's --enable-quota should check the
571 : kernel .config for CONFIG_QUOTA
572 Details : configure is terminated if --enable-quota is passed but
573 : no quota support is in kernel
577 Description: Remove portals compatibility
578 Details : Remove portals compatibility, not interoperable with releases
583 Description: Resolve device initialization race
584 Details : Prevent proc handler from accessing devices added to the
585 obd_devs array but yet be intialized.
587 Severity : enhancement
589 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
591 Severity : enhancement
593 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
598 Description: Using wrong pointer in osc_brw_prep_request
599 Details : Access to array[-1] can produce panic if kernel compiled with
600 CONFIG_PAGE_ALLOC enabled
604 Description: Client runs out of low memory
605 Details : Consider only lowmem when counting initial number of llap pages
609 Description: *optional* service tags registration
610 Details : if the "service tags" package is installed on a Lustre node
611 When the filesystem is mounted, a local-node service tag will
612 be created. See http://inventory.sun.com/ for more information
613 about the Service Tags asset management system.
617 Description: Kernel BUG tries to release flock
618 Details : Lustre does not destroy flock lock before last reference goes
619 away. So always drop flock locks when client is evicted and
620 perform unlock regardless of successfulness of speaking to MDS.
624 Description: add refcount protection for osc callbacks, avoid panic on shutdown
628 Description: sanity test 65a fails if stripecount of -1 is set
629 Details : handle -1 striping on filesystem in ll_dirstripe_verify
634 Description: ASSERTION(CheckWriteback(page,cmd)) failed
635 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
638 Severity : enhancement
640 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
644 Description: do not process already freed flock
645 Details : flock can possibly be freed by another thread before it reaches
646 to ldlm_flock_completion_ast.
650 Description: LBUG during stress test
651 Details : Need properly lock accesses the flock deadlock detection list.
655 Description: oops in page fault handler
656 Details : kernel page fault handler can return two special 'pages' in error
657 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
661 Description: timeout with invalidate import.
662 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
663 should be handled by ptlrpcd. This produce long age waiting and
664 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
666 Severity : enhancement
668 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
672 Description: enable MGS and MDT services start separately
673 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
674 a co-located MGS without starting the MGS, which is a complement
675 to 'nosvc' mount option.
679 Frequency : after recovery
680 Description: precreate to many object's after del orphan.
681 Details : del orphan st in oscc last_id == next_id and this triger growing
682 count of precreated objects. Set flag LOW to skip increase count
683 of precreated objects.
687 Frequency : rare, on clear nid stats
688 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
689 Details : when clean nid stats sometimes try destroy live entry,
690 and this produce panic in free.
694 Description: Stack overflow during MDS log replay
695 ease stack pressure by using a thread dealing llog_process.
699 Description: wait until IO finished before start new when do lock cancel.
700 Details : VM protocol want old IO finished before start new, in this case
701 need wait until PG_writeback is cleared until check dirty flag and
702 call writepages in lock cancel callback.
704 Severity : enhancement
706 Description: using special macro for print time and cleanup in includes.
710 Description: mds_mfd_close() ASSERTION(rc == 0)
711 Details : In mds_mfd_close(), we need protect inode's writecount change
712 within its orphan write semaphore to prevent possible races.
716 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
720 Frequency : rare, on shutdown ost
721 Description: don't hit live lock with umount ost.
722 Details : shrink_dcache_parent can be in long loop with destroy dentries,
723 use shrink_dcache_sb instead.
727 Description: don't panic with use echo client
728 Details : echo client pass NULL as client nid pointer and this produce null
733 Description: fix build on ppc32
734 Details : compile code with -m64 flag produce wrong object file for ppc32.
738 Description: add message levels for liblustreapi
742 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
743 Details : Move the 'good_osts' check before the 'total_bavail' check. This
744 will result in an -EAGAIN and in the exit call path we call
745 alloc_rr() which will with increasing aggressiveness attempt to
746 aquire precreated objects on the minimum number of required OSCs.
750 Description: Use old size assignment to avoid deadlock
751 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
752 the scheduling while holding a spinlock. We do not need locking
753 for size in ll_update_inode() because size is only updated from
754 the MDS for directories or files without objects, so there is no
755 other place to do the update, and concurrent access to such inodes
756 are protected by the inode lock.
760 Description: resolve "_IOWR redefined" build error on SLES10
764 Description: dump the memory debugging after all modules are unloaded to
765 suppress false negative in conf_sanity test 39
767 Severity : enhancement
769 Description: build kernel-ib packages for OFED 1.3 in our release cycle
774 Description: fix SLES kernel versioning
775 Details : the kernel version for our SLES 10 kernel did not include a "-"
776 before the "smp" at the end. while this was not a problem in
777 general, it did mean that software trying to use the kernel
778 version to try to detect a vendor specific kernel would fail.
779 this was most evident by the OFED build scripts.
783 Description: Don't update lov_desc members until making sure they are valid
784 Details : When updating lov_desc members via proc fs, need fix their
785 validities before doing the real update.
789 Description: don't put request into delay list while invalidate in flight.
790 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
791 import in flight. this produce timeout for invalidate and sometimes
792 can cause stale data.
796 Frequency : on ppc only
797 Description: not convert ost objects for directory because it's not exist.
798 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
802 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
804 Severity : enhancement
806 Description: Update to support 2.6.22.14 vanilla kernel.
810 Frequency : rare, on recovery
811 Description: read procfs can produce deadlock in some situation
812 Details : Holding lprocfs lock which send rpc can produce block for destroy
813 obd objects and this also block reconnect with -EALREADY. This
814 isn't fix all lprocfs bugs - but make it rare.
816 Severity : enhancement
818 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
821 Frequency : frequent on X2 node
823 Description: mdc_set_open_replay_data LBUG
824 Details : Set replay data for requests that are eligible for replay.
828 Description: lustre_mgs: operation 101 on unconnected MGS
829 Details : When MGC is disconnected from MGS long enough, MGS will evict the
830 MGC, and late on MGC cannot successfully connect to MGS and a lot
831 of the error messages complaining that MGS is not connected.
835 Frequency : on network error
836 Description: panic with double free request if network error
837 Details : mdc_finish_enqueue is finish request if any network error ocuring,
838 but it's true only for synchronus enqueue, for async enqueue
839 (via ptlrpcd) this incorrect and ptlrpcd want finish request
842 Severity : enhancement
844 Description: client-side metadata stat-ahead during readdir(directory readahead)
845 Details : perform client-side metadata stat-ahead when the client detects
846 readdir and sequential stat of dir entries therein
849 Frequency : on start mds
851 Description: Implement get_info(last_id) in obdfilter.
854 Frequency : occasional
856 Description: Correctly check stale fid, not start epoch if ost not support SOM
857 Details : open with flag O_CREATE need set old fid in op_fid3 because
858 op_fid2 was overwritten with new generated fid, but mds can answer
859 with one of these two fids and both is not stale. Setattr
860 incorrectly started an epoch and assume will be called
861 done_writing, but without SOM done_writing ever being called.
864 Frequency : rare, depends on device drivers and load
866 Description: MDS or OSS nodes crash due to stack overflow
867 Details : Code changes in 1.8.0 increased the stack usage of some functions.
868 In some cases, in conjunction with device drivers that use a lot
869 of stack the MDS (or possibly OSS) service threads could overflow
870 the stack. One change which was identified to consume additional
871 stack has been reworked to avoid the extra stack usage.
874 Frequency : occasional
876 Description: Do not fail import if osc_interpret_create gets -EAGAIN
877 Details : If osc_interpret_create got -EAGAIN it immediately exits and
878 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
879 oscc_has_objects and see OSC has no objests and call
880 oscc_internal_create to resend create request.
882 Severity : enhancement
884 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
886 Severity : enhancement
888 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
891 Frequency : very rare
893 Description: Fix replay if there is an un-replied request and open
894 Details : In some cases, older replay request will revert the
895 mcd->mcd_last_xid on MDS which is used to record the client's
898 Severity : enhancement
900 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
902 Severity : enhancement
904 Description: Add rhel5 support to HEAD.
906 Serverity : enhancement
908 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
913 Description: Don't allow skipping OSTs if index has been specified.
914 Details : Don't allow skipping OSTs if index has been specified, make
915 locking in internal create lots better.
919 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
920 Details : access to bitfield in structure is always rounded to long
921 and this produce problem with not atomic change any bit.
925 Description: Lustre make rpms failed.
926 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
927 builds Lustre rpms from tarball.
930 Frequency : rare on shutdown ost
932 Description: If llog cancel was not send before clean_exports phase, this can
933 produce deadlock in llog code.
934 Details : If llog thread has last reference to obd and call class_import_put
935 this produce deadlock because llog_cleanup_commit_master wait when
936 last llog_commit_thread exited, but this never success because was
937 called from llog_commit_thread.
941 Description: allow userland application know is lost one of stripes.
942 Details : fill lvb_blocks with error code on ost and return it to
943 application if error flag found.
947 Description: NULL lov_tgts causing MDS oops
948 Details : more safe checks for NULL lov_tgts for avoid oops.
950 Severity : enhancement
952 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
956 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
957 Details : If an OST is down the MDS will hang indefinitely in
958 obd_statfs_rqset() waiting for the statfs data. While for
959 MDS QOS usage of statfs, it should not stuck in waiting.
961 Severity : enhancement
963 Description: remote_acl support
964 Details : Support ACL-based permission check for remote user.
965 Support setfacl/getfacl for remote user with the utils
966 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
967 the system "{s,g}etfacl" utils.
969 Severity : enhancement
971 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
973 Severity : enhancement
975 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
979 Description: make mgs_setparam() handle fsname containing dash
980 Details : fsname containing a dash does not work with lctl conf_param
982 Severity : enhancement
984 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
986 Severity : enhancement
988 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
992 Frequency : rare, at shutdown
993 Description: access already free / zero obd_namespace.
994 Details : if client_disconnect_export was called without force flag set,
995 and exist connect request in flight, this can produce access to
996 NULL pointer (or already free pointer) when connect_interpret
997 store ocd flags in obd_namespace.
1001 Frequency : only at startup
1002 Description: not alloc memory with spinlock held.
1003 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
1004 if any spinlock held.
1006 Severity : enhancement
1008 Description: make lustre randomly fail allocating memory
1009 Details : Make lustre randomly failed allocating memory for testing purpose.
1011 Severity : enhancement
1013 Description: lost problems with lov objid file
1014 Details : Fixes some scability and access to not inited memory problems
1015 in work with lov objdid file.
1020 Description: lfs find does not continue on file error
1021 Details : Continue other files processing when a file/dir is absent.
1025 Description: Inconsistent usage of lustre_pack_reply()
1026 Details : Standardize the usage of lustre_pack_reply() such that it
1027 always generate a CERROR on failure.
1030 Frequency : occasional
1032 Description: MDS hang or stay in waiting lock
1033 Details : If client receive lock with CBPENDING flag ldlm need send lock
1034 cancel as separate rpc, to avoid situation when cancel request
1035 can't processed due all i/o threads stay in wait lock.
1039 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
1043 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
1047 Description: Update to SLES9 kernel-2.6.5-7.287.3.
1051 Description: Update to RHEL4 latest kernel.
1053 Severity : enhancement
1055 Description: Build SLES10 patchless client fails
1056 Details : The configure was broken by run ./configure with
1057 --with-linux-obj=.... argument for patchless client. When the
1058 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
1059 can't search header adequately. Use absolute path such as
1060 -I($LINUX)/include instead.
1064 Description: interrupt oig_wait produce painc on resend.
1065 Details : brw_redo_request can be used for resend requests from ptlrpcd and
1066 private set, and this produce situation when rq_ptlrpcd_data not
1067 copyed to new allocated request and triggered LBUG on assert
1068 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
1069 ptlrpcd set if request is changed and can be safety changed to use
1074 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1075 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1076 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1077 these macros and replace them with their difinitions instead.
1081 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1082 Details : Update which_patch & target file for SLES10 latest kernel.
1085 Description: Remove client patches from SLES 10 kernel.
1086 Details : This causes SLES 10 clients to behave as patchless clients
1087 even on a Lustre-patched (server) kernel.
1089 Severity : enhancement
1091 Description: self-adjustable client's lru lists
1092 Details : use adaptive algorithm for managing client cached locks lru
1093 lists according to current server load, other client's work
1094 pattern, memory activities, etc. Both, server and client
1095 side namespaces provide number of proc tunables for controlling
1098 Severity : enhancement
1100 Description: light-weight GSS support
1101 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1102 and reduce performance overhead.
1104 Severity : enhancement
1106 Description: Linux keyring support
1107 Details : Support using service of Linux keyring for Lustre GSS internal
1108 context refresh/cache mechanism.
1112 Description: Fix errors in lfs documentation
1113 Details : Fixes man pages
1117 Description: don't use GFP_* in generic Lustre code.
1118 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1122 Description: obdclass is limited by single OBD_ALLOC(idarray)
1123 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1127 Description: warnings with build patchless client with vanila 2.6.19 and up
1128 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1129 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1133 Frequency : Only for SLES
1134 Description: sanity_quota fail test_1
1135 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1136 file, which makes TSTID[2] inunique.
1140 Frequency : Only for SLES9
1141 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1142 Details : Update target/ChangeLog/which_patch .
1146 Frequency : Only for test_10 in sanity-quota.sh
1147 Description: a bug in quota test code
1148 Details : checking a test flag in a wrong place
1152 Frequency : Only for test_13 in sanity-quota.sh
1153 Description: a bug in quota test script
1154 Details : shouldn't check overfull allocated quota
1158 Description: jbd statistics
1159 Details : Port older jbd statistics patch for sles10
1161 Severity : enhancement
1163 Description: Add printing inode info into message about error in writepage.
1168 Description: reply_lock_interpret crash due to race with it and lock cancel.
1169 Details : Do not replay locks that are being cancelled. Do not reference
1170 locks by their address during replay, just by their handle.
1174 Frequency : When flocks are used.
1175 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1176 Details : Flock locks might destroy just granted lock if it could be merged
1177 with another existing flock, this is done in completion handler,
1178 so teach ldlm_cli_enquque_fini that this is a valid case for
1184 Description: Oops in read and write path when failing to allocate lock.
1185 Details : Check if lock allocation failed and return error back.
1189 Description: lstripe command fails for valid OST index
1190 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1191 instead of lov->desc.ld_active_tgt_count.
1195 Description: sanity.sh failed test 103
1196 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1201 Description: df doesn't work properly if diskfs blocksize != 4K
1202 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1206 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1207 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1209 Severity : enhancement
1211 Description: Adaptive timeouts
1212 Details : RPC timeouts adapt to changing server load and network
1213 conditions to reduce resend attempts and improve recovery time.
1217 Description: llapi_file_create() does not allow some changes
1218 Details : add llapi_file_open() that allows specifying the mode and
1219 open flags, and also returns an open file handle.
1223 Description: merge and cleanup kernel patches.
1224 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1228 Description: Add journal checksum support.(Kernel part)
1229 Details : The journal checksum feature adds two new flags i.e
1230 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1231 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1232 indicates that the commit block contains the checksum for
1233 the blocks described by the descriptor blocks. Now commit
1234 record can be sent to disk without waiting for descriptor
1235 blocks to be written to disk. This behavior is controlled
1236 using JBD2_FEATURE_ASYNC_COMMIT flag.
1240 Description: OSS needs mutliple precreate threads
1241 Details : Add ability to start more than one create thread per OSS.
1245 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1246 Details : execute lfs setstripe on client
1250 Description: mds_obd_create error creating tmp object
1251 Details : When the user sets quota on root, llog will be affected and can't
1252 create files and write files.
1255 Frequency : Always on ia64 patchless client, and possibly others.
1257 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1258 Details : This allows the patchless client to be loaded on architectures
1259 without this export.
1263 Description: RedHat Update kernel for RHEL5
1264 Details : Add the kernel config file for RHEL5.
1268 Description: RedHat Update kernel for RHEL5
1269 Details : Modify the kernel config file more closer RHEL5.
1273 Description: Build failure against Centos5 (RHEL5)
1274 Details : Define PAGE_SIZE when it isn't present.
1278 Description: test_7 on sanity-quota.sh can't be used on separate machines
1279 Details : add facet to handle it
1283 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1284 Details : with NFS, the anon dentry's parent was set to itself in
1285 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1286 corresponding dentry other than use rec->ur_name.
1288 Severity : enhancement
1290 Description: lfs setstripe enhancement
1291 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1295 Description: enable data checksumming by default
1296 Details : enable checksum by default, allow --disable-checksum
1297 configure option and "-o nochecksum" mount option. Checksums
1298 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1302 Description: lustre support for RHEL5
1303 Details : Add support for RHEL5.
1307 Description: Client eviction due to failover config
1308 Details : after a connection loss, the lustre client should attempt to
1309 reconnect to the last active server first before trying the
1310 other potential connections.
1314 Description: when mds and osts use different quota unit(32bit and 64bit),
1315 quota will be released repeatly.
1316 Details : void sending multiple quota reqs to mds, which will keep the
1317 status between the reqs.
1321 Description: osts not allocated evenly to files
1322 Details : change the condition to increase offset_idx
1325 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1327 Description: Data corruption for OSTs that are formatted larger than 2TB
1329 Details : When generating the bio request for lustre file writes the
1330 sector number would overflow a temporary variable before being
1331 used for the IO. The data reads correctly from Lustre (which
1332 will overflow in a similar manner) but other file data or
1333 filesystem metadata may be corrupted in some cases.
1337 Description: Tune the kernel for good SCSI performance.
1338 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1339 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1344 Description: rewrite ext2-derived code in obdclass/uuid.c
1345 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1346 from scratch preserving functionality.
1350 Description: "lfs find -obd UUID" prints directories
1351 Details : "lfs find -obd UUID" will return all directory names instead
1352 of just file names. It is incorrect because the directories
1353 do not reside on the OSTs.
1357 Description: use i_size_read and i_size_write in 2.6 port
1358 Details : replace inode->i_size access with i_size_read/write()
1361 Frequency : when removing large files
1363 Description: scheduling issue during removal of large Lustre files
1364 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1365 It causes scheduling issues when removing large files (17TB in the
1369 Frequency : only with liblustre clients on XT3
1371 Description: evictions taking too long
1372 Details : allow llrd to evict clients directly on OSTs
1375 Frequency : only on ppc
1377 Description: /proc/fs/lustre/devices broken on ppc
1378 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1379 We should make sure the type of 'index' is loff_t and then cast
1380 explicitly as needed below. Do not assign an explicitly cast
1384 Frequency : only for rhel5
1386 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1387 Details : Modify the target file & which_kernel.
1389 Severity : enhancement
1391 Description: omit set fsid for export NFS
1392 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1393 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1397 Frequency : Always, for kernels after 2.6.16
1398 Description: Fix warning idr_remove called for id=.. which is not allocated.
1399 Details : Last kernels save old s_dev before kill super and not allow
1400 to restore from callback - restore it before call kill_anon_super.
1404 Description: buffer overruns could theoretically occur
1405 Details : llapi_semantic_traverse() modifies the "path" argument by
1406 appending values to the end of the origin string, and a
1407 overrun may occur. Adding buffer overrun check in liblustreapi.
1411 Description: Fix error on 'ls .' at the top of the Lustre mount.
1412 Details : Don't revalidate dentry if it is a root dentry.
1416 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1417 Details : Modify vm-tunables-rhel4.patch.
1421 Description: Kernel config for 2.6.18-vanilla.
1422 Details : Modify targets/2.6-vanilla.target.in.
1423 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1424 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1425 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1426 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1430 Description: improve handling recoverable errors
1431 Details : if request processig with error which can be recoverable on server
1432 request should be resend, otherwise page released from cache and
1437 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1438 Details : Modify target file & which_patch.
1439 A flaw was found in the IA32 system call emulation provided
1440 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1441 value could be stored in the %RAX register, which could trigger an
1442 out-of-bounds system call table access. An untrusted local user
1443 could exploit this flaw to run code in the kernel
1444 (ie a root privilege escalation). (CVE-2007-4573).
1448 Description: O_DIRECT bypasses client statistics.
1449 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1450 read_ahead_stats were updated. copied the stats section from
1451 osc_send_oap_rpc() into async_internal().
1455 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1459 Description: change order of libsysio includes
1460 Details : '#include sysio.h' should always come before '#include xtio.h'
1464 Description: handle "serious error: objid * already exists" more gracefully
1465 Details : If LAST_ID value on disk is smaller than the objects existing in
1466 the O/0/d* directories, it indicates disk corruption and causes an
1467 LBUG(). If the object is 0-length, then we should use the existing
1468 object. This will help to avoid a full fsck in most cases.
1470 Severity : enhancement
1472 Description: adapt the lustre_config script to support the upgrade case
1473 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1474 server targets to 1.6 in parallel.
1478 Description: To avoid grant space > avaible space when the disk is almost
1479 full. Without this patch you might see the error "grant XXXX >
1480 available" or some LBUG about grant, when the disk is almost
1482 Details : In filter_check_grant, for non_grant cache write, we should
1483 check the left space by if (*left > ungranted + bytes), instead
1484 of (*left > ungranted), because only we are sure the left space
1485 is enough for another "bytes", then the ungrant space should be
1486 increase. In client, we should update cl_avail_grant only there
1487 is OBD_MD_FLGRANT in the reply.
1491 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1492 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1493 described in CVE-2007-4573. This problem could allow untrusted
1494 local users to gain root access.
1497 Frequency : when using O_DIRECT and quotas
1499 Description: Incorrect file ownership on O_DIRECT output files
1500 Details : block usage reported by 'lfs quota' does not take into account
1501 files that have been written with O_DIRECT.
1506 Description: touch file failed when fs is not full
1507 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1508 otherwise we can get ENOSP while fs is not full.
1512 Description: parallel lock callbacks
1513 Details : Instead of sending blocking and completion callbacks as separated
1514 requests, adding them to a set and sending in parallel.
1517 Frequency : only for Cray XT3
1518 Bugzilla : 12829/13455
1519 Description: Changing primary group doesn't change the group lustre assigns to
1521 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1522 overridden with the primary group provided by the group upcall,
1523 whereas the supplied fsgid can be trusted if it is in the list of
1524 supplementary groups returned by the group upcall.
1526 Severity : enhancement
1528 Description: Allow masking D_WARNING, D_ERROR messages from console
1529 Details : Console messages can now be disabled via lnet.printk.
1533 Description: User code with malformed file open parameter crashes client node
1534 Details : Before packing join_file req, all the related reference should be
1535 checked carefully in case some malformed flags cause fake
1536 join_file req on client.
1540 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1541 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1542 mode is equal to the granted mode. However, it should also check
1543 that LDLM_FL_CP_REQD is not set.
1547 Description: Heavy nfs access might result in deadlocks
1548 Details : After ELC code landed, it is now improper to enqueue any mds
1549 locks under och_sem, because enqueue might want to decide to
1550 cancel open locks for same inode we are holding och_sem for.
1554 Description: Client eviction while running blogbench
1555 Details : A lot of unlink operations with concurrent I/O can lead to a
1556 deadlock causing evictions. To address the problem, the number of
1557 oustanding OST_DESTROY requests is now throttled to
1558 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1559 callbacks are processed in priority.
1563 Description: enable ACLs on MDS by default
1564 Details : ACLs must be enabled on MDS by default.
1567 Frequency : PPC/PPC64 only
1569 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1570 Details : fix duplicated definitions between asm-ppc64/types.h and
1571 lustre_types.h on PPC.
1574 Frequency : PPC/PPC64 only
1576 Description: asm-ppc/segment.h does not exist
1577 Details : fix compile issue on PPC.
1581 Description: better handle error messages in extents code
1584 Frequency : RHEL4 only
1586 Description: mkfs is very slow on IA64/RHEL4
1587 Details : A performance regression has been discovered in the MPT Fusion
1588 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1589 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1590 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1592 Severity : enhancement
1594 Description: SNMP support enhancement
1595 Details : Adding total number of sampled request for an MDS node in snmp
1598 Severity : enhancement
1600 Description: Optimize ldlm waiting list processing for PR extent locks
1601 Details : When processing waiting list for read extent lock and meeting read
1602 lock that is same or wider to it that is not contended, skip
1603 processing rest of the list and immediatelly return current
1604 status of conflictness, since we are guaranteed there are no
1605 conflicting locks in the rest of the list.
1609 Description: Time out and refuse to reconnect
1610 Details : When the failover node is the primary node, it is possible
1611 to have two identical connections in imp_conn_list. We must
1612 compare not conn's pointers but NIDs, otherwise we can defeat
1613 connection throttling.
1617 Description: port llog fixes from b1_6 into HEAD
1618 Details : Port llog reference couting and some llog cleanups from b1_6
1619 (bug 10800) into HEAD, for protect from panic and access to already
1620 free llog structures.
1624 Description: Detect stride IO mode in read-ahead
1625 Details : When a client does stride read, read-ahead should detect that and
1626 read-ahead pages according to the detected stride pattern.
1630 Description: data checksumming impacts single node performance
1631 Details : add support for several checksum algorithm. Currently, only CRC32
1632 and Adler-32 are supported. The checksum type can be changed on
1633 the fly via /proc/fs/lustre/osc/*/checksum_type.
1637 Description: use adler32 for page checksums
1638 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1643 Description: build for x2 fails
1644 Details : fix compile issue on Cray systems.
1648 Description: Properly match for duplicate locks
1649 Details : Due to different lock order from skiplists code, we need to
1650 traverse entire list for now
1653 Frequency : only on PPC/SLES10
1655 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1656 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1657 be defined. Add a hack in mkfs_lustre.c to work around this compile
1662 Description: LASSERT on MDS when client holding flock lock dies
1663 Details : ldlm pool logic depends on number of granted locks equal to
1664 number of released locks which is not true for flock locks, so
1665 just exclude such locks from consideration.
1669 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1670 Details : Use fsfilt_sync() for both the whole filesystem sync and
1671 individual file sync to eliminate dangerous inode locking
1672 with I_LOCK that can lead to a deadlock.
1676 Description: performance in 1.6.3
1677 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1678 sure that 1MB requests are not fragmented by the block layer.
1680 Severity : enhancement
1682 Description: organize the server-side client stats on per-nid basis
1683 Details : Change the structure of stats under obdfilter and mds to
1691 The "uuid"s file would list the uuids of _active_ exports.
1692 And the clear entry is to clear all stats and stale nids.
1694 Severity : enhancement
1696 Description: eliminate client locks in face of contention
1697 Details : file contention detection and lockless i/o implementation
1698 for contended files.
1702 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1705 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1707 Description: MDS may lose file striping (and hence file data) in some cases
1708 Details : If there are additional extended attributes stored on the MDS,
1709 in particular ACLs, SELinux, or user attributes (if user_xattr
1710 is specified for the client mount options) then there is a risk
1711 of attribute loss. Additionally, the Lustre file striping
1712 needs to be larger than default (e.g. striped over all OSTs),
1713 and an additional attribute must be stored initially in the
1714 inode and then increase in size enough to be moved to the
1715 external attribute block (e.g. ACL growing in size) for file
1720 Description: skiplist implementation simplification
1721 Details : skiplists are used to group compatible locks on granted list
1722 that was implemented as tracking first and last lock of each
1723 lock group the patch changes that to using doubly linked lists
1727 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1728 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1729 because LASSERT on both the data supplied by a client, and the
1730 data on disk is dangerous and incorrect.
1732 Severity : enhancement
1734 Description: Slow truncate/writes to huge files at high offsets.
1735 Details : Directly associate cached pages to lock that protect those pages,
1736 this allows us to quickly find what pages to write and remove
1737 once lock callback is received.
1741 Description: more ldlm soft lockups
1742 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1743 starve other threads from the resource lock for a long time in
1744 case of long waiting queue, so change the debug level from
1745 D_OTHER to the less frequently used D_INFO.
1747 Severity : enhancement
1749 Description: add -gid, -group, -uid, -user options to lfs find
1753 Description: Hung threads in invalidate_inode_pages2_range
1754 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1755 once one is completed. As a result, some RPCs are stuck in the
1756 queue and are never sent.
1760 Description: filter threads hungs on waiting journal commit
1761 Details : Cleanup filter group llog code, then only filter group llog will
1762 be only created in the MDS/OST syncing process.
1766 Description: Procfs and llog threads access destoryed import sometimes.
1767 Details : Sync the import destoryed process with procfs and llog threads by
1768 the import refcount and semaphore.
1770 Severity : enhancement
1772 Description: openlock cache of b1_6 port to HEAD
1777 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1778 Details : If insertion of an extent fails, then discard the inode
1779 preallocation and free data blocks else it can lead to duplicate
1784 Description: don't always update ctime in ext3_xattr_set_handle()
1785 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1786 In some cases the ctime should not be updated, for example for
1787 2.0->1.8 compatibility it is necessary to delete an xattr and it
1788 should not update the ctime.
1792 Bugzilla : 15713/16362
1793 Description: Assertion in iopen_connect_dentry in 1.6.3
1794 Details : looking up an inode via iopen with the wrong generation number can
1795 populate the dcache with a disconneced dentry while the inode
1796 number is in the process of being reallocated. This causes an
1797 assertion failure in iopen since the inode's dentry list contains
1798 both a connected and disconnected dentry.
1802 Description: assertion failure in ldlm_handle2lock()
1803 Details : fix a race between class_handle_unhash() and class_handle2object()
1804 introduced in lustre 1.6.5 by bug 13622.
1809 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1810 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1815 Description: oops in sd_iostats_seq_show()
1816 Details : unloading/reloading the scsi low level driver triggers a kernel
1817 bug when trying to access the sd iostat file.
1822 Description: Kernel panics during QLogic driver reload
1823 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1824 patch, causing memory corruption.
1829 Description: journal_dev option does not work in b1_6
1830 Details : pass mount option during pre-mount.
1832 Severity : enhancement
1834 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1835 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1836 extent information of a file. It can be used to map logical blocks
1837 in a file to physical blocks in the block device.
1841 Description: LDLM soft lockups - improvement
1842 Details : It is be possible to send the lock handle along with each read
1843 or write request because the client is already doing a lock match
1844 itself so there isn't any reason the OST should have to re-do that
1848 Frequency : only on Cray X2
1850 Description: X2 build failures
1851 Details : fix build failures on Cray X2.
1855 Description: xid & resent requests
1856 Details : Initialize RPC XID from clock at startup (randomly if clock is
1859 Severity : enhancement
1861 Description: Add lustre_start utility to start or stop multiple Lustre servers
1866 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1867 Details : In case of memory pressure, list_del() can be called twice on
1868 req->rq_history_list, causing a kernel oops.
1872 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1873 Details : fix stack overflow in the distributed lock manager by defering
1874 export eviction after a failed AST to the elt thread instead of
1875 handling it in the dlm interpret routine.
1879 Description: Convert some comments to new format.
1880 Details : Update documenting comments to match doxygen conventions.
1884 Description: Grammar fixes.
1885 Details : A couple of trivial spelling fixes.
1889 Description: OSD_COUNTERS-mandatory
1890 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1891 to simplify the code.
1895 Description: simplify cmm_device freeing logic.
1896 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1900 Description: Add lockdep support to dt_object_operations locking interface.
1901 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1902 parameter indicating lock ordering. Update mdd code to use new
1907 Description: Introduce failloc constants for lockless IO tests.
1908 Details : Add two new failloc constants to test lockless IO. Only one of
1909 them in implemented---another is checked in yet to be landed
1914 Description: Add lockdep support for inode mutex.
1915 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1916 in the situations where Lustre has to lock more than one inode
1921 Description: Add optional invariants checking support.
1922 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1923 configure switch. This macro is to be used for consistency and
1924 sanity checks that are too expensive to be left in `production'
1929 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1930 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1935 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1936 Details : Use spin_lock_nested() in (the only) situation where more than
1937 one ldlm_lock is locked simultaneously. Also, fix possible
1938 dead-lock in ldlm_lock_change_resource() by enforcing particular
1943 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1944 Details : Instead of specifying each ldlm_lock call-back through separate
1945 parameter, wrap them into struct ldlm_callback_suite.
1949 Description: Kill join_lru obd method and its callers.
1950 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1951 in memory---a requirement implemented through ->o_join_lru() obd
1952 method in HEAD. Remove this method and its users.
1956 Description: Add asynchronous ldlm ENQUEUE completion handler.
1957 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1958 case that stock ldlm_completion_ast() cannot handle as it waits
1959 until lock is granted. Introduce new ldlm_completion_ast_async()
1960 for this. Also comment ldlm_completion_ast().
1964 Description: ldlm_error <-> errno conversion.
1965 Details : Add functions to map (rather arbitrary) between LDLM error codes
1966 and standard errno values. CLIO needs this to prevent LDLM specific
1967 constants from escaping ldlm and osc.
1971 Description: Kill unused ldlm_handle2lock_ns() function.
1972 Details : Kill unused ldlm_handle2lock_ns() function.
1976 Description: Add lu_ref support to ldlm_lock
1977 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1978 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1979 and ldlm_resource. LDLM interface has to be changed, because code that
1980 releases a reference on a lock, has to "know" what reference this is.
1981 In the most frequent case
1983 lock = ldlm_handle2lock(handle);
1985 LDLM_LOCK_PUT(lock);
1987 no changes are required. When any other reference (received _not_ from
1988 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1989 instead of LDLM_LOCK_PUT().
1991 Arguably, changes are pervasive, and interface requires some discipline
1992 for proper use. On the other hand, it was very instrumental in finding
1993 a few leaked lock references.
1997 Description: Add ldlm_lock_addref_try().
1998 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1999 attempts to addref a lock that might be being canceled
2004 Description: Add ldlm_weigh_callback().
2005 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
2006 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
2007 hard-coded `number of pages' logic.
2011 Description: Add lockdep annotations to llog code.
2012 Details : Use appropriately tagged _nested() locking calls in the places
2013 where llog takes more than one ->lgh_lock lock.
2017 Description: Add loi_kms_set().
2018 Details : Wrap kms updates into a helper function.
2022 Description: Constify instances of struct lsm_operations.
2023 Details : Constify instances of struct lsm_operations.
2027 Description: lu_conf support.
2028 Details : On a server, a file system object is uniquely identified
2029 by a fid, which is sufficient to locate and load all object
2030 state (inode). On a client, on the other hand, more data are
2031 necessary instantiate an object. Change lu_object_find() and
2032 friends to take additional `lu_conf' argument describing object.
2033 Typically this includes layout information.
2037 Description: lu_context fixes.
2038 Details : Introduce new lu_context functions that are needed on the client
2039 side, where some system threads (ptlrpcd) are shared by multiple
2040 modules, and so cannot be stopped during module shutdown.
2044 Description: Add start and stop methods to lu_device_type_operations.
2045 Details : Introduce two new methods in lu_device_type_operations, that are
2046 invoked when first instance of a given type is created and last one
2047 is destroyed respectively. This is need by CLIO.
2051 Description: Add lu_ref support to struct lu_device.
2052 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
2053 track leaked references.
2057 Description: Introduce lu_kmem_descr.
2058 Details : lu_kmem_descr and its companion interface allow to create
2059 and destroy a number of kmem caches at once.
2063 Description: Fix lu_object finalization race.
2064 Details : Fix a race between lu_object_find() finding an object and its
2065 concurrent finalization. This race is (most likely) not possible
2066 on the server, but might happen on the client.
2070 Description: Introduce lu_ref interface.
2071 Details : lu_ref is a debugging module allowing to track references to
2072 a given object. It is quite cpu expensive, and has to be
2073 explicitly enabled with --enable-lu_ref. See usage description
2078 Description: Factor lu_site procfs stats into a separate function.
2079 Details : Separate lu_site stats printing code into a separate function
2080 that can be reused on a client.
2084 Description: Constify instances of struct {lu,dt,md}_device_operations.
2085 Details : Constify instances of struct {lu,dt,md}_device_operations.
2089 Description: Introduce struct md_site and move meta-data specific parts of
2090 struct lu_site here.
2091 Details : Move md-specific fields out of struct lu_site into special struct
2092 md_site, so that lu_site can be used on a client.
2096 Description: Kill mdd_lov_destroy().
2097 Details : Remove unused mdd code.
2101 Description: Add st_block checking to multistat.c.
2102 Details : Add st_block checking to multistat.c.
2106 Description: Add lu_ref support to struct obd_device.
2107 Details : Add lu_ref tracking to obd_device.
2111 Description: Kill obd_set_fail_loc().
2112 Details : Remove unused code.
2116 Description: Add special type for ptlrpc_request interpret functions.
2117 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2118 there. Actual usage will be in CLIO.
2122 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2123 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2124 doesn't work with lockdep.
2128 Description: Add rwv.c test program.
2129 Details : New testing program exercising readv(2) and writev(2) (Qian).
2133 Description: Add sendfile.c test program.
2134 Details : New testing program exercising sendfile(2) (Jay).
2138 Description: Ratelimit a message that can be very frequent.
2139 Details : Ratelimit a memory allocation failure message that can
2144 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2145 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2149 Description: Liblustre build fixes.
2150 Details : Liblustre build fixes.
2154 Description: libcfs: add cfs_{need,cond}_resched() interface.
2155 Details : libcfs: add cfs_{need,cond}_resched() definition and
2156 implementations for Linux, NT, and liblustre.
2158 Severity : enhancement
2160 Description: More exported tunables for mballoc
2161 Details : Add support for tunable preallocation window and new tunables for
2162 large/small requests
2166 Description: Detect corruption of block bitmap and checking for preallocations
2167 Details : Checks validity of on-disk block bitmap. Also it does better
2168 checking of number of applied preallocations. When corruption is
2169 found, it turns filesystem readonly to prevent further corruptions.
2173 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2174 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2175 window to 0. If it does hit cache miss, and read-pattern is still
2176 stride-io mode, does not reset the stride window, but also does not
2177 increase the stride window length in this case.
2181 Frequency : only for big-endian servers
2182 Description: Check if system is big-endian while mounting fs with extents feature
2183 Details : Mounting a filesystem with extents feature will fail on big-endian
2184 systems since ext3-based ldiskfs is not supported on big-endian
2185 systems. This can be over-riden with "bigendian_extents" mount option.
2187 Severity : enhancement
2189 Description: The root squash functionality
2190 Details : A security feature, which is to prevent users from being able
2191 to mount lustre on their desktop, run as root, and delete
2192 all of the files in the filesystem. The goal is accomplished by
2193 remapping user id (UID) and group id (GID) of the root user to
2194 a UID and GID specified by the system administartor via Lustre
2195 configuration management server (MGS). The functionality also
2196 allows to specify sets of clients for which the remapping does
2201 Description: Excessive recovery window
2202 Details : With AT enabled, the recovery window can be excessively long (6000+
2203 seconds). To address this problem, we no longer use
2204 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2205 timeout no longer depends on the service time, it is set to
2206 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2207 time via pb_service_time.
2211 Description: Watchdog triggered on MDS failover
2212 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2213 the OSTs know that the MDS "UUID" can be reused for the same export
2214 from a different NID, so we do not need to wait for the export to be
2218 Frequency : rare, only if using MMP with Linux RAID
2220 Description: MMP doesn't work with Linux RAID
2221 Details : While using HA for Lustre servers with Linux RAID, it is possible
2222 that MMP will not detect multiple mounts. To make this work we
2223 need to unplug the device queue in RAID when the MMP block is being
2224 written. Also while reading the MMP block, we should read it from
2225 disk and not the cached one.
2227 Severity : enhancement
2229 Description: open file using fid
2230 Details : A file can be opened using just its fid, like
2231 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2234 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2237 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2238 Details : add proper locking for creation of jbd_4k slab cache
2242 Description: MMP check in ext3_remount() fails without displaying any error
2243 Details : When multiple mount protection fails during remount, proper error
2246 Severity : enhancement
2248 Description: Allow stripe size to be up to 4G-64k
2249 Details : Fix math logic to allow large stripe sizes.
2253 Description: add check for >8TB ldiskfs filesystems
2254 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2255 Don't allow >8TB ldiskfs filesystems to be mounted without
2256 force_over_8tb mount option
2261 Descriptoin: OSS complains "too many blocks" and crashes
2262 Details : In filter_commitrw_write():
2263 1. lquota_chkquota();
2264 2. obdo_from_inode();
2265 3. lquota_pending_commit();
2266 "2." has changed uid and gid in this case, then caused
2267 lquota_chkquota() and lquota_pending_commit() to use
2268 different uid and gid. That is the root of the bug.
2272 Descriptoin: getstripe doesn't report stripe count = -1 sanely
2273 Details : Detect and print out the 'stripe count = -1' case properly.
2275 --------------------------------------------------------------------------------
2277 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2279 * Support for kernels:
2280 2.6.5-7.283 (SLES 9),
2281 2.6.9-55.EL (RHEL 4),
2282 2.6.16.46-0.14 (SLES 10),
2283 2.6.18.8 vanilla (kernel.org)
2284 * Client support for unpatched kernels:
2285 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2286 2.6.16 - 2.6.22 vanilla (kernel.org)
2287 * Due to recently discovered recovery problems, we do not recommend
2288 using patchless RHEL 4 clients with this or any earlier release.
2289 * Recommended e2fsprogs version: 1.39.cfs8
2290 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2291 * Starting with this release, the ldiskfs backing filesystem required
2292 by Lustre is now in its own package, lustre-ldiskfs. This package
2293 should be installed. It is versioned separately from Lustre and
2294 may be released separately in future.
2298 Description: block reactivating mgc import until all deactivates complete
2299 Details : Fix race when failing back MDT/MGS to itself (testing)
2302 Frequency : at statup only
2304 Description: mds_lov_synchronize race leads to various problems
2305 Details : simultaneous MDT->OST connections at startup can cause the
2306 sync to abort, leaving the OSC in a bad state.
2308 Severity : enhancement
2310 Description: add optional extra BUILD_VERSION info
2311 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2312 to override the lustre version.
2315 Frequency : 2.6.18 servers only
2317 Description: ll_kern_mount() doesn't release the module reference
2318 Details : The ldiskfs module reference count never drops down to 0
2319 because ll_kern_mount() doesn't release the module reference.
2324 Description: server LBUG when using old ost_num_threads parameter
2325 Details : Accept the old ost_num_threads parameter but warn that it
2326 is deprecated, and fix an off-by-one error that caused an LBUG.
2331 Description: Transient SCSI error results in persistent IO issue
2332 Details : iobuf->dr_error is not reinitialized to 0 between two
2336 Frequency : sometimes when underlying device returns I/O errors
2338 Description: OSTs not going read-only during write failures
2339 Details : OSTs are not remounted read-only when the journal commit threads
2340 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2341 instead of the ext3 wrappers.
2345 Description: poor connect scaling with increasing client count
2346 Details : Don't run filter_grant_sanity_check for more than 100 exports
2347 to improve scaling for large numbers of clients.
2350 Frequency : SLES10 only
2352 Description: sanity-quota.sh quotacheck failed: rc = -22
2353 Details : Quotas cannot be enabled on SLES10.
2356 Frequency : liblustre clients only
2358 Description: getdirentries does not give error when run on compute nodes
2359 Details : getdirentries does not fail when the size specified as an argument
2360 is too small to contain at least one entry
2362 Severity : enhancement
2364 Description: Add LNET router traceability for debug purposes
2365 Details : If a checksum failure occurs with a router as part of the
2366 IO path, the NID of the last router that forwarded the bulk data
2367 is printed so it can be identified.
2372 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2373 Details : Due to a race condition, liblustre clients were occasionally
2374 evicted incorrectly.
2376 Severity : enhancement
2378 Description: lfs setstripe use optional parameters instead of postional
2381 Severity : enhancement
2383 Description: Nanosecond timestamp support for ldiskfs
2384 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2385 resolution timestamps. There is not yet support for this at
2386 the Lustre filesystem level.
2389 Frequency : during server recovery
2391 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2392 Details : request with rq_no_resend flag not awake l_wait_event if they get
2396 Frequency : nfs export on patchless client
2398 Description: connectathon hang when test nfs export over patchless client
2399 Details : Disconnected dentry cannot be found with lookup, so we do not need
2400 to unhash it or make it invalid
2403 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2404 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2405 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2410 Description: open req refcounting wrong on reconnect
2411 Details : If reconnect happened between getting open reply from server and
2412 call to mdc_set_replay_data in ll_file_open, we will schedule
2413 replay for unreferenced request that we are about to free.
2414 Subsequent close will crash in variety of ways.
2415 Check that request is still eligible for replay in
2416 mdc_set_replay_data().
2421 Description: disable writes to filesystem when reading health_check file
2422 Details : the default for reading the health_check proc file has changed
2423 to NOT do a journal transaction and write to disk, because this
2424 can cause reads of the /proc file to hang and block HA state
2425 checking on a healthy but otherwise heavily loaded system. It
2426 is possible to return to the previous behaviour during configure
2427 with --enable-health-write.
2429 Severity : enhancement
2431 Description: 64-bit inode version
2432 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2433 to the inode. This will be required for version-based recovery.
2438 Description: MDS fails to start if a duplicate client export is detected
2439 Details : in some rare cases it was possible for a client to connect to
2440 an MDS multiple times. Upon recovery the MDS would detect this
2441 and fail during startup. Handle this more gracefully.
2443 Severity : enhancement
2445 Description: Add -o localflock option to simulate old noflock
2447 Details : This will achieve local-only flock/fcntl locks
2453 Description: log_commit_thread vs filter_destroy race leads to crash
2454 Details : Take import reference before releasing llog record semaphore
2459 Description: Wrong request locking in request set processing
2460 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2461 imp_delayed_list, in this place should be used imp_lock.
2464 Frequency : when reconnection
2466 Description: Grant Leak when osc reconnect to OST
2467 Details : When osc reconnect ost, OST(filter) should check whether it
2468 should grant more space to client by comparing fed_grant and
2469 cl_avail_grant, and return the granted space to client instead
2470 of "new granted" space, because client will call osc_init_grant
2471 to update the client grant space info.
2474 Frequency : when client reconnect to OST
2476 Description: Grant Leak when osc do resend and replay bulk write
2477 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2478 bulk write request, because the grant info while be sync between
2479 OSC and OST when reconnect, and we should ignore the grant info
2480 these of resend/replay write req.
2485 Description: Grant space more than avaiable left space sometimes.
2486 Details : When then OST is about to be full, if two bulk writing from
2487 different clients came to OST. Accord the avaliable space of the
2488 OST, the first req should be permitted, and the second one
2489 should be denied by ENOSPC. But if the seconde arrived before
2490 the first one is commited. The OST might wrongly permit second
2491 writing, which will cause grant space > avaiable space.
2494 Frequency : when client is evicted
2496 Description: Grant might be wrongly erased when osc is evicted by OST
2497 Details : when the import is evicted by server, it will fork another
2498 thread ptlrpc_invalidate_import_thread to invalidate the
2499 import, where the grant will be set to 0. While the original
2500 thread will update the grant it got when connecting. So if
2501 the former happened latter, the grant will be wrongly errased
2502 because of this race.
2507 Description: Checking Stale with correct fid
2508 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2509 to check whether it is stale, because sometimes, we want the
2510 enqueue happened anyway, and op_data.fid2 will not be initialized.
2512 Severity : enhancement
2514 Description: update patchless client
2515 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2518 Frequency : only with 2.4 kernel
2520 Description: random memory corruption
2521 Details : size of struct ll_inode_info is to big for union inode.u and this
2522 can be cause of random memory corruption.
2527 Description: Memory leak in recovery
2528 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2529 It should also check obd_fail before fsfilt_start, otherwise if
2530 fsfilt_start return -EROFS,(failover mds during mds recovery).
2531 then the req will return with repmsg->transno = 0 and rc = EROFS.
2532 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2533 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2534 be freed no matter whether fsfilt_commit success or not.
2537 Frequency : only with huge count clients
2539 Description: Prevents from taking the superblock lock in llap_from_page for
2541 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2542 for llap_from_page call in ll_removepage prevents from taking the
2543 superblock lock for a soon died page.
2548 Description: Not check open intent error before release open handle
2549 Details : in some rare cases, the open intent error is not checked before
2550 release open handle, which may cause
2551 ASSERTION(open_req->rq_transno != 0), because it tries to release
2552 the failed open handle.
2557 Description: Set cat log bitmap only after create log success.
2558 Details : in some rare cases, the cat log bitmap is set too early. and it
2559 should be set only after create log success.
2563 Description: Accessing a block bevice can re-enable I/O when Lustre is
2564 tearing down a device.
2565 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2570 Description: service threads may hog cpus when there are a lot of requests
2571 Details : Insert cond_resched to give other threads a chance to use some CPU
2576 Description: the cat log was not initialized in recovery
2577 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2578 unlink log on mds will not be initialized until mds post
2579 recovery. And also in mds post recovery, the unlink log will
2580 initialization will be done asynchronausly, so there will be race
2581 between add unlink log and unlink log initialization.
2585 Description: brw_stats were being printed incorrectly
2586 Details : brw_stats were being printed as log2 but all of them were not
2587 recorded as log2. Also remove some code duplication arising from
2588 filter_tally_{read,write}.
2592 Frequency : rare, only in recovery.
2593 Description: ASSERTION(req->rq_type != LI_POISON) failed
2594 Details : imp_lock should be held while iterating over imp_sending_list for
2595 prevent destroy request after get timeout in ptlrpc_queue_wait.
2599 Description: replay-single.sh test 52 fails
2600 Details : A lock's skiplist need to be cleanup when it being unlinked
2601 from its resource list.
2605 Description: Short directio read returns full requested size rather than
2607 Details : Direct I/O operations should return actual amount of bytes
2608 transferred rather than requested size.
2610 Severity : enhancement
2612 Description: metadata RPC reduction (e.g. for rm performance)
2613 Details : decrease the amount of synchronous RPC between clients and servers
2614 by canceling conflicing lock before the operation on the client
2615 and packing thier handles into the main operation RPC to server.
2617 Severity : enhancement
2619 Description: Async OSC create to avoid the blocking unnecessarily.
2620 Details : If a OST has no remain object, system will block on the creating
2621 when need to create a new object on this OST. Now, ways use
2622 pre-created objects when available, instead of blocking on an
2623 empty osc while others are not empty. If we must block, we block
2624 for the shortest possible period of time.
2628 Description: improve handling recoverable errors
2629 Details : if request processig with error which can be recoverable on server
2630 request should be resend, otherwise page released from cache and
2633 Severity : enhancement
2635 Description: refine locking for avoid write wrong info into lov_objid file
2636 Details : fix possible races with add new target and write/update data in
2639 --------------------------------------------------------------------------------
2641 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2646 Frequency : on some architectures
2648 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2649 Details : Uninitialized flags sometimes cause configuration commands to
2653 Frequency : patchless clients only
2655 Description: missing __iget() symbol export
2656 Details : The __iget() symbol export is missing. To avoid the need for
2657 this on patchless clients the deathrow inode reaper is turned
2658 off, and we depend on the VM to clean up old inodes. This
2659 dependency was during via the fix for bug 12181.
2661 --------------------------------------------------------------------------------
2663 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2665 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2666 INTEROPERATE with older versions automatically. In many cases a
2667 special upgrade step is needed. Please read the
2668 user documentation before upgrading any part of a live system.
2669 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2670 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2671 * WARNING: Lustre configuration and startup changes are required with
2672 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2676 Severity : enhancement
2678 Description: MountConf
2679 Details : Lustre configuration is now managed via mkfs and mount
2680 commands instead of lmc and lconf. New obd types (MGS, MGC)
2681 are added for dynamic configuration management. See
2682 https://mail.clusterfs.com/wikis/lustre/MountConf for
2685 Severity : enhancement
2687 Description: dynamic OST addition
2688 Details : OSTs can now be added to a live filesystem
2690 Severity : enhancement
2692 Description: startup order invariance
2693 Details : MDTs and OSTs can be started in any order. Clients only
2694 require the MDT to complete startup.
2696 Severity : enhancement
2698 Description: parallel, asynchronous orphan cleanup
2699 Details : orphan cleanup is now performed in separate threads for each
2700 OST, allowing parallel non-blocking operation.
2702 Severity : enhancement
2704 Description: optimized stripe assignment
2705 Details : stripe assignments are now made based on ost space available,
2706 ost previous usage, and OSS previous usage, in order to try
2707 to optimize storage space and networking resources.
2709 Severity : enhancement
2711 Description: Permanently set tunables
2712 Details : All writable /proc/fs/lustre tunables can now be permanently
2713 set on a per-server basis, at mkfs time or on a live system.
2715 Severity : enhancement
2717 Description: Lustre message v2
2718 Details : Add lustre message format v2.
2720 Severity : enhancement
2722 Description: client OST exclusion list
2723 Details : Clients can be started with a list of OSTs that should be
2724 declared "inactive" for known non-responsive OSTs.
2728 Description: SPEC SFS validation failure on NFS v2 over lustre.
2729 Details : Changes the blocksize for regular files to be 2x RPC size,
2730 and not depend on stripe size.
2732 Severity : enhancement
2734 Description: Multiple MD RPCs in flight.
2735 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2736 To avoid overly-overloading MDS, introduce a limit on number of
2737 MDS RPCs in flight for a single client and add /proc controls
2738 to adjust this limit.
2740 Severity : enhancement
2742 Description: client read/write statistics
2743 Details : Add client read/write call usage stats for performance
2744 analysis of user processes.
2745 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2746 file access. extents_stats shows chunk size distribution.
2747 extents_stats_per_process show chunk size distribution per
2750 Severity : enhancement
2752 Description: mds statistics
2753 Details : Add detailed mds operations statistics in
2754 /proc/fs/lustre/mds/*/stats.
2758 Description: Failure of copying files with lustre special EAs.
2759 Details : Client side always return success for setxattr call for lustre
2760 special xattr (currently only "trusted.lov").
2765 Description: Refcount LNET uuids
2766 Details : The global LNET uuid list grew linearly with every startup;
2767 refcount repeated list entries instead of always adding to
2770 Severity : enhancement
2772 Description: Dynamic service threads
2773 Details : Within a small range, start extra service threads
2774 automatically when the request queue builds up.
2777 Frequency : mixed-endian client/server environments
2779 Description: mixed-endian crashes
2780 Details : The new msg_v2 system had some failures in mixed-endian
2783 Severity : enhancement
2785 Description: Easy OST removal
2786 Details : OSTs can be permanently deactivated with e.g. 'lctl
2787 conf_param lustre-OST0001.osc.active=0'
2789 Severity : enhancement
2791 Description: MGS proc entries
2792 Details : Added basic proc entries for the MGS showing what filesystems
2795 Severity : enhancement
2797 Description: provide MGS failover
2798 Details : Added config lock reacquisition after MGS server failover.
2800 Severity : enhancement
2802 Description: add Linux 2.4 support
2803 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2807 Description: a large application tries to do I/O to the same resource and dies
2808 in the middle of it.
2809 Details : Check the req->rq_arrival time after the call to
2810 ost_brw_lock_get(), but before we do anything about
2811 processing it & sending the BULK transfer request. This
2812 should help move old stale pending locks off the queue as
2813 quickly as obd_timeout.
2816 Frequency : when an incorrect nid is specified during startup
2818 Description: ptlrpc connect to non-existant node causes kernel crash
2819 Details : LNET can't be re-entered from an event callback, which
2820 happened when we expire a message after the export has been
2821 cleaned up. Instead, hand the zombie cleanup off to another
2824 Severity : enhancement
2826 Description: plain/inodebits lock performance improvement
2827 Details : Grouping plain/inodebits in granted list by their request modes
2828 and bits policy, thus improving the performance of search through
2832 Frequency : only if OST filesystem is corrupted
2834 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2835 Details : for a short time RPCs with bulk IO are in the replay list,
2836 but replay of bulk IOs is unimplemented. If the OST filesystem
2837 is corrupted due to disk cache incoherency and then replay is
2838 started it is possible to trip an assertion. Avoid putting
2839 committed RPCs into the replay list at all to avoid this issue.
2842 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2844 Description: System hang on startup
2845 Details : This bug allowed the liblustre (e.g. catamount) client to
2846 return to the app before handling all startup RPCs. This
2847 could leave the node unresponsive to lustre network traffic
2848 and manifested as a server ptllnd timeout.
2850 Severity : enhancement
2852 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2853 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2854 boolean to control whether to print peer debug info when a
2855 client's RPC times out.
2858 Frequency : only for kernels with patches from Lustre below 1.4.3
2860 Description: Remove old rdonly API
2861 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2864 Frequency : only for devices with external journals
2866 Description: Set external device read-only also
2867 Details : During a commanded failover stop, we set the disk device
2868 read-only while the server shuts down. We now also set any
2869 external journal device read-only at the same time.
2872 Frequency : when upgrading from 1.4 while trying to change parameters
2874 Description: The wrong (new) MDC name was used when setting parameters for
2875 upgraded MDT's. Also allows changing of OSC (and MDC)
2876 parameters if --writeconf is specified at tunefs upgrade time.
2879 Frequency : when setting specific ost indicies
2881 Description: QOS code breaks on skipped indicies
2882 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2883 created with --index need not be sequential.
2885 Severity : enhancement
2887 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2888 Details : The uninit_groups feature works in conjunction with the kernel
2889 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2890 up the pass1 processing of e2fsck. This is a read-only feature
2891 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2892 mount filesystems that have had this feature enabled.
2894 Severity : enhancement
2896 Description: Improve multi-block allocation algorithm to avoid fragmentation
2897 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2898 allocation locality and avoid filesystem fragmentation.
2900 ------------------------------------------------------------------------------
2902 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2904 * Support for kernels:
2905 2.4.21-47.0.1.EL (RHEL 3)
2906 2.6.5-7.283 (SLES 9)
2907 2.6.9-42.0.10.EL (RHEL 4)
2908 2.6.12.6 vanilla (kernel.org)
2909 2.6.16.27-0.9 (SLES 10)
2910 * Recommended e2fsprogs version: 1.39.cfs5
2912 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2916 Frequency : occasional, depends on client load and configuration
2917 Bugzilla : 12181, 12203
2918 Description: data loss for recently-modified files
2920 Details : In some cases it is possible that recently written or created
2921 files may not be written to disk in a timely manner (this should
2922 normally be within 30s unless client IO load is very high).
2923 The problem appears as zero-length files or files that are a
2924 multiple of 1MB in size after a client crash or client eviction
2925 that are missing data at the end of the file.
2927 This problem is more likely to be hit on clients where files are
2928 repeatedly created and unlinked in the same directory, clients
2929 have a large amount of RAM, have many CPUs, the filesystem has
2930 many OSTs, the clients are rebooted frequently, and/or the files
2931 are not accessed by other nodes after being written.
2933 The presence of the problem can be detected by looking at
2934 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2935 smaller than the second (nr_unused) then dirty files will not
2936 be flushed automatically to disk. "sync; sleep 10" should be
2937 run several times on the node before unmounting it to update
2938 Lustre (this is also safe to run on nodes without this problem).
2940 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2941 that can cause this same problem, so customers using that kernel
2942 also need to update the kernel in addition to Lustre. In order
2943 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2945 It is normal that files written just before a client crash (less
2946 than 30s) may not yet have been flushed to disk, even for local
2950 Frequency : frequent on thin XT3 nodes
2952 Description: UUID collision on thin XT3 Linux nodes
2953 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2954 randomly, since we relied on an insufficiently-seeded PRNG.
2959 Description: OSS hangs after "All ost request buffers busy"
2960 Details : A deadlock between quota and journal operations caused OSS
2961 hangs after printing "All ost request buffers busy."
2964 Frequency : always on liblustre builds
2966 Description: Cleanup compiler warnings on liblustre
2969 Frequency : always on liblustre builds on XT3
2971 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2975 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2976 > 32000 subdirectories
2977 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2978 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2979 subdirectory count crosses 32000. This will aid e2fsck to
2980 correctly handle more than 32000 subdirectories.
2983 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2985 Description: System hang on startup
2986 Details : This bug allowed the liblustre (e.g. catamount) client to
2987 return to the app before handling all startup RPCs. This
2988 could leave the node unresponsive to lustre network traffic
2989 and manifested as a server ptllnd timeout.
2991 Severity : enhancement
2993 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2994 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2995 boolean to control whether to print peer debug info when a
2996 client's RPC times out.
3001 Description: make O_SYNC working on 2.6 kernels
3002 Details : 2.6 kernels use different method for mark pages for write,
3003 so need add a code to lustre for O_SYNC work.
3008 Description: Failure to close file and release space on NFS
3009 Details : Put inode details into lock acquired in ll_intent_file_open.
3010 Use mdc_intent_lock in ll_intent_open to properly
3011 detect all kind of errors unhandled by mdc_enqueue.
3016 Description: proc file read during shutdown sometimes raced obd removal,
3018 Details : Add lock to prevent obd access after proc file removal.
3021 Frequency : Only for files larger than 4GB on 32-bit clients.
3023 Description: improperly doing page alignment of locks
3024 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
3025 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
3030 Description: RPCs being resent when they shouldn't be
3031 Details : Some RPCs that should not be resent are being resent. This
3032 can cause inconsistencies in the RPC state machine. Do not
3033 resend such requests.
3036 Frequency : rare, only with NFS export
3038 Description: Crash on NFS re-export node
3039 Details : under very unusual load conditions an assertion is hit in
3040 ll_intent_file_open()
3043 Frequency : only if OST filesystem is corrupted
3045 Description: client incorrectly hits assertion in ptlrpc_replay_req()
3046 Details : for a short time RPCs with bulk IO are in the replay list,
3047 but replay of bulk IOs is unimplemented. If the OST filesystem
3048 is corrupted due to disk cache incoherency and then replay is
3049 started it is possible to trip an assertion. Avoid putting
3050 committed RPCs into the replay list at all to avoid this issue.
3055 Description: large O_DIRECT requests fail under memory pressure/fragmentation
3056 Details : Large single O_DIRECT read and write calls can fail to allocate
3057 a sufficiently large buffer to process the request. In case of
3058 allocation failure the allocation is retried with a smaller
3059 buffer and broken into smaller requests.
3061 Severity : enhancement
3063 Description: Add -o localflock option to simulate old noflock behaviour.
3064 Details : This will achieve local-only flock/fcntl locks coherentness.
3069 Description: versioning check is incomplete
3070 Details : Checking the version difference of client vs. server, report
3071 error if the gap is too big.
3076 Description: add support PG_writeback bit
3077 Details : add support for PG_writeback bit for Lustre, for more carefull
3078 work with page cache in 2.6 kernel. This also fix some deadlocks
3079 and remove hack for work O_SYNC with 2.6 kernel.
3081 Severity : enhancement
3083 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3084 Details : The uninit_groups feature works in conjunction with the kernel
3085 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3086 up the pass1 processing of e2fsck. This is a read-only feature
3087 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3088 mount filesystems that have had this feature enabled.
3090 Severity : enhancement
3092 Description: Improve multi-block allocation algorithm to avoid fragmentation
3093 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3094 allocation locality and avoid filesystem fragmentation.
3096 ------------------------------------------------------------------------------
3098 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3100 * Support for kernels:
3101 2.6.9-42.0.3.EL (RHEL 4)
3102 2.6.5-7.276 (SLES 9)
3103 2.4.21-47.0.1.EL (RHEL 3)
3104 2.6.12.6 vanilla (kernel.org)
3105 2.6.16.21-0.8 (SLES10)
3106 * Recommended e2fsprogs version: 1.39.cfs2-0
3108 * The backwards-compatible /proc/sys/portals symlink has been removed
3109 in this release. Before upgrading, please ensure that you change
3110 any configuration scripts or /etc/sysctl.conf files that access
3111 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3112 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3113 in advance of the upgrade on any system running Lustre 1.4.6 or
3114 newer, since /proc/sys/lnet was added in that version.
3115 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3119 Frequency : only when quota is used
3121 Description: avoid scanning export list for quota master
3122 Details : Change the algorithms to avoid scanning export list in order
3123 to improve the efficiency.
3126 Frequency : MDS failover only, very rarely
3128 Description: "went back in time" messages on mds failover
3129 Details : The greatest transno may be lost when the current operation
3130 finishes with an error (transno==0) and the client's last_rcvd
3131 record is over-written. Save the greatest transno in the
3132 mds_last_transno for this case.
3135 Frequency : always for specific kernels and striping counts
3137 Description: client may get "Matching packet too big" without ACL support
3138 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3139 when trying to access files in certain configurations. The
3140 clients should in fact be denied when mounting because they do
3141 not understand ACLs.
3144 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3146 Description: many clients connecting with IO in progress causes connect timeouts
3147 Details : Avoid synchronous journal commits to avoid delays caused by many
3148 clients connecting/disconnecting when bulk IO is in progress.
3149 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3150 OST_IO_PORTAL to avoid delays behind potentially many pending
3154 Frequency : occasionally with multiple writers to a single file
3156 Description: shared writes to file may result in wrong size reported by stat()
3157 Details : Allow growing of kms when extent lock is cancelled
3160 Frequency : always with random mmap IO to multi-striped file
3162 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3163 Details : Only if the hole is at the end of OST object so that kms is too
3164 small. Fix is to increase kms accordingly in ll_nopage.
3167 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3169 Description: writes to a missing object would leak memory on the OST
3170 Details : If there is an inconsistency between the MDS and OST filesystems,
3171 such that the MDS references an object that doesn't exist, writes
3172 to that object will leak memory due to incorrect cleanup in the
3173 error handling path, eventually running out of memory on the OST.
3178 Description: Creating too long symlink causes lustre errors
3179 Details : Check symlink and name lengths before sending requests to MDS.
3182 Frequency : only if flock is enabled (not on by default)
3184 Description: posix locks not released on fd closure on 2.6.9+
3185 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3186 kernels, this caused such locks not to be released on fd close and
3187 then assertions on fs unmount about still used locks.
3190 Frequency : MDS failover only, very rarely
3192 Description: clients may get ASSERTION(granted_lock != NULL)
3193 Details : When request was taking a long time, and a client was resending
3194 a getattr by name lock request. The were multiple lock requests
3195 with the same client lock handle and
3196 mds_getattr_name->fixup_handle_for_resent_request found one of the
3197 lock handles but later failed with ASSERTION(granted_lock != NULL).
3202 Description: handle->h_buffer_credits > 0, assertion failure
3203 Details : h_buffer_credits is zero after truncate, causing assertion
3204 failure. This patch extends the transaction or creates a new
3208 Frequency : NFS re-export or patchless client
3209 Bugzilla : 11179, 10796
3210 Description: Crash on NFS re-export node (__d_move)
3211 Details : We do not want to hash the dentry if we don't have a lock.
3212 But if this dentry is later used in d_move, we'd hit uninitialised
3213 list head d_hash, so we just do this to init d_hash field but
3214 leave dentry unhashed.
3217 Frequency : NFS re-export or patchless client
3219 Description: NFS exports has problem with symbolic link
3220 Details : lustre client didn't properly install dentry when re-exported
3221 to NFS or running patchless client.
3224 Frequency : NFS re-export or patchless client
3226 Description: Various nfs/patchless fixes.
3227 Details : fixes reuse disconected alias for lookup process - this fixes
3228 warning "find_exported_dentry: npd != pd",
3229 fix permission error with open files at nfs.
3230 fix apply umask when do revalidate.
3233 Frequency : occasional
3235 Description: Crash on NFS re-export node
3236 Details : calling clear_page() on the wrong pointer triggered oops in
3237 generic_mapping_read().
3240 Frequency : rarely, using O_DIRECT IO
3242 Description: unaligned directio crashes client with LASSERT
3243 Details : check for unaligned buffers before trying any requests.
3246 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3248 Description: stale data returned from RAID cache
3249 Details : If only a small amount of IO is done to the RAID device before
3250 reading it again it is possible to get stale data from the RAID
3251 cache instead of reading it from disk.
3254 Frequency : always for sles10 kernel
3256 Description: sles10 support
3257 Details : ll_follow_link: compile fixes and using of nd_set_link
3258 under newer kernels.
3261 Frequency : depends on arch, kernel and compiler version, always on sles10
3264 Description: recursive or deep enough symlinks cause stack overflow
3265 Details : getting rid of large stack-allocated variable in
3269 Frequency : depends on hardware
3271 Description: lustre write performance loss in the SLES10 kernel
3272 Details : the performance loss is caused by using of write barriers in the
3273 ext3 code. The SLES10 kernel turns barrier support on by
3274 default. The fix is to undo that change for ldiskfs.
3276 ------------------------------------------------------------------------------
3278 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3280 * Support for kernels:
3281 2.6.9-42.0.3EL (RHEL 4)
3282 2.6.5-7.276 (SLES 9)
3283 2.4.21-47.0.1.EL (RHEL 3)
3284 2.6.12.6 vanilla (kernel.org)
3288 Frequency : quota enabled and large files being deleted
3290 Description: releasing more than 4GB of quota at once hangs OST
3291 Details : If a user deletes more than 4GB of files on a single OST it
3292 will cause the OST to spin in an infinite loop. Release
3293 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3298 Description: statfs data retrieved from /proc may be stale or zero
3299 Details : When reading per-device statfs data from /proc, in the
3300 {kbytes,files}_{total,free,avail} files, it may appear
3301 as zero or be out of date.
3304 Frequency : systems with MD RAID1 external journal devices
3306 Description: lconf's call to blkid is confused by RAID1 journal devices
3307 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3308 of returning all block devices that match the journal UUID.
3311 Frequency : always, for aggregate stripe size over 4GB
3313 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3314 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3315 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3318 Frequency : always if "lfs find" used on a local file/directory
3320 Description: "lfs find" segfaults if used on a local file/directory
3321 Details : The case where a directory component was not specified wasn't
3322 handled correctly. Handle this properly.
3325 Frequency : always on ppc64
3327 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3328 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3329 perform endianness conversion.
3332 Frequency : rarely (truncate to non-zero file size after write under load)
3333 Bugzilla : 10730, 10687
3334 Description: Files padded with zeros to next 4K multiple
3335 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3336 it is possible that files that are truncated to a non-zero size
3337 immediately after being written are filled with zero bytes beyond
3338 the truncated size. No file data is lost.
3340 Severity : enhancement
3342 Description: Allow recovery/failover for liblustre clients.
3343 Details : liblustre clients were unaware of failover configurations until
3346 Severity : enhancement
3348 Description: user file locks should fail when not mounting with flock option
3349 Details : Set up an error-returning stub in ll_file_operations.lock field
3350 to prevent incorrect behaviour when client is mounted without
3351 flock option. Also, set up properly f_op->flock field for
3355 Frequency : always on ia64
3357 Description: "lfs df" loops on printing out MDS statfs information
3358 Details : The obd_ioctl_data was not initialized and in some systems
3359 this caused a failure during the ioctl that did not return
3360 an error. Initialize the struct and return an error on failure.
3363 Frequency : SLES 9 only
3365 Description: Error of copying files with lustre special EAs as root
3366 Details : Client side always return success for setxattr call for lustre
3367 special xattr (currently only "trusted.lov").
3370 Frequency : rarely on clusters with both ia64+i386 clients
3372 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3373 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3374 clients are doing concurrent writes to the same file it is
3375 possible that the ia64 clients may LASSERT because the OST
3376 extent locks are not PAGE_SIZE aligned. Ensure that grown
3377 locks are always aligned on the request boundary.
3380 Frequency : specific use, occasional
3382 Description: Overwriting in use executable truncates on-disk binary image
3383 Details : If one node attempts to overwrite an executable in use by
3384 another node, we now correctly return ETXTBSY instead of
3385 truncating the file.
3387 Severity : enhancement
3389 Description: Async OSC create to avoid the blocking unnecessarily.
3390 Details : If a OST has no remain object, system will block on the creating
3391 when need to create a new object on this OST. Now, ways use
3392 pre-created objects when available, instead of blocking on an
3393 empty osc while others are not empty. If we must block, we block
3394 for the shortest possible period of time.
3399 Description: chmod on Lustre root is propagated to other clients
3400 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3401 invalid by the follow_mount time.
3404 Frequency : liblustre clients only
3406 Description: Race in 'instant cancel' lock handling could lead to such locks
3407 never to be granted in case of SMP MDS
3408 Details : Do not destroy not yet granted but cbpending locks in
3412 Frequency : replay/resend of open
3414 Description: non null lock assetion failure in mds_intent_policy
3415 Details : Trying to replay/resend lockless open requests resulted in
3416 mds_open() returning 0 with no lock. Now it sets a flag if
3417 a lock is going to be returned.
3419 Severity : enhancement
3421 Description: Checksum enhancements
3422 Details : New checksum enhancements allow for resending RPCs that failed
3425 Severity : enhancement
3427 Description: Tunables on number of dirty pages in cacche
3428 Details : Allow to set limit on number of dirty pages cached.
3433 Description: client crash on unmount - lock still has references
3434 Details : In some error handling cases it was possible to leak a lock
3435 reference on a client while accessing a file. This was not
3436 harmful to the client during operation, but would cause the
3437 client to crash when the filesystem is unmounted.
3440 Frequency : specific case, rare
3442 Description: ETXTBSY on mds though file not in use by client
3443 Details : ETXTBSY is no longer incorrectly returned when attempting to
3444 chmod or chown a directory that the user previously tried to
3445 execute or a currently-executing binary.
3448 Frequency : extremely rare except on liblustre-based clients
3450 Description: Lustre space not freed when files are deleted
3451 Details : Clean up open-unlinked files after client eviction. Previously
3452 the unlink was skipped and the files remained as orphans.
3457 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3458 Details : In some cases it was possible to send a blocking callback to
3459 a client doing a glimpse, even though that client didn't get
3460 a lock granted. When the glimpse lock is cancelled on the OST
3461 the freed lock is left on the waiting list and corrupted the list.
3464 Frequency : all core dumps
3466 Description: Broke core dumps to lustre
3467 Details : Negative dentry may be unhashed if parent does not have UPDATE
3468 lock, but some callers, e.g. do_coredump, expect dentry to be
3469 hashed after successful create, hash it in ll_create_it.
3471 ------------------------------------------------------------------------------
3473 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3475 * Support for kernels:
3476 2.6.9-42.0.2.EL (RHEL 4)
3477 2.6.5-7.276 (SLES 9)
3478 2.4.21-40.EL (RHEL 3)
3479 2.6.12.6 vanilla (kernel.org)
3483 Frequency : always on RHEL 3
3485 Description: Number of open files grows over time
3486 Details : The number of open files grows over time, whether or not
3487 Lustre is started. This was due to a filp leak introduced
3488 by one of our kernel patches.
3490 ------------------------------------------------------------------------------
3492 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3494 * Support for kernels:
3495 2.6.9-42.EL (RHEL 4)
3496 2.6.5-7.276 (SLES 9)
3497 2.4.21-40.EL (RHEL 3)
3498 2.6.12.6 vanilla (kernel.org)
3503 Bugzilla : 5719, 9635, 9792, 9684
3504 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3505 Details : If a server is under heavy load and cannot reply to new
3506 connection requests before the client resends the (re)connect,
3507 the connection handling code can behave badly if two service
3508 threads are concurrently handing separate (re)connections from
3509 the same client. Add better locking to the connection handling
3510 code, and ensure that only a single connection will be processed
3511 for a given client UUID, even if the lock is dropped.
3513 Severity : enhancement
3515 Description: add TCP zero-copy support to kernel
3516 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3517 sends if the hardware supports scatter-gather and checksumming.
3518 This allows socklnd to do client-write and server-read more
3519 efficiently and reduce CPU utilization from skbuf copying.
3522 Frequency : only if NFS exporting from client
3524 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3525 Details : If setting attributes on a file created under NFS that had
3526 never been opened it would be possible to oops the client
3527 if the file had no objects.
3530 Frequency : always for liblustre
3532 Description: liblustre client does MDS+OSTs setattr RPC for each write
3533 Details : When doing a write from a liblustre client, the client
3534 incorrectly issued an RPC to the MDS and each OST the file was
3535 striped over in order to update the timestamps. When writing
3536 with small chunks and many clients this could overwhelm the MDS
3537 with RPCs. In all cases it would slow down the write because
3538 these RPCs are unnecessary.
3540 Severity : enhancement
3542 Description: allow number of MDS service threads to be changed at module load
3543 Details : It is now possible to change the number of MDS service threads
3544 running. Adding "options mds mds_num_threads={N}" to the MDS's
3545 /etc/modprobe.conf will set the number of threads for the next
3546 time Lustre is restarted (assuming the "mds" module is also
3547 reloaded at that time). The default number of threads will
3548 stay the same, 32 for most systems.
3553 Description: OST crash if filesystem is unformatted or corrupt
3554 Details : If an OST is started on a device that has never been formatted
3555 or if the filesystem is corrupt and cannot even mount then the
3556 error handling cleanup routines would dereference a NULL pointer.
3561 Description: NULL pointer deref in llap_from_page.
3562 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3563 incorrect) mapping if the page was truncated/reclaimed while it was
3564 searched for. Check for this condition and skip such pages when
3565 doing readahead. Introduce extra check to llap_from_page() to
3566 verify page->mapping->host is non-NULL (so page is not anonymous).
3569 Frequency : Sometimes when using sys_sendfile
3571 Description: "page not covered by a lock" warnings from ll_readpage
3572 Details : sendfile called ll_readpage without right page locks present.
3573 Now we introduced ll_file_sendfile that does necessary locking
3574 around call to generic_file_sendfile() much like we do in
3578 Frequency : with certain MDS communication failures at client mount time
3580 Description: NULL pointer deref after failed client mount
3581 Details : a client connection request may delayed by the network layer
3582 and not be sent until after the PTLRPC layer has timed out the
3583 request. If the client fails the mount immediately it will try
3584 to clean up before the network times out the request. Add a
3585 reference from the request import to the obd device and delay
3586 the cleanup until the network drops the request.
3589 Frequency : occasionally during client (re)connect
3591 Description: assertion failure during client (re)connect
3592 Details : processing a client connection request may be delayed by the
3593 client or server longer than the client connect timeout. This
3594 causes the client to resend the connection request. If the
3595 original connection request is replied in this interval, the
3596 client may trip an assertion failure in ptlrpc_connect_interpret()
3597 which thought it would be the only running connect process.
3600 Frequency : only with obd_echo servers and clients that are rebooted
3602 Description: kernel BUG accessing uninitialized data structure
3603 Details : When running an obd_echo server it did not start the ping_evictor
3604 thread, and when a client was evicted an uninitialized data
3605 structure was accessed. Start the ping_evictor in the RPC
3606 service startup instead of the OBD startup.
3608 Severity : enhancement
3609 Bugzilla : 10193 (patchless)
3610 Description: Remove dependency on various unexported kernel interfaces.
3611 Details : No longer need reparent_to_init, exit_mm, exit_files,
3612 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3615 Frequency : rare (only users of deprecated and unsupported LDAP config)
3617 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3618 Details : LDAP apparently contains 'lustreName' attributes instead of
3619 'name'. A simple remapping of the name is sufficient.
3622 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3624 Description: waiting_locks_callback trips kernel BUG if client is evicted
3625 Details : Running with the dump_on_timeout debug flag turned on makes
3626 it possible that the waiting_locks_callback() can try to dump
3627 the Lustre kernel debug logs from an interrupt handler. Defer
3628 this log dumping to the expired_lock_main() thread.
3630 Severity : enhancement
3632 Description: Support NFS exporting on 2.6 kernels.
3633 Details : Implement non-rawops metadata methods for NFS server to use without
3634 changing NFS server code.
3637 Frequency : very rare (synthetic metadata workload only)
3639 Description: two racing renames might cause an MDS thread to deadlock
3640 Details : Running the "racer" program may cause one MDS thread to rename
3641 a file from being the source of a rename to being the target of
3642 a rename at exactly the same time that another thread is doing
3643 so, and the second thread has already enqueued these locks after
3644 doing a lookup of the target and is trying to relock them in
3645 order. Ensure that we don't try to re-lock the same resource.
3648 Frequency : only very large systems with liblustre clients
3650 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3651 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3652 all OSTs in parallel. This allows the eviction of stale liblustre
3653 clients to proceed much faster than if they were done in series,
3654 and also offers similar improvements for other set_info RPCs.
3659 Description: excessive CPU usage during initial read phase on client
3660 Details : During the initial read phase on a client, it would agressively
3661 retry readahead on the file, consuming too much CPU and impacting
3662 performance (since 1.4.5.8). Improve the readahead algorithm
3663 to avoid this, and also improve some other common cases (read
3664 of small files in particular, where "small" is files smaller than
3665 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3670 Description: MDS crash when receiving packet with unknown intent.
3671 Details : Do not LBUG in unknown intent case, just return -EFAULT
3673 Severity : enhancement
3674 Bugzilla : 9293, 9385
3675 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3676 Details : Do not serialize getattr (non-intent version) and statfs.
3679 Frequency : occasional, when OST network is overloaded/intermittent
3681 Description: client evicted by OST after bulk IO timeout
3682 Details : If a client sends a bulk IO request (read or write) the OST
3683 may evict the client if it is unresposive to its data GET/PUT
3684 request. This is incorrect if the network is overloaded (takes
3685 too long to transfer the RPC data) or dropped the OST GET/PUT
3686 request. There is no need to evict the client at all, since
3687 the pinger and/or lock callbacks will handle this, and the
3688 client can restart the bulk request.
3691 Frequency : Always when mmapping file with no objects
3693 Description: client crashes when mmapping file with no objects
3694 Details : Check that we actually have objects in a file before doing any
3695 operations on objects in ll_vm_open, ll_vm_close and
3701 Description: Request leak when working with deleted CWD
3702 Details : Introduce advanced request refcount tracking for requests
3703 referenced from lustre intent.
3705 Severity : Enhancement
3707 Description: Cache open file handles on client.
3708 Details : MDS now will return special lock along with openhandle, if
3709 requested and client is allowed to hold openhandle, even if unused,
3710 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3711 closing files for every read/write openration.
3713 Severity : Enhancement
3715 Description: Cache open negative dentries on client when possible.
3716 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3717 negative dentries on lock revocation.
3722 Description: Remounting a client read-only wasn't possible with a zconf mount
3723 Details : It wasn't possible to remount a client read-only with llmount.
3725 Severity : enhancement
3726 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3727 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3728 MPICH is included to add Lustre-specific ADIO interfaces.
3729 This is based closely on the UFS ADIO layer and only differs
3730 in file creation, in order to allow the OST striping to be set.
3731 This is user-contributed code and not supported by CFS.
3736 Description: extended inode attributes (immutable, append-only) work improperly
3737 when 2.4 and 2.6 kernels are used on client/server or vice versa
3738 Details : Introduce kernel-independent values for these flags.
3740 Severity : enhancement
3743 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3744 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3745 granularity. Now, Fractional MB granularity is be supported,
3746 this is very useful for low memory system.
3748 Severity : enhancement
3750 Description: Getattr by fid
3751 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3752 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3753 have locks and avoids getattr rpc requests for it on every stat.
3756 Frequency : Always, for filesystems larger than 2TB
3758 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3759 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3760 filesystems larger than 2TB to be mounted with mballoc enabled.
3763 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3765 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3766 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3767 filesystems larger than 2TB is mounted then the kernel will
3768 silently corrupt the start of the filesystem. CONFIG_LBD is
3769 enabled for all CFS-supported kernels, but the possibility of
3770 this happening with a modified kernel config exists.
3772 Severity : enhancement
3774 Description: add client O_DIRECT support for 2.6 kernels
3775 Details : It is now possible to do O_DIRECT reads and writes to files
3776 in the Lustre client mountpoint on 2.6 kernel clients.
3778 Severity : enhancement
3780 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3781 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3782 parallel, not waiting for response from every OST before sending
3783 a rpc to the next OST.
3788 Description: setattr vs write race when updating file timestamps
3789 Details : Client processes that update a file timestamp into the past
3790 right after writing to the file (e.g. tar) it is possible that
3791 the updated file modification time can be reset to the current
3792 time due to a race between processing the setattr and write RPC.
3794 Severity : enhancement
3796 Description: Bring 'lfs find' closer in line with regular Linux find.
3797 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3798 -print0 options and obtains all the needed info through the lustre
3801 Severity : enhancement
3803 Description: support up to 1024 configured devices on one node
3804 Details : change obd_dev array from statically allocated to dynamically
3805 allocated structs as they are first used to reduce memory usage
3810 Description: Flush dirty partially truncated pages during truncate
3811 Details : Immediatelly flush partially truncated pages in filter_setattr,
3812 this way we completely avoid having any pages in page cache on OST
3813 and can retire ugly workarounds during writes to flush such pages.
3818 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3819 Details : protect inode from truncation within vfs_unlink() context
3820 just take a reference before calling vfs_unlink() and release it
3821 when parent's i_sem is free.
3826 Description: last_id value checked outside lock on OST caused LASSERT failure
3827 Details : If there were multiple MDS->OST object precreate requests in
3828 flight, it was possible that the OST's last object id was checked
3829 outside a lock and incorrectly tripped an assertion. Move checks
3830 inside locks, and discard old precreate requests.
3833 Frequency : always, if extents are used on OSTs
3835 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3836 Details : OSTs using the extents format would not zero the high 16 bits of
3837 the index physical block number. This is not a problem for any
3838 OST filesystems smaller than 16TB, and no kernels support ext3
3839 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3840 new/modified files) and can be fixed for existing filesystems
3841 with e2fsprogs-1.39-cfs1.
3846 Description: import connection selection may be incorrect if timer wraps
3847 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3848 import connections to be ignored if the 32-bit jiffies counter
3849 wraps. Use a 64-bit jiffies counter.
3852 Frequency : very large clusters immediately after boot
3854 Description: LNET request buffers exhausted under heavy short-term load
3855 Details : If a large number of client requests are generated on a service
3856 that has previously never seen so many requests it is possible
3857 that the request buffer growth cannot keep up with the spike in
3858 demand. Instead of dropping incoming requests, they are held in
3859 the LND until the RPC service can accept more requests.
3862 Frequency : Sometimes during replay
3864 Description: Assertion failure in ll_local_open after replay.
3865 Details : If replay happened on an open request reply before we were able
3866 to set replay handler, reply will become not swabbed tripping the
3867 assertion in ll_local_open. Now we set the handler right after
3868 recognising of open request
3871 Frequency : very rare
3873 Description: kernel reports "badness in vsnprintf"
3874 Details : Reading from the "recovery_status" /proc file in small chunks
3875 may cause a negative length in lprocfs_obd_rd_recovery_status()
3876 call to vsnprintf() (which is otherwise harmless). Exit early
3877 if there is no more space in the output buffer.
3879 Severity : enhancement
3881 Description: clear OBD RPC statistics by writing to them
3882 Details : It is now possible to clear the OBD RPC statistics by writing
3883 to the "stats" file.
3888 Description: Client mtime is not the same on different clients after utimes
3889 Details : In some cases, the client was using the utimes() syscall on
3890 a file cached on another node. The clients now validate the
3891 ctime from the MDS + OSTs to determine which one is right.
3896 Description: Inability to activate failout mode
3897 Details : lconf script incorrectly assumed that in pythong string's numeric
3898 value is used in comparisons.
3901 Frequency : always with multiple stripes per file
3903 Description: Inefficient object allocation for mutli-stripe files
3904 Details : When selecting which OSTs to stripe files over, for files with
3905 a stripe count that divides evenly into the number of OSTs,
3906 the MDS is always picking the same starting OST for each file.
3907 Return the OST selection heuristic to the original design.
3912 Description: mount failures may take full timeout to return an error
3913 Details : Under some heavy load conditions it is possible that a
3914 failed mount can wait for the full obd_timeout interval,
3915 possibly several minutes, before reporting an error.
3916 Instead return an error as soon as the status is known.
3918 Frequency : quota enabled and large files being deleted
3920 Description: releasing more than 4GB of quota at once hangs OST
3921 Details : If a user deletes more than 4GB of files on a single OST it
3922 will cause the OST to spin in an infinite loop. Release
3923 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3928 Description: statfs data retrieved from /proc may be stale or zero
3929 Details : When reading per-device statfs data from /proc, in the
3930 {kbytes,files}_{total,free,avail} files, it may appear
3931 as zero or be out of date.
3934 Frequency : systems with MD RAID1 external journal devices
3936 Description: lconf's call to blkid is confused by RAID1 journal devices
3937 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3938 of returning all block devices that match the journal UUID.
3941 Frequency : always, for aggregate stripe size over 4GB
3943 Description: assertion fails when trying to use 4GB stripe size
3944 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3945 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3948 Frequency : always on ppc64
3950 Description: the first write on an ext3 filesystem with mballoc got stuck
3951 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3952 perform endianness conversion.
3954 ------------------------------------------------------------------------------
3956 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3958 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3959 INTEROPERATE with older versions automatically. Please read the
3960 user documentation before upgrading any part of a live system.
3961 * WARNING: Lustre networking configuration changes are required with
3962 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3965 * Support for kernels:
3966 2.6.9-22.0.2.EL (RHEL 4)
3967 2.6.5-7.244 (SLES 9)
3968 2.6.12.6 vanilla (kernel.org)
3971 Severity : enhancement
3972 Bugzilla : 7981/8208
3973 Description: Introduced Lustre Networking (LNET)
3974 Details : LNET is new networking infrastructure for Lustre, it includes
3975 a reorganized network configuration mode (see the user
3976 documentation for full details) as well as support for routing
3977 between different network fabrics. Lustre Networking Devices
3978 (LNDs) for the supported network fabrics have also been
3979 created for this new infrastructure.
3981 Severity : enhancement
3982 Description: Introduced Access control lists
3983 Details : clients can set ACLs on files and directories in order to have
3984 more fine-grained permissions than the standard Unix UGO+RWX.
3985 The MDS must be started with the "-o acl" mount option.
3987 Severity : enhancement
3988 Description: Introduced filesystem quotas
3989 Details : Administrators may now establish per-user quotas on the
3992 Severity : enhancement
3994 Description: Configuration change for the XT3
3995 The PTLLND is now used to run Lustre over Portals on the XT3
3996 The configure option(s) --with-cray-portals are no longer used.
3997 Rather --with-portals=<path-to-portals-includes> is used to
3998 enable building on the XT3. In addition to enable XT3 specific
3999 features the option --enable-cray-xt3 must be used.
4004 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
4005 Details : A race between allocating a new llcd and re-getting the llcd_lock
4006 allowed another thread to grab newly-allocated llcd.
4008 Severity : enhancement
4010 Description: 2.6 OST async journal commit and locking fix to improve performance
4011 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
4012 2.6 kernels are now async as they already were in 2.4 kernels so
4013 that they can commit concurrently with the network bulk transfer.
4014 For block-allocated files the filter allocation semaphore is held
4015 to avoid filesystem fragmentation during allocation. BKL lock
4016 removed for 2.6 xattr operations where it is no longer needed.
4021 Description: lconf incorrectly determined whether two IP networks could talk
4022 Details : In some more complicated routing and multiple-network
4023 configurations, lconf will avoid trying to make a network
4024 connection to a disjoint part of the IP space. It was doing the
4025 math incorrectly for one set of cases.
4030 Description: Fix for potential infinite loop processing records in an llog.
4031 Details : If an llog record is corrupted/zeroed, it is possible to loop
4032 forever in llog_process(). Validate the llog record length
4033 and skip the remainder of the block on error.
4036 Frequency : occasional (liblustre only)
4038 Description: liblustre could not open files whose last component is a symlink
4039 Details : sysio_path_walk() would incorrectly pass the open intent to
4040 intermediate path components.
4043 Frequency : rare (liblustre only with non-standard tuning)
4044 Bugzilla : 7201 (7350)
4045 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
4046 Details : llu_lookup_finish_locks() tries to set lock data on a lock
4047 after it has been released, only do this for referenced locks
4049 Severity : enhancement
4051 Description: specifying an (invalid) directory default stripe_size of -1
4052 would reset the directory default striping
4053 Details : stripe_size -1 was used internally to signal directory stripe
4054 removal, now use "all default" to signal dir stripe removal
4055 as a directory striping of "all default" is not useful
4058 Frequency : common for large clusters running liblustre clients
4060 Description: doing an ls when liblustre clients are running is slow
4061 Details : sending a glimpse AST to a liblustre client waits for every AST
4062 to time out, as liblustre clients will not respond. Since they
4063 cannot cache data we refresh the OST lock LVB from disk instead.
4065 Severity : enhancement
4067 Description: doing an ls at the same time as file IO can be slow
4068 Details : enqueue and other "small" requests can be blocked behind many
4069 large IO requests. Create a new OST IO portal for non-IO
4070 requests so they can be processed faster.
4073 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
4075 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
4076 created file with mode 000
4077 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4078 allow an ftruncate() as a non-root user to a file with mode 000.
4079 Lustre now respects this flag to disable mode checks when
4080 truncating a file owned by the user
4083 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4086 Description: Revoking locks from clients that went dead or catatonic might take
4088 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4089 cancellation of such locks instant on servers without waiting for
4090 any reply from clients. Clients drops these locks when cancel
4091 notification from server is received without replying.
4094 Frequency : liblustre-only, when liblustre client dies or becomes busy
4096 Description: Doing ls on Linux clients can take a long time with active
4098 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4099 granting such locks to it in the first place if possible. Locks
4100 are taken by proxy on the OST during the read or write and
4101 dropped immediately afterward. Add connect flags handling, do
4102 not grant locks to liblustre clients for glimpse ASTs.
4104 Severity : enhancement
4106 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4107 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4108 sequential reads and behaves badly with applications doing random
4109 reads. Improve it by reading ahead at least read region, and
4110 avoiding excessive large RPC for small reads.
4112 Severity : enhancement
4114 Description: Creating more than 1000 files for a single job may cause a load
4115 imbalance on the OSTs if there are also a large number of OSTs.
4116 Details : qos_prep_create() uses an OST index reseed value that is an
4117 even multiple of the number of available OSTs so that if the
4118 reseed happens in the middle of the object allocation it will
4119 still utilize the OSTs as uniformly as possible.
4124 Description: OST or MDS may oops in ping_evictor_main()
4125 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4126 but doesn't restart at beginning of obd_exports_timed list
4129 Severity : enhancement
4131 Description: improve by-nid export eviction on the MDS and OST
4132 Details : allow multiple exports with the same NID to be evicted at one
4133 time without re-searching the exports list.
4136 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4138 Description: MDS may oops in groups_free()
4139 Details : in rare race conditions a newly allocated group_info struct is
4140 freed again, and this can be NULL. The 2.4 compatibility code
4141 for groups_free() checked for a NULL pointer, but 2.6 did not.
4144 Frequency : common for liblustre clients doing little filesystem IO
4145 Bugzilla : 9352, 7313
4146 Description: server may evict liblustre clients accessing contended locks
4147 Details : if a client is granted a lock or receives a completion AST
4148 with a blocking AST already set it would not reply to the AST
4149 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4152 Frequency : lfs setstripe, only systems with more than 160 OSTs
4154 Description: unable to set striping with a starting offset beyond OST 160
4155 Details : llapi_create_file() incorrectly limited the starting stripe
4156 index to the maximum single-file stripe count.
4159 Frequency : LDAP users only
4161 Description: lconf did not handle in-kernel recovery with LDAP properly
4162 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4164 Severity : enhancement
4166 Description: bind OST threads to NUMA nodes to improve performance
4167 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4168 node and do their allocations there to localize memory access
4170 Severity : enhancement
4172 Description: llmount can determine client NID directly from Myrinet (GM)
4173 Details : the client NID code from gmnalnid was moved directly into
4174 llmount, removing the need to use this or specifying the
4175 client NID explicitly when mounting GM clients with zeroconf
4178 Frequency : if client is started with down MDS
4180 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4181 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4182 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4183 ptlrpc_import_delayed_req() because we didn't check if the
4184 request was interrupted, and we also didn't break out of the
4185 event loop if there was no timeout
4190 Description: data loss during non-page-aligned writes to a single file from
4191 both multiple nodes and multiple threads on one node at same time
4192 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4193 inconsistency led to false short-reads, that were cached and later
4194 used by ->prepare_write() to fill in partially written page,
4195 leading to data loss.
4198 Frequency : always, if lconf --abort_recovery used
4200 Description: lconf --abort_recovery fails with 'Operation not supported'
4201 Details : lconf was attempting to abort recovery on the MDT device and not
4204 Severity : enhancement
4206 Description: remove cleanup logs
4207 Details : replace lconf-generated cleanup logs with lustre internal
4208 cleanup routines. Eliminates the need for client-cleanup and
4211 Severity : enhancement
4213 Description: add support for EAs (user and system) on lustre filesystems
4214 Details : it is now possible to store extended attributes in the Lustre
4215 client filesystem, and with the user_xattr mount option it
4216 is possible to allow users to store EAs on their files also
4218 Severity : enhancement
4220 Description: Add possibility (config option) to show minimal available OST free
4222 Details : When compiled with --enable-mindf configure option, statfs(2)
4223 (and so, df) will return least minimal free space available from
4224 all OSTs as amount of free space on FS, instead of summary of
4225 free spaces of all OSTs.
4227 Severity : enhancement
4229 Description: do not expand extent locks acquired on OST-side
4230 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4231 by server: they are not cached anyway.
4234 Frequency : when mmap is used/binaries executed from Lustre
4236 Description: Unmmap pages before throwing them away from read cache.
4237 Details : llap_shrink cache now attempts to unmap pages before discarding
4238 them (if unmapping failed - do not discard). SLES9 kernel has
4239 extra checks that trigger if this unmapping is not done first.
4244 Description: lconf didn't resolve symlinks before checking to see whether a
4245 given mountpoint was already in use
4248 Frequency : when migrating failover services
4249 Bugzilla : 6395, 9514
4250 Description: When migrating a subset of services from a node (e.g. failback
4251 from a failover service node) the remaining services would
4252 time out and evict clients.
4253 Details : lconf --force (implied by --failover) sets the global obd_timeout
4254 to 5 seconds in order to quickly disconnect, but this caused
4255 other RPCs to time out too quickly. Do not change the global
4256 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4258 Severity : enhancement
4259 Frequency : if MDS is started with down OST
4260 Bugzilla : 9439,5706
4261 Description: Allow startup/shutdown of an MDS without depending on the
4262 availability of the OSTs.
4263 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4264 Add appropriate locking and lov-osc refcounts for safe
4265 cleaning. Add osc abort_inflight calls in case the
4266 synchronize never started.
4269 Frequency : occasional (Cray XT3 only)
4271 Description: root not authorized to access files in CRAY_PORTALS environment
4272 Details : The client process capabilities were not honoured on the MDS in
4273 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4274 been accessed by an authorized user then root was able to access
4275 the file on the local client also. The root user capabilities
4276 are now allowed on the MDS, as this environment has secure UID.
4279 Frequency : occasional
4281 Description: ldiskfs "too long searching" message happens too often
4282 Details : A debugging message (otherwise harmless) prints too often on
4283 the OST console. This has been reduced to only happen when
4284 there are fragmentation problems on the filesystem.
4289 Description: Division by zero in statfs when all OSCs are inactive
4290 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4291 lov_statfs divides by value returned by lov_get_stripecnt().
4295 Bugzilla : 9489, 3273
4296 Description: First write from each client to each OST was only 4kB in size,
4297 to initialize client writeback cache, which caused sub-optimal
4298 RPCs and poor layout on disk for the first writen file.
4299 Details : Clients now request an initial cache grant at (re)connect time
4300 and so that they can start streaming writes to the cache right
4301 away and always do full-sized RPCs if there is enough data.
4302 If the OST is rebooted the client also re-establishes its grant
4303 so that client cached writes will be honoured under the grant.
4308 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4309 Details : Now I/O RPCs go to different portal number and (presumably) fast
4310 lock requests (and glimses) and other RPCs get their own service
4311 threads pool that should be able to service those RPCs
4314 Severity : enhancement
4316 Description: Ability to exchange lustre version between client and servers and
4317 issue warnings at client side if client is too old. Also for
4318 liblustre clients there is ability to refuse connection of too old
4320 Details : New 'version' field is added to connect data structure that is
4321 filled with version info. That info is later checked by server and
4325 Frequency : rare, liblustre only.
4326 Bugzilla : 9296, 9581
4327 Description: Two simultaneous writes from liblustre at offset within same page
4328 might proceed at the same time overwriting eachother with stale
4330 Details : I/O lock withing llu_file_prwv was released too early, before data
4331 actually was hitting the wire. Extended lock-holding time until
4332 server acknowledges receiving data.
4335 Frequency : extremely rare. Never observed in practice.
4337 Description: avoid generating lustre_handle cookie of 0.
4338 Details : class_handle_hash() generates handle cookies by incrementing
4339 global counter, and can hit 0 occasionaly (this is unlikely, but
4340 not impossible, because initial value of cookie counter is
4341 selected randonly). Value of 0 is used as a sentinel meaning
4342 "unassigned handle" --- avoid it. Also coalesce two critical
4343 sections in this function into one.
4345 Severity : enhancement
4347 Description: allow liblustre clients to delegate truncate locking to OST
4348 Details : To avoid overhead of locking, liblustre client instructs OST to
4349 take extent lock in ost_punch() on client's behalf. New connection
4350 flag is added to handle backward compatibility.
4352 Severity : enhancement
4353 Bugzilla : 4928, 7341, 9758
4354 Description: allow number of OST service threads to be specified
4355 Details : a module parameter allows the number of OST service threads
4356 to be specified via "options ost ost_num_threads={N}" in the
4357 OSS's /etc/modules.conf or /etc/modprobe.conf.
4361 Bugzilla : 6146, 9635, 9895
4362 Description: servers crash with bad pointer in target_handle_connect()
4363 Details : In rare cases when a client is reconnecting it was possible that
4364 the connection request was the last reference for that export.
4365 We would temporarily drop the export reference and get a new
4366 one, but this may have been the last reference and the export
4367 was just destroyed. Get new reference before dropping old one.
4369 Severity : enhancement
4370 Frequency : if client is started with failover MDS
4372 Description: Allow multiple MDS hostnames in the mount command
4373 Details : Try to read the configuration from all specified MDS
4374 hostnames during a client mount in case the "primary"
4377 Severity : enhancement
4379 Description: Stop sending data to evicted clients as soon as possible.
4380 Details : Check if the client we are about to send or are sending data to
4381 was evicted already. (Check is done every second of waiting,
4382 for which l_wait_event interface was extended to allow checking
4383 of exit condition at specified intervals).
4386 Frequency : rare, normally only when NFS exporting is done from client
4388 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4390 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4393 Frequency : always, if srand() is called before liblustre initialization
4395 Description: Liblustre uses system PRNG disturbing its usage by user application
4396 Details : Introduce internal to lustre fast and high-quality PRNG for
4397 lustre usage and make liblustre and some other places in generic
4398 lustre code to use it.
4400 Severity : enhancement
4401 Bugzilla : 9477, 9557, 9870
4402 Description: Verify that the MDS configuration logs are updated when xml is
4403 Details : Check if the .xml configuration logs are newer than the config
4404 logs stored on the MDS and report an error if this is the case.
4405 Request --write-conf, or allow starting with --old_conf.
4407 Severity : enhancement
4409 Description: Handle symlinks in the path when checking if Lustre is mounted.
4410 Details : Resolve intermediate symlinks when checking if a client has
4411 mounted a filesystem to avoid duplicate client mounts.
4416 Description: lconf can hit an error exception but still return success.
4417 Details : The lconf command catches the Command error exception at the top
4418 level script context and will exit with the associated exit
4419 status, but doesn't ensure that this exit status is non-zero.
4424 Description: failure of ptlrpc thread startup can cause oops
4425 Details : Starting a ptlrpc service thread can fail if there are a large
4426 number of threads or the server memory is very fragmented.
4427 Handle this without oopsing.
4430 Frequency : always, only if liblustre and non-default acceptor port was used
4432 Description: liblustre cannot connect to servers with non-default acceptor port
4433 Details : tcpnal_set_default_params() was not called and was therefore
4434 ignoring the environment varaible TCPNAL_PORT, as well as other
4435 TCPNAL_ environment variables
4440 Description: two objects could be created on the same OST for a single file
4441 Details : If an OST is down, in some cases it was possible to create two
4442 objects on a single OST for a single file. No problems other
4443 than potential performance impact and spurious error messages.
4447 Bugzilla : 5681, 9562
4448 Description: Client may oops in ll_unhash_aliases
4449 Details : Client dcache may become inconsistent in race condition.
4450 In some cases "getcwd" can fail if the current directory is
4456 Description: Inode refcounting problems in NFS export code
4457 Details : link_raw functions used to call d_instantiate without obtaining
4458 extra inode reference first.
4462 Bugzilla : 9942, 9903
4463 Description: Referencing freed requests leading to crash, memleaks with NFS.
4464 Details : We used to require that call to ll_revalidate_it was always
4465 followed by ll_lookup_it. Also with revalidate_special() it is
4466 possible to call ll_revalidate_it() twice for the same dentry
4467 even if first occurence returned success. This fix changes semantic
4468 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4469 reference on a request referred from the intent.
4470 ll_intent_release() then releases such a request.
4473 Frequency : rare, normally benchmark loads only
4475 Description: unlinked inodes were kept in memory on the client
4476 Details : If a client is repeatedly creating and unlinking files it
4477 can accumulate a lot of stale inodes in the inode slab cache.
4478 If there is no other client load running this can cause the
4479 client node to run out of memory. Instead flush old inodes
4480 from client cache that have the same inode number as a new inode.
4483 Frequency : SLES9 2.6.5 kernel and long filenames only
4484 Bugzilla : 9969, 10379
4485 Description: utime reports stale NFS file handle
4486 Details : SLES9 uses out-of-dentry names in some cases, which confused
4487 the lustre dentry revalidation. Change it to always use the
4491 Frequency : rare, unless heavy write-truncate concurrency is continuous
4492 Bugzilla : 4180, 6984, 7171, 9963, 9331
4493 Description: OST becomes very slow and/or deadlocked during object unlink
4494 Details : filter_destroy() was holding onto the parent directory lock
4495 while truncating+unlinking objects. For very large objects this
4496 may block other threads for a long time and slow overall OST
4497 responsiveness. It may also be possible to get a lock ordering
4498 deadlock in this case, or run out of journal credits because of
4499 the combined truncate+unlink. Solution is to do object truncate
4500 first in one transaction without parent lock, and then do the
4501 final unlink in a new transaction with the parent lock. This
4502 reduces the lock hold time dramatically.
4505 Frequency : rare, 2.4 kernels only
4507 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4508 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4509 before unmounting their filesystem in order to avoid deadlock.
4510 The kernel_locked() function in 2.4 kernels only checks whether
4511 the kernel lock is held, not whether it is this process that is
4512 holding it as 2.6 kernels do.
4517 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4518 Details : The client ptlrpc code may be trying to reconnect to a down
4519 server before a previous connection attempt has timed out.
4520 Increase the reconnect interval to be longer than the connection
4521 timeout interval to avoid sending duplicate connections to
4525 Frequency : echo_client brw_test command
4527 Description: fix echo_client to work with OST preallocated code
4528 Details : OST preallocation code (5137) didn't take echo_client IO path
4529 into account: echo_client calls filter methods outside of any
4530 OST thread and, hence, there is no per-thread preallocated
4531 pages and buffers to use. Solution: hijack pga pages for IO. As
4532 a byproduct, this avoids unnecessary data copying.
4536 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4537 Description: Client can oops in mdc_commit_close() after open replay
4538 Details : It was possible for the MDS to return an open request with no
4539 transaction number in mds_finish_transno() if the client was
4540 evicted, but without actually returning an error. Clients
4541 would later try to replay that open and may trip an assertion
4542 Simplify the client close codepath, and always return an error
4543 from the MDS in case the open is not successful.
4546 Frequency : rare, 2.6 OSTs only
4548 Description: OST may deadlock under high load on fragmented files
4549 Details : If there was a heavy load and highly-fragmented OST filesystems
4550 it was possible to have all the OST threads deadlock waiting on
4551 allocation of biovecs, because the biovecs were not released
4552 until the entire RPC IO was completed. Instead, release biovecs
4553 as soon as they are complete to ensure forward IO progress.
4555 Severity : enhancement
4557 Description: Support for specifying external journal device at mount
4558 Details : If an OST or MDS device is formatted with an external journal
4559 device, this device major/minor is stored in the ext3 superblock
4560 and may not be valid for failover. Allow detecting and
4561 specifying the external journal at mount time.
4566 Description: Mounting an MDS with pending unlinked files may cause oops
4567 Details : target_finish_recovery() calls mds_postrecov() which returned
4568 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4569 considers this an error and immediately begins cleaning up the
4570 lov, just after starting the mds_lov process
4572 Severity : enhancement
4574 Description: Implement 'lfs df' to report actual free space on per-OST basis
4575 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4576 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4577 usage of objects; '-h' to report in human readable format.
4579 ------------------------------------------------------------------------------
4581 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4588 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4589 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4590 at startup during orphan unlinking. Instead, skip buddy update
4591 before setup, as it will be regenerated after recovery is complete.
4596 Description: If an OST is inactive, its locks might reference stale inodes.
4597 Details : lov_change_cbdata() must iterate over all namespaces, even if
4598 they are inactive to clear inode references from the lock.
4600 Severity : enhancement
4601 Frequency : occasional, if non-standard max_dirty_mb used
4603 Description: Client will block write RPCs if not enough grant
4604 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4605 then the client will block writes while waiting for another RPC
4606 to complete instead of consuming its dirty limit. With change
4607 we get improved performance when max_dirty_mb is small.
4609 Severity : enhancement
4610 Bugzilla : 3389, 6253
4611 Description: Add support for supplementary groups on the MDS.
4612 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4613 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4614 lookups for user supplementary groups into a cache.
4618 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4619 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4620 for liblustre O_EXCL is enforced by MDS.
4624 Description: readdir never returns NULL in liblustre.
4625 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4626 dentry in struct dirent.
4630 Frequency : liblustre only, depends on application IO pattern
4631 Description: liblustre clients evicted if not contacting servers
4632 Details : Don't put liblustre clients into the ping_evictor list, so
4633 they will not be evicted by the pinger ever.
4635 Severity : enhancement
4637 Description: Add ability to evict clients by NID from MDS.
4638 Details : By echoing "nid:$NID" string into
4639 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4640 $NID would be instantly evicted from this MDS and from all active
4641 OSTs connected to it.
4645 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4646 Details : lookup_it_finish() used to query file size from OSTs that was not
4651 Description: service threads change working directory to that of init
4652 Details : Starting lustre service threads may pin the working directory
4653 of the parent thread, making that filesystem busy. Threads
4654 now change to the working directory of init to avoid this.
4658 Frequency : during shutdown only
4659 Description: shutdown with a failed MDS or OST can cause unmount to hang
4660 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4665 Frequency : 2.6 only
4666 Description: chmod/chown may include an extra supplementary group
4667 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4668 supplementary group and if none is specified this is used.
4671 Bugzilla : 5479 (6816)
4673 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4674 Details : If lookup is in progress on a file that is unlinked we might try
4675 to revalidate the inode and fail in revalidate after lookup is
4676 complete and ll_file_open() enqueues the open again but
4677 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4680 Frequency : always, if lconf --abort_recovery used
4682 Description: lconf --abort_recovery fails with 'Operation not supported'
4683 Details : lconf was attempting to abort recovery on the MDT device and not
4686 ------------------------------------------------------------------------------
4688 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4693 Frequency : rare (only unsupported configurations with a node running as an
4695 Bugzilla : 6514, 5137
4696 Description: Mounting a Lustre file system on a node running as an OST could
4698 Details : OSTs now preallocates memory needed to write out data at
4699 startup, instead of when needed, to avoid having to
4700 allocate memory in possibly low memory situations.
4701 Specifically, if the file system is mounted on on OST,
4702 memory pressure could force it to try to write out data,
4703 which it needed to allocate memory to do. Due to the low
4704 memory, it would be unable to do so and the node would
4705 become unresponsive.
4707 Severity : enhancement
4709 Description: Addition of lconf --service command line option
4710 Details : lconf now accepts a '--service <arg>' option, which is
4711 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4713 Severity : enhancement
4715 Description: Failover mode is now the default for OSTs.
4716 Details : By default, OSTs will now run in failover mode. To return to
4717 the old behaviour, add '--failout' to the lmc line for OSTs.
4719 Severity : enhancement
4721 Description: Health checks are now provided for MDS and OSTs
4722 Details : Additional detailed health check information on MSD and OSTs
4723 is now provided through the procfs health_check value.
4726 Frequency : occasional, depends on IO load
4728 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4729 after numerous create/delete cycles
4730 Details : The ext3 inode allocation policy would not allocate new inodes
4731 very well on the OSTs because there are no new directories
4732 being created. Instead we look for groups with free space if
4733 the parent directories are nearly full.
4738 Description: Network or server problems during mount may cause partially
4739 mounted clients instead of returning an error.
4740 Details : The config llog parsing code may overwrite the error return
4741 code during mount error handling, returning success instead
4747 Description: MDS can fail to allocate large reply buffers
4748 Details : After long uptimes the MDS can fail to allocate large reply
4749 buffers (e.g. zconf client mount config records) due to memory
4750 fragmentation or consumption by the buffer cache. Preallocate
4751 some large reply buffers so that these replies can be sent even
4752 under memory pressure.
4756 Frequency : rare (liblustre)
4757 Description: fsx running with liblustre complained that using truncate() to
4758 extend the file doesn't work. This patch corrects that issue.
4759 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4760 ATTR_SIZE and lsm use in llu_setattr_raw.
4764 Frequency : rare, only 2.6 kernels
4765 Description: Unusual file access patterns on the MDS may result in inode
4766 data being lost in very rare circumstances.
4767 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4768 code in the 2.6 kernel caused the inode and/or EA data not to
4769 be read from disk, causing single-file corruption.
4773 Frequency : rare, only 2.6 filesystems using extents
4774 Description: Heavy concurrent write and delete load may cause data corruption.
4775 Details : It was possible under high-load situations to have an extent
4776 metadata block in the block device cache from a just-unlinked
4777 file overwrite a newly-allocated data block. We now unmap any
4778 metadata buffers that alias just-allocated data blocks.
4782 Frequency : filesystems with default stripe_count larger than 77
4783 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4784 Details : lconf specifies an inode size of 4096 bytes when the default
4785 stripe_count is larger than 77. This conflicts with the default
4786 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4787 this case to avoid pinning too much memory for large EAs.
4789 ------------------------------------------------------------------------------
4791 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4796 Frequency : rare (extremely heavy IO load with hundreds of clients)
4798 Description: Client is evicted, gets IO error writing to file
4799 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4800 caused clients to be evicted for AST timeouts. The fixes in
4801 bug 5192 mean we no longer need to have such short AST timeouts
4802 so ldlm_timeout has been increased.
4805 Frequency : occasional during --force or --failover shutdown under load
4806 Bugzilla : 5949, 4834
4807 Description: Server oops/LBUG if stopped with --force or --failover under load
4808 Details : a collection of import/export refcount and cleanup ordering
4809 issues fixed for safer force cleanup
4812 Frequency : only filesystems larger than 120 OSTs
4813 Bugzilla : 5990, 6223
4814 Description: lfs getstripe would oops on a very large filesystem
4815 Details : lov_getconfig used kfree on vmalloc'd memory
4818 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4819 Bugzilla : 6242, 6243
4820 Description: reading from files that had been truncated to a non-zero size
4821 but never opened returned no data
4822 Details : ll_file_read() reads zeros from no-object files to EOF
4827 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4828 Details : There's little harm in aborting MDS/OSS recovery and letting it
4829 try again, so I removed the LASSERT and return an error instead.
4831 Severity : enhancement
4833 Description: New debugging infrastructure for tracking down data corruption
4834 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4835 (b) cover more of the client-side code path, and (c) try to narrow
4836 down where problems occurred
4840 Bugzilla : 3819, 4364, 4397, 6313
4841 Description: Racing close and eviction MDS could cause assertion in mds_close
4842 Details : It was possible to get multiple mfd references during close and
4843 client eviction, leading to one thread referencing a freed mfd.
4845 Severity: : enhancement
4846 Bugzilla : 3262, 6359
4847 Description: Attempts to reconnect to servers are now more aggressive.
4848 Details : This builds on the enhanced upcall-less recovery that was added
4849 in 1.4.2. When trying to reconnect to servers, clients will
4850 now try each server in the failover group every 10 seconds. By
4851 default, clients would previously try one server every 25 seconds.
4856 Description: After recovery, certain operations trigger a failed
4857 assertion on a client.
4858 Details : Failing over an mds, using lconf -d --failover, while a
4859 client was doing a readdir() call would cause the client to
4860 LBUG after recovery completed and the readdir() was resent.
4862 Severity : enhancement
4864 Description: Default groups are now added by lconf
4865 Details : You can now run lconf --group <servicename> without having to
4866 manually add groups with lmc.
4869 Frequency : occasional
4871 Description: Nodes with an elan id of 0 trigger a failed assertion
4874 Frequency : always when accessing e.g. tty/console device nodes
4876 Description: tty and some other devices nodes cannot be used on lustre
4877 Details : file's private_data field is used by device data and lustre
4878 values in there got lost. New field was added to struct file to
4879 store fs-specific private data.
4882 Frequency : when exporting Lustre via NFS
4884 Description: NFSD failed occasionally when looking up a path component
4885 Details : NFSD is looking up ".." which was broken in ext3 directories
4886 that had grown large enough to become hashed.
4889 Frequency : Clusters with multiple interfaces not on the same subnet
4891 Description: Nodes will repeatedly try to reconnect to an interface which it
4892 cannot reach and report an error to the log.
4893 Details : Extra peer list entries will be created by lconf with some peers
4894 unreachable. lconf now validates the peer before adding it.
4897 Frequency : Only if a default stripe is set on the filesystem root.
4899 Description: Setting a default stripe on the filesystem root prevented the
4900 filesystem from being remounted.
4901 Details : The client was sending extra request flags in the root getattr
4902 request and did not allocate a reply buffer for the dir EA.
4905 Frequency : occasional, higher if lots of files are accessed by one client
4906 Bugzilla : 6159, 6097
4907 Description: Client trips assertion regarding lsm mismatch/magic
4908 Details : While revalidating inodes the VFS looks up inodes with ifind()
4909 and in rare cases can find an inode that is being freed.
4910 The ll_test_inode() code will free the lsm during ifind()
4911 when it finds an existing inode and then the VFS later attaches
4912 this free lsm to a new inode.
4916 Bugzilla : 6422, 7030
4917 Description: MDS deadlock between mkdir and client eviction
4918 Details : Creating a new file via mkdir or mknod (starting a transaction
4919 and getting the ns lock) can deadlock with client eviction
4920 (gets ns lock and trying to finish a synchronous transaction).
4923 Frequency : occasional
4924 Description: While starting a server, the fsfilt_ext3 module could not be
4926 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4927 kernels. Previously, lconf would still use the ext3 name
4928 when trying to load modules. Now, it will correctly use
4929 ext3 on 2.4 and ldiskfs on 2.6.
4931 Severity : enhancement
4932 Description: The default stripe count has been changed to 1
4933 Details : The interpretation of the default stripe count (0, to lfs
4934 or lmc) has been changed to mean striping across a single
4935 OST, rather than all available. For general usage we have
4936 found a stripe count of 1 or 2 works best.
4938 Severity : enhancement
4939 Description: Add support for compiling against Cray portals.
4940 Details : Conditional compiling for some areas that are different
4944 Frequency : occasional
4945 Bugzilla : 6409, 6834
4946 Description: Creating files with an explicit stripe count may lead to
4947 a failed assertion on the MDS
4948 Details : If some OSTs are full or unavailable, creating files may
4949 trigger a failed assertion on the MDS. Now, Lustre will
4950 try to use other servers or return an error to the
4954 Frequency : occasional
4956 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4957 caused lock timeout during readahead (since 1.4.2).
4958 Details : Processes doing readahead might match a lock that hasn't been
4959 granted yet if there are overlapping and conflicting lock
4960 requests. The readahead process waits on ungranted lock
4961 (original lock is CBPENDING), while OST waits for that process
4962 to cancel CBPENDING read lock and eventually evicts client.
4964 Severity : enhancement
4966 Description: Initial enabling of flock support for clients
4967 Details : Implements fcntl advisory locking and file status functions.
4968 This feature is provided as an optional mount flag (default
4969 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4970 locking are implemented yet, and those that are are not guaranteed
4971 to be completely correct in production environments.
4972 mount -t lustre -o [flock|noflock] ...
4975 Frequency : occasional
4977 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4978 assertion in the ext3 JBD (journaling) layer.
4979 Details : The b_committed_data struct is protected by the big kernel lock
4980 in 2.4 kernels, serializing journal_commit_transaction() and
4981 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4982 access to this struct. In 2.6 kernels there is finer grained
4983 locking to improve SMP performance of the JBD layer.
4987 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4992 Description: merge and cleanup kernel patches.
4995 -----------------------------------------------------------------------------
4997 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4999 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
5000 versions of Lustre. Please update all servers and clients to
5001 version 1.4.2 or later at the same time. You must also run
5002 "lconf --write-conf {config}.xml" on the MDS while it is stopped
5003 to update the configuration logs.
5005 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
5006 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
5007 - swab llog records of type '0' so we get proper header size/idx (5861)
5008 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
5009 - fix rename of one directory over another leaking an inode (5953)
5010 - avoid SetPageDirty on 2.6 (5981)
5011 - don't re-add just-being-destroyed locks to the waiting list (5653)
5012 - when creating new directories, inherit the parent's custom
5013 striping settings if present parent (3048)
5014 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
5015 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
5016 - don't hold client locks on temporary worklist from l_lru (5666)
5017 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
5018 - automatically evict dead clients (5921)
5019 - Update file size properly in create+truncate+fstat case (6196)
5020 - Do not unhash mountpoint dentries, do not allow removal of
5022 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
5023 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
5024 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
5025 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
5026 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
5027 - Correctly handle reads of files with no objects (6243)
5028 - lctl recover will also mark a device active if deactivate used (5933)
5030 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
5031 - allow --write-conf on an MDS with different nettype than client (5619)
5032 - don't write config llogs to MDS for mounts not from that MDS (5617)
5033 - lconf should create multiple TCP connections from a client (5201)
5034 - init scripts are now turned off by default; run chkconfig --on
5035 lustre and chkconfig --on lustrefs to use them
5036 - upcalls are no longer needed for clients to recover to failover
5038 - add --abort-recovery option to lconf to abort recovery on device
5040 - add support for an arbitrary number of OSTs (3026)
5041 - Quota support protocol changes.
5042 - forward compatibility changes to wire structs (6007)
5043 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
5044 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
5045 - improved "lustre status" script
5046 - initialize blocksize for non-regular files (6062)
5047 - added --disable-server and --disable-client configure options (5782)
5048 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
5049 - Vanilla 2.4.29 support
5050 - increase maximum number of obd devices to 520 (6242)
5051 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
5052 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
5053 SLES 9 2.6.5 kernels
5054 - limit stripes per file to 160 (the maximum EA size) (6093)
5056 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5059 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
5060 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
5061 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
5062 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
5063 - small ext3 extents cleanups and fixes (5733)
5064 - improved mballoc code, several small races and bugs fixed (5733, 5638)
5065 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
5066 - avoid needless client->OST connect, fix handle mismatch (5317)
5067 - fix DLM error path that led to out-of-sync client, long delays (5779)
5068 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
5069 - fix several locking issues related to i_size (5492,5624,5654,5672)
5070 - don't move pending lock onto export if it is already evicted (5683)
5071 - fix kernel oops when creating .foo in unlinked directory (5548)
5072 - fix deadlock in obdfilter statistics vs. object create (5811)
5073 - use time_{before,after} to avoid timer jiffies wrap (5882)
5074 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
5075 - Do not leak request if server was not able to process it (5154)
5076 - If mds_open unable to find parent dir, make that negative lookup(5154)
5077 - don't create new directories with extent-mapping (5909, 5936)
5079 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5080 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5081 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5082 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5083 - support for RHEL4 kernel on i686 (5773)
5084 - provide error messages when incompatible logs are encountered (5898)
5086 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5087 * version 1.4.0.10 (1.4.1 release candidate 1)
5089 - don't keep a lock reference when lock is not granted (4238)
5090 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5091 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5092 - drop import inflight refcount on signal_completed_replay error (5255)
5093 - unlock page after async write error during send (3677)
5094 - handle missing objects in filter_preprw_read properly (5265)
5095 - no transno return for symlink open, don't save no-trasno open (3440)
5096 - don't try to complete elan receive that already failed (4012)
5097 - free RPC server reply state on error (5406)
5098 - clean up thread from ptlrpc_start_thread() on error (5160)
5099 - readahead could read extra page into cache that wasn't ejected (5388)
5100 - prevent races in class_attach/setup/cleanup/detach (5260)
5101 - don't dereference de->d_inode after l_dput of de (5458)
5102 - use "int" for stripe value returned from lock_to_stripe (5544)
5103 - mballoc allocation and error-checking fixes in 2.6 (5504)
5104 - block device patches to fix I/O request sizes in 2.6 (5482)
5105 - look up hostnames for IB nals (5602)
5106 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5107 - don't start multiple acceptors for the same port (5277)
5108 - fix incorrect LASSERT in mds_getattr_name (5635)
5109 - export a proc file for general "ping" checking (5628)
5110 - fix "lfs check" to not block when the MDS is down (5628)
5112 - service request history (4965)
5113 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5114 - create an "evict_client" /proc entry on OSTs, like the MDS has
5115 - fix mount usage message, return errors per mount(8) (5168)
5116 - change grep [] to grep "[]" in tests so they work in more UMLs
5117 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5118 - remove ancient LOV_MAGIC_V0 EA support (5047)
5119 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5120 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5121 - allow readahead allocations to fail when low on memory (5383)
5122 - mmap locking landed again, after considerable improvement (2828)
5123 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5125 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5128 - send OST transaction number in read/write reply to free req (4966)
5129 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5130 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5131 - the watchdog thread now runs as interruptible (5246)
5132 - flock/lockf fixes (but it's still disabled, pending 5135)
5133 - don't use EXT3 constants in llite code (5094)
5134 - memory shortage at startup could cause assertion (5176)
5136 - reorganization of lov code
5137 - single portals codebase
5139 - add extents/mballoc support (5025)
5140 - direct I/O reads in the obdfilter (4048)
5141 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5143 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5146 - send OST transaction number in read/write reply to free req (4966)
5147 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5148 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5149 - the watchdog thread now runs as interruptible (5246)
5150 - handle missing objects in filter_preprw_read properly (5265)
5151 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5152 - drop import inflight refcount on signal_completed_replay error (5255)
5153 - unlock page after async write error during send (3677)
5154 - return original error code on reconstructed replies (3761)
5155 - no transno return for symlink open, don't save no-trasno open (3440)
5157 - add pid to ldlm debugging output (4922)
5158 - bump the watchdog timeouts -- we can't handle 30sec yet
5159 - extra debugging for orphan dentry/inode bug (5259)
5161 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5164 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5165 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5166 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5167 - fix NULL dereference in /proc/sys/portals/routes (4827)
5168 - allow failed mdc_close() operations to be interrupted (4561)
5169 - stop precreate on OST before MDS would time out on it (4778)
5170 - don't send partial-page writes before EOF from client (4410)
5171 - discard client grant for sub-page writes on large-page clients (4520)
5172 - don't free dentries not owned by NFS code, check generation (4806)
5173 - fix lsm leak if mds_create_objects() fails (4801)
5174 - limit debug_daemon file size, always print CERROR messages (4789)
5175 - use transno after validating reply (3892)
5176 - process timed out requests if import state changes (3754)
5177 - update mtime on OST during writes, return in glimpse (4829)
5178 - add mkfsoptions to LDAP (4679)
5179 - use ->max_readahead method instead of zapping global ra (5039)
5180 - don't interrupt __l_wait_event() during strace
5182 - add software watchdogs to catch hung threads quickly (4941)
5183 - make lustrefs init script start after nfs is mounted
5184 - fix CWARN/ERROR duplication (4930)
5185 - return async write errors to application if possible (2248)
5186 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5187 - print NAL number in %x format (4645)
5188 - update barely-supported suse-2.4.21-171 series (4842)
5189 - support for sles 9 %post scripts
5190 - support for building 2.6 kernel-source packages
5191 - support for sles km_* packages
5193 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5196 - ignore -ENOENT errors in osc_destroy (3639)
5197 - notify osc create thread that OSC is being cleaned up (4600)
5198 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5199 - reconstruct ost_handle() like mds_handle() (4657)
5200 - create a new thread to do import eviction to avoid deadlock (3969)
5201 - let lconf resolve symlinked-to devices (4629)
5202 - don't unlink "objects" from directory with default EA (4554)
5203 - hold socknal file ref over connect in case target is down (4394)
5204 - allow more than 32000 subdirectories in a single directory (3244)
5205 - fix blocks count for O_DIRECT writes (3751)
5206 - OST returns ENOSPC from object create when no space left (4539)
5207 - don't send truncate RPC if file size isn't changing (4410)
5208 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5209 - bind to privileged port in socknal and tcpnal (3689)
5211 - rate limit CERROR/CWARN console message to avoid overload (4519)
5212 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5213 - basic mmap support (3918)
5214 - kernel patch series update from b1_4 (4711)
5216 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5219 - avoid crash during MDS cleanup with OST shut down (2775)
5220 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5221 - don't use bad inodes on the MDS (3744)
5222 - dynamic object preallocation to improve recovery speed (4236)
5223 - don't hold spinlock over lock dumping or change debug flags (4401)
5224 - don't zero obd_dev when it is force cleaned (3651)
5225 - print grants to console if they go negative (4431)
5226 - "lctl deactivate" will stop automatic recovery attempts (3406)
5227 - look for existing locks in ldlm_handle_enqueue() (3764)
5228 - don't resolve lock handle twice in recovery avoiding race (4401)
5229 - revalidate should check working dir is a directory (4134)
5231 - don't always mark "slow" obdfilter messages as errors (4418)
5233 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5236 - don't close LustreDB during write_conf until it is done (3860)
5237 - fix typo in lconf for_each_profile (3821)
5238 - allow dumping logs from multiple threads at one time (3820)
5239 - don't allow multiple threads in OSC recovery (3812)
5240 - fix debug_size parameters (3864)
5241 - fix mds_postrecov to initialize import for llog ctxt (3121)
5242 - replace config semaphore with spinlock (3306)
5243 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5244 - don't allow enqueue to complete on a destroyed export (3822)
5245 - down write_lock before checking llog header bitmap (3825)
5246 - recover from lock replay timeout (3764)
5247 - up llog sem before sending rpc (3652)
5248 - reduce ns lock hold times when setting kms (3267)
5249 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5250 - fix NULL deref and obd_dev leak on setup error (3312)
5251 - replace some LBUG about llog ops with error handling (3841)
5252 - don't match INVALID dentries from d_lookup and spin (3784)
5253 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5254 - fix invalid assertion in ptlrpc_set_wait (3880)
5256 - add libwrap support for the TCP acceptor (3996)
5257 - add /proc/sys/portals/routes for non-root route listing (3994)
5258 - allow setting MDS UUID in .xml (2580)
5259 - print the stack of a process that LBUGs (4228)
5261 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5264 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5265 - make sure to unset replay flag from failed open requests (3440)
5266 - if default stripe count is 0, use OST count for inode size (3636)
5267 - update parent mtime/ctime on client for create/unlink (2611)
5268 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5269 - free recovery state on server during a forced cleanup (3571)
5270 - unregister_reply for resent reqs (3063)
5271 - loop back devices mounting and status check on 2.6 (3563)
5272 - fix resource-creation race that can provoke i_size == 0 (3513)
5273 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5274 - more debugging for page-accounting assertion (3746)
5275 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5276 - avoid deadlock after precreation failure (3758)
5277 - fix race and lock order deadlock in orphan handling (3450, 3750)
5278 - add validity checks when grabbing inodes from l_ast_data (3599)
5280 - add /proc/.../recovery_status to obdfilter (3428)
5281 - lightweight CDEBUG infrastructure, debug daemon (3668)
5282 - change default OSC RPC parameters to be better on small clusters
5283 - turn off OST read cache for files smaller than 32MB
5284 - install man pages and include them in rpms (3100)
5285 - add new init script for (un)mounting lustre filesystems (2593)
5286 - run chkconfig in %post for init scripts (3701)
5287 - drop scimac NAL (unmaintained)
5289 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5292 - clean kiobufs before and after use (3485)
5293 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5294 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5295 - change init script to fail more clearly if not run as root (1528)
5296 - allow clients to reconnect during replay (1742)
5297 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5298 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5299 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5300 - fix cond_resched (3554)
5301 - only evict unfinished clients after recovery (3515)
5302 - allow bulk resend, prevent data loss (3570)
5303 - dynamic ptlrpc request buffer allocation (2102)
5304 - don't allow unlinking open directory if it isn't empty (2904)
5305 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5306 - remove extraneous obd dereference causing LASSERT failure (3334)
5307 - don't use get_cycles() when creating temp. files on the mds (3156)
5308 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5309 - handle EEXIST for set-stripe, set proper directory name (3336)
5311 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5312 - fix ksocknal_fmb_callback() error messages (2918)
5314 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5317 - don't copy lvb into (possibly NULL) reply on error (2983)
5318 - don't deref dentry after dput, don't free lvb on error (2922)
5319 - use the kms to determine writeback rpc length (2947)
5320 - increment oti_logcookies when osc is inactive (2948)
5321 - update client's i_blocks count via lvb messages (2543)
5322 - handle intent open/close of special files properly (1557)
5323 - mount MDS with errors=remount-ro, like obdfilter (2009)
5324 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5325 - don't use cancelling-locks' kms values (2947)
5326 - use highest lock extent for kms, not last one (2925)
5327 - don't dereference ERR_PTR() dentry in error handling path (3107)
5328 - fix thread race in portals_debug_dumplog() (3122)
5329 - create lprocfs device entries at setup instead of at attach (1519)
5330 - common AST error handler, don't evict client on completion race (3145)
5331 - zero nameidata in detach_mnt in 2.6 (3118)
5332 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5333 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5334 - initialize RPC timeout timer earlier for 2.6 (3219)
5335 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5336 - print nal/nid for unknown nid (3258)
5337 - additional checks for oscc recovery before doing precreate (3284)
5338 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5339 - don't crash in mdc_close for bad permissions on open (3285)
5340 - zero i_rdev for non-device files (3147)
5341 - clear page->private before handing to FS, better assertion (3119)
5342 - tune the read pipeline (3236)
5343 - fix incorrect decref of invalidated dentry (2350)
5344 - provide read-ahead stats and refine rpc in flight stats (3328)
5345 - don't hold journal transaction open across create RPC (3313)
5346 - update atime on MDS at close time (3265)
5347 - close LDAP connection when recovering to avoid server load (3315)
5348 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5349 - don't leak open file on MDS after open resend (3325)
5350 - serialize filter_precreate and filter_destroy_precreated (3329)
5351 - loop device shouldn't call sync_dev() for nul device (3092)
5352 - clear page cache after eviction (2766)
5353 - resynchronize MDS->OST in background (2824)
5354 - refuse to mount the same filesystem twice on same mountpoint (3394)
5355 - allow llmount to create routes for mounting behind routers (3320)
5356 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5357 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5358 - fix rare problem with rename on htree directories (3417)
5360 - allow default OST striping configuration per directory (1414)
5361 - fix compilation for qswnal for 2.6 kernels (3125)
5362 - increase maximum number of MDS request buffers for large systems
5363 - change liblustreapi to be useful for external progs like lfsck (3098)
5364 - increase local configuration timeout for slow disks (3353)
5365 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5367 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5370 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5371 - don't overwrite extent policy data in reply if lock was blocked (2901)
5372 - drop filter export grants atomically with removal from device (2663)
5373 - del obd_self_export from work_list in class_disconnect_exports (2908)
5374 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5375 - swab reply message in mdc_close, other PPC fixes (2464)
5376 - fix destroying of named logs (2325)
5377 - overwrite old logs when running lconf --write_conf (2264)
5378 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5379 - fix race in target_handle_connect (2898)
5380 - mds_reint_create() should take same inode create lock (2926)
5381 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5382 - don't close files for self_export to avoid uninitialized obd (2936)
5383 - allow MDS with the same name as client node (2939)
5384 - hold dentry reference for closed log files for unlink (2325)
5385 - reserve space for all logs during transactions (2059)
5386 - don't evict page beyond end of stripe extent (2925)
5387 - don't oops on a deleted current working directory (2399)
5388 - handle hard links to targets without a parent properly (2517)
5389 - don't dereference NULL lock when racing during eviction (2867)
5390 - don't grow lock extents when lots of conflicting locks (2919)
5392 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5395 - account for cache space usage on clients to avoid data loss (974)
5396 - lfsck support in lustre kernel code (2349)
5397 - reduce journal credits needed for BRW writes (2370)
5398 - orphan handling to avoid losing space on client/server crashes
5399 - ptlrpcd can be blocked, stopping ALL progress (2477)
5400 - use lock value blocks to assist in proper KMS, faster stat (1021)
5401 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5402 - recovery for initial connections (2355)
5403 - fixes for mds_cleanup_orphans (1934)
5404 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5405 - block all file creations until orphan recovery completes (1901)
5406 - client remove rq_connection from request struct (2423)
5407 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5408 - recovery timer race (2670)
5409 - mdc_close recovey bug (2532)
5410 - ptlrpc cleanup bug (2710)
5411 - mds timeout on local locks (2588)
5412 - namespace lock held during RPCs (2431)
5413 - handle interrupted sync write properly (2503)
5414 - don't try to handle a message that hasn't been replied to (2699)
5415 - client assert failure during cleanup after abort recovery (2701)
5416 - leak mdc device after failed mount (2712)
5417 - ptlrpc_check_set allows timedout requests to complete (2714)
5418 - wait for inflight reqs when ptlrpcd finishes (2710)
5419 - make sure unregistered services are removed from the srv_list
5420 - reset bulk XID's when resending them (caught by 1138 test)
5421 - unregister_bulk after timeout
5422 - fix lconf error (2694)
5423 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5424 - readahead locks pages, leaves pending causing memory pressure (2673)
5425 - increase OST request buffers to 4096 on large machines (2729)
5426 - fix up permission of existing directories in simple_mkdir (2661)
5427 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5428 - don't assign transno to errored transactions (2742)
5429 - don't delete objects on OST if given a bogus objid from MDS (2751)
5430 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5431 - if rq_no_resend, then timeout request after recovery (2432)
5432 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5433 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5434 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5435 - MDS recovery shouldn't delete objects that it has given out (2730)
5436 - if enqueue arrives after completion, don't clobber LVB (2819)
5437 - don't unlock pages twice when trigger_group_io returns error (2814)
5438 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5439 - don't write pages to disk if there was an error (1450)
5440 - don't ping imports that have recovery disabled (2676)
5441 - take buffered bytes into account when balancing socknal conn (2817)
5442 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5443 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5444 - remove little-endian swabbing of llog records (1987)
5445 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5446 - retry reposting request buffers if they fail (1191)
5447 - grow extent at grant time to avoid granting a revoked lock (2809)
5448 - lock revoke doesn't evict page if covered by a second lock (2765)
5449 - disable VM readahead to avoid reading outside lock extents (2805)
5451 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5452 - updated kernel patches for hp-2.4.20 kernel (2681)
5454 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5457 - fix truncated write corruption (2366)
5458 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5460 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5461 - ensure len(uuid) < 37 in lmc (1171)
5462 - fix ia64 OOPS in llog_test (2255)
5463 - zero end of page at obdfilter for partial page writes (2648)
5464 - don't leave stale dentries around after renames (bug 2428)
5465 - fix timeouts when evicting a client with a single lock held (2642)
5466 - set deadline for the initial HELLO message to drain (2634)
5467 - print out dotted-quad IP addresses in the socknal (2302)
5469 - additional debugging for MDS client eviction problem (2443)
5470 - fix mkfsoptions support for osts (2603, 2604)
5472 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5475 - add series for the vanilla 2.6.0 kernel
5476 - add series for the vanilla 2.4.24 kernel
5477 - add series for a cray x86/64 UL kernel drop
5478 - fix xattr patches for the vanilla 2.4.19 series
5480 - generate true UUIDs in lmc (1171)
5481 - have portals stack dumping break in UML (2466)
5482 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5483 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5484 - wake callback waiting threads on client eviction (2460)
5485 - Add --ptldebug and --subsystem to lmc (1719)
5486 - update assertion to allow safe interrupt allocation
5487 - set rq_no_resend for cancel requests (2432)
5488 - recalculate ptlrpcd timeout after resend (2494)
5489 - call vfs_rmdir when removing pending directories (2368)
5490 - fix renaming a file to itself (2429)
5491 - lmc creates a default one-stripe lov (2454)
5492 - expand procfs space to handle large clusters (2326)
5493 - increase UML stack to avoid overflow
5494 - update lconf's list of debug and subsystem masks
5495 - fix lfs find --obd (2510)
5496 - /proc tunable for disabling filter read caching (2591)
5497 - stop rpm packages from altering slapd.conf (2301)
5498 - disable nagle in the socknal under 0conf (2578)
5499 - choose mds inode size based on stripe count (2572)
5500 - fix kernel-source rpm problems (2516)
5502 - add --disable-doc to avoid pdf generation (2421)
5503 - update documentation, tests, type-os, comments
5504 - avoid format warnings on ia64
5505 - remove the TOE NAL
5506 - tiny code cleanups by removing unused fields
5508 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5511 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5512 - remove the most blatant lies from BUILDING file (2371)
5513 - change default debug level to reasonable production setting
5514 - reduce client side cache size to reduce cache flush time
5515 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5516 - make TCP zerocopy and pinger support enabled by default (2476)
5517 - sync writes completed after process exits caused crashes (2319)
5518 - maintain correct mount count on the MDS (2356)
5519 - backout 1557, because 2316 wasn't really fixed
5520 - better file I/O statistics gathering in /proc
5521 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5522 - another kernel patch to fix zero-copy TCP function export
5523 - don't take duplicate lock when processing re-sent getattr (2420)
5524 - lctl uses obd_self_export instead of creating new conn (2353)
5525 - MDS/OST recovery case which requires object creation asserted (2425)
5526 - move lfs from /usr/sbin to /usr/bin in packages
5527 - fix race between mds_client_add and mds_client_free (2417)
5528 - use kmalloc instead of slabs in portals (2430)
5529 - don't create duplicate records when a failover MDS is present (2442)
5530 - remove unnecessary mount age check (2332)
5531 - don't remove directory inodes from locks prematurely (2451)
5532 - don't break if MDS service name is the same as hostname (2103)
5533 - fix races in client write RPC generation when cache full (2482)
5535 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5538 - remove now-unused request->rq_obd (278)
5539 - if an allocation fails, print out how much memory we've used (1933)
5540 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5541 - add the "configurable stack size" patch to most series files (1256)
5542 - ability to write large log records, for 100+ OST configs (2306)
5543 - fix NULL deref when filter_prep fails (2314)
5544 - fix operator precedence error in filter_sync
5545 - dynamic allocation of socknal TX descriptors (2315)
5546 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5547 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5548 - ensure that CWARN messages in Portals always get to the syslog
5549 - __init/__exit are not for prototype decls (ldlm_init/exit)
5550 - x86-64 compile warning fixes
5551 - fix gateway LMC keyword conflict (2318)
5552 - fix MDS lock inversions in getattr/reint paths (1844)
5553 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5554 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5555 - DLM race condition prevented some lock evictions (2328)
5556 - ENOMEM detection and retry on socknal sends (2230)
5557 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5558 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5559 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5560 - better error messages when a client is rejected during recovery (1505)
5561 - avoid cancelling locks which were never granted, after failure (2330)
5562 - fix i_sem/journal inversion in mds_client_add (2333)
5563 - fix truncate/getattr lock cycle deadlock (2334)
5564 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5565 - fix two rare exit paths which could leak an l_lock() ref (2321)
5566 - fencepost error in MDS/OST orphan recovery (2226)
5567 - make log record alignment 8 bytes (1988)
5568 - lstripe now fails when requested offset > ost_count (2237)
5569 - ensure that all kernel series have a complete list.h (1607)
5570 - fix crashes in special-file operations (2316)
5571 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5573 - allow configurable automake binary, for testing new versions
5574 - small update to the lfs documentation
5576 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5578 * fix negative export reference count in fsfilt_sync (2312)
5580 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5581 * release candidate 0.9.1
5583 - orphans are moved into the PENDING directory for possible recovery
5584 - replayed opens now open by fid for orphan/rename safety (1042)
5585 - last close of an orphan inode generates a transno (683)
5586 - chdir() and mount() now pin the directory entry (1020)
5587 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5588 - discard very old requests without processing them (1502)
5589 - remove some common, well-understood CERRORs (1505)
5590 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5591 - clear "grant" flags in OST replies until OST grant code lands (1644)
5592 - fix read performance by not clobbering i_blksize on client (1598)
5593 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5594 - make LRU size a /proc tunable, clears locks when reduced (707)
5595 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5596 - server locks take references on exports now (1558)
5597 - build fixes for 2.4.20-rh trees (1663)
5598 - return an error from lov_create if all OSCs are inactive (1751)
5599 - fix import levels when a reconnect happens without a timeout (1597)
5600 - exit early from mds_open if we get a lookup error (1749)
5601 - partial page read at EOF wouldn't wait for disk before sending (1642)
5602 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5603 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5604 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5605 - write/truncate lock inversion (1639)
5606 - Don't auto-load obdclass, portals modules during cleanup (1495)
5607 - fix timestamps from jumping to "now" (1763)
5608 - extra journal assertions (1648)
5609 - add an extra multiunlink test (1771)
5610 - fix read_record/write_record API (1776)
5611 - fix leak of offset_extent, possible incorrect i_size later (1772)
5612 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5613 - Debugging for the kqswnal_get_idle_tx problems (1820)
5614 - Allow recovery to be attempted multiple times (1536)
5615 - Write out MDS last_rcvd file after it is first created (1600)
5616 - Fix tx_descriptor leak in failed transmit situations (1827)
5617 - ext3 journaling fixes for assertion failure after IO error (1871)
5618 - class_export_put() on freed export after completion AST error (1896)
5619 - Fix revalidate looping in VFS (1322)
5620 - Don't access a freed export during MDS_REINT timeout (1521)
5621 - Add open-unlink recovery support on the MDS (1673,1764)
5622 - Return an error if no MDS data was read from last_rcvd (1946)
5623 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5624 - Don't setup a disk device that doesn't match exported UUID (317)
5625 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5626 - avoid committing NULL handle in force close
5627 - local.sh is now a one-stripe LOV configuration
5628 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5629 - fix ext3 htree duplicate directory entry corruption (1516)
5630 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5631 - update to new LOV EA format (2097)
5632 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5634 2003-06-15 Phil Schwan <phil@clusterfs.com>
5637 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5638 - per-import/export recovery handling (958, 931, 959)
5639 - multiple last-rcvd slots, for serving multiple FSes (949)
5640 - connections are again shared between multiple imp/exports (963, 964)
5641 - "umount -f" would hang if any requests needed to be sent (393, 978)
5642 - avoid pinning large req buffer by copying for queued messages (989)
5643 - add "uuid" to "lctl device" command to help upcalls (991)
5644 - "open" RPCs with transnos would confuse recovery counters (1037)
5645 - do proper endian conversion of all wire messages (288, 340, 891)
5646 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5647 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5648 - fix ptlrpc_connection leak in target_handle_connect (1174)
5649 - fix import refcounting bug in OST and MDS cleanup (1134)
5650 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5651 - fix very unlikely obd_types race condition (501)
5652 - remove osc_open hack for echo_client (1187)
5653 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5654 - a failure in read_inode2 leads to deadlock (1139)
5655 - cancel ack-locks as soon as transaction is committed (1072)
5656 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5657 - make sure to commitrw after any preprw to avoid deadlock (1162)
5658 - failing to execute a file in a lustre FS would lock inode (1203)
5659 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5660 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5661 - obd_uuid2tgt would walk off the end of the list (1255)
5662 - on IA64 the osc would give portals incorrect bulk size (1258)
5663 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5664 - fix lock inversion caused by new llite matching code (1282)
5665 - limit the number of dirty pages on a client to 10MB (1286)
5666 - timed out locks were not being corrected cancelled (1289)
5667 - fix O_DIRECT above 4GB on IA-32 (1292)
5668 * major user-visible changes
5669 - fail out/fail over policy now controlled by the upcall (993)
5671 - add OBD_PING to check server availability and failure (954)
5672 - lustre messages are now sent in sending host order (288, 340, 891)
5673 - add eadatalen to MDS getattr reply (340)
5674 - OST read replies may contain second buffer, with per-page status (593)
5676 2003-03-11 Phil Schwan <phil@clusterfs.com>
5679 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5680 - failed open()s could cause deadlock; fixed (867, 869)
5681 - stop cancelling OST locks when files are closed (481)
5682 - overlapping XID spaces caused network corruption (851, 853)
5683 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5684 - setattr_raw added, to do single-RPC, server-side setattrs
5685 - lmc/lconf syntax change for OST UUIDs
5686 - fix crashy race condition between ptlrpc_free_req and osc_close
5687 - don't use request in mdc_enqueue if we hit a timeout (889)
5688 - don't set the inode i_size for regular files from the MDS (896)
5689 - handle out of order completion AST (842)
5690 - don't LBUG if a lock request times out after receiving AST (913)
5691 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5692 - if a bad lock AST arrives, send an error instead of dropping entirely
5693 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5694 - fix leak in bulk IO when only partially completed (899, 900, 926)
5695 - fix O_DIRECT for ia64 (55)
5696 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5697 - C-z after timeout could hang a process forever; fixed (977)
5699 - client-side I/O cache (678, 924, 929, 941, 970)
5701 - READPAGE and SETATTRs which don't take server-side locks get
5704 2003-02-11 Phil Schwan <phil@clusterfs.com>
5707 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5708 - Fix fsfilt_extN_readpage() to read a full page of directory
5709 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5710 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5711 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5712 - Most intent operations were removed, in favour of a new RPC mode
5713 that does a single RPC to the server and bypasses most of the VFS
5714 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5715 - Aggressively cancel local locks on DLM servers
5716 - mds_reint_unlink sends EA to the client if it's the last nlink.
5717 client uses that EA to unlink OST objects.
5718 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5719 - recursive symlinks were fixed (439)
5720 - fixed NULL deref in DEBUG_REQ
5721 - filter_update_lastobjid no longer calls sync, which annoyed extN
5722 - fixed multi-client small-writes to a single file problem (445)
5723 - fixed mtime updates during file writes (607)
5724 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5725 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5726 - fixed Linux OST size reporting problem (444, 656)
5727 - OST now updates object mtime with writes or setattr (607, 619)
5728 - client verifies file size before zeroing page past EOF (445)
5729 - OST now writes last allocated objid to disk with allocation (108)
5730 - LOV on echo now works (409)
5732 - mds_reint_unlink sends a new buffer, with the EA included. this
5733 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5734 set if a regular file was being unlinked, and it was the last link
5735 - use PtlGet from the target for bulk writes (315)
5736 - OST now updates object mtime with writes or setattr (607, 619)
5737 - LDLM now has a grant-time callback to revalidate locked items, if
5739 - Many MDS operations were reorganized to combat race conditions
5741 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5742 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5744 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5747 - Fully reactivate OST imports after reconnection (512, others)
5748 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5749 - More graceful error handling for truncating on dead OST (515)
5750 - Don't error out unless we're actually accessing dead stripes (474)
5751 - Fix garbage sizes when stripes are missing (410)
5752 - LRU counters were broken, causing constant lock purge (433, 432)
5753 - garbage on read from stripes with failed OSTs (441)
5754 - mark OSCs as active before reconnecting during recovery (438)
5755 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5756 - lfind did not preserve OST order in output (443)
5757 - symlinks cause hung clients, incorrect data (439)
5758 - stop dereferencing request after dropping refcount (457)
5759 - don't LASSERT(spin_is_locked) on non-SMP (455)
5760 - fixes for many rename() bugs
5761 - fstat didn't correctly synchronize attributes (399)
5762 - server must handle lock cancellation during blocking AST prep (487)
5763 - bulk descriptors were free()d too soon (511)
5764 - fix paths in lconf, which would load incorrect modules (451, 507)
5765 - fix confusing lconf 'host not found' error message (386)
5766 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5767 - fix race condition in mdc_blocking_ast() for inode access (526)
5768 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5769 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5770 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5771 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5772 - fix lstripe in several ways (kernel side) (527)
5773 - fix request leak in ldlm_cli_enqueue (262)
5774 - incorrect OSC was marked inactive after OST failure
5775 - call mds_fs_cleanup before unmounting filesystem (524)
5776 - fix races between taking ns_lock and ldlm_lock_change_resource
5777 - fix races updating LOV export open file list
5778 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5779 - fix recovery NULL deref in ldlm_cli_cancel_unused
5780 - fix some DLM races by using new hash table for lock handles (419)
5781 - permit the client to specify desired inodes, at replay
5782 - duplicate requests when we queue them for replay reintegration
5783 - fix last_rcvd offset calculation
5784 - sync after each recovered transaction, so we always make progress
5785 - never, not always, ERESTART requests without transnos
5786 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5788 - skip replay if the MDS says that the client is already connected
5789 - don't check for a recovery-enabled export to match lctl's UUID
5790 - don't INC_USE_COUNT for phantom exports
5791 - don't crash when cleaning up phantom exports (567)
5792 - don't double-finish or set replay data for errored mdc_open requests
5793 - abort requests when they time out, so we don't get old replies
5794 - send/receive replies for AST messages again
5795 - if the client says that it doesn't have the lock, cancel it on the
5797 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5798 mark it as destroyed, it will all work out when decref is called
5799 - fix module use counts (22, 581)
5801 - ASTs now expect a reply (server cancels lock on error reply)
5803 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5806 - fix many simultaneous client startup (392)
5807 - fix dentry->d_it clobbering
5808 - credentials weren't being shipped for readdir/getattr operations
5809 - remove invalid assertions triggered during some concurrent MD
5811 - proper Lustre versions added (336, 389)
5812 - fix memory leak for create error case (398)
5813 - fix LOV locking bug that would get cli/srv out of sync
5814 - fix echo client over LOV (409)
5815 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5816 - fix double-O_EXCL intent crash (424)
5817 - avoid sending multiple lock CANCELs (352)
5819 - MDS can do multi-client recovery (modulo bugs in new code)
5821 - many updates, edits, cleanups
5823 2002-11-18 Phil Schwan <phil@clusterfs.com>
5826 - fix null d_it dereference (346)
5827 - fix full OST/dbench hang (333)
5828 - fix permission problem with file removal (286)
5829 - fix removal of OSCs from LOV when they fail
5830 - fix NULL deref during bulk timeout (214)
5831 - fix problems related to multiple filesystems on one MDS (241)
5832 - fixed serious subtle metadata locking bugs
5833 - free locks on clients when inodes are removed due to memory
5835 - fix inode pointer in lock data (285)
5836 - partial support for multiple MDS on a single host (241)
5837 - data locks weren't cancelled at clear_inode time (290, 311)
5838 - intent locks could lead to unbounded lock growth (205)
5839 - added a maximum lock count, an LRU list, and a flusher
5840 - fix multiple rename (365)
5841 - properly abstracted the echo client
5842 - OSC locked 1 byte too many; fixed
5843 - rewrote brw callback code:
5844 - fixed recovery bugs related to LOVs (306)
5845 - fixed too-many-pages-in-one-write crash (191)
5846 - fixed (again) crash in sync_io_timeout (214)
5847 - probably fixed callback-related race (385)
5849 - Add capability to MDS protocol
5850 - LDLM cancellations and callbacks on different portals
5852 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5855 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5856 - defer open object destruction to close time (601981, 138)
5857 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5858 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5859 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5860 - don't call obd_setattr in ll_file_release for destroyed objects
5861 * protocol change to lustre_msg: move |version| and add |flags|
5862 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5863 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5864 * added replay of create, unlink, link and rename operations during
5865 MDS failover; recovery should be much more robust now
5866 * remove failed OSCs from LOVs (only lov_create uses this so far)
5867 * the lustre-HOWTO was brought (more) up to date (582544)
5869 2002-10-23 Phil Schwan <phil@clusterfs.com>
5872 - in-use dentries weren't being reused properly (617851)
5873 - prevent multiple LDLM setup (599178)
5874 - fix LOV size calculations for truncate (617853)
5875 - fix client handling of MDS intent errors (POSIX)
5876 - fix permission bug in lovstripe.c test (624321)
5877 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5878 - truncate past end of file could corrupt data
5879 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5880 - a race in recovery could return ETIMEDOUT to apps (623947)
5881 - building outside the source directory was fixed
5882 * the lustre-HOWTO was brought (more) up to date (582544)
5883 * major progress was made on recovery functionality
5885 2002-10-10 Phil Schwan <phil@clusterfs.com>
5888 - recovery deadlock fix
5889 - rm -rf causes LBUG fix (617817)
5890 - file open by multiple tasks fix (618962)
5891 - directory permissions bugs (602707 and 620007)
5892 - journal_stop fixed with locking (611313)
5893 - O_APPEND failures resolved (618273, perhaps 614459)
5894 - lconf PATH fix (619770)
5895 - IA64 build fix (621450)
5896 - RPC buffer sizes scale with amount of memory
5898 2002-10-01 Phil Schwan <phil@clusterfs.com>
5901 - locks would be cancelled without throwing away data pages,
5902 resulting in inconsistent data (605627)
5903 - inode attributes were not always being refreshed (605627, 612449)
5904 - lconf now continues to cleanup after lctl reports an error
5905 - MDS now enforces user permissions (602707)
5906 - lprocfs cleanup fixed, but not yet enabled (614157)
5907 - fixed infinite server hang, should a client not respond to an AST
5908 - avoid going into recovery if user calls readlink() with a buffer
5909 that's too small (613941)
5910 - AST RPCs no longer require replies (614867) -- this may be changed
5911 - don't crash server if client sends an IOV that's too big (611336)
5912 - fixed lock conversion deadlock (611892)
5913 - fixed the following of symlinks (614622)
5914 * recovery: the server can remove locks from a client that dies, other
5915 clients can make progress
5916 * more extN patch fixes
5917 * compile-time configurable ptlrpc buffer allocations
5919 - collaborative read cache document
5920 - Lustre Lite Performance CDR document-in-progress
5922 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5925 - fix typo in patch-2.4.18
5927 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5930 - clear ptlrpc request each time in handle_incoming_request()
5931 - unlink of files now destroys the object on the OST
5933 2002-09-19 Peter Braam <braam@clusterfs.com>
5935 * add hard link support
5936 * change obdfile creation method
5937 * kernel patch changed
5939 2002-09-19 Peter Braam <braam@clusterfs.com>
5942 - stack overflow bug in extN fixed
5944 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5946 * documentation updates
5947 - add man pages for config tools
5948 - update tests/README to describe testing with new config tools
5949 - finish metadata API descriptions
5950 * bug fixes and cleanups
5951 - statfs workaround for 16TB limit
5952 - LOV stripe allocation improved, can stripe on subset of OSTs
5953 - LOV file size/IO offset was wrong for files > 4GB in size
5954 - object EA data was being dropped, caused files to be unreadable
5955 - memory overflow with non-LOV OST caused memory corruption
5956 - fixed regression tests to work with new config tools, obdfilter
5957 - fixed bug when directory size became larger than 1 block
5958 - fixed bug (for single client case) when PWD was deleted
5959 - invalidate local directory pages when doing intent-based ops
5960 - avoid LDLM oops when lock callback contained bad data
5962 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5964 * documentation updates
5965 * bug fixes and cleanups
5966 - configuration tools
5969 - 64-bit compile warnings
5970 - 64-bit internal statfs data
5972 * test_brw on persistent OST devices
5974 * lprocfs (disabled)
5976 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5978 * documentation updates
5979 * bug fixes and cleanups
5980 * configuration tools
5982 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5985 * many small fixes to 0_5_4
5986 * io/network handling
5987 * thinkos in MDS operations
5989 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5992 * crucial basic fixes to 0.5.3
5993 * IOR, Iozone work over Elan
5996 2002-08-07 Phil Schwan <phil@clusterfs.com>
5997 * version 0_5_3, our first alpha
5998 * we use the new Portals iovs
5999 * documentation updates
6000 * bug fixes and cleanups
6001 * small changes in the DLM wire protocol
6003 2002-07-25 Peter J. Braam <braam@clusterfs.com>
6004 * version 0_5_1 with some initial stability,
6005 * locking on MD and file I/O.
6006 * documentation updates
6007 * several bug fixes since 0.5.0
6008 * small changes in wire protocol
6010 2002-07-18 Phil Schwan <phil@clusterfs.com>
6012 * delivered as Lustre Light Alpha
6013 * fixed a crash after handling invalid MDS requests
6014 * fixed directory pages for architectures with non-4k pages sizes
6016 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
6017 * release version v0_4_4
6018 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
6020 * Fixes a number of interruption problems with OST operations.
6021 * Update documentation for portals header changes
6022 * Move all wire protocol structs/defines to lustre_idl.h
6023 * Fixes symlink length bug.
6024 * Add tcpdump to repository.
6026 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
6027 * release version v0_4_3
6028 * Fixes statfs for inodes on extN.
6029 * Fixes bug in runtests which would delete /etc/hosts.
6030 * Use 64-bit object IDs wherever possible (not into VFS though)
6031 Remove ost_get_info, which is unused by lustre, and out of date.
6033 2002-07-03 Peter Braam <braam@clusterfs.com>
6034 * release version v0_4_2 Fixes a lookup error (type not passed)
6035 * move forward to head of Portals
6036 * move forward to latest Lustre kernel
6038 2002-06-25 Peter Braam <braam@clusterfs.com>
6039 * release version v0_4_1. Hopefully stable on single node use.