1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.39.3 (SLES 10),
5 2.6.18-128.1.14.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.41.6.sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
18 Description: Prevent inconsistences between linux and lustre mount structures.
19 Details : Wait indefinitely in server_wait_finished() until mnt_count drops.
20 Make the sleep interruptible.
22 Severity : enhancement
24 Description: provide server to client comms path
25 Details : server to client communications path via new LDLM_SET_INFO rpc
29 Description: truncate starts GFP_FS allocation under transaction causing deadlock
30 Details : ldiskfs_truncate calls grab_cache_page which may start page
31 allocation under an open transaction. This may lead to
32 calling prune_icache with consequent lustre reentrance.
36 Description: send correctly lsm on open replay
37 Details : MDS is trust to LSM size on replay open, but client can set wrong size
42 Description: enable client interface failover
43 Details : When a child reconnects from another NID, properly update export
44 nid hash position and ldlm reverse import.
46 Severity : enhancement
48 Description: Communicate OST degraded/readonly state via statfs to MDS
49 Details : Flags in the statfs returned from OSTs indicate whether the
50 OST is in a degraded RAID state, or if the filesystem has
51 turned read-only after a filesystem error is detected.
56 Description: don't panic if EPROTO was hit when reading symlink
57 Details : correctly handling request reference in error cases.
59 Severity : enhancement
61 Description: Add LustreNetLink, a kernel-userspace communcation path.
63 Severity : enhancement
65 Description: Update kernel to SLES10 SP2 2.6.16.60-0.39.3.
70 Description: access to llog context before init.
71 Details : move handling CATALOGS file at osc layer and forbid access to llog
75 Frequency : rare, only when client is writing to OST on local node
77 Description: Avoid deadlock for local client writes
78 Details : Use new OBD_BRW_MEMALLOC flag to notify OST about writes in the
79 memory freeing context. This allows OST threads to set the
80 PF_MEMALLOC flag on task structures in order to allocate memory
81 from reserved pools and complete IO.
82 Use GFP_HIGHUSER for OST allocations for non-local client writes,
83 so that the OST threads generate memory pressure and allow
84 inactive pages to be reclaimed.
86 Severity : enhancement
88 Description: Update kernel to RHEL5.3 2.6.18-128.1.14.el5.
93 Description: lock ordering violation between &cli->cl_sem and _lprocfs_lock
94 Details : move ldlm namespace creation in setup phase to avoid grab
95 _lprocfs_lock with cli_sem held.
99 Description: Temporarily disable grant shrink.
100 Details : Disable the feature for debugging.
103 Frequency : only when formatting test filesystems
105 Description: Unable to run several mkfs.lustre on loop devices at the same time
106 Details : mkfs.lustre returns error 256 on the concurrent loop devices
107 formatting. The solution is to proper handle the error.
109 Severity : enhancement
111 Description: Update kernel to RHEL5.3 2.6.18-128.1.6.el5.
113 Severity : enhancement
115 Description: Update kernel to SLES10 SP2 2.6.16.60-0.37.
120 Description: resolve race between obd_disconnect and class_disconnect_exports
121 Details : if obd_disconnect will be called to already disconnected export he
122 forget release one reference and osc module can't unloaded.
124 Severity : enhancement
126 Description: Allow tuning service thread via /proc
127 Details : For each service a new
128 /proc/fs/lustre/{service}/*/thread_{min,max,started} entry is
129 created that can be used to set min/max thread counts, and get the
130 current number of running threads.
135 Descriptoin: don't return error if have partially created objects for file.
136 Details : lov_update_create_set uses set->set_success as index for created
137 objects, so if some requests will be failed, they will have hole
138 at end of array and qos_shrink_lsm for allocate correct lsm.
140 Severity : enhancement
142 Description: Update OFED support to 1.4.1
147 Description: Reduce small size read RPC
148 Details : Set read-ahead limit for every file and only do read-ahead when
149 available read-ahead pages are bigger than 1M to avoid small size
152 Severity : enhancement
154 Description: move AT tunable parameters for more consistent usage
155 Details : add AT tunables under /proc/sys/lustre, add to conf_param parsing
157 Severity : enhancement
159 Description: add lazystatfs mount option to allow statfs(2) to skip down OSTs
160 Details : allow skip disconnected ost for send statfs request and hide error
166 Description: fix racy locking of mballoc block bitmaps causing BUG
167 Details : The locking of the mballoc buddy bitmap and the in-memory
168 block bitmap was using two different spin locks in some
169 cases. This made it possible to incorrectly access the
170 mballoc bitmap while another process was modifying it,
171 causing a sanity assertion to fail. While no on-disk corruption
172 was reported, there was some risk of this happening.
175 Frequency : rare, on llog test 6
177 Descriptoin: don't allow connect to already connected import
178 Details : allowing connect to already connected import is hide connecting problem.
183 Descriptoin: allow kill process which wait statahead result
184 Details : for some reasons 'ls' can stick in waiting result from statahead,
185 in this case need way for kill this process.
187 Severity : enhancement
189 Description: Add state history info file, enhance import info file
190 Details : Track import connection state changes in a new osc/mdc proc file;
191 add overview-type data to the osc/mdc import proc file.
193 Severity : enhancement
195 Description: MDS create should not wait for statfs RPC while holding DLM lock.
197 Severity : enhancement
199 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
204 Descriptoin: OST grant too much space to client even there are not enough space.
205 Details : Client will shrink its grant cache to OST if there are no write
206 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
207 this grant cache if there are already not enough avaible space
208 (left_space < total_clients * 32M).
211 Frequency : start MDS on uncleanly shutdowned MDS device
213 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
214 Details : stay in waiting mds<>ost recovery finished produce random bugs
215 due race between two ll_sync thread for one lov target. send
216 ACTIVATE event only if connect realy finished and import have
220 Frequency : rare, connect and disconnect target at same time
222 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
223 Details : don't call obd_disconnect under lov_lock. this long time
224 operation and can block ptlrpcd which answer to connect request.
229 Descriptoin: don't lose wakeup for imp_recovery_waitq
230 Details : recover_import_no_retry or invalidate_import and import_close can
231 both sleep on imp_recovery_waitq, but we was send only one wakeup
235 Frequency : always with long access acl
237 Descriptoin: mds can't pack reply with long acl.
238 Details : mds don't control size of acl but they limited by reint/getattr
241 Severity : enhancement
243 Description: Update to SLES10 kernel-2.6.16.60-0.33.
245 Severity : enhancement
247 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
250 Frequency : start MDS on uncleanly shutdowned MDS device
252 Descriptoin: aborting recovery hang on MDS
253 Details : don't throttle destroy RPCs for the MDT.
256 Frequency : on remount
258 Description: external journal device not working after the remount
259 Details : clear dev_rdonly flag for external journal devices in
265 Description: shutdown vs evict race
266 Details : client_disconnect_export vs connect request race.
267 if client will evicted at this time - we start invalidate
268 thread without referece to import and import can be freed
272 Frequency : rare, need acl's on inode.
274 Description: client can't handle ost additional correctly
275 Details : if ost was added after client connected to mds client can have
276 hit lnet_try_match_md ... to big messages to wide striped files.
277 in this case need teach client to handle config events about add
278 lov target and update client max ea size at that event.
280 Severity : enhancement
282 Description: Changelogs
283 Details : Changelogs are a lightweight mechanism to track filesystem
284 metadata and namespace changes. The changelog is recorded
285 permanently on the MDTs, and is periodically "consumed" / purged
286 when records are no longer needed.
288 Severity : enhancement
290 Description: compact fld format with extents
291 Details : Store range of seq rather than every seq in FLD. Seq
292 controller update FLD rather than clients. In Case of CMD, mdt0
293 has FLD, all other metadata server act as non persistent proxy
294 for FLD queries and cache fld entries in fld cache.
299 Description: don't skip ost target if they assigned to file
300 Details : Drop slow OSCs if we can, but not for requested start idx.
301 This means "if OSC is slow and it is not the requested
302 start OST, then it can be skipped, otherwise skip it only
303 if it is inactive/recovering/out-of-space.
307 Description: more cleanup in mds_lov
308 Details : not send LOV EA under replay, we can't know about they size at this
309 time. Don't allow client connect to mds before any ost connected,
310 for avoid problems with LOV EA size and returning EIO to client.
312 Severity : enhancement
314 Description: Interoperability at server side (Disk interoperability)
316 Severity : enhancement
318 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
320 Severity : enhancement
322 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
324 Severity : enhancement
326 Description: New client IO stack (CLIO).
328 Severity : enhancement
330 Description: Commit on sharing. Eliminate inter-client dependencies between
331 uncommitted transactions by doing transaction commits.
332 Thereby clients may recovery independently.
335 Frequency : Create a symlink file with a very long name
337 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
338 Details : If there is no extra space in the request for early cancels,
339 ldlm_req_handles_avail() returns 0 instead of a negative value.
341 Severity : enhancement
343 Description: Add /proc entry for import status
344 Details : The mdc, osc, and mgc import directories now have
345 an import directory that contains useful import data for debugging
348 Severity : enhancement
350 Description: Re-disable certain /proc logging
351 Details : Enable and disable client's offset_stats, extents_stats and
352 extents_stats_per_process stats logging on the fly.
355 Frequency : Only on FC kernels 2.6.22+
357 Description: oops in statahead
358 Details : Do not drop reference count for the dentry from VFS when lookup,
359 VFS will do that by itself.
361 Severity : enhancement
363 Description: Generic /proc file permissions
364 Details : Set /Proc file permissions in a more generic way to enable non-
365 root users operate on some /proc files.
369 Description: Hitting mdc_commit_close() ASSERTION
370 Details : Properly handle request reference release in
371 ll_release_openhandle().
375 Description: quota recovery deadlock during mds failover
376 Details : This patch includes att18982, att18236, att18237 in bz14840.
378 1. fix osts hang when mds does failover with quotaon
379 2. prevent watchdog storm when osts threads wait for the
384 Frequency : only patchless client
385 Description: add workaround for race between add/remove dentry from hash
387 Severity : enhancement
389 Description: Allow OST glimpses to return PW locks
393 Description: LBUG when llog conf file is full
394 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
398 Description: Prevent import from entering FULL state when server in recovery
402 Description: service mount cannot take device name with ":"
403 Details : Only when device name contains ":/" will mount treat it as
409 Description: replace ptlrpcd with the statahead thread to interpret the async
410 statahead RPC callback
414 Frequency : on recovery
415 Description: I/O failures after umount during fail back
416 Details : if client reconnected to restarted server we need join to recovery
417 instead of find server handler is changed and process self
418 eviction with cancel all locks.
420 Severity : enhancement
422 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
424 Severity : enhancement
426 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
428 Severity : enhancement
430 Description: Upcall on Lustre log has been dumped
431 Details : Allow for a user mode script to be called once a Lustre log has
432 been dumped. It passes the filename of the dumped log to the
433 script, the location of the script can be specified via
434 /proc/sys/lnet/debug_log_upcall.
439 Description: avoid idr_remove called for id which is not allocated.
440 Details : Move assigment s_dev for clustered nfs to end of initialization,
441 to avoid problem with error handling.
446 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
447 Details : When connection is reused this not moved from CONN_UNUSED_HASH
448 into CONN_USED_HASH and this prodice warning when put connection
449 again in unused hash.
452 Severity : enhancement
454 Description: File striping can now be set to use an arbitrary pool of OSTs.
456 Severity : enhancement
458 Description: Export bytes_read/bytes_write count on OSC/OST.
462 Description: Early reply size mismatch, MGC loses connection
463 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
464 the connect flags are properly negotiated.
469 Description: quotas are not honored with O_DIRECT
470 Details : all writes with the flag O_DIRECT will use grants which leads to
471 this problem. Now using OBD_BRW_SYNC to guard this.
475 Description: add quota statistics
476 Details : 1. sort out quota proc entries and proc code.
477 2. add quota statistics
479 Severity : enhancement
481 Description: enable quota support for HEAD.
485 Description: Properly propagate oinfo flags from lov to osc for statfs
486 Details : restore missing copy oi_flags to lov requests.
488 Severity : enhancement
490 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
491 : l_getgroups(8), lst(8), routerstat(8)
492 Details : included man pages for llobdstat(8), llstat(8),
493 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
495 Severity : enhancement
497 Description: configure's --enable-quota should check the
498 : kernel .config for CONFIG_QUOTA
499 Details : configure is terminated if --enable-quota is passed but
500 : no quota support is in kernel
504 Description: Remove portals compatibility
505 Details : Remove portals compatibility, not interoperable with releases
510 Description: Resolve device initialization race
511 Details : Prevent proc handler from accessing devices added to the
512 obd_devs array but yet be intialized.
514 Severity : enhancement
516 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
518 Severity : enhancement
520 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
525 Description: Using wrong pointer in osc_brw_prep_request
526 Details : Access to array[-1] can produce panic if kernel compiled with
527 CONFIG_PAGE_ALLOC enabled
531 Description: Client runs out of low memory
532 Details : Consider only lowmem when counting initial number of llap pages
536 Description: *optional* service tags registration
537 Details : if the "service tags" package is installed on a Lustre node
538 When the filesystem is mounted, a local-node service tag will
539 be created. See http://inventory.sun.com/ for more information
540 about the Service Tags asset management system.
544 Description: Kernel BUG tries to release flock
545 Details : Lustre does not destroy flock lock before last reference goes
546 away. So always drop flock locks when client is evicted and
547 perform unlock regardless of successfulness of speaking to MDS.
551 Description: add refcount protection for osc callbacks, avoid panic on shutdown
555 Description: sanity test 65a fails if stripecount of -1 is set
556 Details : handle -1 striping on filesystem in ll_dirstripe_verify
561 Description: ASSERTION(CheckWriteback(page,cmd)) failed
562 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
565 Severity : enhancement
567 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
571 Description: do not process already freed flock
572 Details : flock can possibly be freed by another thread before it reaches
573 to ldlm_flock_completion_ast.
577 Description: LBUG during stress test
578 Details : Need properly lock accesses the flock deadlock detection list.
582 Description: oops in page fault handler
583 Details : kernel page fault handler can return two special 'pages' in error
584 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
588 Description: timeout with invalidate import.
589 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
590 should be handled by ptlrpcd. This produce long age waiting and
591 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
593 Severity : enhancement
595 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
599 Description: enable MGS and MDT services start separately
600 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
601 a co-located MGS without starting the MGS, which is a complement
602 to 'nosvc' mount option.
606 Frequency : after recovery
607 Description: precreate to many object's after del orphan.
608 Details : del orphan st in oscc last_id == next_id and this triger growing
609 count of precreated objects. Set flag LOW to skip increase count
610 of precreated objects.
614 Frequency : rare, on clear nid stats
615 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
616 Details : when clean nid stats sometimes try destroy live entry,
617 and this produce panic in free.
621 Description: Stack overflow during MDS log replay
622 ease stack pressure by using a thread dealing llog_process.
626 Description: wait until IO finished before start new when do lock cancel.
627 Details : VM protocol want old IO finished before start new, in this case
628 need wait until PG_writeback is cleared until check dirty flag and
629 call writepages in lock cancel callback.
631 Severity : enhancement
633 Description: using special macro for print time and cleanup in includes.
637 Description: mds_mfd_close() ASSERTION(rc == 0)
638 Details : In mds_mfd_close(), we need protect inode's writecount change
639 within its orphan write semaphore to prevent possible races.
643 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
647 Frequency : rare, on shutdown ost
648 Description: don't hit live lock with umount ost.
649 Details : shrink_dcache_parent can be in long loop with destroy dentries,
650 use shrink_dcache_sb instead.
654 Description: don't panic with use echo client
655 Details : echo client pass NULL as client nid pointer and this produce null
660 Description: fix build on ppc32
661 Details : compile code with -m64 flag produce wrong object file for ppc32.
665 Description: add message levels for liblustreapi
669 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
670 Details : Move the 'good_osts' check before the 'total_bavail' check. This
671 will result in an -EAGAIN and in the exit call path we call
672 alloc_rr() which will with increasing aggressiveness attempt to
673 aquire precreated objects on the minimum number of required OSCs.
677 Description: Use old size assignment to avoid deadlock
678 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
679 the scheduling while holding a spinlock. We do not need locking
680 for size in ll_update_inode() because size is only updated from
681 the MDS for directories or files without objects, so there is no
682 other place to do the update, and concurrent access to such inodes
683 are protected by the inode lock.
687 Description: resolve "_IOWR redefined" build error on SLES10
691 Description: dump the memory debugging after all modules are unloaded to
692 suppress false negative in conf_sanity test 39
694 Severity : enhancement
696 Description: build kernel-ib packages for OFED 1.3 in our release cycle
701 Description: fix SLES kernel versioning
702 Details : the kernel version for our SLES 10 kernel did not include a "-"
703 before the "smp" at the end. while this was not a problem in
704 general, it did mean that software trying to use the kernel
705 version to try to detect a vendor specific kernel would fail.
706 this was most evident by the OFED build scripts.
710 Description: Don't update lov_desc members until making sure they are valid
711 Details : When updating lov_desc members via proc fs, need fix their
712 validities before doing the real update.
716 Description: don't put request into delay list while invalidate in flight.
717 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
718 import in flight. this produce timeout for invalidate and sometimes
719 can cause stale data.
723 Frequency : on ppc only
724 Description: not convert ost objects for directory because it's not exist.
725 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
729 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
731 Severity : enhancement
733 Description: Update to support 2.6.22.14 vanilla kernel.
737 Frequency : rare, on recovery
738 Description: read procfs can produce deadlock in some situation
739 Details : Holding lprocfs lock which send rpc can produce block for destroy
740 obd objects and this also block reconnect with -EALREADY. This
741 isn't fix all lprocfs bugs - but make it rare.
743 Severity : enhancement
745 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
748 Frequency : frequent on X2 node
750 Description: mdc_set_open_replay_data LBUG
751 Details : Set replay data for requests that are eligible for replay.
755 Description: lustre_mgs: operation 101 on unconnected MGS
756 Details : When MGC is disconnected from MGS long enough, MGS will evict the
757 MGC, and late on MGC cannot successfully connect to MGS and a lot
758 of the error messages complaining that MGS is not connected.
762 Frequency : on network error
763 Description: panic with double free request if network error
764 Details : mdc_finish_enqueue is finish request if any network error ocuring,
765 but it's true only for synchronus enqueue, for async enqueue
766 (via ptlrpcd) this incorrect and ptlrpcd want finish request
769 Severity : enhancement
771 Description: client-side metadata stat-ahead during readdir(directory readahead)
772 Details : perform client-side metadata stat-ahead when the client detects
773 readdir and sequential stat of dir entries therein
776 Frequency : on start mds
778 Description: Implement get_info(last_id) in obdfilter.
781 Frequency : occasional
783 Description: Correctly check stale fid, not start epoch if ost not support SOM
784 Details : open with flag O_CREATE need set old fid in op_fid3 because
785 op_fid2 was overwritten with new generated fid, but mds can answer
786 with one of these two fids and both is not stale. Setattr
787 incorrectly started an epoch and assume will be called
788 done_writing, but without SOM done_writing ever being called.
791 Frequency : rare, depends on device drivers and load
793 Description: MDS or OSS nodes crash due to stack overflow
794 Details : Code changes in 1.8.0 increased the stack usage of some functions.
795 In some cases, in conjunction with device drivers that use a lot
796 of stack the MDS (or possibly OSS) service threads could overflow
797 the stack. One change which was identified to consume additional
798 stack has been reworked to avoid the extra stack usage.
801 Frequency : occasional
803 Description: Do not fail import if osc_interpret_create gets -EAGAIN
804 Details : If osc_interpret_create got -EAGAIN it immediately exits and
805 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
806 oscc_has_objects and see OSC has no objests and call
807 oscc_internal_create to resend create request.
809 Severity : enhancement
811 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
813 Severity : enhancement
815 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
818 Frequency : very rare
820 Description: Fix replay if there is an un-replied request and open
821 Details : In some cases, older replay request will revert the
822 mcd->mcd_last_xid on MDS which is used to record the client's
825 Severity : enhancement
827 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
829 Severity : enhancement
831 Description: Add rhel5 support to HEAD.
833 Serverity : enhancement
835 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
840 Description: Don't allow skipping OSTs if index has been specified.
841 Details : Don't allow skipping OSTs if index has been specified, make
842 locking in internal create lots better.
846 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
847 Details : access to bitfield in structure is always rounded to long
848 and this produce problem with not atomic change any bit.
852 Description: Lustre make rpms failed.
853 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
854 builds Lustre rpms from tarball.
857 Frequency : rare on shutdown ost
859 Description: If llog cancel was not send before clean_exports phase, this can
860 produce deadlock in llog code.
861 Details : If llog thread has last reference to obd and call class_import_put
862 this produce deadlock because llog_cleanup_commit_master wait when
863 last llog_commit_thread exited, but this never success because was
864 called from llog_commit_thread.
868 Description: allow userland application know is lost one of stripes.
869 Details : fill lvb_blocks with error code on ost and return it to
870 application if error flag found.
874 Description: NULL lov_tgts causing MDS oops
875 Details : more safe checks for NULL lov_tgts for avoid oops.
877 Severity : enhancement
879 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
883 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
884 Details : If an OST is down the MDS will hang indefinitely in
885 obd_statfs_rqset() waiting for the statfs data. While for
886 MDS QOS usage of statfs, it should not stuck in waiting.
888 Severity : enhancement
890 Description: remote_acl support
891 Details : Support ACL-based permission check for remote user.
892 Support setfacl/getfacl for remote user with the utils
893 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
894 the system "{s,g}etfacl" utils.
896 Severity : enhancement
898 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
900 Severity : enhancement
902 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
906 Description: make mgs_setparam() handle fsname containing dash
907 Details : fsname containing a dash does not work with lctl conf_param
909 Severity : enhancement
911 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
913 Severity : enhancement
915 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
919 Frequency : rare, at shutdown
920 Description: access already free / zero obd_namespace.
921 Details : if client_disconnect_export was called without force flag set,
922 and exist connect request in flight, this can produce access to
923 NULL pointer (or already free pointer) when connect_interpret
924 store ocd flags in obd_namespace.
928 Frequency : only at startup
929 Description: not alloc memory with spinlock held.
930 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
931 if any spinlock held.
933 Severity : enhancement
935 Description: make lustre randomly fail allocating memory
936 Details : Make lustre randomly failed allocating memory for testing purpose.
938 Severity : enhancement
940 Description: lost problems with lov objid file
941 Details : Fixes some scability and access to not inited memory problems
942 in work with lov objdid file.
947 Description: lfs find does not continue on file error
948 Details : Continue other files processing when a file/dir is absent.
952 Description: Inconsistent usage of lustre_pack_reply()
953 Details : Standardize the usage of lustre_pack_reply() such that it
954 always generate a CERROR on failure.
957 Frequency : occasional
959 Description: MDS hang or stay in waiting lock
960 Details : If client receive lock with CBPENDING flag ldlm need send lock
961 cancel as separate rpc, to avoid situation when cancel request
962 can't processed due all i/o threads stay in wait lock.
966 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
970 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
974 Description: Update to SLES9 kernel-2.6.5-7.287.3.
978 Description: Update to RHEL4 latest kernel.
980 Severity : enhancement
982 Description: Build SLES10 patchless client fails
983 Details : The configure was broken by run ./configure with
984 --with-linux-obj=.... argument for patchless client. When the
985 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
986 can't search header adequately. Use absolute path such as
987 -I($LINUX)/include instead.
991 Description: interrupt oig_wait produce painc on resend.
992 Details : brw_redo_request can be used for resend requests from ptlrpcd and
993 private set, and this produce situation when rq_ptlrpcd_data not
994 copyed to new allocated request and triggered LBUG on assert
995 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
996 ptlrpcd set if request is changed and can be safety changed to use
1001 Description: LASSERT_{REQ,REP}SWAB macros are buggy
1002 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
1003 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
1004 these macros and replace them with their difinitions instead.
1008 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
1009 Details : Update which_patch & target file for SLES10 latest kernel.
1012 Description: Remove client patches from SLES 10 kernel.
1013 Details : This causes SLES 10 clients to behave as patchless clients
1014 even on a Lustre-patched (server) kernel.
1016 Severity : enhancement
1018 Description: self-adjustable client's lru lists
1019 Details : use adaptive algorithm for managing client cached locks lru
1020 lists according to current server load, other client's work
1021 pattern, memory activities, etc. Both, server and client
1022 side namespaces provide number of proc tunables for controlling
1025 Severity : enhancement
1027 Description: light-weight GSS support
1028 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
1029 and reduce performance overhead.
1031 Severity : enhancement
1033 Description: Linux keyring support
1034 Details : Support using service of Linux keyring for Lustre GSS internal
1035 context refresh/cache mechanism.
1039 Description: Fix errors in lfs documentation
1040 Details : Fixes man pages
1044 Description: don't use GFP_* in generic Lustre code.
1045 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1049 Description: obdclass is limited by single OBD_ALLOC(idarray)
1050 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1054 Description: warnings with build patchless client with vanila 2.6.19 and up
1055 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1056 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1060 Frequency : Only for SLES
1061 Description: sanity_quota fail test_1
1062 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1063 file, which makes TSTID[2] inunique.
1067 Frequency : Only for SLES9
1068 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1069 Details : Update target/ChangeLog/which_patch .
1073 Frequency : Only for test_10 in sanity-quota.sh
1074 Description: a bug in quota test code
1075 Details : checking a test flag in a wrong place
1079 Frequency : Only for test_13 in sanity-quota.sh
1080 Description: a bug in quota test script
1081 Details : shouldn't check overfull allocated quota
1085 Description: jbd statistics
1086 Details : Port older jbd statistics patch for sles10
1088 Severity : enhancement
1090 Description: Add printing inode info into message about error in writepage.
1095 Description: reply_lock_interpret crash due to race with it and lock cancel.
1096 Details : Do not replay locks that are being cancelled. Do not reference
1097 locks by their address during replay, just by their handle.
1101 Frequency : When flocks are used.
1102 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1103 Details : Flock locks might destroy just granted lock if it could be merged
1104 with another existing flock, this is done in completion handler,
1105 so teach ldlm_cli_enquque_fini that this is a valid case for
1111 Description: Oops in read and write path when failing to allocate lock.
1112 Details : Check if lock allocation failed and return error back.
1116 Description: lstripe command fails for valid OST index
1117 Details : The stripe offset is compared to lov->desc.ld_tgt_count
1118 instead of lov->desc.ld_active_tgt_count.
1122 Description: sanity.sh failed test 103
1123 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
1128 Description: df doesn't work properly if diskfs blocksize != 4K
1129 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1133 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1134 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1136 Severity : enhancement
1138 Description: Adaptive timeouts
1139 Details : RPC timeouts adapt to changing server load and network
1140 conditions to reduce resend attempts and improve recovery time.
1144 Description: llapi_file_create() does not allow some changes
1145 Details : add llapi_file_open() that allows specifying the mode and
1146 open flags, and also returns an open file handle.
1150 Description: merge and cleanup kernel patches.
1151 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1155 Description: Add journal checksum support.(Kernel part)
1156 Details : The journal checksum feature adds two new flags i.e
1157 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1158 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1159 indicates that the commit block contains the checksum for
1160 the blocks described by the descriptor blocks. Now commit
1161 record can be sent to disk without waiting for descriptor
1162 blocks to be written to disk. This behavior is controlled
1163 using JBD2_FEATURE_ASYNC_COMMIT flag.
1167 Description: OSS needs mutliple precreate threads
1168 Details : Add ability to start more than one create thread per OSS.
1172 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1173 Details : execute lfs setstripe on client
1177 Description: mds_obd_create error creating tmp object
1178 Details : When the user sets quota on root, llog will be affected and can't
1179 create files and write files.
1182 Frequency : Always on ia64 patchless client, and possibly others.
1184 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1185 Details : This allows the patchless client to be loaded on architectures
1186 without this export.
1190 Description: RedHat Update kernel for RHEL5
1191 Details : Add the kernel config file for RHEL5.
1195 Description: RedHat Update kernel for RHEL5
1196 Details : Modify the kernel config file more closer RHEL5.
1200 Description: Build failure against Centos5 (RHEL5)
1201 Details : Define PAGE_SIZE when it isn't present.
1205 Description: test_7 on sanity-quota.sh can't be used on separate machines
1206 Details : add facet to handle it
1210 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1211 Details : with NFS, the anon dentry's parent was set to itself in
1212 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1213 corresponding dentry other than use rec->ur_name.
1215 Severity : enhancement
1217 Description: lfs setstripe enhancement
1218 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1222 Description: enable data checksumming by default
1223 Details : enable checksum by default, allow --disable-checksum
1224 configure option and "-o nochecksum" mount option. Checksums
1225 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1229 Description: lustre support for RHEL5
1230 Details : Add support for RHEL5.
1234 Description: Client eviction due to failover config
1235 Details : after a connection loss, the lustre client should attempt to
1236 reconnect to the last active server first before trying the
1237 other potential connections.
1241 Description: when mds and osts use different quota unit(32bit and 64bit),
1242 quota will be released repeatly.
1243 Details : void sending multiple quota reqs to mds, which will keep the
1244 status between the reqs.
1248 Description: osts not allocated evenly to files
1249 Details : change the condition to increase offset_idx
1252 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1254 Description: Data corruption for OSTs that are formatted larger than 2TB
1256 Details : When generating the bio request for lustre file writes the
1257 sector number would overflow a temporary variable before being
1258 used for the IO. The data reads correctly from Lustre (which
1259 will overflow in a similar manner) but other file data or
1260 filesystem metadata may be corrupted in some cases.
1264 Description: Tune the kernel for good SCSI performance.
1265 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1266 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1271 Description: rewrite ext2-derived code in obdclass/uuid.c
1272 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1273 from scratch preserving functionality.
1277 Description: "lfs find -obd UUID" prints directories
1278 Details : "lfs find -obd UUID" will return all directory names instead
1279 of just file names. It is incorrect because the directories
1280 do not reside on the OSTs.
1284 Description: use i_size_read and i_size_write in 2.6 port
1285 Details : replace inode->i_size access with i_size_read/write()
1288 Frequency : when removing large files
1290 Description: scheduling issue during removal of large Lustre files
1291 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1292 It causes scheduling issues when removing large files (17TB in the
1296 Frequency : only with liblustre clients on XT3
1298 Description: evictions taking too long
1299 Details : allow llrd to evict clients directly on OSTs
1302 Frequency : only on ppc
1304 Description: /proc/fs/lustre/devices broken on ppc
1305 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1306 We should make sure the type of 'index' is loff_t and then cast
1307 explicitly as needed below. Do not assign an explicitly cast
1311 Frequency : only for rhel5
1313 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1314 Details : Modify the target file & which_kernel.
1316 Severity : enhancement
1318 Description: omit set fsid for export NFS
1319 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1320 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1324 Frequency : Always, for kernels after 2.6.16
1325 Description: Fix warning idr_remove called for id=.. which is not allocated.
1326 Details : Last kernels save old s_dev before kill super and not allow
1327 to restore from callback - restore it before call kill_anon_super.
1331 Description: buffer overruns could theoretically occur
1332 Details : llapi_semantic_traverse() modifies the "path" argument by
1333 appending values to the end of the origin string, and a
1334 overrun may occur. Adding buffer overrun check in liblustreapi.
1338 Description: Fix error on 'ls .' at the top of the Lustre mount.
1339 Details : Don't revalidate dentry if it is a root dentry.
1343 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1344 Details : Modify vm-tunables-rhel4.patch.
1348 Description: Kernel config for 2.6.18-vanilla.
1349 Details : Modify targets/2.6-vanilla.target.in.
1350 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1351 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1352 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1353 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1357 Description: improve handling recoverable errors
1358 Details : if request processig with error which can be recoverable on server
1359 request should be resend, otherwise page released from cache and
1364 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1365 Details : Modify target file & which_patch.
1366 A flaw was found in the IA32 system call emulation provided
1367 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1368 value could be stored in the %RAX register, which could trigger an
1369 out-of-bounds system call table access. An untrusted local user
1370 could exploit this flaw to run code in the kernel
1371 (ie a root privilege escalation). (CVE-2007-4573).
1375 Description: O_DIRECT bypasses client statistics.
1376 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1377 read_ahead_stats were updated. copied the stats section from
1378 osc_send_oap_rpc() into async_internal().
1382 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1386 Description: change order of libsysio includes
1387 Details : '#include sysio.h' should always come before '#include xtio.h'
1391 Description: handle "serious error: objid * already exists" more gracefully
1392 Details : If LAST_ID value on disk is smaller than the objects existing in
1393 the O/0/d* directories, it indicates disk corruption and causes an
1394 LBUG(). If the object is 0-length, then we should use the existing
1395 object. This will help to avoid a full fsck in most cases.
1397 Severity : enhancement
1399 Description: adapt the lustre_config script to support the upgrade case
1400 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1401 server targets to 1.6 in parallel.
1405 Description: To avoid grant space > avaible space when the disk is almost
1406 full. Without this patch you might see the error "grant XXXX >
1407 available" or some LBUG about grant, when the disk is almost
1409 Details : In filter_check_grant, for non_grant cache write, we should
1410 check the left space by if (*left > ungranted + bytes), instead
1411 of (*left > ungranted), because only we are sure the left space
1412 is enough for another "bytes", then the ungrant space should be
1413 increase. In client, we should update cl_avail_grant only there
1414 is OBD_MD_FLGRANT in the reply.
1418 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1419 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1420 described in CVE-2007-4573. This problem could allow untrusted
1421 local users to gain root access.
1424 Frequency : when using O_DIRECT and quotas
1426 Description: Incorrect file ownership on O_DIRECT output files
1427 Details : block usage reported by 'lfs quota' does not take into account
1428 files that have been written with O_DIRECT.
1433 Description: touch file failed when fs is not full
1434 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1435 otherwise we can get ENOSP while fs is not full.
1439 Description: parallel lock callbacks
1440 Details : Instead of sending blocking and completion callbacks as separated
1441 requests, adding them to a set and sending in parallel.
1444 Frequency : only for Cray XT3
1445 Bugzilla : 12829/13455
1446 Description: Changing primary group doesn't change the group lustre assigns to
1448 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1449 overridden with the primary group provided by the group upcall,
1450 whereas the supplied fsgid can be trusted if it is in the list of
1451 supplementary groups returned by the group upcall.
1453 Severity : enhancement
1455 Description: Allow masking D_WARNING, D_ERROR messages from console
1456 Details : Console messages can now be disabled via lnet.printk.
1460 Description: User code with malformed file open parameter crashes client node
1461 Details : Before packing join_file req, all the related reference should be
1462 checked carefully in case some malformed flags cause fake
1463 join_file req on client.
1467 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1468 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1469 mode is equal to the granted mode. However, it should also check
1470 that LDLM_FL_CP_REQD is not set.
1474 Description: Heavy nfs access might result in deadlocks
1475 Details : After ELC code landed, it is now improper to enqueue any mds
1476 locks under och_sem, because enqueue might want to decide to
1477 cancel open locks for same inode we are holding och_sem for.
1481 Description: Client eviction while running blogbench
1482 Details : A lot of unlink operations with concurrent I/O can lead to a
1483 deadlock causing evictions. To address the problem, the number of
1484 oustanding OST_DESTROY requests is now throttled to
1485 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1486 callbacks are processed in priority.
1490 Description: enable ACLs on MDS by default
1491 Details : ACLs must be enabled on MDS by default.
1494 Frequency : PPC/PPC64 only
1496 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1497 Details : fix duplicated definitions between asm-ppc64/types.h and
1498 lustre_types.h on PPC.
1501 Frequency : PPC/PPC64 only
1503 Description: asm-ppc/segment.h does not exist
1504 Details : fix compile issue on PPC.
1508 Description: better handle error messages in extents code
1511 Frequency : RHEL4 only
1513 Description: mkfs is very slow on IA64/RHEL4
1514 Details : A performance regression has been discovered in the MPT Fusion
1515 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1516 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1517 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1519 Severity : enhancement
1521 Description: SNMP support enhancement
1522 Details : Adding total number of sampled request for an MDS node in snmp
1525 Severity : enhancement
1527 Description: Optimize ldlm waiting list processing for PR extent locks
1528 Details : When processing waiting list for read extent lock and meeting read
1529 lock that is same or wider to it that is not contended, skip
1530 processing rest of the list and immediatelly return current
1531 status of conflictness, since we are guaranteed there are no
1532 conflicting locks in the rest of the list.
1536 Description: Time out and refuse to reconnect
1537 Details : When the failover node is the primary node, it is possible
1538 to have two identical connections in imp_conn_list. We must
1539 compare not conn's pointers but NIDs, otherwise we can defeat
1540 connection throttling.
1544 Description: port llog fixes from b1_6 into HEAD
1545 Details : Port llog reference couting and some llog cleanups from b1_6
1546 (bug 10800) into HEAD, for protect from panic and access to already
1547 free llog structures.
1551 Description: Detect stride IO mode in read-ahead
1552 Details : When a client does stride read, read-ahead should detect that and
1553 read-ahead pages according to the detected stride pattern.
1557 Description: data checksumming impacts single node performance
1558 Details : add support for several checksum algorithm. Currently, only CRC32
1559 and Adler-32 are supported. The checksum type can be changed on
1560 the fly via /proc/fs/lustre/osc/*/checksum_type.
1564 Description: use adler32 for page checksums
1565 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1570 Description: build for x2 fails
1571 Details : fix compile issue on Cray systems.
1575 Description: Properly match for duplicate locks
1576 Details : Due to different lock order from skiplists code, we need to
1577 traverse entire list for now
1580 Frequency : only on PPC/SLES10
1582 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1583 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1584 be defined. Add a hack in mkfs_lustre.c to work around this compile
1589 Description: LASSERT on MDS when client holding flock lock dies
1590 Details : ldlm pool logic depends on number of granted locks equal to
1591 number of released locks which is not true for flock locks, so
1592 just exclude such locks from consideration.
1596 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1597 Details : Use fsfilt_sync() for both the whole filesystem sync and
1598 individual file sync to eliminate dangerous inode locking
1599 with I_LOCK that can lead to a deadlock.
1603 Description: performance in 1.6.3
1604 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1605 sure that 1MB requests are not fragmented by the block layer.
1607 Severity : enhancement
1609 Description: organize the server-side client stats on per-nid basis
1610 Details : Change the structure of stats under obdfilter and mds to
1618 The "uuid"s file would list the uuids of _active_ exports.
1619 And the clear entry is to clear all stats and stale nids.
1621 Severity : enhancement
1623 Description: eliminate client locks in face of contention
1624 Details : file contention detection and lockless i/o implementation
1625 for contended files.
1629 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1632 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1634 Description: MDS may lose file striping (and hence file data) in some cases
1635 Details : If there are additional extended attributes stored on the MDS,
1636 in particular ACLs, SELinux, or user attributes (if user_xattr
1637 is specified for the client mount options) then there is a risk
1638 of attribute loss. Additionally, the Lustre file striping
1639 needs to be larger than default (e.g. striped over all OSTs),
1640 and an additional attribute must be stored initially in the
1641 inode and then increase in size enough to be moved to the
1642 external attribute block (e.g. ACL growing in size) for file
1647 Description: skiplist implementation simplification
1648 Details : skiplists are used to group compatible locks on granted list
1649 that was implemented as tracking first and last lock of each
1650 lock group the patch changes that to using doubly linked lists
1654 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1655 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1656 because LASSERT on both the data supplied by a client, and the
1657 data on disk is dangerous and incorrect.
1659 Severity : enhancement
1661 Description: Slow truncate/writes to huge files at high offsets.
1662 Details : Directly associate cached pages to lock that protect those pages,
1663 this allows us to quickly find what pages to write and remove
1664 once lock callback is received.
1668 Description: more ldlm soft lockups
1669 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1670 starve other threads from the resource lock for a long time in
1671 case of long waiting queue, so change the debug level from
1672 D_OTHER to the less frequently used D_INFO.
1674 Severity : enhancement
1676 Description: add -gid, -group, -uid, -user options to lfs find
1680 Description: Hung threads in invalidate_inode_pages2_range
1681 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1682 once one is completed. As a result, some RPCs are stuck in the
1683 queue and are never sent.
1687 Description: filter threads hungs on waiting journal commit
1688 Details : Cleanup filter group llog code, then only filter group llog will
1689 be only created in the MDS/OST syncing process.
1693 Description: Procfs and llog threads access destoryed import sometimes.
1694 Details : Sync the import destoryed process with procfs and llog threads by
1695 the import refcount and semaphore.
1697 Severity : enhancement
1699 Description: openlock cache of b1_6 port to HEAD
1704 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1705 Details : If insertion of an extent fails, then discard the inode
1706 preallocation and free data blocks else it can lead to duplicate
1711 Description: don't always update ctime in ext3_xattr_set_handle()
1712 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1713 In some cases the ctime should not be updated, for example for
1714 2.0->1.8 compatibility it is necessary to delete an xattr and it
1715 should not update the ctime.
1719 Bugzilla : 15713/16362
1720 Description: Assertion in iopen_connect_dentry in 1.6.3
1721 Details : looking up an inode via iopen with the wrong generation number can
1722 populate the dcache with a disconneced dentry while the inode
1723 number is in the process of being reallocated. This causes an
1724 assertion failure in iopen since the inode's dentry list contains
1725 both a connected and disconnected dentry.
1729 Description: assertion failure in ldlm_handle2lock()
1730 Details : fix a race between class_handle_unhash() and class_handle2object()
1731 introduced in lustre 1.6.5 by bug 13622.
1736 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1737 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1742 Description: oops in sd_iostats_seq_show()
1743 Details : unloading/reloading the scsi low level driver triggers a kernel
1744 bug when trying to access the sd iostat file.
1749 Description: Kernel panics during QLogic driver reload
1750 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1751 patch, causing memory corruption.
1756 Description: journal_dev option does not work in b1_6
1757 Details : pass mount option during pre-mount.
1759 Severity : enhancement
1761 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1762 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1763 extent information of a file. It can be used to map logical blocks
1764 in a file to physical blocks in the block device.
1768 Description: LDLM soft lockups - improvement
1769 Details : It is be possible to send the lock handle along with each read
1770 or write request because the client is already doing a lock match
1771 itself so there isn't any reason the OST should have to re-do that
1775 Frequency : only on Cray X2
1777 Description: X2 build failures
1778 Details : fix build failures on Cray X2.
1782 Description: xid & resent requests
1783 Details : Initialize RPC XID from clock at startup (randomly if clock is
1786 Severity : enhancement
1788 Description: Add lustre_start utility to start or stop multiple Lustre servers
1793 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1794 Details : In case of memory pressure, list_del() can be called twice on
1795 req->rq_history_list, causing a kernel oops.
1799 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1800 Details : fix stack overflow in the distributed lock manager by defering
1801 export eviction after a failed AST to the elt thread instead of
1802 handling it in the dlm interpret routine.
1806 Description: Convert some comments to new format.
1807 Details : Update documenting comments to match doxygen conventions.
1811 Description: Grammar fixes.
1812 Details : A couple of trivial spelling fixes.
1816 Description: OSD_COUNTERS-mandatory
1817 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1818 to simplify the code.
1822 Description: simplify cmm_device freeing logic.
1823 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1827 Description: Add lockdep support to dt_object_operations locking interface.
1828 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1829 parameter indicating lock ordering. Update mdd code to use new
1834 Description: Introduce failloc constants for lockless IO tests.
1835 Details : Add two new failloc constants to test lockless IO. Only one of
1836 them in implemented---another is checked in yet to be landed
1841 Description: Add lockdep support for inode mutex.
1842 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1843 in the situations where Lustre has to lock more than one inode
1848 Description: Add optional invariants checking support.
1849 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1850 configure switch. This macro is to be used for consistency and
1851 sanity checks that are too expensive to be left in `production'
1856 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1857 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1862 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1863 Details : Use spin_lock_nested() in (the only) situation where more than
1864 one ldlm_lock is locked simultaneously. Also, fix possible
1865 dead-lock in ldlm_lock_change_resource() by enforcing particular
1870 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1871 Details : Instead of specifying each ldlm_lock call-back through separate
1872 parameter, wrap them into struct ldlm_callback_suite.
1876 Description: Kill join_lru obd method and its callers.
1877 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1878 in memory---a requirement implemented through ->o_join_lru() obd
1879 method in HEAD. Remove this method and its users.
1883 Description: Add asynchronous ldlm ENQUEUE completion handler.
1884 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1885 case that stock ldlm_completion_ast() cannot handle as it waits
1886 until lock is granted. Introduce new ldlm_completion_ast_async()
1887 for this. Also comment ldlm_completion_ast().
1891 Description: ldlm_error <-> errno conversion.
1892 Details : Add functions to map (rather arbitrary) between LDLM error codes
1893 and standard errno values. CLIO needs this to prevent LDLM specific
1894 constants from escaping ldlm and osc.
1898 Description: Kill unused ldlm_handle2lock_ns() function.
1899 Details : Kill unused ldlm_handle2lock_ns() function.
1903 Description: Add lu_ref support to ldlm_lock
1904 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1905 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1906 and ldlm_resource. LDLM interface has to be changed, because code that
1907 releases a reference on a lock, has to "know" what reference this is.
1908 In the most frequent case
1910 lock = ldlm_handle2lock(handle);
1912 LDLM_LOCK_PUT(lock);
1914 no changes are required. When any other reference (received _not_ from
1915 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1916 instead of LDLM_LOCK_PUT().
1918 Arguably, changes are pervasive, and interface requires some discipline
1919 for proper use. On the other hand, it was very instrumental in finding
1920 a few leaked lock references.
1924 Description: Add ldlm_lock_addref_try().
1925 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1926 attempts to addref a lock that might be being canceled
1931 Description: Add ldlm_weigh_callback().
1932 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1933 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1934 hard-coded `number of pages' logic.
1938 Description: Add lockdep annotations to llog code.
1939 Details : Use appropriately tagged _nested() locking calls in the places
1940 where llog takes more than one ->lgh_lock lock.
1944 Description: Add loi_kms_set().
1945 Details : Wrap kms updates into a helper function.
1949 Description: Constify instances of struct lsm_operations.
1950 Details : Constify instances of struct lsm_operations.
1954 Description: lu_conf support.
1955 Details : On a server, a file system object is uniquely identified
1956 by a fid, which is sufficient to locate and load all object
1957 state (inode). On a client, on the other hand, more data are
1958 necessary instantiate an object. Change lu_object_find() and
1959 friends to take additional `lu_conf' argument describing object.
1960 Typically this includes layout information.
1964 Description: lu_context fixes.
1965 Details : Introduce new lu_context functions that are needed on the client
1966 side, where some system threads (ptlrpcd) are shared by multiple
1967 modules, and so cannot be stopped during module shutdown.
1971 Description: Add start and stop methods to lu_device_type_operations.
1972 Details : Introduce two new methods in lu_device_type_operations, that are
1973 invoked when first instance of a given type is created and last one
1974 is destroyed respectively. This is need by CLIO.
1978 Description: Add lu_ref support to struct lu_device.
1979 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1980 track leaked references.
1984 Description: Introduce lu_kmem_descr.
1985 Details : lu_kmem_descr and its companion interface allow to create
1986 and destroy a number of kmem caches at once.
1990 Description: Fix lu_object finalization race.
1991 Details : Fix a race between lu_object_find() finding an object and its
1992 concurrent finalization. This race is (most likely) not possible
1993 on the server, but might happen on the client.
1997 Description: Introduce lu_ref interface.
1998 Details : lu_ref is a debugging module allowing to track references to
1999 a given object. It is quite cpu expensive, and has to be
2000 explicitly enabled with --enable-lu_ref. See usage description
2005 Description: Factor lu_site procfs stats into a separate function.
2006 Details : Separate lu_site stats printing code into a separate function
2007 that can be reused on a client.
2011 Description: Constify instances of struct {lu,dt,md}_device_operations.
2012 Details : Constify instances of struct {lu,dt,md}_device_operations.
2016 Description: Introduce struct md_site and move meta-data specific parts of
2017 struct lu_site here.
2018 Details : Move md-specific fields out of struct lu_site into special struct
2019 md_site, so that lu_site can be used on a client.
2023 Description: Kill mdd_lov_destroy().
2024 Details : Remove unused mdd code.
2028 Description: Add st_block checking to multistat.c.
2029 Details : Add st_block checking to multistat.c.
2033 Description: Add lu_ref support to struct obd_device.
2034 Details : Add lu_ref tracking to obd_device.
2038 Description: Kill obd_set_fail_loc().
2039 Details : Remove unused code.
2043 Description: Add special type for ptlrpc_request interpret functions.
2044 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
2045 there. Actual usage will be in CLIO.
2049 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
2050 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
2051 doesn't work with lockdep.
2055 Description: Add rwv.c test program.
2056 Details : New testing program exercising readv(2) and writev(2) (Qian).
2060 Description: Add sendfile.c test program.
2061 Details : New testing program exercising sendfile(2) (Jay).
2065 Description: Ratelimit a message that can be very frequent.
2066 Details : Ratelimit a memory allocation failure message that can
2071 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2072 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
2076 Description: Liblustre build fixes.
2077 Details : Liblustre build fixes.
2081 Description: libcfs: add cfs_{need,cond}_resched() interface.
2082 Details : libcfs: add cfs_{need,cond}_resched() definition and
2083 implementations for Linux, NT, and liblustre.
2085 Severity : enhancement
2087 Description: More exported tunables for mballoc
2088 Details : Add support for tunable preallocation window and new tunables for
2089 large/small requests
2093 Description: Detect corruption of block bitmap and checking for preallocations
2094 Details : Checks validity of on-disk block bitmap. Also it does better
2095 checking of number of applied preallocations. When corruption is
2096 found, it turns filesystem readonly to prevent further corruptions.
2100 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
2101 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
2102 window to 0. If it does hit cache miss, and read-pattern is still
2103 stride-io mode, does not reset the stride window, but also does not
2104 increase the stride window length in this case.
2108 Frequency : only for big-endian servers
2109 Description: Check if system is big-endian while mounting fs with extents feature
2110 Details : Mounting a filesystem with extents feature will fail on big-endian
2111 systems since ext3-based ldiskfs is not supported on big-endian
2112 systems. This can be over-riden with "bigendian_extents" mount option.
2114 Severity : enhancement
2116 Description: The root squash functionality
2117 Details : A security feature, which is to prevent users from being able
2118 to mount lustre on their desktop, run as root, and delete
2119 all of the files in the filesystem. The goal is accomplished by
2120 remapping user id (UID) and group id (GID) of the root user to
2121 a UID and GID specified by the system administartor via Lustre
2122 configuration management server (MGS). The functionality also
2123 allows to specify sets of clients for which the remapping does
2128 Description: Excessive recovery window
2129 Details : With AT enabled, the recovery window can be excessively long (6000+
2130 seconds). To address this problem, we no longer use
2131 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
2132 timeout no longer depends on the service time, it is set to
2133 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
2134 time via pb_service_time.
2138 Description: Watchdog triggered on MDS failover
2139 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
2140 the OSTs know that the MDS "UUID" can be reused for the same export
2141 from a different NID, so we do not need to wait for the export to be
2145 Frequency : rare, only if using MMP with Linux RAID
2147 Description: MMP doesn't work with Linux RAID
2148 Details : While using HA for Lustre servers with Linux RAID, it is possible
2149 that MMP will not detect multiple mounts. To make this work we
2150 need to unplug the device queue in RAID when the MMP block is being
2151 written. Also while reading the MMP block, we should read it from
2152 disk and not the cached one.
2154 Severity : enhancement
2156 Description: open file using fid
2157 Details : A file can be opened using just its fid, like
2158 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2161 Frequency : Only in RHEL5 when mounting multiple ext3 filesystems
2164 Description: "kmem_cache_create: duplicate cache jbd_4k" error message
2165 Details : add proper locking for creation of jbd_4k slab cache
2169 Description: MMP check in ext3_remount() fails without displaying any error
2170 Details : When multiple mount protection fails during remount, proper error
2173 Severity : enhancement
2175 Description: Allow stripe size to be up to 4G-64k
2176 Details : Fix math logic to allow large stripe sizes.
2180 Description: add check for >8TB ldiskfs filesystems
2181 Details : ext3-based ldiskfs does not support greater than 8TB LUNs.
2182 Don't allow >8TB ldiskfs filesystems to be mounted without
2183 force_over_8tb mount option
2188 Descriptoin: OSS complains "too many blocks" and crashes
2189 Details : In filter_commitrw_write():
2190 1. lquota_chkquota();
2191 2. obdo_from_inode();
2192 3. lquota_pending_commit();
2193 "2." has changed uid and gid in this case, then caused
2194 lquota_chkquota() and lquota_pending_commit() to use
2195 different uid and gid. That is the root of the bug.
2197 --------------------------------------------------------------------------------
2199 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2201 * Support for kernels:
2202 2.6.5-7.283 (SLES 9),
2203 2.6.9-55.EL (RHEL 4),
2204 2.6.16.46-0.14 (SLES 10),
2205 2.6.18.8 vanilla (kernel.org)
2206 * Client support for unpatched kernels:
2207 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2208 2.6.16 - 2.6.22 vanilla (kernel.org)
2209 * Due to recently discovered recovery problems, we do not recommend
2210 using patchless RHEL 4 clients with this or any earlier release.
2211 * Recommended e2fsprogs version: 1.39.cfs8
2212 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2213 * Starting with this release, the ldiskfs backing filesystem required
2214 by Lustre is now in its own package, lustre-ldiskfs. This package
2215 should be installed. It is versioned separately from Lustre and
2216 may be released separately in future.
2220 Description: block reactivating mgc import until all deactivates complete
2221 Details : Fix race when failing back MDT/MGS to itself (testing)
2224 Frequency : at statup only
2226 Description: mds_lov_synchronize race leads to various problems
2227 Details : simultaneous MDT->OST connections at startup can cause the
2228 sync to abort, leaving the OSC in a bad state.
2230 Severity : enhancement
2232 Description: add optional extra BUILD_VERSION info
2233 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2234 to override the lustre version.
2237 Frequency : 2.6.18 servers only
2239 Description: ll_kern_mount() doesn't release the module reference
2240 Details : The ldiskfs module reference count never drops down to 0
2241 because ll_kern_mount() doesn't release the module reference.
2246 Description: server LBUG when using old ost_num_threads parameter
2247 Details : Accept the old ost_num_threads parameter but warn that it
2248 is deprecated, and fix an off-by-one error that caused an LBUG.
2253 Description: Transient SCSI error results in persistent IO issue
2254 Details : iobuf->dr_error is not reinitialized to 0 between two
2258 Frequency : sometimes when underlying device returns I/O errors
2260 Description: OSTs not going read-only during write failures
2261 Details : OSTs are not remounted read-only when the journal commit threads
2262 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2263 instead of the ext3 wrappers.
2267 Description: poor connect scaling with increasing client count
2268 Details : Don't run filter_grant_sanity_check for more than 100 exports
2269 to improve scaling for large numbers of clients.
2272 Frequency : SLES10 only
2274 Description: sanity-quota.sh quotacheck failed: rc = -22
2275 Details : Quotas cannot be enabled on SLES10.
2278 Frequency : liblustre clients only
2280 Description: getdirentries does not give error when run on compute nodes
2281 Details : getdirentries does not fail when the size specified as an argument
2282 is too small to contain at least one entry
2284 Severity : enhancement
2286 Description: Add LNET router traceability for debug purposes
2287 Details : If a checksum failure occurs with a router as part of the
2288 IO path, the NID of the last router that forwarded the bulk data
2289 is printed so it can be identified.
2294 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2295 Details : Due to a race condition, liblustre clients were occasionally
2296 evicted incorrectly.
2298 Severity : enhancement
2300 Description: lfs setstripe use optional parameters instead of postional
2303 Severity : enhancement
2305 Description: Nanosecond timestamp support for ldiskfs
2306 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2307 resolution timestamps. There is not yet support for this at
2308 the Lustre filesystem level.
2311 Frequency : during server recovery
2313 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2314 Details : request with rq_no_resend flag not awake l_wait_event if they get
2318 Frequency : nfs export on patchless client
2320 Description: connectathon hang when test nfs export over patchless client
2321 Details : Disconnected dentry cannot be found with lookup, so we do not need
2322 to unhash it or make it invalid
2325 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2326 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2327 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2332 Description: open req refcounting wrong on reconnect
2333 Details : If reconnect happened between getting open reply from server and
2334 call to mdc_set_replay_data in ll_file_open, we will schedule
2335 replay for unreferenced request that we are about to free.
2336 Subsequent close will crash in variety of ways.
2337 Check that request is still eligible for replay in
2338 mdc_set_replay_data().
2343 Description: disable writes to filesystem when reading health_check file
2344 Details : the default for reading the health_check proc file has changed
2345 to NOT do a journal transaction and write to disk, because this
2346 can cause reads of the /proc file to hang and block HA state
2347 checking on a healthy but otherwise heavily loaded system. It
2348 is possible to return to the previous behaviour during configure
2349 with --enable-health-write.
2351 Severity : enhancement
2353 Description: 64-bit inode version
2354 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2355 to the inode. This will be required for version-based recovery.
2360 Description: MDS fails to start if a duplicate client export is detected
2361 Details : in some rare cases it was possible for a client to connect to
2362 an MDS multiple times. Upon recovery the MDS would detect this
2363 and fail during startup. Handle this more gracefully.
2365 Severity : enhancement
2367 Description: Add -o localflock option to simulate old noflock
2369 Details : This will achieve local-only flock/fcntl locks
2375 Description: log_commit_thread vs filter_destroy race leads to crash
2376 Details : Take import reference before releasing llog record semaphore
2381 Description: Wrong request locking in request set processing
2382 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2383 imp_delayed_list, in this place should be used imp_lock.
2386 Frequency : when reconnection
2388 Description: Grant Leak when osc reconnect to OST
2389 Details : When osc reconnect ost, OST(filter) should check whether it
2390 should grant more space to client by comparing fed_grant and
2391 cl_avail_grant, and return the granted space to client instead
2392 of "new granted" space, because client will call osc_init_grant
2393 to update the client grant space info.
2396 Frequency : when client reconnect to OST
2398 Description: Grant Leak when osc do resend and replay bulk write
2399 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2400 bulk write request, because the grant info while be sync between
2401 OSC and OST when reconnect, and we should ignore the grant info
2402 these of resend/replay write req.
2407 Description: Grant space more than avaiable left space sometimes.
2408 Details : When then OST is about to be full, if two bulk writing from
2409 different clients came to OST. Accord the avaliable space of the
2410 OST, the first req should be permitted, and the second one
2411 should be denied by ENOSPC. But if the seconde arrived before
2412 the first one is commited. The OST might wrongly permit second
2413 writing, which will cause grant space > avaiable space.
2416 Frequency : when client is evicted
2418 Description: Grant might be wrongly erased when osc is evicted by OST
2419 Details : when the import is evicted by server, it will fork another
2420 thread ptlrpc_invalidate_import_thread to invalidate the
2421 import, where the grant will be set to 0. While the original
2422 thread will update the grant it got when connecting. So if
2423 the former happened latter, the grant will be wrongly errased
2424 because of this race.
2429 Description: Checking Stale with correct fid
2430 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2431 to check whether it is stale, because sometimes, we want the
2432 enqueue happened anyway, and op_data.fid2 will not be initialized.
2434 Severity : enhancement
2436 Description: update patchless client
2437 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2440 Frequency : only with 2.4 kernel
2442 Description: random memory corruption
2443 Details : size of struct ll_inode_info is to big for union inode.u and this
2444 can be cause of random memory corruption.
2449 Description: Memory leak in recovery
2450 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2451 It should also check obd_fail before fsfilt_start, otherwise if
2452 fsfilt_start return -EROFS,(failover mds during mds recovery).
2453 then the req will return with repmsg->transno = 0 and rc = EROFS.
2454 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2455 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2456 be freed no matter whether fsfilt_commit success or not.
2459 Frequency : only with huge count clients
2461 Description: Prevents from taking the superblock lock in llap_from_page for
2463 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2464 for llap_from_page call in ll_removepage prevents from taking the
2465 superblock lock for a soon died page.
2470 Description: Not check open intent error before release open handle
2471 Details : in some rare cases, the open intent error is not checked before
2472 release open handle, which may cause
2473 ASSERTION(open_req->rq_transno != 0), because it tries to release
2474 the failed open handle.
2479 Description: Set cat log bitmap only after create log success.
2480 Details : in some rare cases, the cat log bitmap is set too early. and it
2481 should be set only after create log success.
2485 Description: Accessing a block bevice can re-enable I/O when Lustre is
2486 tearing down a device.
2487 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2492 Description: service threads may hog cpus when there are a lot of requests
2493 Details : Insert cond_resched to give other threads a chance to use some CPU
2498 Description: the cat log was not initialized in recovery
2499 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2500 unlink log on mds will not be initialized until mds post
2501 recovery. And also in mds post recovery, the unlink log will
2502 initialization will be done asynchronausly, so there will be race
2503 between add unlink log and unlink log initialization.
2507 Description: brw_stats were being printed incorrectly
2508 Details : brw_stats were being printed as log2 but all of them were not
2509 recorded as log2. Also remove some code duplication arising from
2510 filter_tally_{read,write}.
2514 Frequency : rare, only in recovery.
2515 Description: ASSERTION(req->rq_type != LI_POISON) failed
2516 Details : imp_lock should be held while iterating over imp_sending_list for
2517 prevent destroy request after get timeout in ptlrpc_queue_wait.
2521 Description: replay-single.sh test 52 fails
2522 Details : A lock's skiplist need to be cleanup when it being unlinked
2523 from its resource list.
2527 Description: Short directio read returns full requested size rather than
2529 Details : Direct I/O operations should return actual amount of bytes
2530 transferred rather than requested size.
2532 Severity : enhancement
2534 Description: metadata RPC reduction (e.g. for rm performance)
2535 Details : decrease the amount of synchronous RPC between clients and servers
2536 by canceling conflicing lock before the operation on the client
2537 and packing thier handles into the main operation RPC to server.
2539 Severity : enhancement
2541 Description: Async OSC create to avoid the blocking unnecessarily.
2542 Details : If a OST has no remain object, system will block on the creating
2543 when need to create a new object on this OST. Now, ways use
2544 pre-created objects when available, instead of blocking on an
2545 empty osc while others are not empty. If we must block, we block
2546 for the shortest possible period of time.
2550 Description: improve handling recoverable errors
2551 Details : if request processig with error which can be recoverable on server
2552 request should be resend, otherwise page released from cache and
2555 Severity : enhancement
2557 Description: refine locking for avoid write wrong info into lov_objid file
2558 Details : fix possible races with add new target and write/update data in
2561 --------------------------------------------------------------------------------
2563 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2568 Frequency : on some architectures
2570 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2571 Details : Uninitialized flags sometimes cause configuration commands to
2575 Frequency : patchless clients only
2577 Description: missing __iget() symbol export
2578 Details : The __iget() symbol export is missing. To avoid the need for
2579 this on patchless clients the deathrow inode reaper is turned
2580 off, and we depend on the VM to clean up old inodes. This
2581 dependency was during via the fix for bug 12181.
2583 --------------------------------------------------------------------------------
2585 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2587 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2588 INTEROPERATE with older versions automatically. In many cases a
2589 special upgrade step is needed. Please read the
2590 user documentation before upgrading any part of a live system.
2591 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2592 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2593 * WARNING: Lustre configuration and startup changes are required with
2594 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2598 Severity : enhancement
2600 Description: MountConf
2601 Details : Lustre configuration is now managed via mkfs and mount
2602 commands instead of lmc and lconf. New obd types (MGS, MGC)
2603 are added for dynamic configuration management. See
2604 https://mail.clusterfs.com/wikis/lustre/MountConf for
2607 Severity : enhancement
2609 Description: dynamic OST addition
2610 Details : OSTs can now be added to a live filesystem
2612 Severity : enhancement
2614 Description: startup order invariance
2615 Details : MDTs and OSTs can be started in any order. Clients only
2616 require the MDT to complete startup.
2618 Severity : enhancement
2620 Description: parallel, asynchronous orphan cleanup
2621 Details : orphan cleanup is now performed in separate threads for each
2622 OST, allowing parallel non-blocking operation.
2624 Severity : enhancement
2626 Description: optimized stripe assignment
2627 Details : stripe assignments are now made based on ost space available,
2628 ost previous usage, and OSS previous usage, in order to try
2629 to optimize storage space and networking resources.
2631 Severity : enhancement
2633 Description: Permanently set tunables
2634 Details : All writable /proc/fs/lustre tunables can now be permanently
2635 set on a per-server basis, at mkfs time or on a live system.
2637 Severity : enhancement
2639 Description: Lustre message v2
2640 Details : Add lustre message format v2.
2642 Severity : enhancement
2644 Description: client OST exclusion list
2645 Details : Clients can be started with a list of OSTs that should be
2646 declared "inactive" for known non-responsive OSTs.
2650 Description: SPEC SFS validation failure on NFS v2 over lustre.
2651 Details : Changes the blocksize for regular files to be 2x RPC size,
2652 and not depend on stripe size.
2654 Severity : enhancement
2656 Description: Multiple MD RPCs in flight.
2657 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2658 To avoid overly-overloading MDS, introduce a limit on number of
2659 MDS RPCs in flight for a single client and add /proc controls
2660 to adjust this limit.
2662 Severity : enhancement
2664 Description: client read/write statistics
2665 Details : Add client read/write call usage stats for performance
2666 analysis of user processes.
2667 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2668 file access. extents_stats shows chunk size distribution.
2669 extents_stats_per_process show chunk size distribution per
2672 Severity : enhancement
2674 Description: mds statistics
2675 Details : Add detailed mds operations statistics in
2676 /proc/fs/lustre/mds/*/stats.
2680 Description: Failure of copying files with lustre special EAs.
2681 Details : Client side always return success for setxattr call for lustre
2682 special xattr (currently only "trusted.lov").
2687 Description: Refcount LNET uuids
2688 Details : The global LNET uuid list grew linearly with every startup;
2689 refcount repeated list entries instead of always adding to
2692 Severity : enhancement
2694 Description: Dynamic service threads
2695 Details : Within a small range, start extra service threads
2696 automatically when the request queue builds up.
2699 Frequency : mixed-endian client/server environments
2701 Description: mixed-endian crashes
2702 Details : The new msg_v2 system had some failures in mixed-endian
2705 Severity : enhancement
2707 Description: Easy OST removal
2708 Details : OSTs can be permanently deactivated with e.g. 'lctl
2709 conf_param lustre-OST0001.osc.active=0'
2711 Severity : enhancement
2713 Description: MGS proc entries
2714 Details : Added basic proc entries for the MGS showing what filesystems
2717 Severity : enhancement
2719 Description: provide MGS failover
2720 Details : Added config lock reacquisition after MGS server failover.
2722 Severity : enhancement
2724 Description: add Linux 2.4 support
2725 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2729 Description: a large application tries to do I/O to the same resource and dies
2730 in the middle of it.
2731 Details : Check the req->rq_arrival time after the call to
2732 ost_brw_lock_get(), but before we do anything about
2733 processing it & sending the BULK transfer request. This
2734 should help move old stale pending locks off the queue as
2735 quickly as obd_timeout.
2738 Frequency : when an incorrect nid is specified during startup
2740 Description: ptlrpc connect to non-existant node causes kernel crash
2741 Details : LNET can't be re-entered from an event callback, which
2742 happened when we expire a message after the export has been
2743 cleaned up. Instead, hand the zombie cleanup off to another
2746 Severity : enhancement
2748 Description: plain/inodebits lock performance improvement
2749 Details : Grouping plain/inodebits in granted list by their request modes
2750 and bits policy, thus improving the performance of search through
2754 Frequency : only if OST filesystem is corrupted
2756 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2757 Details : for a short time RPCs with bulk IO are in the replay list,
2758 but replay of bulk IOs is unimplemented. If the OST filesystem
2759 is corrupted due to disk cache incoherency and then replay is
2760 started it is possible to trip an assertion. Avoid putting
2761 committed RPCs into the replay list at all to avoid this issue.
2764 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2766 Description: System hang on startup
2767 Details : This bug allowed the liblustre (e.g. catamount) client to
2768 return to the app before handling all startup RPCs. This
2769 could leave the node unresponsive to lustre network traffic
2770 and manifested as a server ptllnd timeout.
2772 Severity : enhancement
2774 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2775 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2776 boolean to control whether to print peer debug info when a
2777 client's RPC times out.
2780 Frequency : only for kernels with patches from Lustre below 1.4.3
2782 Description: Remove old rdonly API
2783 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2786 Frequency : only for devices with external journals
2788 Description: Set external device read-only also
2789 Details : During a commanded failover stop, we set the disk device
2790 read-only while the server shuts down. We now also set any
2791 external journal device read-only at the same time.
2794 Frequency : when upgrading from 1.4 while trying to change parameters
2796 Description: The wrong (new) MDC name was used when setting parameters for
2797 upgraded MDT's. Also allows changing of OSC (and MDC)
2798 parameters if --writeconf is specified at tunefs upgrade time.
2801 Frequency : when setting specific ost indicies
2803 Description: QOS code breaks on skipped indicies
2804 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2805 created with --index need not be sequential.
2807 Severity : enhancement
2809 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2810 Details : The uninit_groups feature works in conjunction with the kernel
2811 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2812 up the pass1 processing of e2fsck. This is a read-only feature
2813 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2814 mount filesystems that have had this feature enabled.
2816 Severity : enhancement
2818 Description: Improve multi-block allocation algorithm to avoid fragmentation
2819 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2820 allocation locality and avoid filesystem fragmentation.
2822 ------------------------------------------------------------------------------
2824 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2826 * Support for kernels:
2827 2.4.21-47.0.1.EL (RHEL 3)
2828 2.6.5-7.283 (SLES 9)
2829 2.6.9-42.0.10.EL (RHEL 4)
2830 2.6.12.6 vanilla (kernel.org)
2831 2.6.16.27-0.9 (SLES 10)
2832 * Recommended e2fsprogs version: 1.39.cfs5
2834 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2838 Frequency : occasional, depends on client load and configuration
2839 Bugzilla : 12181, 12203
2840 Description: data loss for recently-modified files
2842 Details : In some cases it is possible that recently written or created
2843 files may not be written to disk in a timely manner (this should
2844 normally be within 30s unless client IO load is very high).
2845 The problem appears as zero-length files or files that are a
2846 multiple of 1MB in size after a client crash or client eviction
2847 that are missing data at the end of the file.
2849 This problem is more likely to be hit on clients where files are
2850 repeatedly created and unlinked in the same directory, clients
2851 have a large amount of RAM, have many CPUs, the filesystem has
2852 many OSTs, the clients are rebooted frequently, and/or the files
2853 are not accessed by other nodes after being written.
2855 The presence of the problem can be detected by looking at
2856 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2857 smaller than the second (nr_unused) then dirty files will not
2858 be flushed automatically to disk. "sync; sleep 10" should be
2859 run several times on the node before unmounting it to update
2860 Lustre (this is also safe to run on nodes without this problem).
2862 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2863 that can cause this same problem, so customers using that kernel
2864 also need to update the kernel in addition to Lustre. In order
2865 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2867 It is normal that files written just before a client crash (less
2868 than 30s) may not yet have been flushed to disk, even for local
2872 Frequency : frequent on thin XT3 nodes
2874 Description: UUID collision on thin XT3 Linux nodes
2875 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2876 randomly, since we relied on an insufficiently-seeded PRNG.
2881 Description: OSS hangs after "All ost request buffers busy"
2882 Details : A deadlock between quota and journal operations caused OSS
2883 hangs after printing "All ost request buffers busy."
2886 Frequency : always on liblustre builds
2888 Description: Cleanup compiler warnings on liblustre
2891 Frequency : always on liblustre builds on XT3
2893 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2897 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2898 > 32000 subdirectories
2899 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2900 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2901 subdirectory count crosses 32000. This will aid e2fsck to
2902 correctly handle more than 32000 subdirectories.
2905 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2907 Description: System hang on startup
2908 Details : This bug allowed the liblustre (e.g. catamount) client to
2909 return to the app before handling all startup RPCs. This
2910 could leave the node unresponsive to lustre network traffic
2911 and manifested as a server ptllnd timeout.
2913 Severity : enhancement
2915 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2916 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2917 boolean to control whether to print peer debug info when a
2918 client's RPC times out.
2923 Description: make O_SYNC working on 2.6 kernels
2924 Details : 2.6 kernels use different method for mark pages for write,
2925 so need add a code to lustre for O_SYNC work.
2930 Description: Failure to close file and release space on NFS
2931 Details : Put inode details into lock acquired in ll_intent_file_open.
2932 Use mdc_intent_lock in ll_intent_open to properly
2933 detect all kind of errors unhandled by mdc_enqueue.
2938 Description: proc file read during shutdown sometimes raced obd removal,
2940 Details : Add lock to prevent obd access after proc file removal.
2943 Frequency : Only for files larger than 4GB on 32-bit clients.
2945 Description: improperly doing page alignment of locks
2946 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2947 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2952 Description: RPCs being resent when they shouldn't be
2953 Details : Some RPCs that should not be resent are being resent. This
2954 can cause inconsistencies in the RPC state machine. Do not
2955 resend such requests.
2958 Frequency : rare, only with NFS export
2960 Description: Crash on NFS re-export node
2961 Details : under very unusual load conditions an assertion is hit in
2962 ll_intent_file_open()
2965 Frequency : only if OST filesystem is corrupted
2967 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2968 Details : for a short time RPCs with bulk IO are in the replay list,
2969 but replay of bulk IOs is unimplemented. If the OST filesystem
2970 is corrupted due to disk cache incoherency and then replay is
2971 started it is possible to trip an assertion. Avoid putting
2972 committed RPCs into the replay list at all to avoid this issue.
2977 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2978 Details : Large single O_DIRECT read and write calls can fail to allocate
2979 a sufficiently large buffer to process the request. In case of
2980 allocation failure the allocation is retried with a smaller
2981 buffer and broken into smaller requests.
2983 Severity : enhancement
2985 Description: Add -o localflock option to simulate old noflock behaviour.
2986 Details : This will achieve local-only flock/fcntl locks coherentness.
2991 Description: versioning check is incomplete
2992 Details : Checking the version difference of client vs. server, report
2993 error if the gap is too big.
2998 Description: add support PG_writeback bit
2999 Details : add support for PG_writeback bit for Lustre, for more carefull
3000 work with page cache in 2.6 kernel. This also fix some deadlocks
3001 and remove hack for work O_SYNC with 2.6 kernel.
3003 Severity : enhancement
3005 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
3006 Details : The uninit_groups feature works in conjunction with the kernel
3007 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
3008 up the pass1 processing of e2fsck. This is a read-only feature
3009 in ldiskfs2 only, so older kernels and current ldiskfs cannot
3010 mount filesystems that have had this feature enabled.
3012 Severity : enhancement
3014 Description: Improve multi-block allocation algorithm to avoid fragmentation
3015 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
3016 allocation locality and avoid filesystem fragmentation.
3018 ------------------------------------------------------------------------------
3020 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
3022 * Support for kernels:
3023 2.6.9-42.0.3.EL (RHEL 4)
3024 2.6.5-7.276 (SLES 9)
3025 2.4.21-47.0.1.EL (RHEL 3)
3026 2.6.12.6 vanilla (kernel.org)
3027 2.6.16.21-0.8 (SLES10)
3028 * Recommended e2fsprogs version: 1.39.cfs2-0
3030 * The backwards-compatible /proc/sys/portals symlink has been removed
3031 in this release. Before upgrading, please ensure that you change
3032 any configuration scripts or /etc/sysctl.conf files that access
3033 /proc/sys/portals/* or sysctl portals.* to use the corresponding
3034 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
3035 in advance of the upgrade on any system running Lustre 1.4.6 or
3036 newer, since /proc/sys/lnet was added in that version.
3037 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
3041 Frequency : only when quota is used
3043 Description: avoid scanning export list for quota master
3044 Details : Change the algorithms to avoid scanning export list in order
3045 to improve the efficiency.
3048 Frequency : MDS failover only, very rarely
3050 Description: "went back in time" messages on mds failover
3051 Details : The greatest transno may be lost when the current operation
3052 finishes with an error (transno==0) and the client's last_rcvd
3053 record is over-written. Save the greatest transno in the
3054 mds_last_transno for this case.
3057 Frequency : always for specific kernels and striping counts
3059 Description: client may get "Matching packet too big" without ACL support
3060 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
3061 when trying to access files in certain configurations. The
3062 clients should in fact be denied when mounting because they do
3063 not understand ACLs.
3066 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
3068 Description: many clients connecting with IO in progress causes connect timeouts
3069 Details : Avoid synchronous journal commits to avoid delays caused by many
3070 clients connecting/disconnecting when bulk IO is in progress.
3071 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
3072 OST_IO_PORTAL to avoid delays behind potentially many pending
3076 Frequency : occasionally with multiple writers to a single file
3078 Description: shared writes to file may result in wrong size reported by stat()
3079 Details : Allow growing of kms when extent lock is cancelled
3082 Frequency : always with random mmap IO to multi-striped file
3084 Description: mmap write might be lost if we are writing to a 'hole' in stripe
3085 Details : Only if the hole is at the end of OST object so that kms is too
3086 small. Fix is to increase kms accordingly in ll_nopage.
3089 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
3091 Description: writes to a missing object would leak memory on the OST
3092 Details : If there is an inconsistency between the MDS and OST filesystems,
3093 such that the MDS references an object that doesn't exist, writes
3094 to that object will leak memory due to incorrect cleanup in the
3095 error handling path, eventually running out of memory on the OST.
3100 Description: Creating too long symlink causes lustre errors
3101 Details : Check symlink and name lengths before sending requests to MDS.
3104 Frequency : only if flock is enabled (not on by default)
3106 Description: posix locks not released on fd closure on 2.6.9+
3107 Details : We failed to add posix locks to list of inode locks on 2.6.9+
3108 kernels, this caused such locks not to be released on fd close and
3109 then assertions on fs unmount about still used locks.
3112 Frequency : MDS failover only, very rarely
3114 Description: clients may get ASSERTION(granted_lock != NULL)
3115 Details : When request was taking a long time, and a client was resending
3116 a getattr by name lock request. The were multiple lock requests
3117 with the same client lock handle and
3118 mds_getattr_name->fixup_handle_for_resent_request found one of the
3119 lock handles but later failed with ASSERTION(granted_lock != NULL).
3124 Description: handle->h_buffer_credits > 0, assertion failure
3125 Details : h_buffer_credits is zero after truncate, causing assertion
3126 failure. This patch extends the transaction or creates a new
3130 Frequency : NFS re-export or patchless client
3131 Bugzilla : 11179, 10796
3132 Description: Crash on NFS re-export node (__d_move)
3133 Details : We do not want to hash the dentry if we don't have a lock.
3134 But if this dentry is later used in d_move, we'd hit uninitialised
3135 list head d_hash, so we just do this to init d_hash field but
3136 leave dentry unhashed.
3139 Frequency : NFS re-export or patchless client
3141 Description: NFS exports has problem with symbolic link
3142 Details : lustre client didn't properly install dentry when re-exported
3143 to NFS or running patchless client.
3146 Frequency : NFS re-export or patchless client
3148 Description: Various nfs/patchless fixes.
3149 Details : fixes reuse disconected alias for lookup process - this fixes
3150 warning "find_exported_dentry: npd != pd",
3151 fix permission error with open files at nfs.
3152 fix apply umask when do revalidate.
3155 Frequency : occasional
3157 Description: Crash on NFS re-export node
3158 Details : calling clear_page() on the wrong pointer triggered oops in
3159 generic_mapping_read().
3162 Frequency : rarely, using O_DIRECT IO
3164 Description: unaligned directio crashes client with LASSERT
3165 Details : check for unaligned buffers before trying any requests.
3168 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
3170 Description: stale data returned from RAID cache
3171 Details : If only a small amount of IO is done to the RAID device before
3172 reading it again it is possible to get stale data from the RAID
3173 cache instead of reading it from disk.
3176 Frequency : always for sles10 kernel
3178 Description: sles10 support
3179 Details : ll_follow_link: compile fixes and using of nd_set_link
3180 under newer kernels.
3183 Frequency : depends on arch, kernel and compiler version, always on sles10
3186 Description: recursive or deep enough symlinks cause stack overflow
3187 Details : getting rid of large stack-allocated variable in
3191 Frequency : depends on hardware
3193 Description: lustre write performance loss in the SLES10 kernel
3194 Details : the performance loss is caused by using of write barriers in the
3195 ext3 code. The SLES10 kernel turns barrier support on by
3196 default. The fix is to undo that change for ldiskfs.
3198 ------------------------------------------------------------------------------
3200 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3202 * Support for kernels:
3203 2.6.9-42.0.3EL (RHEL 4)
3204 2.6.5-7.276 (SLES 9)
3205 2.4.21-47.0.1.EL (RHEL 3)
3206 2.6.12.6 vanilla (kernel.org)
3210 Frequency : quota enabled and large files being deleted
3212 Description: releasing more than 4GB of quota at once hangs OST
3213 Details : If a user deletes more than 4GB of files on a single OST it
3214 will cause the OST to spin in an infinite loop. Release
3215 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3220 Description: statfs data retrieved from /proc may be stale or zero
3221 Details : When reading per-device statfs data from /proc, in the
3222 {kbytes,files}_{total,free,avail} files, it may appear
3223 as zero or be out of date.
3226 Frequency : systems with MD RAID1 external journal devices
3228 Description: lconf's call to blkid is confused by RAID1 journal devices
3229 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3230 of returning all block devices that match the journal UUID.
3233 Frequency : always, for aggregate stripe size over 4GB
3235 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3236 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3237 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3240 Frequency : always if "lfs find" used on a local file/directory
3242 Description: "lfs find" segfaults if used on a local file/directory
3243 Details : The case where a directory component was not specified wasn't
3244 handled correctly. Handle this properly.
3247 Frequency : always on ppc64
3249 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3250 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3251 perform endianness conversion.
3254 Frequency : rarely (truncate to non-zero file size after write under load)
3255 Bugzilla : 10730, 10687
3256 Description: Files padded with zeros to next 4K multiple
3257 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3258 it is possible that files that are truncated to a non-zero size
3259 immediately after being written are filled with zero bytes beyond
3260 the truncated size. No file data is lost.
3262 Severity : enhancement
3264 Description: Allow recovery/failover for liblustre clients.
3265 Details : liblustre clients were unaware of failover configurations until
3268 Severity : enhancement
3270 Description: user file locks should fail when not mounting with flock option
3271 Details : Set up an error-returning stub in ll_file_operations.lock field
3272 to prevent incorrect behaviour when client is mounted without
3273 flock option. Also, set up properly f_op->flock field for
3277 Frequency : always on ia64
3279 Description: "lfs df" loops on printing out MDS statfs information
3280 Details : The obd_ioctl_data was not initialized and in some systems
3281 this caused a failure during the ioctl that did not return
3282 an error. Initialize the struct and return an error on failure.
3285 Frequency : SLES 9 only
3287 Description: Error of copying files with lustre special EAs as root
3288 Details : Client side always return success for setxattr call for lustre
3289 special xattr (currently only "trusted.lov").
3292 Frequency : rarely on clusters with both ia64+i386 clients
3294 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3295 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3296 clients are doing concurrent writes to the same file it is
3297 possible that the ia64 clients may LASSERT because the OST
3298 extent locks are not PAGE_SIZE aligned. Ensure that grown
3299 locks are always aligned on the request boundary.
3302 Frequency : specific use, occasional
3304 Description: Overwriting in use executable truncates on-disk binary image
3305 Details : If one node attempts to overwrite an executable in use by
3306 another node, we now correctly return ETXTBSY instead of
3307 truncating the file.
3309 Severity : enhancement
3311 Description: Async OSC create to avoid the blocking unnecessarily.
3312 Details : If a OST has no remain object, system will block on the creating
3313 when need to create a new object on this OST. Now, ways use
3314 pre-created objects when available, instead of blocking on an
3315 empty osc while others are not empty. If we must block, we block
3316 for the shortest possible period of time.
3321 Description: chmod on Lustre root is propagated to other clients
3322 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3323 invalid by the follow_mount time.
3326 Frequency : liblustre clients only
3328 Description: Race in 'instant cancel' lock handling could lead to such locks
3329 never to be granted in case of SMP MDS
3330 Details : Do not destroy not yet granted but cbpending locks in
3334 Frequency : replay/resend of open
3336 Description: non null lock assetion failure in mds_intent_policy
3337 Details : Trying to replay/resend lockless open requests resulted in
3338 mds_open() returning 0 with no lock. Now it sets a flag if
3339 a lock is going to be returned.
3341 Severity : enhancement
3343 Description: Checksum enhancements
3344 Details : New checksum enhancements allow for resending RPCs that failed
3347 Severity : enhancement
3349 Description: Tunables on number of dirty pages in cacche
3350 Details : Allow to set limit on number of dirty pages cached.
3355 Description: client crash on unmount - lock still has references
3356 Details : In some error handling cases it was possible to leak a lock
3357 reference on a client while accessing a file. This was not
3358 harmful to the client during operation, but would cause the
3359 client to crash when the filesystem is unmounted.
3362 Frequency : specific case, rare
3364 Description: ETXTBSY on mds though file not in use by client
3365 Details : ETXTBSY is no longer incorrectly returned when attempting to
3366 chmod or chown a directory that the user previously tried to
3367 execute or a currently-executing binary.
3370 Frequency : extremely rare except on liblustre-based clients
3372 Description: Lustre space not freed when files are deleted
3373 Details : Clean up open-unlinked files after client eviction. Previously
3374 the unlink was skipped and the files remained as orphans.
3379 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3380 Details : In some cases it was possible to send a blocking callback to
3381 a client doing a glimpse, even though that client didn't get
3382 a lock granted. When the glimpse lock is cancelled on the OST
3383 the freed lock is left on the waiting list and corrupted the list.
3386 Frequency : all core dumps
3388 Description: Broke core dumps to lustre
3389 Details : Negative dentry may be unhashed if parent does not have UPDATE
3390 lock, but some callers, e.g. do_coredump, expect dentry to be
3391 hashed after successful create, hash it in ll_create_it.
3393 ------------------------------------------------------------------------------
3395 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3397 * Support for kernels:
3398 2.6.9-42.0.2.EL (RHEL 4)
3399 2.6.5-7.276 (SLES 9)
3400 2.4.21-40.EL (RHEL 3)
3401 2.6.12.6 vanilla (kernel.org)
3405 Frequency : always on RHEL 3
3407 Description: Number of open files grows over time
3408 Details : The number of open files grows over time, whether or not
3409 Lustre is started. This was due to a filp leak introduced
3410 by one of our kernel patches.
3412 ------------------------------------------------------------------------------
3414 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3416 * Support for kernels:
3417 2.6.9-42.EL (RHEL 4)
3418 2.6.5-7.276 (SLES 9)
3419 2.4.21-40.EL (RHEL 3)
3420 2.6.12.6 vanilla (kernel.org)
3425 Bugzilla : 5719, 9635, 9792, 9684
3426 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3427 Details : If a server is under heavy load and cannot reply to new
3428 connection requests before the client resends the (re)connect,
3429 the connection handling code can behave badly if two service
3430 threads are concurrently handing separate (re)connections from
3431 the same client. Add better locking to the connection handling
3432 code, and ensure that only a single connection will be processed
3433 for a given client UUID, even if the lock is dropped.
3435 Severity : enhancement
3437 Description: add TCP zero-copy support to kernel
3438 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3439 sends if the hardware supports scatter-gather and checksumming.
3440 This allows socklnd to do client-write and server-read more
3441 efficiently and reduce CPU utilization from skbuf copying.
3444 Frequency : only if NFS exporting from client
3446 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3447 Details : If setting attributes on a file created under NFS that had
3448 never been opened it would be possible to oops the client
3449 if the file had no objects.
3452 Frequency : always for liblustre
3454 Description: liblustre client does MDS+OSTs setattr RPC for each write
3455 Details : When doing a write from a liblustre client, the client
3456 incorrectly issued an RPC to the MDS and each OST the file was
3457 striped over in order to update the timestamps. When writing
3458 with small chunks and many clients this could overwhelm the MDS
3459 with RPCs. In all cases it would slow down the write because
3460 these RPCs are unnecessary.
3462 Severity : enhancement
3464 Description: allow number of MDS service threads to be changed at module load
3465 Details : It is now possible to change the number of MDS service threads
3466 running. Adding "options mds mds_num_threads={N}" to the MDS's
3467 /etc/modprobe.conf will set the number of threads for the next
3468 time Lustre is restarted (assuming the "mds" module is also
3469 reloaded at that time). The default number of threads will
3470 stay the same, 32 for most systems.
3475 Description: OST crash if filesystem is unformatted or corrupt
3476 Details : If an OST is started on a device that has never been formatted
3477 or if the filesystem is corrupt and cannot even mount then the
3478 error handling cleanup routines would dereference a NULL pointer.
3483 Description: NULL pointer deref in llap_from_page.
3484 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3485 incorrect) mapping if the page was truncated/reclaimed while it was
3486 searched for. Check for this condition and skip such pages when
3487 doing readahead. Introduce extra check to llap_from_page() to
3488 verify page->mapping->host is non-NULL (so page is not anonymous).
3491 Frequency : Sometimes when using sys_sendfile
3493 Description: "page not covered by a lock" warnings from ll_readpage
3494 Details : sendfile called ll_readpage without right page locks present.
3495 Now we introduced ll_file_sendfile that does necessary locking
3496 around call to generic_file_sendfile() much like we do in
3500 Frequency : with certain MDS communication failures at client mount time
3502 Description: NULL pointer deref after failed client mount
3503 Details : a client connection request may delayed by the network layer
3504 and not be sent until after the PTLRPC layer has timed out the
3505 request. If the client fails the mount immediately it will try
3506 to clean up before the network times out the request. Add a
3507 reference from the request import to the obd device and delay
3508 the cleanup until the network drops the request.
3511 Frequency : occasionally during client (re)connect
3513 Description: assertion failure during client (re)connect
3514 Details : processing a client connection request may be delayed by the
3515 client or server longer than the client connect timeout. This
3516 causes the client to resend the connection request. If the
3517 original connection request is replied in this interval, the
3518 client may trip an assertion failure in ptlrpc_connect_interpret()
3519 which thought it would be the only running connect process.
3522 Frequency : only with obd_echo servers and clients that are rebooted
3524 Description: kernel BUG accessing uninitialized data structure
3525 Details : When running an obd_echo server it did not start the ping_evictor
3526 thread, and when a client was evicted an uninitialized data
3527 structure was accessed. Start the ping_evictor in the RPC
3528 service startup instead of the OBD startup.
3530 Severity : enhancement
3531 Bugzilla : 10193 (patchless)
3532 Description: Remove dependency on various unexported kernel interfaces.
3533 Details : No longer need reparent_to_init, exit_mm, exit_files,
3534 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3537 Frequency : rare (only users of deprecated and unsupported LDAP config)
3539 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3540 Details : LDAP apparently contains 'lustreName' attributes instead of
3541 'name'. A simple remapping of the name is sufficient.
3544 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3546 Description: waiting_locks_callback trips kernel BUG if client is evicted
3547 Details : Running with the dump_on_timeout debug flag turned on makes
3548 it possible that the waiting_locks_callback() can try to dump
3549 the Lustre kernel debug logs from an interrupt handler. Defer
3550 this log dumping to the expired_lock_main() thread.
3552 Severity : enhancement
3554 Description: Support NFS exporting on 2.6 kernels.
3555 Details : Implement non-rawops metadata methods for NFS server to use without
3556 changing NFS server code.
3559 Frequency : very rare (synthetic metadata workload only)
3561 Description: two racing renames might cause an MDS thread to deadlock
3562 Details : Running the "racer" program may cause one MDS thread to rename
3563 a file from being the source of a rename to being the target of
3564 a rename at exactly the same time that another thread is doing
3565 so, and the second thread has already enqueued these locks after
3566 doing a lookup of the target and is trying to relock them in
3567 order. Ensure that we don't try to re-lock the same resource.
3570 Frequency : only very large systems with liblustre clients
3572 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3573 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3574 all OSTs in parallel. This allows the eviction of stale liblustre
3575 clients to proceed much faster than if they were done in series,
3576 and also offers similar improvements for other set_info RPCs.
3581 Description: excessive CPU usage during initial read phase on client
3582 Details : During the initial read phase on a client, it would agressively
3583 retry readahead on the file, consuming too much CPU and impacting
3584 performance (since 1.4.5.8). Improve the readahead algorithm
3585 to avoid this, and also improve some other common cases (read
3586 of small files in particular, where "small" is files smaller than
3587 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3592 Description: MDS crash when receiving packet with unknown intent.
3593 Details : Do not LBUG in unknown intent case, just return -EFAULT
3595 Severity : enhancement
3596 Bugzilla : 9293, 9385
3597 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3598 Details : Do not serialize getattr (non-intent version) and statfs.
3601 Frequency : occasional, when OST network is overloaded/intermittent
3603 Description: client evicted by OST after bulk IO timeout
3604 Details : If a client sends a bulk IO request (read or write) the OST
3605 may evict the client if it is unresposive to its data GET/PUT
3606 request. This is incorrect if the network is overloaded (takes
3607 too long to transfer the RPC data) or dropped the OST GET/PUT
3608 request. There is no need to evict the client at all, since
3609 the pinger and/or lock callbacks will handle this, and the
3610 client can restart the bulk request.
3613 Frequency : Always when mmapping file with no objects
3615 Description: client crashes when mmapping file with no objects
3616 Details : Check that we actually have objects in a file before doing any
3617 operations on objects in ll_vm_open, ll_vm_close and
3623 Description: Request leak when working with deleted CWD
3624 Details : Introduce advanced request refcount tracking for requests
3625 referenced from lustre intent.
3627 Severity : Enhancement
3629 Description: Cache open file handles on client.
3630 Details : MDS now will return special lock along with openhandle, if
3631 requested and client is allowed to hold openhandle, even if unused,
3632 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3633 closing files for every read/write openration.
3635 Severity : Enhancement
3637 Description: Cache open negative dentries on client when possible.
3638 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3639 negative dentries on lock revocation.
3644 Description: Remounting a client read-only wasn't possible with a zconf mount
3645 Details : It wasn't possible to remount a client read-only with llmount.
3647 Severity : enhancement
3648 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3649 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3650 MPICH is included to add Lustre-specific ADIO interfaces.
3651 This is based closely on the UFS ADIO layer and only differs
3652 in file creation, in order to allow the OST striping to be set.
3653 This is user-contributed code and not supported by CFS.
3658 Description: extended inode attributes (immutable, append-only) work improperly
3659 when 2.4 and 2.6 kernels are used on client/server or vice versa
3660 Details : Introduce kernel-independent values for these flags.
3662 Severity : enhancement
3665 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3666 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3667 granularity. Now, Fractional MB granularity is be supported,
3668 this is very useful for low memory system.
3670 Severity : enhancement
3672 Description: Getattr by fid
3673 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3674 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3675 have locks and avoids getattr rpc requests for it on every stat.
3678 Frequency : Always, for filesystems larger than 2TB
3680 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3681 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3682 filesystems larger than 2TB to be mounted with mballoc enabled.
3685 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3687 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3688 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3689 filesystems larger than 2TB is mounted then the kernel will
3690 silently corrupt the start of the filesystem. CONFIG_LBD is
3691 enabled for all CFS-supported kernels, but the possibility of
3692 this happening with a modified kernel config exists.
3694 Severity : enhancement
3696 Description: add client O_DIRECT support for 2.6 kernels
3697 Details : It is now possible to do O_DIRECT reads and writes to files
3698 in the Lustre client mountpoint on 2.6 kernel clients.
3700 Severity : enhancement
3702 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3703 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3704 parallel, not waiting for response from every OST before sending
3705 a rpc to the next OST.
3710 Description: setattr vs write race when updating file timestamps
3711 Details : Client processes that update a file timestamp into the past
3712 right after writing to the file (e.g. tar) it is possible that
3713 the updated file modification time can be reset to the current
3714 time due to a race between processing the setattr and write RPC.
3716 Severity : enhancement
3718 Description: Bring 'lfs find' closer in line with regular Linux find.
3719 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3720 -print0 options and obtains all the needed info through the lustre
3723 Severity : enhancement
3725 Description: support up to 1024 configured devices on one node
3726 Details : change obd_dev array from statically allocated to dynamically
3727 allocated structs as they are first used to reduce memory usage
3732 Description: Flush dirty partially truncated pages during truncate
3733 Details : Immediatelly flush partially truncated pages in filter_setattr,
3734 this way we completely avoid having any pages in page cache on OST
3735 and can retire ugly workarounds during writes to flush such pages.
3740 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3741 Details : protect inode from truncation within vfs_unlink() context
3742 just take a reference before calling vfs_unlink() and release it
3743 when parent's i_sem is free.
3748 Description: last_id value checked outside lock on OST caused LASSERT failure
3749 Details : If there were multiple MDS->OST object precreate requests in
3750 flight, it was possible that the OST's last object id was checked
3751 outside a lock and incorrectly tripped an assertion. Move checks
3752 inside locks, and discard old precreate requests.
3755 Frequency : always, if extents are used on OSTs
3757 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3758 Details : OSTs using the extents format would not zero the high 16 bits of
3759 the index physical block number. This is not a problem for any
3760 OST filesystems smaller than 16TB, and no kernels support ext3
3761 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3762 new/modified files) and can be fixed for existing filesystems
3763 with e2fsprogs-1.39-cfs1.
3768 Description: import connection selection may be incorrect if timer wraps
3769 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3770 import connections to be ignored if the 32-bit jiffies counter
3771 wraps. Use a 64-bit jiffies counter.
3774 Frequency : very large clusters immediately after boot
3776 Description: LNET request buffers exhausted under heavy short-term load
3777 Details : If a large number of client requests are generated on a service
3778 that has previously never seen so many requests it is possible
3779 that the request buffer growth cannot keep up with the spike in
3780 demand. Instead of dropping incoming requests, they are held in
3781 the LND until the RPC service can accept more requests.
3784 Frequency : Sometimes during replay
3786 Description: Assertion failure in ll_local_open after replay.
3787 Details : If replay happened on an open request reply before we were able
3788 to set replay handler, reply will become not swabbed tripping the
3789 assertion in ll_local_open. Now we set the handler right after
3790 recognising of open request
3793 Frequency : very rare
3795 Description: kernel reports "badness in vsnprintf"
3796 Details : Reading from the "recovery_status" /proc file in small chunks
3797 may cause a negative length in lprocfs_obd_rd_recovery_status()
3798 call to vsnprintf() (which is otherwise harmless). Exit early
3799 if there is no more space in the output buffer.
3801 Severity : enhancement
3803 Description: clear OBD RPC statistics by writing to them
3804 Details : It is now possible to clear the OBD RPC statistics by writing
3805 to the "stats" file.
3810 Description: Client mtime is not the same on different clients after utimes
3811 Details : In some cases, the client was using the utimes() syscall on
3812 a file cached on another node. The clients now validate the
3813 ctime from the MDS + OSTs to determine which one is right.
3818 Description: Inability to activate failout mode
3819 Details : lconf script incorrectly assumed that in pythong string's numeric
3820 value is used in comparisons.
3823 Frequency : always with multiple stripes per file
3825 Description: Inefficient object allocation for mutli-stripe files
3826 Details : When selecting which OSTs to stripe files over, for files with
3827 a stripe count that divides evenly into the number of OSTs,
3828 the MDS is always picking the same starting OST for each file.
3829 Return the OST selection heuristic to the original design.
3834 Description: mount failures may take full timeout to return an error
3835 Details : Under some heavy load conditions it is possible that a
3836 failed mount can wait for the full obd_timeout interval,
3837 possibly several minutes, before reporting an error.
3838 Instead return an error as soon as the status is known.
3840 Frequency : quota enabled and large files being deleted
3842 Description: releasing more than 4GB of quota at once hangs OST
3843 Details : If a user deletes more than 4GB of files on a single OST it
3844 will cause the OST to spin in an infinite loop. Release
3845 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3850 Description: statfs data retrieved from /proc may be stale or zero
3851 Details : When reading per-device statfs data from /proc, in the
3852 {kbytes,files}_{total,free,avail} files, it may appear
3853 as zero or be out of date.
3856 Frequency : systems with MD RAID1 external journal devices
3858 Description: lconf's call to blkid is confused by RAID1 journal devices
3859 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3860 of returning all block devices that match the journal UUID.
3863 Frequency : always, for aggregate stripe size over 4GB
3865 Description: assertion fails when trying to use 4GB stripe size
3866 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3867 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3870 Frequency : always on ppc64
3872 Description: the first write on an ext3 filesystem with mballoc got stuck
3873 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3874 perform endianness conversion.
3876 ------------------------------------------------------------------------------
3878 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3880 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3881 INTEROPERATE with older versions automatically. Please read the
3882 user documentation before upgrading any part of a live system.
3883 * WARNING: Lustre networking configuration changes are required with
3884 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3887 * Support for kernels:
3888 2.6.9-22.0.2.EL (RHEL 4)
3889 2.6.5-7.244 (SLES 9)
3890 2.6.12.6 vanilla (kernel.org)
3893 Severity : enhancement
3894 Bugzilla : 7981/8208
3895 Description: Introduced Lustre Networking (LNET)
3896 Details : LNET is new networking infrastructure for Lustre, it includes
3897 a reorganized network configuration mode (see the user
3898 documentation for full details) as well as support for routing
3899 between different network fabrics. Lustre Networking Devices
3900 (LNDs) for the supported network fabrics have also been
3901 created for this new infrastructure.
3903 Severity : enhancement
3904 Description: Introduced Access control lists
3905 Details : clients can set ACLs on files and directories in order to have
3906 more fine-grained permissions than the standard Unix UGO+RWX.
3907 The MDS must be started with the "-o acl" mount option.
3909 Severity : enhancement
3910 Description: Introduced filesystem quotas
3911 Details : Administrators may now establish per-user quotas on the
3914 Severity : enhancement
3916 Description: Configuration change for the XT3
3917 The PTLLND is now used to run Lustre over Portals on the XT3
3918 The configure option(s) --with-cray-portals are no longer used.
3919 Rather --with-portals=<path-to-portals-includes> is used to
3920 enable building on the XT3. In addition to enable XT3 specific
3921 features the option --enable-cray-xt3 must be used.
3926 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3927 Details : A race between allocating a new llcd and re-getting the llcd_lock
3928 allowed another thread to grab newly-allocated llcd.
3930 Severity : enhancement
3932 Description: 2.6 OST async journal commit and locking fix to improve performance
3933 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3934 2.6 kernels are now async as they already were in 2.4 kernels so
3935 that they can commit concurrently with the network bulk transfer.
3936 For block-allocated files the filter allocation semaphore is held
3937 to avoid filesystem fragmentation during allocation. BKL lock
3938 removed for 2.6 xattr operations where it is no longer needed.
3943 Description: lconf incorrectly determined whether two IP networks could talk
3944 Details : In some more complicated routing and multiple-network
3945 configurations, lconf will avoid trying to make a network
3946 connection to a disjoint part of the IP space. It was doing the
3947 math incorrectly for one set of cases.
3952 Description: Fix for potential infinite loop processing records in an llog.
3953 Details : If an llog record is corrupted/zeroed, it is possible to loop
3954 forever in llog_process(). Validate the llog record length
3955 and skip the remainder of the block on error.
3958 Frequency : occasional (liblustre only)
3960 Description: liblustre could not open files whose last component is a symlink
3961 Details : sysio_path_walk() would incorrectly pass the open intent to
3962 intermediate path components.
3965 Frequency : rare (liblustre only with non-standard tuning)
3966 Bugzilla : 7201 (7350)
3967 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3968 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3969 after it has been released, only do this for referenced locks
3971 Severity : enhancement
3973 Description: specifying an (invalid) directory default stripe_size of -1
3974 would reset the directory default striping
3975 Details : stripe_size -1 was used internally to signal directory stripe
3976 removal, now use "all default" to signal dir stripe removal
3977 as a directory striping of "all default" is not useful
3980 Frequency : common for large clusters running liblustre clients
3982 Description: doing an ls when liblustre clients are running is slow
3983 Details : sending a glimpse AST to a liblustre client waits for every AST
3984 to time out, as liblustre clients will not respond. Since they
3985 cannot cache data we refresh the OST lock LVB from disk instead.
3987 Severity : enhancement
3989 Description: doing an ls at the same time as file IO can be slow
3990 Details : enqueue and other "small" requests can be blocked behind many
3991 large IO requests. Create a new OST IO portal for non-IO
3992 requests so they can be processed faster.
3995 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3997 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3998 created file with mode 000
3999 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
4000 allow an ftruncate() as a non-root user to a file with mode 000.
4001 Lustre now respects this flag to disable mode checks when
4002 truncating a file owned by the user
4005 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
4008 Description: Revoking locks from clients that went dead or catatonic might take
4010 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
4011 cancellation of such locks instant on servers without waiting for
4012 any reply from clients. Clients drops these locks when cancel
4013 notification from server is received without replying.
4016 Frequency : liblustre-only, when liblustre client dies or becomes busy
4018 Description: Doing ls on Linux clients can take a long time with active
4020 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
4021 granting such locks to it in the first place if possible. Locks
4022 are taken by proxy on the OST during the read or write and
4023 dropped immediately afterward. Add connect flags handling, do
4024 not grant locks to liblustre clients for glimpse ASTs.
4026 Severity : enhancement
4028 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
4029 Details : Existing read-ahead algorithm is tuned for the case of streamlined
4030 sequential reads and behaves badly with applications doing random
4031 reads. Improve it by reading ahead at least read region, and
4032 avoiding excessive large RPC for small reads.
4034 Severity : enhancement
4036 Description: Creating more than 1000 files for a single job may cause a load
4037 imbalance on the OSTs if there are also a large number of OSTs.
4038 Details : qos_prep_create() uses an OST index reseed value that is an
4039 even multiple of the number of available OSTs so that if the
4040 reseed happens in the middle of the object allocation it will
4041 still utilize the OSTs as uniformly as possible.
4046 Description: OST or MDS may oops in ping_evictor_main()
4047 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
4048 but doesn't restart at beginning of obd_exports_timed list
4051 Severity : enhancement
4053 Description: improve by-nid export eviction on the MDS and OST
4054 Details : allow multiple exports with the same NID to be evicted at one
4055 time without re-searching the exports list.
4058 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
4060 Description: MDS may oops in groups_free()
4061 Details : in rare race conditions a newly allocated group_info struct is
4062 freed again, and this can be NULL. The 2.4 compatibility code
4063 for groups_free() checked for a NULL pointer, but 2.6 did not.
4066 Frequency : common for liblustre clients doing little filesystem IO
4067 Bugzilla : 9352, 7313
4068 Description: server may evict liblustre clients accessing contended locks
4069 Details : if a client is granted a lock or receives a completion AST
4070 with a blocking AST already set it would not reply to the AST
4071 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
4074 Frequency : lfs setstripe, only systems with more than 160 OSTs
4076 Description: unable to set striping with a starting offset beyond OST 160
4077 Details : llapi_create_file() incorrectly limited the starting stripe
4078 index to the maximum single-file stripe count.
4081 Frequency : LDAP users only
4083 Description: lconf did not handle in-kernel recovery with LDAP properly
4084 Details : lconf/LustreDB get_refs() is searching the wrong namespace
4086 Severity : enhancement
4088 Description: bind OST threads to NUMA nodes to improve performance
4089 Details : all OST threads are uniformly bound to CPUs on a single NUMA
4090 node and do their allocations there to localize memory access
4092 Severity : enhancement
4094 Description: llmount can determine client NID directly from Myrinet (GM)
4095 Details : the client NID code from gmnalnid was moved directly into
4096 llmount, removing the need to use this or specifying the
4097 client NID explicitly when mounting GM clients with zeroconf
4100 Frequency : if client is started with down MDS
4102 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
4103 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
4104 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
4105 ptlrpc_import_delayed_req() because we didn't check if the
4106 request was interrupted, and we also didn't break out of the
4107 event loop if there was no timeout
4112 Description: data loss during non-page-aligned writes to a single file from
4113 both multiple nodes and multiple threads on one node at same time
4114 Details : updates to KMS and lsm weren't protected by common lock. Resulting
4115 inconsistency led to false short-reads, that were cached and later
4116 used by ->prepare_write() to fill in partially written page,
4117 leading to data loss.
4120 Frequency : always, if lconf --abort_recovery used
4122 Description: lconf --abort_recovery fails with 'Operation not supported'
4123 Details : lconf was attempting to abort recovery on the MDT device and not
4126 Severity : enhancement
4128 Description: remove cleanup logs
4129 Details : replace lconf-generated cleanup logs with lustre internal
4130 cleanup routines. Eliminates the need for client-cleanup and
4133 Severity : enhancement
4135 Description: add support for EAs (user and system) on lustre filesystems
4136 Details : it is now possible to store extended attributes in the Lustre
4137 client filesystem, and with the user_xattr mount option it
4138 is possible to allow users to store EAs on their files also
4140 Severity : enhancement
4142 Description: Add possibility (config option) to show minimal available OST free
4144 Details : When compiled with --enable-mindf configure option, statfs(2)
4145 (and so, df) will return least minimal free space available from
4146 all OSTs as amount of free space on FS, instead of summary of
4147 free spaces of all OSTs.
4149 Severity : enhancement
4151 Description: do not expand extent locks acquired on OST-side
4152 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
4153 by server: they are not cached anyway.
4156 Frequency : when mmap is used/binaries executed from Lustre
4158 Description: Unmmap pages before throwing them away from read cache.
4159 Details : llap_shrink cache now attempts to unmap pages before discarding
4160 them (if unmapping failed - do not discard). SLES9 kernel has
4161 extra checks that trigger if this unmapping is not done first.
4166 Description: lconf didn't resolve symlinks before checking to see whether a
4167 given mountpoint was already in use
4170 Frequency : when migrating failover services
4171 Bugzilla : 6395, 9514
4172 Description: When migrating a subset of services from a node (e.g. failback
4173 from a failover service node) the remaining services would
4174 time out and evict clients.
4175 Details : lconf --force (implied by --failover) sets the global obd_timeout
4176 to 5 seconds in order to quickly disconnect, but this caused
4177 other RPCs to time out too quickly. Do not change the global
4178 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
4180 Severity : enhancement
4181 Frequency : if MDS is started with down OST
4182 Bugzilla : 9439,5706
4183 Description: Allow startup/shutdown of an MDS without depending on the
4184 availability of the OSTs.
4185 Details : Asynchronously call mds_lov_synchronize during MDS startup.
4186 Add appropriate locking and lov-osc refcounts for safe
4187 cleaning. Add osc abort_inflight calls in case the
4188 synchronize never started.
4191 Frequency : occasional (Cray XT3 only)
4193 Description: root not authorized to access files in CRAY_PORTALS environment
4194 Details : The client process capabilities were not honoured on the MDS in
4195 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4196 been accessed by an authorized user then root was able to access
4197 the file on the local client also. The root user capabilities
4198 are now allowed on the MDS, as this environment has secure UID.
4201 Frequency : occasional
4203 Description: ldiskfs "too long searching" message happens too often
4204 Details : A debugging message (otherwise harmless) prints too often on
4205 the OST console. This has been reduced to only happen when
4206 there are fragmentation problems on the filesystem.
4211 Description: Division by zero in statfs when all OSCs are inactive
4212 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4213 lov_statfs divides by value returned by lov_get_stripecnt().
4217 Bugzilla : 9489, 3273
4218 Description: First write from each client to each OST was only 4kB in size,
4219 to initialize client writeback cache, which caused sub-optimal
4220 RPCs and poor layout on disk for the first writen file.
4221 Details : Clients now request an initial cache grant at (re)connect time
4222 and so that they can start streaming writes to the cache right
4223 away and always do full-sized RPCs if there is enough data.
4224 If the OST is rebooted the client also re-establishes its grant
4225 so that client cached writes will be honoured under the grant.
4230 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4231 Details : Now I/O RPCs go to different portal number and (presumably) fast
4232 lock requests (and glimses) and other RPCs get their own service
4233 threads pool that should be able to service those RPCs
4236 Severity : enhancement
4238 Description: Ability to exchange lustre version between client and servers and
4239 issue warnings at client side if client is too old. Also for
4240 liblustre clients there is ability to refuse connection of too old
4242 Details : New 'version' field is added to connect data structure that is
4243 filled with version info. That info is later checked by server and
4247 Frequency : rare, liblustre only.
4248 Bugzilla : 9296, 9581
4249 Description: Two simultaneous writes from liblustre at offset within same page
4250 might proceed at the same time overwriting eachother with stale
4252 Details : I/O lock withing llu_file_prwv was released too early, before data
4253 actually was hitting the wire. Extended lock-holding time until
4254 server acknowledges receiving data.
4257 Frequency : extremely rare. Never observed in practice.
4259 Description: avoid generating lustre_handle cookie of 0.
4260 Details : class_handle_hash() generates handle cookies by incrementing
4261 global counter, and can hit 0 occasionaly (this is unlikely, but
4262 not impossible, because initial value of cookie counter is
4263 selected randonly). Value of 0 is used as a sentinel meaning
4264 "unassigned handle" --- avoid it. Also coalesce two critical
4265 sections in this function into one.
4267 Severity : enhancement
4269 Description: allow liblustre clients to delegate truncate locking to OST
4270 Details : To avoid overhead of locking, liblustre client instructs OST to
4271 take extent lock in ost_punch() on client's behalf. New connection
4272 flag is added to handle backward compatibility.
4274 Severity : enhancement
4275 Bugzilla : 4928, 7341, 9758
4276 Description: allow number of OST service threads to be specified
4277 Details : a module parameter allows the number of OST service threads
4278 to be specified via "options ost ost_num_threads={N}" in the
4279 OSS's /etc/modules.conf or /etc/modprobe.conf.
4283 Bugzilla : 6146, 9635, 9895
4284 Description: servers crash with bad pointer in target_handle_connect()
4285 Details : In rare cases when a client is reconnecting it was possible that
4286 the connection request was the last reference for that export.
4287 We would temporarily drop the export reference and get a new
4288 one, but this may have been the last reference and the export
4289 was just destroyed. Get new reference before dropping old one.
4291 Severity : enhancement
4292 Frequency : if client is started with failover MDS
4294 Description: Allow multiple MDS hostnames in the mount command
4295 Details : Try to read the configuration from all specified MDS
4296 hostnames during a client mount in case the "primary"
4299 Severity : enhancement
4301 Description: Stop sending data to evicted clients as soon as possible.
4302 Details : Check if the client we are about to send or are sending data to
4303 was evicted already. (Check is done every second of waiting,
4304 for which l_wait_event interface was extended to allow checking
4305 of exit condition at specified intervals).
4308 Frequency : rare, normally only when NFS exporting is done from client
4310 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4312 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4315 Frequency : always, if srand() is called before liblustre initialization
4317 Description: Liblustre uses system PRNG disturbing its usage by user application
4318 Details : Introduce internal to lustre fast and high-quality PRNG for
4319 lustre usage and make liblustre and some other places in generic
4320 lustre code to use it.
4322 Severity : enhancement
4323 Bugzilla : 9477, 9557, 9870
4324 Description: Verify that the MDS configuration logs are updated when xml is
4325 Details : Check if the .xml configuration logs are newer than the config
4326 logs stored on the MDS and report an error if this is the case.
4327 Request --write-conf, or allow starting with --old_conf.
4329 Severity : enhancement
4331 Description: Handle symlinks in the path when checking if Lustre is mounted.
4332 Details : Resolve intermediate symlinks when checking if a client has
4333 mounted a filesystem to avoid duplicate client mounts.
4338 Description: lconf can hit an error exception but still return success.
4339 Details : The lconf command catches the Command error exception at the top
4340 level script context and will exit with the associated exit
4341 status, but doesn't ensure that this exit status is non-zero.
4346 Description: failure of ptlrpc thread startup can cause oops
4347 Details : Starting a ptlrpc service thread can fail if there are a large
4348 number of threads or the server memory is very fragmented.
4349 Handle this without oopsing.
4352 Frequency : always, only if liblustre and non-default acceptor port was used
4354 Description: liblustre cannot connect to servers with non-default acceptor port
4355 Details : tcpnal_set_default_params() was not called and was therefore
4356 ignoring the environment varaible TCPNAL_PORT, as well as other
4357 TCPNAL_ environment variables
4362 Description: two objects could be created on the same OST for a single file
4363 Details : If an OST is down, in some cases it was possible to create two
4364 objects on a single OST for a single file. No problems other
4365 than potential performance impact and spurious error messages.
4369 Bugzilla : 5681, 9562
4370 Description: Client may oops in ll_unhash_aliases
4371 Details : Client dcache may become inconsistent in race condition.
4372 In some cases "getcwd" can fail if the current directory is
4378 Description: Inode refcounting problems in NFS export code
4379 Details : link_raw functions used to call d_instantiate without obtaining
4380 extra inode reference first.
4384 Bugzilla : 9942, 9903
4385 Description: Referencing freed requests leading to crash, memleaks with NFS.
4386 Details : We used to require that call to ll_revalidate_it was always
4387 followed by ll_lookup_it. Also with revalidate_special() it is
4388 possible to call ll_revalidate_it() twice for the same dentry
4389 even if first occurence returned success. This fix changes semantic
4390 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4391 reference on a request referred from the intent.
4392 ll_intent_release() then releases such a request.
4395 Frequency : rare, normally benchmark loads only
4397 Description: unlinked inodes were kept in memory on the client
4398 Details : If a client is repeatedly creating and unlinking files it
4399 can accumulate a lot of stale inodes in the inode slab cache.
4400 If there is no other client load running this can cause the
4401 client node to run out of memory. Instead flush old inodes
4402 from client cache that have the same inode number as a new inode.
4405 Frequency : SLES9 2.6.5 kernel and long filenames only
4406 Bugzilla : 9969, 10379
4407 Description: utime reports stale NFS file handle
4408 Details : SLES9 uses out-of-dentry names in some cases, which confused
4409 the lustre dentry revalidation. Change it to always use the
4413 Frequency : rare, unless heavy write-truncate concurrency is continuous
4414 Bugzilla : 4180, 6984, 7171, 9963, 9331
4415 Description: OST becomes very slow and/or deadlocked during object unlink
4416 Details : filter_destroy() was holding onto the parent directory lock
4417 while truncating+unlinking objects. For very large objects this
4418 may block other threads for a long time and slow overall OST
4419 responsiveness. It may also be possible to get a lock ordering
4420 deadlock in this case, or run out of journal credits because of
4421 the combined truncate+unlink. Solution is to do object truncate
4422 first in one transaction without parent lock, and then do the
4423 final unlink in a new transaction with the parent lock. This
4424 reduces the lock hold time dramatically.
4427 Frequency : rare, 2.4 kernels only
4429 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4430 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4431 before unmounting their filesystem in order to avoid deadlock.
4432 The kernel_locked() function in 2.4 kernels only checks whether
4433 the kernel lock is held, not whether it is this process that is
4434 holding it as 2.6 kernels do.
4439 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4440 Details : The client ptlrpc code may be trying to reconnect to a down
4441 server before a previous connection attempt has timed out.
4442 Increase the reconnect interval to be longer than the connection
4443 timeout interval to avoid sending duplicate connections to
4447 Frequency : echo_client brw_test command
4449 Description: fix echo_client to work with OST preallocated code
4450 Details : OST preallocation code (5137) didn't take echo_client IO path
4451 into account: echo_client calls filter methods outside of any
4452 OST thread and, hence, there is no per-thread preallocated
4453 pages and buffers to use. Solution: hijack pga pages for IO. As
4454 a byproduct, this avoids unnecessary data copying.
4458 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4459 Description: Client can oops in mdc_commit_close() after open replay
4460 Details : It was possible for the MDS to return an open request with no
4461 transaction number in mds_finish_transno() if the client was
4462 evicted, but without actually returning an error. Clients
4463 would later try to replay that open and may trip an assertion
4464 Simplify the client close codepath, and always return an error
4465 from the MDS in case the open is not successful.
4468 Frequency : rare, 2.6 OSTs only
4470 Description: OST may deadlock under high load on fragmented files
4471 Details : If there was a heavy load and highly-fragmented OST filesystems
4472 it was possible to have all the OST threads deadlock waiting on
4473 allocation of biovecs, because the biovecs were not released
4474 until the entire RPC IO was completed. Instead, release biovecs
4475 as soon as they are complete to ensure forward IO progress.
4477 Severity : enhancement
4479 Description: Support for specifying external journal device at mount
4480 Details : If an OST or MDS device is formatted with an external journal
4481 device, this device major/minor is stored in the ext3 superblock
4482 and may not be valid for failover. Allow detecting and
4483 specifying the external journal at mount time.
4488 Description: Mounting an MDS with pending unlinked files may cause oops
4489 Details : target_finish_recovery() calls mds_postrecov() which returned
4490 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4491 considers this an error and immediately begins cleaning up the
4492 lov, just after starting the mds_lov process
4494 Severity : enhancement
4496 Description: Implement 'lfs df' to report actual free space on per-OST basis
4497 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4498 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4499 usage of objects; '-h' to report in human readable format.
4501 ------------------------------------------------------------------------------
4503 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4510 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4511 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4512 at startup during orphan unlinking. Instead, skip buddy update
4513 before setup, as it will be regenerated after recovery is complete.
4518 Description: If an OST is inactive, its locks might reference stale inodes.
4519 Details : lov_change_cbdata() must iterate over all namespaces, even if
4520 they are inactive to clear inode references from the lock.
4522 Severity : enhancement
4523 Frequency : occasional, if non-standard max_dirty_mb used
4525 Description: Client will block write RPCs if not enough grant
4526 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4527 then the client will block writes while waiting for another RPC
4528 to complete instead of consuming its dirty limit. With change
4529 we get improved performance when max_dirty_mb is small.
4531 Severity : enhancement
4532 Bugzilla : 3389, 6253
4533 Description: Add support for supplementary groups on the MDS.
4534 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4535 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4536 lookups for user supplementary groups into a cache.
4540 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4541 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4542 for liblustre O_EXCL is enforced by MDS.
4546 Description: readdir never returns NULL in liblustre.
4547 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4548 dentry in struct dirent.
4552 Frequency : liblustre only, depends on application IO pattern
4553 Description: liblustre clients evicted if not contacting servers
4554 Details : Don't put liblustre clients into the ping_evictor list, so
4555 they will not be evicted by the pinger ever.
4557 Severity : enhancement
4559 Description: Add ability to evict clients by NID from MDS.
4560 Details : By echoing "nid:$NID" string into
4561 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4562 $NID would be instantly evicted from this MDS and from all active
4563 OSTs connected to it.
4567 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4568 Details : lookup_it_finish() used to query file size from OSTs that was not
4573 Description: service threads change working directory to that of init
4574 Details : Starting lustre service threads may pin the working directory
4575 of the parent thread, making that filesystem busy. Threads
4576 now change to the working directory of init to avoid this.
4580 Frequency : during shutdown only
4581 Description: shutdown with a failed MDS or OST can cause unmount to hang
4582 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4587 Frequency : 2.6 only
4588 Description: chmod/chown may include an extra supplementary group
4589 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4590 supplementary group and if none is specified this is used.
4593 Bugzilla : 5479 (6816)
4595 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4596 Details : If lookup is in progress on a file that is unlinked we might try
4597 to revalidate the inode and fail in revalidate after lookup is
4598 complete and ll_file_open() enqueues the open again but
4599 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4602 Frequency : always, if lconf --abort_recovery used
4604 Description: lconf --abort_recovery fails with 'Operation not supported'
4605 Details : lconf was attempting to abort recovery on the MDT device and not
4608 ------------------------------------------------------------------------------
4610 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4615 Frequency : rare (only unsupported configurations with a node running as an
4617 Bugzilla : 6514, 5137
4618 Description: Mounting a Lustre file system on a node running as an OST could
4620 Details : OSTs now preallocates memory needed to write out data at
4621 startup, instead of when needed, to avoid having to
4622 allocate memory in possibly low memory situations.
4623 Specifically, if the file system is mounted on on OST,
4624 memory pressure could force it to try to write out data,
4625 which it needed to allocate memory to do. Due to the low
4626 memory, it would be unable to do so and the node would
4627 become unresponsive.
4629 Severity : enhancement
4631 Description: Addition of lconf --service command line option
4632 Details : lconf now accepts a '--service <arg>' option, which is
4633 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4635 Severity : enhancement
4637 Description: Failover mode is now the default for OSTs.
4638 Details : By default, OSTs will now run in failover mode. To return to
4639 the old behaviour, add '--failout' to the lmc line for OSTs.
4641 Severity : enhancement
4643 Description: Health checks are now provided for MDS and OSTs
4644 Details : Additional detailed health check information on MSD and OSTs
4645 is now provided through the procfs health_check value.
4648 Frequency : occasional, depends on IO load
4650 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4651 after numerous create/delete cycles
4652 Details : The ext3 inode allocation policy would not allocate new inodes
4653 very well on the OSTs because there are no new directories
4654 being created. Instead we look for groups with free space if
4655 the parent directories are nearly full.
4660 Description: Network or server problems during mount may cause partially
4661 mounted clients instead of returning an error.
4662 Details : The config llog parsing code may overwrite the error return
4663 code during mount error handling, returning success instead
4669 Description: MDS can fail to allocate large reply buffers
4670 Details : After long uptimes the MDS can fail to allocate large reply
4671 buffers (e.g. zconf client mount config records) due to memory
4672 fragmentation or consumption by the buffer cache. Preallocate
4673 some large reply buffers so that these replies can be sent even
4674 under memory pressure.
4678 Frequency : rare (liblustre)
4679 Description: fsx running with liblustre complained that using truncate() to
4680 extend the file doesn't work. This patch corrects that issue.
4681 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4682 ATTR_SIZE and lsm use in llu_setattr_raw.
4686 Frequency : rare, only 2.6 kernels
4687 Description: Unusual file access patterns on the MDS may result in inode
4688 data being lost in very rare circumstances.
4689 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4690 code in the 2.6 kernel caused the inode and/or EA data not to
4691 be read from disk, causing single-file corruption.
4695 Frequency : rare, only 2.6 filesystems using extents
4696 Description: Heavy concurrent write and delete load may cause data corruption.
4697 Details : It was possible under high-load situations to have an extent
4698 metadata block in the block device cache from a just-unlinked
4699 file overwrite a newly-allocated data block. We now unmap any
4700 metadata buffers that alias just-allocated data blocks.
4704 Frequency : filesystems with default stripe_count larger than 77
4705 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4706 Details : lconf specifies an inode size of 4096 bytes when the default
4707 stripe_count is larger than 77. This conflicts with the default
4708 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4709 this case to avoid pinning too much memory for large EAs.
4711 ------------------------------------------------------------------------------
4713 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4718 Frequency : rare (extremely heavy IO load with hundreds of clients)
4720 Description: Client is evicted, gets IO error writing to file
4721 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4722 caused clients to be evicted for AST timeouts. The fixes in
4723 bug 5192 mean we no longer need to have such short AST timeouts
4724 so ldlm_timeout has been increased.
4727 Frequency : occasional during --force or --failover shutdown under load
4728 Bugzilla : 5949, 4834
4729 Description: Server oops/LBUG if stopped with --force or --failover under load
4730 Details : a collection of import/export refcount and cleanup ordering
4731 issues fixed for safer force cleanup
4734 Frequency : only filesystems larger than 120 OSTs
4735 Bugzilla : 5990, 6223
4736 Description: lfs getstripe would oops on a very large filesystem
4737 Details : lov_getconfig used kfree on vmalloc'd memory
4740 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4741 Bugzilla : 6242, 6243
4742 Description: reading from files that had been truncated to a non-zero size
4743 but never opened returned no data
4744 Details : ll_file_read() reads zeros from no-object files to EOF
4749 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4750 Details : There's little harm in aborting MDS/OSS recovery and letting it
4751 try again, so I removed the LASSERT and return an error instead.
4753 Severity : enhancement
4755 Description: New debugging infrastructure for tracking down data corruption
4756 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4757 (b) cover more of the client-side code path, and (c) try to narrow
4758 down where problems occurred
4762 Bugzilla : 3819, 4364, 4397, 6313
4763 Description: Racing close and eviction MDS could cause assertion in mds_close
4764 Details : It was possible to get multiple mfd references during close and
4765 client eviction, leading to one thread referencing a freed mfd.
4767 Severity: : enhancement
4768 Bugzilla : 3262, 6359
4769 Description: Attempts to reconnect to servers are now more aggressive.
4770 Details : This builds on the enhanced upcall-less recovery that was added
4771 in 1.4.2. When trying to reconnect to servers, clients will
4772 now try each server in the failover group every 10 seconds. By
4773 default, clients would previously try one server every 25 seconds.
4778 Description: After recovery, certain operations trigger a failed
4779 assertion on a client.
4780 Details : Failing over an mds, using lconf -d --failover, while a
4781 client was doing a readdir() call would cause the client to
4782 LBUG after recovery completed and the readdir() was resent.
4784 Severity : enhancement
4786 Description: Default groups are now added by lconf
4787 Details : You can now run lconf --group <servicename> without having to
4788 manually add groups with lmc.
4791 Frequency : occasional
4793 Description: Nodes with an elan id of 0 trigger a failed assertion
4796 Frequency : always when accessing e.g. tty/console device nodes
4798 Description: tty and some other devices nodes cannot be used on lustre
4799 Details : file's private_data field is used by device data and lustre
4800 values in there got lost. New field was added to struct file to
4801 store fs-specific private data.
4804 Frequency : when exporting Lustre via NFS
4806 Description: NFSD failed occasionally when looking up a path component
4807 Details : NFSD is looking up ".." which was broken in ext3 directories
4808 that had grown large enough to become hashed.
4811 Frequency : Clusters with multiple interfaces not on the same subnet
4813 Description: Nodes will repeatedly try to reconnect to an interface which it
4814 cannot reach and report an error to the log.
4815 Details : Extra peer list entries will be created by lconf with some peers
4816 unreachable. lconf now validates the peer before adding it.
4819 Frequency : Only if a default stripe is set on the filesystem root.
4821 Description: Setting a default stripe on the filesystem root prevented the
4822 filesystem from being remounted.
4823 Details : The client was sending extra request flags in the root getattr
4824 request and did not allocate a reply buffer for the dir EA.
4827 Frequency : occasional, higher if lots of files are accessed by one client
4828 Bugzilla : 6159, 6097
4829 Description: Client trips assertion regarding lsm mismatch/magic
4830 Details : While revalidating inodes the VFS looks up inodes with ifind()
4831 and in rare cases can find an inode that is being freed.
4832 The ll_test_inode() code will free the lsm during ifind()
4833 when it finds an existing inode and then the VFS later attaches
4834 this free lsm to a new inode.
4838 Bugzilla : 6422, 7030
4839 Description: MDS deadlock between mkdir and client eviction
4840 Details : Creating a new file via mkdir or mknod (starting a transaction
4841 and getting the ns lock) can deadlock with client eviction
4842 (gets ns lock and trying to finish a synchronous transaction).
4845 Frequency : occasional
4846 Description: While starting a server, the fsfilt_ext3 module could not be
4848 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4849 kernels. Previously, lconf would still use the ext3 name
4850 when trying to load modules. Now, it will correctly use
4851 ext3 on 2.4 and ldiskfs on 2.6.
4853 Severity : enhancement
4854 Description: The default stripe count has been changed to 1
4855 Details : The interpretation of the default stripe count (0, to lfs
4856 or lmc) has been changed to mean striping across a single
4857 OST, rather than all available. For general usage we have
4858 found a stripe count of 1 or 2 works best.
4860 Severity : enhancement
4861 Description: Add support for compiling against Cray portals.
4862 Details : Conditional compiling for some areas that are different
4866 Frequency : occasional
4867 Bugzilla : 6409, 6834
4868 Description: Creating files with an explicit stripe count may lead to
4869 a failed assertion on the MDS
4870 Details : If some OSTs are full or unavailable, creating files may
4871 trigger a failed assertion on the MDS. Now, Lustre will
4872 try to use other servers or return an error to the
4876 Frequency : occasional
4878 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4879 caused lock timeout during readahead (since 1.4.2).
4880 Details : Processes doing readahead might match a lock that hasn't been
4881 granted yet if there are overlapping and conflicting lock
4882 requests. The readahead process waits on ungranted lock
4883 (original lock is CBPENDING), while OST waits for that process
4884 to cancel CBPENDING read lock and eventually evicts client.
4886 Severity : enhancement
4888 Description: Initial enabling of flock support for clients
4889 Details : Implements fcntl advisory locking and file status functions.
4890 This feature is provided as an optional mount flag (default
4891 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4892 locking are implemented yet, and those that are are not guaranteed
4893 to be completely correct in production environments.
4894 mount -t lustre -o [flock|noflock] ...
4897 Frequency : occasional
4899 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4900 assertion in the ext3 JBD (journaling) layer.
4901 Details : The b_committed_data struct is protected by the big kernel lock
4902 in 2.4 kernels, serializing journal_commit_transaction() and
4903 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4904 access to this struct. In 2.6 kernels there is finer grained
4905 locking to improve SMP performance of the JBD layer.
4909 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4914 Description: merge and cleanup kernel patches.
4917 -----------------------------------------------------------------------------
4919 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4921 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4922 versions of Lustre. Please update all servers and clients to
4923 version 1.4.2 or later at the same time. You must also run
4924 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4925 to update the configuration logs.
4927 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4928 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4929 - swab llog records of type '0' so we get proper header size/idx (5861)
4930 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4931 - fix rename of one directory over another leaking an inode (5953)
4932 - avoid SetPageDirty on 2.6 (5981)
4933 - don't re-add just-being-destroyed locks to the waiting list (5653)
4934 - when creating new directories, inherit the parent's custom
4935 striping settings if present parent (3048)
4936 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4937 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4938 - don't hold client locks on temporary worklist from l_lru (5666)
4939 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4940 - automatically evict dead clients (5921)
4941 - Update file size properly in create+truncate+fstat case (6196)
4942 - Do not unhash mountpoint dentries, do not allow removal of
4944 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4945 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4946 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4947 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4948 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4949 - Correctly handle reads of files with no objects (6243)
4950 - lctl recover will also mark a device active if deactivate used (5933)
4952 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4953 - allow --write-conf on an MDS with different nettype than client (5619)
4954 - don't write config llogs to MDS for mounts not from that MDS (5617)
4955 - lconf should create multiple TCP connections from a client (5201)
4956 - init scripts are now turned off by default; run chkconfig --on
4957 lustre and chkconfig --on lustrefs to use them
4958 - upcalls are no longer needed for clients to recover to failover
4960 - add --abort-recovery option to lconf to abort recovery on device
4962 - add support for an arbitrary number of OSTs (3026)
4963 - Quota support protocol changes.
4964 - forward compatibility changes to wire structs (6007)
4965 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4966 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4967 - improved "lustre status" script
4968 - initialize blocksize for non-regular files (6062)
4969 - added --disable-server and --disable-client configure options (5782)
4970 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4971 - Vanilla 2.4.29 support
4972 - increase maximum number of obd devices to 520 (6242)
4973 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4974 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4975 SLES 9 2.6.5 kernels
4976 - limit stripes per file to 160 (the maximum EA size) (6093)
4978 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4981 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4982 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4983 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4984 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4985 - small ext3 extents cleanups and fixes (5733)
4986 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4987 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4988 - avoid needless client->OST connect, fix handle mismatch (5317)
4989 - fix DLM error path that led to out-of-sync client, long delays (5779)
4990 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4991 - fix several locking issues related to i_size (5492,5624,5654,5672)
4992 - don't move pending lock onto export if it is already evicted (5683)
4993 - fix kernel oops when creating .foo in unlinked directory (5548)
4994 - fix deadlock in obdfilter statistics vs. object create (5811)
4995 - use time_{before,after} to avoid timer jiffies wrap (5882)
4996 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4997 - Do not leak request if server was not able to process it (5154)
4998 - If mds_open unable to find parent dir, make that negative lookup(5154)
4999 - don't create new directories with extent-mapping (5909, 5936)
5001 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
5002 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
5003 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
5004 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
5005 - support for RHEL4 kernel on i686 (5773)
5006 - provide error messages when incompatible logs are encountered (5898)
5008 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
5009 * version 1.4.0.10 (1.4.1 release candidate 1)
5011 - don't keep a lock reference when lock is not granted (4238)
5012 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5013 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
5014 - drop import inflight refcount on signal_completed_replay error (5255)
5015 - unlock page after async write error during send (3677)
5016 - handle missing objects in filter_preprw_read properly (5265)
5017 - no transno return for symlink open, don't save no-trasno open (3440)
5018 - don't try to complete elan receive that already failed (4012)
5019 - free RPC server reply state on error (5406)
5020 - clean up thread from ptlrpc_start_thread() on error (5160)
5021 - readahead could read extra page into cache that wasn't ejected (5388)
5022 - prevent races in class_attach/setup/cleanup/detach (5260)
5023 - don't dereference de->d_inode after l_dput of de (5458)
5024 - use "int" for stripe value returned from lock_to_stripe (5544)
5025 - mballoc allocation and error-checking fixes in 2.6 (5504)
5026 - block device patches to fix I/O request sizes in 2.6 (5482)
5027 - look up hostnames for IB nals (5602)
5028 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
5029 - don't start multiple acceptors for the same port (5277)
5030 - fix incorrect LASSERT in mds_getattr_name (5635)
5031 - export a proc file for general "ping" checking (5628)
5032 - fix "lfs check" to not block when the MDS is down (5628)
5034 - service request history (4965)
5035 - put {ll,lov,osc}_async_page structs in a single slab (4699)
5036 - create an "evict_client" /proc entry on OSTs, like the MDS has
5037 - fix mount usage message, return errors per mount(8) (5168)
5038 - change grep [] to grep "[]" in tests so they work in more UMLs
5039 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
5040 - remove ancient LOV_MAGIC_V0 EA support (5047)
5041 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
5042 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
5043 - allow readahead allocations to fail when low on memory (5383)
5044 - mmap locking landed again, after considerable improvement (2828)
5045 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
5047 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
5050 - send OST transaction number in read/write reply to free req (4966)
5051 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5052 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5053 - the watchdog thread now runs as interruptible (5246)
5054 - flock/lockf fixes (but it's still disabled, pending 5135)
5055 - don't use EXT3 constants in llite code (5094)
5056 - memory shortage at startup could cause assertion (5176)
5058 - reorganization of lov code
5059 - single portals codebase
5061 - add extents/mballoc support (5025)
5062 - direct I/O reads in the obdfilter (4048)
5063 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
5065 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5068 - send OST transaction number in read/write reply to free req (4966)
5069 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
5070 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
5071 - the watchdog thread now runs as interruptible (5246)
5072 - handle missing objects in filter_preprw_read properly (5265)
5073 - unsafe list practices (rarely) led to infinite eviction loop (4908)
5074 - drop import inflight refcount on signal_completed_replay error (5255)
5075 - unlock page after async write error during send (3677)
5076 - return original error code on reconstructed replies (3761)
5077 - no transno return for symlink open, don't save no-trasno open (3440)
5079 - add pid to ldlm debugging output (4922)
5080 - bump the watchdog timeouts -- we can't handle 30sec yet
5081 - extra debugging for orphan dentry/inode bug (5259)
5083 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
5086 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
5087 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
5088 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
5089 - fix NULL dereference in /proc/sys/portals/routes (4827)
5090 - allow failed mdc_close() operations to be interrupted (4561)
5091 - stop precreate on OST before MDS would time out on it (4778)
5092 - don't send partial-page writes before EOF from client (4410)
5093 - discard client grant for sub-page writes on large-page clients (4520)
5094 - don't free dentries not owned by NFS code, check generation (4806)
5095 - fix lsm leak if mds_create_objects() fails (4801)
5096 - limit debug_daemon file size, always print CERROR messages (4789)
5097 - use transno after validating reply (3892)
5098 - process timed out requests if import state changes (3754)
5099 - update mtime on OST during writes, return in glimpse (4829)
5100 - add mkfsoptions to LDAP (4679)
5101 - use ->max_readahead method instead of zapping global ra (5039)
5102 - don't interrupt __l_wait_event() during strace
5104 - add software watchdogs to catch hung threads quickly (4941)
5105 - make lustrefs init script start after nfs is mounted
5106 - fix CWARN/ERROR duplication (4930)
5107 - return async write errors to application if possible (2248)
5108 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
5109 - print NAL number in %x format (4645)
5110 - update barely-supported suse-2.4.21-171 series (4842)
5111 - support for sles 9 %post scripts
5112 - support for building 2.6 kernel-source packages
5113 - support for sles km_* packages
5115 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
5118 - ignore -ENOENT errors in osc_destroy (3639)
5119 - notify osc create thread that OSC is being cleaned up (4600)
5120 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
5121 - reconstruct ost_handle() like mds_handle() (4657)
5122 - create a new thread to do import eviction to avoid deadlock (3969)
5123 - let lconf resolve symlinked-to devices (4629)
5124 - don't unlink "objects" from directory with default EA (4554)
5125 - hold socknal file ref over connect in case target is down (4394)
5126 - allow more than 32000 subdirectories in a single directory (3244)
5127 - fix blocks count for O_DIRECT writes (3751)
5128 - OST returns ENOSPC from object create when no space left (4539)
5129 - don't send truncate RPC if file size isn't changing (4410)
5130 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
5131 - bind to privileged port in socknal and tcpnal (3689)
5133 - rate limit CERROR/CWARN console message to avoid overload (4519)
5134 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
5135 - basic mmap support (3918)
5136 - kernel patch series update from b1_4 (4711)
5138 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
5141 - avoid crash during MDS cleanup with OST shut down (2775)
5142 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
5143 - don't use bad inodes on the MDS (3744)
5144 - dynamic object preallocation to improve recovery speed (4236)
5145 - don't hold spinlock over lock dumping or change debug flags (4401)
5146 - don't zero obd_dev when it is force cleaned (3651)
5147 - print grants to console if they go negative (4431)
5148 - "lctl deactivate" will stop automatic recovery attempts (3406)
5149 - look for existing locks in ldlm_handle_enqueue() (3764)
5150 - don't resolve lock handle twice in recovery avoiding race (4401)
5151 - revalidate should check working dir is a directory (4134)
5153 - don't always mark "slow" obdfilter messages as errors (4418)
5155 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
5158 - don't close LustreDB during write_conf until it is done (3860)
5159 - fix typo in lconf for_each_profile (3821)
5160 - allow dumping logs from multiple threads at one time (3820)
5161 - don't allow multiple threads in OSC recovery (3812)
5162 - fix debug_size parameters (3864)
5163 - fix mds_postrecov to initialize import for llog ctxt (3121)
5164 - replace config semaphore with spinlock (3306)
5165 - be sure to send a reply for a CANCEL rpc with bad export (3863)
5166 - don't allow enqueue to complete on a destroyed export (3822)
5167 - down write_lock before checking llog header bitmap (3825)
5168 - recover from lock replay timeout (3764)
5169 - up llog sem before sending rpc (3652)
5170 - reduce ns lock hold times when setting kms (3267)
5171 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
5172 - fix NULL deref and obd_dev leak on setup error (3312)
5173 - replace some LBUG about llog ops with error handling (3841)
5174 - don't match INVALID dentries from d_lookup and spin (3784)
5175 - hold dcache_lock while marking dentries INVALID and hashing (4255)
5176 - fix invalid assertion in ptlrpc_set_wait (3880)
5178 - add libwrap support for the TCP acceptor (3996)
5179 - add /proc/sys/portals/routes for non-root route listing (3994)
5180 - allow setting MDS UUID in .xml (2580)
5181 - print the stack of a process that LBUGs (4228)
5183 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
5186 - don't cleanup request in ll_file_open() on failed MDS open (3430)
5187 - make sure to unset replay flag from failed open requests (3440)
5188 - if default stripe count is 0, use OST count for inode size (3636)
5189 - update parent mtime/ctime on client for create/unlink (2611)
5190 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
5191 - free recovery state on server during a forced cleanup (3571)
5192 - unregister_reply for resent reqs (3063)
5193 - loop back devices mounting and status check on 2.6 (3563)
5194 - fix resource-creation race that can provoke i_size == 0 (3513)
5195 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5196 - more debugging for page-accounting assertion (3746)
5197 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5198 - avoid deadlock after precreation failure (3758)
5199 - fix race and lock order deadlock in orphan handling (3450, 3750)
5200 - add validity checks when grabbing inodes from l_ast_data (3599)
5202 - add /proc/.../recovery_status to obdfilter (3428)
5203 - lightweight CDEBUG infrastructure, debug daemon (3668)
5204 - change default OSC RPC parameters to be better on small clusters
5205 - turn off OST read cache for files smaller than 32MB
5206 - install man pages and include them in rpms (3100)
5207 - add new init script for (un)mounting lustre filesystems (2593)
5208 - run chkconfig in %post for init scripts (3701)
5209 - drop scimac NAL (unmaintained)
5211 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5214 - clean kiobufs before and after use (3485)
5215 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5216 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5217 - change init script to fail more clearly if not run as root (1528)
5218 - allow clients to reconnect during replay (1742)
5219 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5220 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5221 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5222 - fix cond_resched (3554)
5223 - only evict unfinished clients after recovery (3515)
5224 - allow bulk resend, prevent data loss (3570)
5225 - dynamic ptlrpc request buffer allocation (2102)
5226 - don't allow unlinking open directory if it isn't empty (2904)
5227 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5228 - remove extraneous obd dereference causing LASSERT failure (3334)
5229 - don't use get_cycles() when creating temp. files on the mds (3156)
5230 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5231 - handle EEXIST for set-stripe, set proper directory name (3336)
5233 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5234 - fix ksocknal_fmb_callback() error messages (2918)
5236 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5239 - don't copy lvb into (possibly NULL) reply on error (2983)
5240 - don't deref dentry after dput, don't free lvb on error (2922)
5241 - use the kms to determine writeback rpc length (2947)
5242 - increment oti_logcookies when osc is inactive (2948)
5243 - update client's i_blocks count via lvb messages (2543)
5244 - handle intent open/close of special files properly (1557)
5245 - mount MDS with errors=remount-ro, like obdfilter (2009)
5246 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5247 - don't use cancelling-locks' kms values (2947)
5248 - use highest lock extent for kms, not last one (2925)
5249 - don't dereference ERR_PTR() dentry in error handling path (3107)
5250 - fix thread race in portals_debug_dumplog() (3122)
5251 - create lprocfs device entries at setup instead of at attach (1519)
5252 - common AST error handler, don't evict client on completion race (3145)
5253 - zero nameidata in detach_mnt in 2.6 (3118)
5254 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5255 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5256 - initialize RPC timeout timer earlier for 2.6 (3219)
5257 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5258 - print nal/nid for unknown nid (3258)
5259 - additional checks for oscc recovery before doing precreate (3284)
5260 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5261 - don't crash in mdc_close for bad permissions on open (3285)
5262 - zero i_rdev for non-device files (3147)
5263 - clear page->private before handing to FS, better assertion (3119)
5264 - tune the read pipeline (3236)
5265 - fix incorrect decref of invalidated dentry (2350)
5266 - provide read-ahead stats and refine rpc in flight stats (3328)
5267 - don't hold journal transaction open across create RPC (3313)
5268 - update atime on MDS at close time (3265)
5269 - close LDAP connection when recovering to avoid server load (3315)
5270 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5271 - don't leak open file on MDS after open resend (3325)
5272 - serialize filter_precreate and filter_destroy_precreated (3329)
5273 - loop device shouldn't call sync_dev() for nul device (3092)
5274 - clear page cache after eviction (2766)
5275 - resynchronize MDS->OST in background (2824)
5276 - refuse to mount the same filesystem twice on same mountpoint (3394)
5277 - allow llmount to create routes for mounting behind routers (3320)
5278 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5279 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5280 - fix rare problem with rename on htree directories (3417)
5282 - allow default OST striping configuration per directory (1414)
5283 - fix compilation for qswnal for 2.6 kernels (3125)
5284 - increase maximum number of MDS request buffers for large systems
5285 - change liblustreapi to be useful for external progs like lfsck (3098)
5286 - increase local configuration timeout for slow disks (3353)
5287 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5289 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5292 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5293 - don't overwrite extent policy data in reply if lock was blocked (2901)
5294 - drop filter export grants atomically with removal from device (2663)
5295 - del obd_self_export from work_list in class_disconnect_exports (2908)
5296 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5297 - swab reply message in mdc_close, other PPC fixes (2464)
5298 - fix destroying of named logs (2325)
5299 - overwrite old logs when running lconf --write_conf (2264)
5300 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5301 - fix race in target_handle_connect (2898)
5302 - mds_reint_create() should take same inode create lock (2926)
5303 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5304 - don't close files for self_export to avoid uninitialized obd (2936)
5305 - allow MDS with the same name as client node (2939)
5306 - hold dentry reference for closed log files for unlink (2325)
5307 - reserve space for all logs during transactions (2059)
5308 - don't evict page beyond end of stripe extent (2925)
5309 - don't oops on a deleted current working directory (2399)
5310 - handle hard links to targets without a parent properly (2517)
5311 - don't dereference NULL lock when racing during eviction (2867)
5312 - don't grow lock extents when lots of conflicting locks (2919)
5314 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5317 - account for cache space usage on clients to avoid data loss (974)
5318 - lfsck support in lustre kernel code (2349)
5319 - reduce journal credits needed for BRW writes (2370)
5320 - orphan handling to avoid losing space on client/server crashes
5321 - ptlrpcd can be blocked, stopping ALL progress (2477)
5322 - use lock value blocks to assist in proper KMS, faster stat (1021)
5323 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5324 - recovery for initial connections (2355)
5325 - fixes for mds_cleanup_orphans (1934)
5326 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5327 - block all file creations until orphan recovery completes (1901)
5328 - client remove rq_connection from request struct (2423)
5329 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5330 - recovery timer race (2670)
5331 - mdc_close recovey bug (2532)
5332 - ptlrpc cleanup bug (2710)
5333 - mds timeout on local locks (2588)
5334 - namespace lock held during RPCs (2431)
5335 - handle interrupted sync write properly (2503)
5336 - don't try to handle a message that hasn't been replied to (2699)
5337 - client assert failure during cleanup after abort recovery (2701)
5338 - leak mdc device after failed mount (2712)
5339 - ptlrpc_check_set allows timedout requests to complete (2714)
5340 - wait for inflight reqs when ptlrpcd finishes (2710)
5341 - make sure unregistered services are removed from the srv_list
5342 - reset bulk XID's when resending them (caught by 1138 test)
5343 - unregister_bulk after timeout
5344 - fix lconf error (2694)
5345 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5346 - readahead locks pages, leaves pending causing memory pressure (2673)
5347 - increase OST request buffers to 4096 on large machines (2729)
5348 - fix up permission of existing directories in simple_mkdir (2661)
5349 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5350 - don't assign transno to errored transactions (2742)
5351 - don't delete objects on OST if given a bogus objid from MDS (2751)
5352 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5353 - if rq_no_resend, then timeout request after recovery (2432)
5354 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5355 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5356 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5357 - MDS recovery shouldn't delete objects that it has given out (2730)
5358 - if enqueue arrives after completion, don't clobber LVB (2819)
5359 - don't unlock pages twice when trigger_group_io returns error (2814)
5360 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5361 - don't write pages to disk if there was an error (1450)
5362 - don't ping imports that have recovery disabled (2676)
5363 - take buffered bytes into account when balancing socknal conn (2817)
5364 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5365 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5366 - remove little-endian swabbing of llog records (1987)
5367 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5368 - retry reposting request buffers if they fail (1191)
5369 - grow extent at grant time to avoid granting a revoked lock (2809)
5370 - lock revoke doesn't evict page if covered by a second lock (2765)
5371 - disable VM readahead to avoid reading outside lock extents (2805)
5373 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5374 - updated kernel patches for hp-2.4.20 kernel (2681)
5376 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5379 - fix truncated write corruption (2366)
5380 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5382 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5383 - ensure len(uuid) < 37 in lmc (1171)
5384 - fix ia64 OOPS in llog_test (2255)
5385 - zero end of page at obdfilter for partial page writes (2648)
5386 - don't leave stale dentries around after renames (bug 2428)
5387 - fix timeouts when evicting a client with a single lock held (2642)
5388 - set deadline for the initial HELLO message to drain (2634)
5389 - print out dotted-quad IP addresses in the socknal (2302)
5391 - additional debugging for MDS client eviction problem (2443)
5392 - fix mkfsoptions support for osts (2603, 2604)
5394 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5397 - add series for the vanilla 2.6.0 kernel
5398 - add series for the vanilla 2.4.24 kernel
5399 - add series for a cray x86/64 UL kernel drop
5400 - fix xattr patches for the vanilla 2.4.19 series
5402 - generate true UUIDs in lmc (1171)
5403 - have portals stack dumping break in UML (2466)
5404 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5405 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5406 - wake callback waiting threads on client eviction (2460)
5407 - Add --ptldebug and --subsystem to lmc (1719)
5408 - update assertion to allow safe interrupt allocation
5409 - set rq_no_resend for cancel requests (2432)
5410 - recalculate ptlrpcd timeout after resend (2494)
5411 - call vfs_rmdir when removing pending directories (2368)
5412 - fix renaming a file to itself (2429)
5413 - lmc creates a default one-stripe lov (2454)
5414 - expand procfs space to handle large clusters (2326)
5415 - increase UML stack to avoid overflow
5416 - update lconf's list of debug and subsystem masks
5417 - fix lfs find --obd (2510)
5418 - /proc tunable for disabling filter read caching (2591)
5419 - stop rpm packages from altering slapd.conf (2301)
5420 - disable nagle in the socknal under 0conf (2578)
5421 - choose mds inode size based on stripe count (2572)
5422 - fix kernel-source rpm problems (2516)
5424 - add --disable-doc to avoid pdf generation (2421)
5425 - update documentation, tests, type-os, comments
5426 - avoid format warnings on ia64
5427 - remove the TOE NAL
5428 - tiny code cleanups by removing unused fields
5430 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5433 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5434 - remove the most blatant lies from BUILDING file (2371)
5435 - change default debug level to reasonable production setting
5436 - reduce client side cache size to reduce cache flush time
5437 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5438 - make TCP zerocopy and pinger support enabled by default (2476)
5439 - sync writes completed after process exits caused crashes (2319)
5440 - maintain correct mount count on the MDS (2356)
5441 - backout 1557, because 2316 wasn't really fixed
5442 - better file I/O statistics gathering in /proc
5443 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5444 - another kernel patch to fix zero-copy TCP function export
5445 - don't take duplicate lock when processing re-sent getattr (2420)
5446 - lctl uses obd_self_export instead of creating new conn (2353)
5447 - MDS/OST recovery case which requires object creation asserted (2425)
5448 - move lfs from /usr/sbin to /usr/bin in packages
5449 - fix race between mds_client_add and mds_client_free (2417)
5450 - use kmalloc instead of slabs in portals (2430)
5451 - don't create duplicate records when a failover MDS is present (2442)
5452 - remove unnecessary mount age check (2332)
5453 - don't remove directory inodes from locks prematurely (2451)
5454 - don't break if MDS service name is the same as hostname (2103)
5455 - fix races in client write RPC generation when cache full (2482)
5457 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5460 - remove now-unused request->rq_obd (278)
5461 - if an allocation fails, print out how much memory we've used (1933)
5462 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5463 - add the "configurable stack size" patch to most series files (1256)
5464 - ability to write large log records, for 100+ OST configs (2306)
5465 - fix NULL deref when filter_prep fails (2314)
5466 - fix operator precedence error in filter_sync
5467 - dynamic allocation of socknal TX descriptors (2315)
5468 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5469 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5470 - ensure that CWARN messages in Portals always get to the syslog
5471 - __init/__exit are not for prototype decls (ldlm_init/exit)
5472 - x86-64 compile warning fixes
5473 - fix gateway LMC keyword conflict (2318)
5474 - fix MDS lock inversions in getattr/reint paths (1844)
5475 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5476 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5477 - DLM race condition prevented some lock evictions (2328)
5478 - ENOMEM detection and retry on socknal sends (2230)
5479 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5480 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5481 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5482 - better error messages when a client is rejected during recovery (1505)
5483 - avoid cancelling locks which were never granted, after failure (2330)
5484 - fix i_sem/journal inversion in mds_client_add (2333)
5485 - fix truncate/getattr lock cycle deadlock (2334)
5486 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5487 - fix two rare exit paths which could leak an l_lock() ref (2321)
5488 - fencepost error in MDS/OST orphan recovery (2226)
5489 - make log record alignment 8 bytes (1988)
5490 - lstripe now fails when requested offset > ost_count (2237)
5491 - ensure that all kernel series have a complete list.h (1607)
5492 - fix crashes in special-file operations (2316)
5493 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5495 - allow configurable automake binary, for testing new versions
5496 - small update to the lfs documentation
5498 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5500 * fix negative export reference count in fsfilt_sync (2312)
5502 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5503 * release candidate 0.9.1
5505 - orphans are moved into the PENDING directory for possible recovery
5506 - replayed opens now open by fid for orphan/rename safety (1042)
5507 - last close of an orphan inode generates a transno (683)
5508 - chdir() and mount() now pin the directory entry (1020)
5509 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5510 - discard very old requests without processing them (1502)
5511 - remove some common, well-understood CERRORs (1505)
5512 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5513 - clear "grant" flags in OST replies until OST grant code lands (1644)
5514 - fix read performance by not clobbering i_blksize on client (1598)
5515 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5516 - make LRU size a /proc tunable, clears locks when reduced (707)
5517 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5518 - server locks take references on exports now (1558)
5519 - build fixes for 2.4.20-rh trees (1663)
5520 - return an error from lov_create if all OSCs are inactive (1751)
5521 - fix import levels when a reconnect happens without a timeout (1597)
5522 - exit early from mds_open if we get a lookup error (1749)
5523 - partial page read at EOF wouldn't wait for disk before sending (1642)
5524 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5525 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5526 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5527 - write/truncate lock inversion (1639)
5528 - Don't auto-load obdclass, portals modules during cleanup (1495)
5529 - fix timestamps from jumping to "now" (1763)
5530 - extra journal assertions (1648)
5531 - add an extra multiunlink test (1771)
5532 - fix read_record/write_record API (1776)
5533 - fix leak of offset_extent, possible incorrect i_size later (1772)
5534 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5535 - Debugging for the kqswnal_get_idle_tx problems (1820)
5536 - Allow recovery to be attempted multiple times (1536)
5537 - Write out MDS last_rcvd file after it is first created (1600)
5538 - Fix tx_descriptor leak in failed transmit situations (1827)
5539 - ext3 journaling fixes for assertion failure after IO error (1871)
5540 - class_export_put() on freed export after completion AST error (1896)
5541 - Fix revalidate looping in VFS (1322)
5542 - Don't access a freed export during MDS_REINT timeout (1521)
5543 - Add open-unlink recovery support on the MDS (1673,1764)
5544 - Return an error if no MDS data was read from last_rcvd (1946)
5545 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5546 - Don't setup a disk device that doesn't match exported UUID (317)
5547 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5548 - avoid committing NULL handle in force close
5549 - local.sh is now a one-stripe LOV configuration
5550 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5551 - fix ext3 htree duplicate directory entry corruption (1516)
5552 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5553 - update to new LOV EA format (2097)
5554 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5556 2003-06-15 Phil Schwan <phil@clusterfs.com>
5559 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5560 - per-import/export recovery handling (958, 931, 959)
5561 - multiple last-rcvd slots, for serving multiple FSes (949)
5562 - connections are again shared between multiple imp/exports (963, 964)
5563 - "umount -f" would hang if any requests needed to be sent (393, 978)
5564 - avoid pinning large req buffer by copying for queued messages (989)
5565 - add "uuid" to "lctl device" command to help upcalls (991)
5566 - "open" RPCs with transnos would confuse recovery counters (1037)
5567 - do proper endian conversion of all wire messages (288, 340, 891)
5568 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5569 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5570 - fix ptlrpc_connection leak in target_handle_connect (1174)
5571 - fix import refcounting bug in OST and MDS cleanup (1134)
5572 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5573 - fix very unlikely obd_types race condition (501)
5574 - remove osc_open hack for echo_client (1187)
5575 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5576 - a failure in read_inode2 leads to deadlock (1139)
5577 - cancel ack-locks as soon as transaction is committed (1072)
5578 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5579 - make sure to commitrw after any preprw to avoid deadlock (1162)
5580 - failing to execute a file in a lustre FS would lock inode (1203)
5581 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5582 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5583 - obd_uuid2tgt would walk off the end of the list (1255)
5584 - on IA64 the osc would give portals incorrect bulk size (1258)
5585 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5586 - fix lock inversion caused by new llite matching code (1282)
5587 - limit the number of dirty pages on a client to 10MB (1286)
5588 - timed out locks were not being corrected cancelled (1289)
5589 - fix O_DIRECT above 4GB on IA-32 (1292)
5590 * major user-visible changes
5591 - fail out/fail over policy now controlled by the upcall (993)
5593 - add OBD_PING to check server availability and failure (954)
5594 - lustre messages are now sent in sending host order (288, 340, 891)
5595 - add eadatalen to MDS getattr reply (340)
5596 - OST read replies may contain second buffer, with per-page status (593)
5598 2003-03-11 Phil Schwan <phil@clusterfs.com>
5601 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5602 - failed open()s could cause deadlock; fixed (867, 869)
5603 - stop cancelling OST locks when files are closed (481)
5604 - overlapping XID spaces caused network corruption (851, 853)
5605 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5606 - setattr_raw added, to do single-RPC, server-side setattrs
5607 - lmc/lconf syntax change for OST UUIDs
5608 - fix crashy race condition between ptlrpc_free_req and osc_close
5609 - don't use request in mdc_enqueue if we hit a timeout (889)
5610 - don't set the inode i_size for regular files from the MDS (896)
5611 - handle out of order completion AST (842)
5612 - don't LBUG if a lock request times out after receiving AST (913)
5613 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5614 - if a bad lock AST arrives, send an error instead of dropping entirely
5615 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5616 - fix leak in bulk IO when only partially completed (899, 900, 926)
5617 - fix O_DIRECT for ia64 (55)
5618 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5619 - C-z after timeout could hang a process forever; fixed (977)
5621 - client-side I/O cache (678, 924, 929, 941, 970)
5623 - READPAGE and SETATTRs which don't take server-side locks get
5626 2003-02-11 Phil Schwan <phil@clusterfs.com>
5629 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5630 - Fix fsfilt_extN_readpage() to read a full page of directory
5631 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5632 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5633 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5634 - Most intent operations were removed, in favour of a new RPC mode
5635 that does a single RPC to the server and bypasses most of the VFS
5636 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5637 - Aggressively cancel local locks on DLM servers
5638 - mds_reint_unlink sends EA to the client if it's the last nlink.
5639 client uses that EA to unlink OST objects.
5640 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5641 - recursive symlinks were fixed (439)
5642 - fixed NULL deref in DEBUG_REQ
5643 - filter_update_lastobjid no longer calls sync, which annoyed extN
5644 - fixed multi-client small-writes to a single file problem (445)
5645 - fixed mtime updates during file writes (607)
5646 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5647 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5648 - fixed Linux OST size reporting problem (444, 656)
5649 - OST now updates object mtime with writes or setattr (607, 619)
5650 - client verifies file size before zeroing page past EOF (445)
5651 - OST now writes last allocated objid to disk with allocation (108)
5652 - LOV on echo now works (409)
5654 - mds_reint_unlink sends a new buffer, with the EA included. this
5655 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5656 set if a regular file was being unlinked, and it was the last link
5657 - use PtlGet from the target for bulk writes (315)
5658 - OST now updates object mtime with writes or setattr (607, 619)
5659 - LDLM now has a grant-time callback to revalidate locked items, if
5661 - Many MDS operations were reorganized to combat race conditions
5663 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5664 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5666 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5669 - Fully reactivate OST imports after reconnection (512, others)
5670 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5671 - More graceful error handling for truncating on dead OST (515)
5672 - Don't error out unless we're actually accessing dead stripes (474)
5673 - Fix garbage sizes when stripes are missing (410)
5674 - LRU counters were broken, causing constant lock purge (433, 432)
5675 - garbage on read from stripes with failed OSTs (441)
5676 - mark OSCs as active before reconnecting during recovery (438)
5677 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5678 - lfind did not preserve OST order in output (443)
5679 - symlinks cause hung clients, incorrect data (439)
5680 - stop dereferencing request after dropping refcount (457)
5681 - don't LASSERT(spin_is_locked) on non-SMP (455)
5682 - fixes for many rename() bugs
5683 - fstat didn't correctly synchronize attributes (399)
5684 - server must handle lock cancellation during blocking AST prep (487)
5685 - bulk descriptors were free()d too soon (511)
5686 - fix paths in lconf, which would load incorrect modules (451, 507)
5687 - fix confusing lconf 'host not found' error message (386)
5688 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5689 - fix race condition in mdc_blocking_ast() for inode access (526)
5690 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5691 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5692 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5693 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5694 - fix lstripe in several ways (kernel side) (527)
5695 - fix request leak in ldlm_cli_enqueue (262)
5696 - incorrect OSC was marked inactive after OST failure
5697 - call mds_fs_cleanup before unmounting filesystem (524)
5698 - fix races between taking ns_lock and ldlm_lock_change_resource
5699 - fix races updating LOV export open file list
5700 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5701 - fix recovery NULL deref in ldlm_cli_cancel_unused
5702 - fix some DLM races by using new hash table for lock handles (419)
5703 - permit the client to specify desired inodes, at replay
5704 - duplicate requests when we queue them for replay reintegration
5705 - fix last_rcvd offset calculation
5706 - sync after each recovered transaction, so we always make progress
5707 - never, not always, ERESTART requests without transnos
5708 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5710 - skip replay if the MDS says that the client is already connected
5711 - don't check for a recovery-enabled export to match lctl's UUID
5712 - don't INC_USE_COUNT for phantom exports
5713 - don't crash when cleaning up phantom exports (567)
5714 - don't double-finish or set replay data for errored mdc_open requests
5715 - abort requests when they time out, so we don't get old replies
5716 - send/receive replies for AST messages again
5717 - if the client says that it doesn't have the lock, cancel it on the
5719 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5720 mark it as destroyed, it will all work out when decref is called
5721 - fix module use counts (22, 581)
5723 - ASTs now expect a reply (server cancels lock on error reply)
5725 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5728 - fix many simultaneous client startup (392)
5729 - fix dentry->d_it clobbering
5730 - credentials weren't being shipped for readdir/getattr operations
5731 - remove invalid assertions triggered during some concurrent MD
5733 - proper Lustre versions added (336, 389)
5734 - fix memory leak for create error case (398)
5735 - fix LOV locking bug that would get cli/srv out of sync
5736 - fix echo client over LOV (409)
5737 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5738 - fix double-O_EXCL intent crash (424)
5739 - avoid sending multiple lock CANCELs (352)
5741 - MDS can do multi-client recovery (modulo bugs in new code)
5743 - many updates, edits, cleanups
5745 2002-11-18 Phil Schwan <phil@clusterfs.com>
5748 - fix null d_it dereference (346)
5749 - fix full OST/dbench hang (333)
5750 - fix permission problem with file removal (286)
5751 - fix removal of OSCs from LOV when they fail
5752 - fix NULL deref during bulk timeout (214)
5753 - fix problems related to multiple filesystems on one MDS (241)
5754 - fixed serious subtle metadata locking bugs
5755 - free locks on clients when inodes are removed due to memory
5757 - fix inode pointer in lock data (285)
5758 - partial support for multiple MDS on a single host (241)
5759 - data locks weren't cancelled at clear_inode time (290, 311)
5760 - intent locks could lead to unbounded lock growth (205)
5761 - added a maximum lock count, an LRU list, and a flusher
5762 - fix multiple rename (365)
5763 - properly abstracted the echo client
5764 - OSC locked 1 byte too many; fixed
5765 - rewrote brw callback code:
5766 - fixed recovery bugs related to LOVs (306)
5767 - fixed too-many-pages-in-one-write crash (191)
5768 - fixed (again) crash in sync_io_timeout (214)
5769 - probably fixed callback-related race (385)
5771 - Add capability to MDS protocol
5772 - LDLM cancellations and callbacks on different portals
5774 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5777 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5778 - defer open object destruction to close time (601981, 138)
5779 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5780 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5781 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5782 - don't call obd_setattr in ll_file_release for destroyed objects
5783 * protocol change to lustre_msg: move |version| and add |flags|
5784 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5785 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5786 * added replay of create, unlink, link and rename operations during
5787 MDS failover; recovery should be much more robust now
5788 * remove failed OSCs from LOVs (only lov_create uses this so far)
5789 * the lustre-HOWTO was brought (more) up to date (582544)
5791 2002-10-23 Phil Schwan <phil@clusterfs.com>
5794 - in-use dentries weren't being reused properly (617851)
5795 - prevent multiple LDLM setup (599178)
5796 - fix LOV size calculations for truncate (617853)
5797 - fix client handling of MDS intent errors (POSIX)
5798 - fix permission bug in lovstripe.c test (624321)
5799 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5800 - truncate past end of file could corrupt data
5801 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5802 - a race in recovery could return ETIMEDOUT to apps (623947)
5803 - building outside the source directory was fixed
5804 * the lustre-HOWTO was brought (more) up to date (582544)
5805 * major progress was made on recovery functionality
5807 2002-10-10 Phil Schwan <phil@clusterfs.com>
5810 - recovery deadlock fix
5811 - rm -rf causes LBUG fix (617817)
5812 - file open by multiple tasks fix (618962)
5813 - directory permissions bugs (602707 and 620007)
5814 - journal_stop fixed with locking (611313)
5815 - O_APPEND failures resolved (618273, perhaps 614459)
5816 - lconf PATH fix (619770)
5817 - IA64 build fix (621450)
5818 - RPC buffer sizes scale with amount of memory
5820 2002-10-01 Phil Schwan <phil@clusterfs.com>
5823 - locks would be cancelled without throwing away data pages,
5824 resulting in inconsistent data (605627)
5825 - inode attributes were not always being refreshed (605627, 612449)
5826 - lconf now continues to cleanup after lctl reports an error
5827 - MDS now enforces user permissions (602707)
5828 - lprocfs cleanup fixed, but not yet enabled (614157)
5829 - fixed infinite server hang, should a client not respond to an AST
5830 - avoid going into recovery if user calls readlink() with a buffer
5831 that's too small (613941)
5832 - AST RPCs no longer require replies (614867) -- this may be changed
5833 - don't crash server if client sends an IOV that's too big (611336)
5834 - fixed lock conversion deadlock (611892)
5835 - fixed the following of symlinks (614622)
5836 * recovery: the server can remove locks from a client that dies, other
5837 clients can make progress
5838 * more extN patch fixes
5839 * compile-time configurable ptlrpc buffer allocations
5841 - collaborative read cache document
5842 - Lustre Lite Performance CDR document-in-progress
5844 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5847 - fix typo in patch-2.4.18
5849 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5852 - clear ptlrpc request each time in handle_incoming_request()
5853 - unlink of files now destroys the object on the OST
5855 2002-09-19 Peter Braam <braam@clusterfs.com>
5857 * add hard link support
5858 * change obdfile creation method
5859 * kernel patch changed
5861 2002-09-19 Peter Braam <braam@clusterfs.com>
5864 - stack overflow bug in extN fixed
5866 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5868 * documentation updates
5869 - add man pages for config tools
5870 - update tests/README to describe testing with new config tools
5871 - finish metadata API descriptions
5872 * bug fixes and cleanups
5873 - statfs workaround for 16TB limit
5874 - LOV stripe allocation improved, can stripe on subset of OSTs
5875 - LOV file size/IO offset was wrong for files > 4GB in size
5876 - object EA data was being dropped, caused files to be unreadable
5877 - memory overflow with non-LOV OST caused memory corruption
5878 - fixed regression tests to work with new config tools, obdfilter
5879 - fixed bug when directory size became larger than 1 block
5880 - fixed bug (for single client case) when PWD was deleted
5881 - invalidate local directory pages when doing intent-based ops
5882 - avoid LDLM oops when lock callback contained bad data
5884 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5886 * documentation updates
5887 * bug fixes and cleanups
5888 - configuration tools
5891 - 64-bit compile warnings
5892 - 64-bit internal statfs data
5894 * test_brw on persistent OST devices
5896 * lprocfs (disabled)
5898 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5900 * documentation updates
5901 * bug fixes and cleanups
5902 * configuration tools
5904 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5907 * many small fixes to 0_5_4
5908 * io/network handling
5909 * thinkos in MDS operations
5911 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5914 * crucial basic fixes to 0.5.3
5915 * IOR, Iozone work over Elan
5918 2002-08-07 Phil Schwan <phil@clusterfs.com>
5919 * version 0_5_3, our first alpha
5920 * we use the new Portals iovs
5921 * documentation updates
5922 * bug fixes and cleanups
5923 * small changes in the DLM wire protocol
5925 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5926 * version 0_5_1 with some initial stability,
5927 * locking on MD and file I/O.
5928 * documentation updates
5929 * several bug fixes since 0.5.0
5930 * small changes in wire protocol
5932 2002-07-18 Phil Schwan <phil@clusterfs.com>
5934 * delivered as Lustre Light Alpha
5935 * fixed a crash after handling invalid MDS requests
5936 * fixed directory pages for architectures with non-4k pages sizes
5938 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5939 * release version v0_4_4
5940 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5942 * Fixes a number of interruption problems with OST operations.
5943 * Update documentation for portals header changes
5944 * Move all wire protocol structs/defines to lustre_idl.h
5945 * Fixes symlink length bug.
5946 * Add tcpdump to repository.
5948 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5949 * release version v0_4_3
5950 * Fixes statfs for inodes on extN.
5951 * Fixes bug in runtests which would delete /etc/hosts.
5952 * Use 64-bit object IDs wherever possible (not into VFS though)
5953 Remove ost_get_info, which is unused by lustre, and out of date.
5955 2002-07-03 Peter Braam <braam@clusterfs.com>
5956 * release version v0_4_2 Fixes a lookup error (type not passed)
5957 * move forward to head of Portals
5958 * move forward to latest Lustre kernel
5960 2002-06-25 Peter Braam <braam@clusterfs.com>
5961 * release version v0_4_1. Hopefully stable on single node use.