1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.33 (SLES 10),
5 2.6.18-128.1.1.el5 (RHEL 5),
6 2.6.22.14 vanilla (kernel.org).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.11-sun1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
14 * File join has been disabled in this release, refer to Bugzilla 16929.
17 Frequency : rare, on llog test 6
19 Descriptoin: don't allow connect to already connected import
20 Details : allowing connect to already connected import is hide connecting problem.
25 Descriptoin: allow kill process which wait statahead result
26 Details : for some reasons 'ls' can stick in waiting result from statahead,
27 in this case need way for kill this process.
29 Severity : enhancement
31 Description: Add state history info file, enhance import info file
32 Details : Track import connection state changes in a new osc/mdc proc file;
33 add overview-type data to the osc/mdc import proc file.
35 Severity : enhancement
37 Description: MDS create should not wait for statfs RPC while holding DLM lock.
39 Severity : enhancement
41 Description: Update to RHEL5U3 kernel-2.6.18-128.1.1.el5.
46 Descriptoin: OST grant too much space to client even there are not enough space.
47 Details : Client will shrink its grant cache to OST if there are no write
48 activity over 6 mins (GRANT_SHRINK_INTERVAL), and OST will retrieve
49 this grant cache if there are already not enough avaible space
50 (left_space < total_clients * 32M).
53 Frequency : start MDS on uncleanly shutdowned MDS device
55 Descriptoin: ll_sync thread stay in waiting mds<>ost recovery finished
56 Details : stay in waiting mds<>ost recovery finished produce random bugs
57 due race between two ll_sync thread for one lov target. send
58 ACTIVATE event only if connect realy finished and import have
62 Frequency : rare, connect and disconnect target at same time
64 Descriptoin: ASSERTION(atomic_read(&imp->imp_inflight) == 0
65 Details : don't call obd_disconnect under lov_lock. this long time
66 operation and can block ptlrpcd which answer to connect request.
71 Descriptoin: don't lose wakeup for imp_recovery_waitq
72 Details : recover_import_no_retry or invalidate_import and import_close can
73 both sleep on imp_recovery_waitq, but we was send only one wakeup
77 Frequency : always with long access acl
79 Descriptoin: mds can't pack reply with long acl.
80 Details : mds don't control size of acl but they limited by reint/getattr
83 Severity : enhancement
85 Description: Update to SLES10 kernel-2.6.16.60-0.33.
87 Severity : enhancement
89 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
92 Frequency : start MDS on uncleanly shutdowned MDS device
94 Descriptoin: aborting recovery hang on MDS
95 Details : don't throttle destroy RPCs for the MDT.
98 Frequency : on remount
100 Description: external journal device not working after the remount
101 Details : clear dev_rdonly flag for external journal devices in
107 Description: shutdown vs evict race
108 Details : client_disconnect_export vs connect request race.
109 if client will evicted at this time - we start invalidate
110 thread without referece to import and import can be freed
114 Frequency : rare, need acl's on inode.
116 Description: client can't handle ost additional correctly
117 Details : if ost was added after client connected to mds client can have
118 hit lnet_try_match_md ... to big messages to wide striped files.
119 in this case need teach client to handle config events about add
120 lov target and update client max ea size at that event.
122 Severity : enhancement
124 Description: Changelogs
125 Details : Changelogs are a lightweight mechanism to track filesystem
126 metadata and namespace changes. The changelog is recorded
127 permanently on the MDTs, and is periodically "consumed" / purged
128 when records are no longer needed.
130 Severity : enhancement
132 Description: compact fld format with extents
133 Details : Store range of seq rather than every seq in FLD. Seq
134 controller update FLD rather than clients. In Case of CMD, mdt0
135 has FLD, all other metadata server act as non persistent proxy
136 for FLD queries and cache fld entries in fld cache.
141 Description: don't skip ost target if they assigned to file
142 Details : Drop slow OSCs if we can, but not for requested start idx.
143 This means "if OSC is slow and it is not the requested
144 start OST, then it can be skipped, otherwise skip it only
145 if it is inactive/recovering/out-of-space.
149 Description: more cleanup in mds_lov
150 Details : not send LOV EA under replay, we can't know about they size at this
151 time. Don't allow client connect to mds before any ost connected,
152 for avoid problems with LOV EA size and returning EIO to client.
154 Severity : enhancement
156 Description: Interoperability at server side (Disk interoperability)
158 Severity : enhancement
160 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
162 Severity : enhancement
164 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
166 Severity : enhancement
168 Description: New client IO stack (CLIO).
170 Severity : enhancement
172 Description: Commit on sharing. Eliminate inter-client dependencies between
173 uncommitted transactions by doing transaction commits.
174 Thereby clients may recovery independently.
177 Frequency : Create a symlink file with a very long name
179 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
180 Details : If there is no extra space in the request for early cancels,
181 ldlm_req_handles_avail() returns 0 instead of a negative value.
183 Severity : enhancement
185 Description: Add /proc entry for import status
186 Details : The mdc, osc, and mgc import directories now have
187 an import directory that contains useful import data for debugging
190 Severity : enhancement
192 Description: Re-disable certain /proc logging
193 Details : Enable and disable client's offset_stats, extents_stats and
194 extents_stats_per_process stats logging on the fly.
197 Frequency : Only on FC kernels 2.6.22+
199 Description: oops in statahead
200 Details : Do not drop reference count for the dentry from VFS when lookup,
201 VFS will do that by itself.
203 Severity : enhancement
205 Description: Generic /proc file permissions
206 Details : Set /Proc file permissions in a more generic way to enable non-
207 root users operate on some /proc files.
211 Description: Hitting mdc_commit_close() ASSERTION
212 Details : Properly handle request reference release in
213 ll_release_openhandle().
217 Description: quota recovery deadlock during mds failover
218 Details : This patch includes att18982, att18236, att18237 in bz14840.
220 1. fix osts hang when mds does failover with quotaon
221 2. prevent watchdog storm when osts threads wait for the
226 Frequency : only patchless client
227 Description: add workaround for race between add/remove dentry from hash
229 Severity : enhancement
231 Description: Allow OST glimpses to return PW locks
235 Description: LBUG when llog conf file is full
236 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
240 Description: Prevent import from entering FULL state when server in recovery
244 Description: service mount cannot take device name with ":"
245 Details : Only when device name contains ":/" will mount treat it as
251 Description: replace ptlrpcd with the statahead thread to interpret the async
252 statahead RPC callback
256 Frequency : on recovery
257 Description: I/O failures after umount during fail back
258 Details : if client reconnected to restarted server we need join to recovery
259 instead of find server handler is changed and process self
260 eviction with cancel all locks.
262 Severity : enhancement
264 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
266 Severity : enhancement
268 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
270 Severity : enhancement
272 Description: Upcall on Lustre log has been dumped
273 Details : Allow for a user mode script to be called once a Lustre log has
274 been dumped. It passes the filename of the dumped log to the
275 script, the location of the script can be specified via
276 /proc/sys/lnet/debug_log_upcall.
281 Description: avoid idr_remove called for id which is not allocated.
282 Details : Move assigment s_dev for clustered nfs to end of initialization,
283 to avoid problem with error handling.
288 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
289 Details : When connection is reused this not moved from CONN_UNUSED_HASH
290 into CONN_USED_HASH and this prodice warning when put connection
291 again in unused hash.
294 Severity : enhancement
296 Description: File striping can now be set to use an arbitrary pool of OSTs.
298 Severity : enhancement
300 Description: Export bytes_read/bytes_write count on OSC/OST.
304 Description: Early reply size mismatch, MGC loses connection
305 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
306 the connect flags are properly negotiated.
311 Description: quotas are not honored with O_DIRECT
312 Details : all writes with the flag O_DIRECT will use grants which leads to
313 this problem. Now using OBD_BRW_SYNC to guard this.
317 Description: add quota statistics
318 Details : 1. sort out quota proc entries and proc code.
319 2. add quota statistics
321 Severity : enhancement
323 Description: enable quota support for HEAD.
327 Description: Properly propagate oinfo flags from lov to osc for statfs
328 Details : restore missing copy oi_flags to lov requests.
330 Severity : enhancement
332 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
333 : l_getgroups(8), lst(8), routerstat(8)
334 Details : included man pages for llobdstat(8), llstat(8),
335 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
337 Severity : enhancement
339 Description: configure's --enable-quota should check the
340 : kernel .config for CONFIG_QUOTA
341 Details : configure is terminated if --enable-quota is passed but
342 : no quota support is in kernel
346 Description: Remove portals compatibility
347 Details : Remove portals compatibility, not interoperable with releases
352 Description: Resolve device initialization race
353 Details : Prevent proc handler from accessing devices added to the
354 obd_devs array but yet be intialized.
356 Severity : enhancement
358 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
360 Severity : enhancement
362 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
367 Description: Using wrong pointer in osc_brw_prep_request
368 Details : Access to array[-1] can produce panic if kernel compiled with
369 CONFIG_PAGE_ALLOC enabled
373 Description: Client runs out of low memory
374 Details : Consider only lowmem when counting initial number of llap pages
378 Description: *optional* service tags registration
379 Details : if the "service tags" package is installed on a Lustre node
380 When the filesystem is mounted, a local-node service tag will
381 be created. See http://inventory.sun.com/ for more information
382 about the Service Tags asset management system.
386 Description: Kernel BUG tries to release flock
387 Details : Lustre does not destroy flock lock before last reference goes
388 away. So always drop flock locks when client is evicted and
389 perform unlock regardless of successfulness of speaking to MDS.
393 Description: add refcount protection for osc callbacks, avoid panic on shutdown
397 Description: sanity test 65a fails if stripecount of -1 is set
398 Details : handle -1 striping on filesystem in ll_dirstripe_verify
403 Description: ASSERTION(CheckWriteback(page,cmd)) failed
404 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
407 Severity : enhancement
409 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
413 Description: do not process already freed flock
414 Details : flock can possibly be freed by another thread before it reaches
415 to ldlm_flock_completion_ast.
419 Description: LBUG during stress test
420 Details : Need properly lock accesses the flock deadlock detection list.
424 Description: oops in page fault handler
425 Details : kernel page fault handler can return two special 'pages' in error
426 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
430 Description: timeout with invalidate import.
431 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
432 should be handled by ptlrpcd. This produce long age waiting and
433 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
435 Severity : enhancement
437 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
441 Description: enable MGS and MDT services start separately
442 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
443 a co-located MGS without starting the MGS, which is a complement
444 to 'nosvc' mount option.
448 Frequency : after recovery
449 Description: precreate to many object's after del orphan.
450 Details : del orphan st in oscc last_id == next_id and this triger growing
451 count of precreated objects. Set flag LOW to skip increase count
452 of precreated objects.
456 Frequency : rare, on clear nid stats
457 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
458 Details : when clean nid stats sometimes try destroy live entry,
459 and this produce panic in free.
463 Description: Stack overflow during MDS log replay
464 ease stack pressure by using a thread dealing llog_process.
468 Description: wait until IO finished before start new when do lock cancel.
469 Details : VM protocol want old IO finished before start new, in this case
470 need wait until PG_writeback is cleared until check dirty flag and
471 call writepages in lock cancel callback.
473 Severity : enhancement
475 Description: using special macro for print time and cleanup in includes.
479 Description: mds_mfd_close() ASSERTION(rc == 0)
480 Details : In mds_mfd_close(), we need protect inode's writecount change
481 within its orphan write semaphore to prevent possible races.
485 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
489 Frequency : rare, on shutdown ost
490 Description: don't hit live lock with umount ost.
491 Details : shrink_dcache_parent can be in long loop with destroy dentries,
492 use shrink_dcache_sb instead.
496 Description: don't panic with use echo client
497 Details : echo client pass NULL as client nid pointer and this produce null
502 Description: fix build on ppc32
503 Details : compile code with -m64 flag produce wrong object file for ppc32.
507 Description: add message levels for liblustreapi
511 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
512 Details : Move the 'good_osts' check before the 'total_bavail' check. This
513 will result in an -EAGAIN and in the exit call path we call
514 alloc_rr() which will with increasing aggressiveness attempt to
515 aquire precreated objects on the minimum number of required OSCs.
519 Description: Use old size assignment to avoid deadlock
520 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
521 the scheduling while holding a spinlock. We do not need locking
522 for size in ll_update_inode() because size is only updated from
523 the MDS for directories or files without objects, so there is no
524 other place to do the update, and concurrent access to such inodes
525 are protected by the inode lock.
529 Description: resolve "_IOWR redefined" build error on SLES10
533 Description: dump the memory debugging after all modules are unloaded to
534 suppress false negative in conf_sanity test 39
536 Severity : enhancement
538 Description: build kernel-ib packages for OFED 1.3 in our release cycle
543 Description: fix SLES kernel versioning
544 Details : the kernel version for our SLES 10 kernel did not include a "-"
545 before the "smp" at the end. while this was not a problem in
546 general, it did mean that software trying to use the kernel
547 version to try to detect a vendor specific kernel would fail.
548 this was most evident by the OFED build scripts.
552 Description: Don't update lov_desc members until making sure they are valid
553 Details : When updating lov_desc members via proc fs, need fix their
554 validities before doing the real update.
558 Description: don't put request into delay list while invalidate in flight.
559 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
560 import in flight. this produce timeout for invalidate and sometimes
561 can cause stale data.
565 Frequency : on ppc only
566 Description: not convert ost objects for directory because it's not exist.
567 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
571 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
573 Severity : enhancement
575 Description: Update to support 2.6.22.14 vanilla kernel.
579 Frequency : rare, on recovery
580 Description: read procfs can produce deadlock in some situation
581 Details : Holding lprocfs lock which send rpc can produce block for destroy
582 obd objects and this also block reconnect with -EALREADY. This
583 isn't fix all lprocfs bugs - but make it rare.
585 Severity : enhancement
587 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
590 Frequency : frequent on X2 node
592 Description: mdc_set_open_replay_data LBUG
593 Details : Set replay data for requests that are eligible for replay.
597 Description: lustre_mgs: operation 101 on unconnected MGS
598 Details : When MGC is disconnected from MGS long enough, MGS will evict the
599 MGC, and late on MGC cannot successfully connect to MGS and a lot
600 of the error messages complaining that MGS is not connected.
604 Frequency : on network error
605 Description: panic with double free request if network error
606 Details : mdc_finish_enqueue is finish request if any network error ocuring,
607 but it's true only for synchronus enqueue, for async enqueue
608 (via ptlrpcd) this incorrect and ptlrpcd want finish request
611 Severity : enhancement
613 Description: client-side metadata stat-ahead during readdir(directory readahead)
614 Details : perform client-side metadata stat-ahead when the client detects
615 readdir and sequential stat of dir entries therein
618 Frequency : on start mds
620 Description: Implement get_info(last_id) in obdfilter.
623 Frequency : occasional
625 Description: Correctly check stale fid, not start epoch if ost not support SOM
626 Details : open with flag O_CREATE need set old fid in op_fid3 because
627 op_fid2 was overwritten with new generated fid, but mds can answer
628 with one of these two fids and both is not stale. Setattr
629 incorrectly started an epoch and assume will be called
630 done_writing, but without SOM done_writing ever being called.
633 Frequency : rare, depends on device drivers and load
635 Description: MDS or OSS nodes crash due to stack overflow
636 Details : Code changes in 1.8.0 increased the stack usage of some functions.
637 In some cases, in conjunction with device drivers that use a lot
638 of stack the MDS (or possibly OSS) service threads could overflow
639 the stack. One change which was identified to consume additional
640 stack has been reworked to avoid the extra stack usage.
643 Frequency : occasional
645 Description: Do not fail import if osc_interpret_create gets -EAGAIN
646 Details : If osc_interpret_create got -EAGAIN it immediately exits and
647 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
648 oscc_has_objects and see OSC has no objests and call
649 oscc_internal_create to resend create request.
651 Severity : enhancement
653 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
655 Severity : enhancement
657 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
660 Frequency : very rare
662 Description: Fix replay if there is an un-replied request and open
663 Details : In some cases, older replay request will revert the
664 mcd->mcd_last_xid on MDS which is used to record the client's
667 Severity : enhancement
669 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
671 Severity : enhancement
673 Description: Add rhel5 support to HEAD.
675 Serverity : enhancement
677 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
682 Description: Don't allow skipping OSTs if index has been specified.
683 Details : Don't allow skipping OSTs if index has been specified, make
684 locking in internal create lots better.
688 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
689 Details : access to bitfield in structure is always rounded to long
690 and this produce problem with not atomic change any bit.
694 Description: Lustre make rpms failed.
695 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
696 builds Lustre rpms from tarball.
699 Frequency : rare on shutdown ost
701 Description: If llog cancel was not send before clean_exports phase, this can
702 produce deadlock in llog code.
703 Details : If llog thread has last reference to obd and call class_import_put
704 this produce deadlock because llog_cleanup_commit_master wait when
705 last llog_commit_thread exited, but this never success because was
706 called from llog_commit_thread.
710 Description: allow userland application know is lost one of stripes.
711 Details : fill lvb_blocks with error code on ost and return it to
712 application if error flag found.
716 Description: NULL lov_tgts causing MDS oops
717 Details : more safe checks for NULL lov_tgts for avoid oops.
719 Severity : enhancement
721 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
725 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
726 Details : If an OST is down the MDS will hang indefinitely in
727 obd_statfs_rqset() waiting for the statfs data. While for
728 MDS QOS usage of statfs, it should not stuck in waiting.
730 Severity : enhancement
732 Description: remote_acl support
733 Details : Support ACL-based permission check for remote user.
734 Support setfacl/getfacl for remote user with the utils
735 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
736 the system "{s,g}etfacl" utils.
738 Severity : enhancement
740 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
742 Severity : enhancement
744 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
748 Description: make mgs_setparam() handle fsname containing dash
749 Details : fsname containing a dash does not work with lctl conf_param
751 Severity : enhancement
753 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
755 Severity : enhancement
757 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
761 Frequency : rare, at shutdown
762 Description: access already free / zero obd_namespace.
763 Details : if client_disconnect_export was called without force flag set,
764 and exist connect request in flight, this can produce access to
765 NULL pointer (or already free pointer) when connect_interpret
766 store ocd flags in obd_namespace.
770 Frequency : only at startup
771 Description: not alloc memory with spinlock held.
772 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
773 if any spinlock held.
775 Severity : enhancement
777 Description: make lustre randomly fail allocating memory
778 Details : Make lustre randomly failed allocating memory for testing purpose.
780 Severity : enhancement
782 Description: lost problems with lov objid file
783 Details : Fixes some scability and access to not inited memory problems
784 in work with lov objdid file.
789 Description: lfs find does not continue on file error
790 Details : Continue other files processing when a file/dir is absent.
794 Description: Inconsistent usage of lustre_pack_reply()
795 Details : Standardize the usage of lustre_pack_reply() such that it
796 always generate a CERROR on failure.
799 Frequency : occasional
801 Description: MDS hang or stay in waiting lock
802 Details : If client receive lock with CBPENDING flag ldlm need send lock
803 cancel as separate rpc, to avoid situation when cancel request
804 can't processed due all i/o threads stay in wait lock.
808 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
812 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
816 Description: Update to SLES9 kernel-2.6.5-7.287.3.
820 Description: Update to RHEL4 latest kernel.
822 Severity : enhancement
824 Description: Build SLES10 patchless client fails
825 Details : The configure was broken by run ./configure with
826 --with-linux-obj=.... argument for patchless client. When the
827 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
828 can't search header adequately. Use absolute path such as
829 -I($LINUX)/include instead.
833 Description: interrupt oig_wait produce painc on resend.
834 Details : brw_redo_request can be used for resend requests from ptlrpcd and
835 private set, and this produce situation when rq_ptlrpcd_data not
836 copyed to new allocated request and triggered LBUG on assert
837 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
838 ptlrpcd set if request is changed and can be safety changed to use
843 Description: LASSERT_{REQ,REP}SWAB macros are buggy
844 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
845 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
846 these macros and replace them with their difinitions instead.
850 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
851 Details : Update which_patch & target file for SLES10 latest kernel.
854 Description: Remove client patches from SLES 10 kernel.
855 Details : This causes SLES 10 clients to behave as patchless clients
856 even on a Lustre-patched (server) kernel.
858 Severity : enhancement
860 Description: self-adjustable client's lru lists
861 Details : use adaptive algorithm for managing client cached locks lru
862 lists according to current server load, other client's work
863 pattern, memory activities, etc. Both, server and client
864 side namespaces provide number of proc tunables for controlling
867 Severity : enhancement
869 Description: light-weight GSS support
870 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
871 and reduce performance overhead.
873 Severity : enhancement
875 Description: Linux keyring support
876 Details : Support using service of Linux keyring for Lustre GSS internal
877 context refresh/cache mechanism.
881 Description: Fix errors in lfs documentation
882 Details : Fixes man pages
886 Description: don't use GFP_* in generic Lustre code.
887 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
891 Description: obdclass is limited by single OBD_ALLOC(idarray)
892 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
896 Description: warnings with build patchless client with vanila 2.6.19 and up
897 Details : change old ctl_table style and replace ctl_table/ctl_table_header
898 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
902 Frequency : Only for SLES
903 Description: sanity_quota fail test_1
904 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
905 file, which makes TSTID[2] inunique.
909 Frequency : Only for SLES9
910 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
911 Details : Update target/ChangeLog/which_patch .
915 Frequency : Only for test_10 in sanity-quota.sh
916 Description: a bug in quota test code
917 Details : checking a test flag in a wrong place
921 Frequency : Only for test_13 in sanity-quota.sh
922 Description: a bug in quota test script
923 Details : shouldn't check overfull allocated quota
927 Description: jbd statistics
928 Details : Port older jbd statistics patch for sles10
930 Severity : enhancement
932 Description: Add printing inode info into message about error in writepage.
937 Description: reply_lock_interpret crash due to race with it and lock cancel.
938 Details : Do not replay locks that are being cancelled. Do not reference
939 locks by their address during replay, just by their handle.
943 Frequency : When flocks are used.
944 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
945 Details : Flock locks might destroy just granted lock if it could be merged
946 with another existing flock, this is done in completion handler,
947 so teach ldlm_cli_enquque_fini that this is a valid case for
953 Description: Oops in read and write path when failing to allocate lock.
954 Details : Check if lock allocation failed and return error back.
958 Description: lstripe command fails for valid OST index
959 Details : The stripe offset is compared to lov->desc.ld_tgt_count
960 instead of lov->desc.ld_active_tgt_count.
964 Description: sanity.sh failed test 103
965 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
970 Description: df doesn't work properly if diskfs blocksize != 4K
971 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
975 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
976 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
978 Severity : enhancement
980 Description: Adaptive timeouts
981 Details : RPC timeouts adapt to changing server load and network
982 conditions to reduce resend attempts and improve recovery time.
986 Description: llapi_file_create() does not allow some changes
987 Details : add llapi_file_open() that allows specifying the mode and
988 open flags, and also returns an open file handle.
992 Description: merge and cleanup kernel patches.
993 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
997 Description: Add journal checksum support.(Kernel part)
998 Details : The journal checksum feature adds two new flags i.e
999 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
1000 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
1001 indicates that the commit block contains the checksum for
1002 the blocks described by the descriptor blocks. Now commit
1003 record can be sent to disk without waiting for descriptor
1004 blocks to be written to disk. This behavior is controlled
1005 using JBD2_FEATURE_ASYNC_COMMIT flag.
1009 Description: OSS needs mutliple precreate threads
1010 Details : Add ability to start more than one create thread per OSS.
1014 Description: Sanity.sh test_65k failed: setstripe should have succeeded
1015 Details : execute lfs setstripe on client
1019 Description: mds_obd_create error creating tmp object
1020 Details : When the user sets quota on root, llog will be affected and can't
1021 create files and write files.
1024 Frequency : Always on ia64 patchless client, and possibly others.
1026 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1027 Details : This allows the patchless client to be loaded on architectures
1028 without this export.
1032 Description: RedHat Update kernel for RHEL5
1033 Details : Add the kernel config file for RHEL5.
1037 Description: RedHat Update kernel for RHEL5
1038 Details : Modify the kernel config file more closer RHEL5.
1042 Description: Build failure against Centos5 (RHEL5)
1043 Details : Define PAGE_SIZE when it isn't present.
1047 Description: test_7 on sanity-quota.sh can't be used on separate machines
1048 Details : add facet to handle it
1052 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1053 Details : with NFS, the anon dentry's parent was set to itself in
1054 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1055 corresponding dentry other than use rec->ur_name.
1057 Severity : enhancement
1059 Description: lfs setstripe enhancement
1060 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1064 Description: enable data checksumming by default
1065 Details : enable checksum by default, allow --disable-checksum
1066 configure option and "-o nochecksum" mount option. Checksums
1067 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1071 Description: lustre support for RHEL5
1072 Details : Add support for RHEL5.
1076 Description: Client eviction due to failover config
1077 Details : after a connection loss, the lustre client should attempt to
1078 reconnect to the last active server first before trying the
1079 other potential connections.
1083 Description: when mds and osts use different quota unit(32bit and 64bit),
1084 quota will be released repeatly.
1085 Details : void sending multiple quota reqs to mds, which will keep the
1086 status between the reqs.
1090 Description: osts not allocated evenly to files
1091 Details : change the condition to increase offset_idx
1094 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1096 Description: Data corruption for OSTs that are formatted larger than 2TB
1098 Details : When generating the bio request for lustre file writes the
1099 sector number would overflow a temporary variable before being
1100 used for the IO. The data reads correctly from Lustre (which
1101 will overflow in a similar manner) but other file data or
1102 filesystem metadata may be corrupted in some cases.
1106 Description: Tune the kernel for good SCSI performance.
1107 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1108 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1113 Description: rewrite ext2-derived code in obdclass/uuid.c
1114 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1115 from scratch preserving functionality.
1119 Description: "lfs find -obd UUID" prints directories
1120 Details : "lfs find -obd UUID" will return all directory names instead
1121 of just file names. It is incorrect because the directories
1122 do not reside on the OSTs.
1126 Description: use i_size_read and i_size_write in 2.6 port
1127 Details : replace inode->i_size access with i_size_read/write()
1130 Frequency : when removing large files
1132 Description: scheduling issue during removal of large Lustre files
1133 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1134 It causes scheduling issues when removing large files (17TB in the
1138 Frequency : only with liblustre clients on XT3
1140 Description: evictions taking too long
1141 Details : allow llrd to evict clients directly on OSTs
1144 Frequency : only on ppc
1146 Description: /proc/fs/lustre/devices broken on ppc
1147 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1148 We should make sure the type of 'index' is loff_t and then cast
1149 explicitly as needed below. Do not assign an explicitly cast
1153 Frequency : only for rhel5
1155 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1156 Details : Modify the target file & which_kernel.
1158 Severity : enhancement
1160 Description: omit set fsid for export NFS
1161 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1162 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1166 Frequency : Always, for kernels after 2.6.16
1167 Description: Fix warning idr_remove called for id=.. which is not allocated.
1168 Details : Last kernels save old s_dev before kill super and not allow
1169 to restore from callback - restore it before call kill_anon_super.
1173 Description: buffer overruns could theoretically occur
1174 Details : llapi_semantic_traverse() modifies the "path" argument by
1175 appending values to the end of the origin string, and a
1176 overrun may occur. Adding buffer overrun check in liblustreapi.
1180 Description: Fix error on 'ls .' at the top of the Lustre mount.
1181 Details : Don't revalidate dentry if it is a root dentry.
1185 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1186 Details : Modify vm-tunables-rhel4.patch.
1190 Description: Kernel config for 2.6.18-vanilla.
1191 Details : Modify targets/2.6-vanilla.target.in.
1192 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1193 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1194 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1195 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1199 Description: improve handling recoverable errors
1200 Details : if request processig with error which can be recoverable on server
1201 request should be resend, otherwise page released from cache and
1206 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1207 Details : Modify target file & which_patch.
1208 A flaw was found in the IA32 system call emulation provided
1209 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1210 value could be stored in the %RAX register, which could trigger an
1211 out-of-bounds system call table access. An untrusted local user
1212 could exploit this flaw to run code in the kernel
1213 (ie a root privilege escalation). (CVE-2007-4573).
1217 Description: O_DIRECT bypasses client statistics.
1218 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1219 read_ahead_stats were updated. copied the stats section from
1220 osc_send_oap_rpc() into async_internal().
1224 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1228 Description: change order of libsysio includes
1229 Details : '#include sysio.h' should always come before '#include xtio.h'
1233 Description: handle "serious error: objid * already exists" more gracefully
1234 Details : If LAST_ID value on disk is smaller than the objects existing in
1235 the O/0/d* directories, it indicates disk corruption and causes an
1236 LBUG(). If the object is 0-length, then we should use the existing
1237 object. This will help to avoid a full fsck in most cases.
1239 Severity : enhancement
1241 Description: adapt the lustre_config script to support the upgrade case
1242 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1243 server targets to 1.6 in parallel.
1247 Description: To avoid grant space > avaible space when the disk is almost
1248 full. Without this patch you might see the error "grant XXXX >
1249 available" or some LBUG about grant, when the disk is almost
1251 Details : In filter_check_grant, for non_grant cache write, we should
1252 check the left space by if (*left > ungranted + bytes), instead
1253 of (*left > ungranted), because only we are sure the left space
1254 is enough for another "bytes", then the ungrant space should be
1255 increase. In client, we should update cl_avail_grant only there
1256 is OBD_MD_FLGRANT in the reply.
1260 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1261 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1262 described in CVE-2007-4573. This problem could allow untrusted
1263 local users to gain root access.
1266 Frequency : when using O_DIRECT and quotas
1268 Description: Incorrect file ownership on O_DIRECT output files
1269 Details : block usage reported by 'lfs quota' does not take into account
1270 files that have been written with O_DIRECT.
1275 Description: touch file failed when fs is not full
1276 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1277 otherwise we can get ENOSP while fs is not full.
1281 Description: parallel lock callbacks
1282 Details : Instead of sending blocking and completion callbacks as separated
1283 requests, adding them to a set and sending in parallel.
1286 Frequency : only for Cray XT3
1287 Bugzilla : 12829/13455
1288 Description: Changing primary group doesn't change the group lustre assigns to
1290 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1291 overridden with the primary group provided by the group upcall,
1292 whereas the supplied fsgid can be trusted if it is in the list of
1293 supplementary groups returned by the group upcall.
1295 Severity : enhancement
1297 Description: Allow masking D_WARNING, D_ERROR messages from console
1298 Details : Console messages can now be disabled via lnet.printk.
1302 Description: User code with malformed file open parameter crashes client node
1303 Details : Before packing join_file req, all the related reference should be
1304 checked carefully in case some malformed flags cause fake
1305 join_file req on client.
1309 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1310 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1311 mode is equal to the granted mode. However, it should also check
1312 that LDLM_FL_CP_REQD is not set.
1316 Description: Heavy nfs access might result in deadlocks
1317 Details : After ELC code landed, it is now improper to enqueue any mds
1318 locks under och_sem, because enqueue might want to decide to
1319 cancel open locks for same inode we are holding och_sem for.
1323 Description: Client eviction while running blogbench
1324 Details : A lot of unlink operations with concurrent I/O can lead to a
1325 deadlock causing evictions. To address the problem, the number of
1326 oustanding OST_DESTROY requests is now throttled to
1327 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1328 callbacks are processed in priority.
1332 Description: enable ACLs on MDS by default
1333 Details : ACLs must be enabled on MDS by default.
1336 Frequency : PPC/PPC64 only
1338 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1339 Details : fix duplicated definitions between asm-ppc64/types.h and
1340 lustre_types.h on PPC.
1343 Frequency : PPC/PPC64 only
1345 Description: asm-ppc/segment.h does not exist
1346 Details : fix compile issue on PPC.
1350 Description: better handle error messages in extents code
1353 Frequency : RHEL4 only
1355 Description: mkfs is very slow on IA64/RHEL4
1356 Details : A performance regression has been discovered in the MPT Fusion
1357 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1358 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1359 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1361 Severity : enhancement
1363 Description: SNMP support enhancement
1364 Details : Adding total number of sampled request for an MDS node in snmp
1367 Severity : enhancement
1369 Description: Optimize ldlm waiting list processing for PR extent locks
1370 Details : When processing waiting list for read extent lock and meeting read
1371 lock that is same or wider to it that is not contended, skip
1372 processing rest of the list and immediatelly return current
1373 status of conflictness, since we are guaranteed there are no
1374 conflicting locks in the rest of the list.
1378 Description: Time out and refuse to reconnect
1379 Details : When the failover node is the primary node, it is possible
1380 to have two identical connections in imp_conn_list. We must
1381 compare not conn's pointers but NIDs, otherwise we can defeat
1382 connection throttling.
1386 Description: port llog fixes from b1_6 into HEAD
1387 Details : Port llog reference couting and some llog cleanups from b1_6
1388 (bug 10800) into HEAD, for protect from panic and access to already
1389 free llog structures.
1393 Description: Detect stride IO mode in read-ahead
1394 Details : When a client does stride read, read-ahead should detect that and
1395 read-ahead pages according to the detected stride pattern.
1399 Description: data checksumming impacts single node performance
1400 Details : add support for several checksum algorithm. Currently, only CRC32
1401 and Adler-32 are supported. The checksum type can be changed on
1402 the fly via /proc/fs/lustre/osc/*/checksum_type.
1406 Description: use adler32 for page checksums
1407 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1412 Description: build for x2 fails
1413 Details : fix compile issue on Cray systems.
1417 Description: Properly match for duplicate locks
1418 Details : Due to different lock order from skiplists code, we need to
1419 traverse entire list for now
1422 Frequency : only on PPC/SLES10
1424 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1425 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1426 be defined. Add a hack in mkfs_lustre.c to work around this compile
1431 Description: LASSERT on MDS when client holding flock lock dies
1432 Details : ldlm pool logic depends on number of granted locks equal to
1433 number of released locks which is not true for flock locks, so
1434 just exclude such locks from consideration.
1438 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1439 Details : Use fsfilt_sync() for both the whole filesystem sync and
1440 individual file sync to eliminate dangerous inode locking
1441 with I_LOCK that can lead to a deadlock.
1445 Description: performance in 1.6.3
1446 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1447 sure that 1MB requests are not fragmented by the block layer.
1449 Severity : enhancement
1451 Description: organize the server-side client stats on per-nid basis
1452 Details : Change the structure of stats under obdfilter and mds to
1460 The "uuid"s file would list the uuids of _active_ exports.
1461 And the clear entry is to clear all stats and stale nids.
1463 Severity : enhancement
1465 Description: eliminate client locks in face of contention
1466 Details : file contention detection and lockless i/o implementation
1467 for contended files.
1471 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1474 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1476 Description: MDS may lose file striping (and hence file data) in some cases
1477 Details : If there are additional extended attributes stored on the MDS,
1478 in particular ACLs, SELinux, or user attributes (if user_xattr
1479 is specified for the client mount options) then there is a risk
1480 of attribute loss. Additionally, the Lustre file striping
1481 needs to be larger than default (e.g. striped over all OSTs),
1482 and an additional attribute must be stored initially in the
1483 inode and then increase in size enough to be moved to the
1484 external attribute block (e.g. ACL growing in size) for file
1489 Description: skiplist implementation simplification
1490 Details : skiplists are used to group compatible locks on granted list
1491 that was implemented as tracking first and last lock of each
1492 lock group the patch changes that to using doubly linked lists
1496 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1497 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1498 because LASSERT on both the data supplied by a client, and the
1499 data on disk is dangerous and incorrect.
1501 Severity : enhancement
1503 Description: Slow truncate/writes to huge files at high offsets.
1504 Details : Directly associate cached pages to lock that protect those pages,
1505 this allows us to quickly find what pages to write and remove
1506 once lock callback is received.
1510 Description: more ldlm soft lockups
1511 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1512 starve other threads from the resource lock for a long time in
1513 case of long waiting queue, so change the debug level from
1514 D_OTHER to the less frequently used D_INFO.
1516 Severity : enhancement
1518 Description: add -gid, -group, -uid, -user options to lfs find
1522 Description: Hung threads in invalidate_inode_pages2_range
1523 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1524 once one is completed. As a result, some RPCs are stuck in the
1525 queue and are never sent.
1529 Description: filter threads hungs on waiting journal commit
1530 Details : Cleanup filter group llog code, then only filter group llog will
1531 be only created in the MDS/OST syncing process.
1535 Description: Procfs and llog threads access destoryed import sometimes.
1536 Details : Sync the import destoryed process with procfs and llog threads by
1537 the import refcount and semaphore.
1539 Severity : enhancement
1541 Description: openlock cache of b1_6 port to HEAD
1546 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1547 Details : If insertion of an extent fails, then discard the inode
1548 preallocation and free data blocks else it can lead to duplicate
1553 Description: don't always update ctime in ext3_xattr_set_handle()
1554 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1555 In some cases the ctime should not be updated, for example for
1556 2.0->1.8 compatibility it is necessary to delete an xattr and it
1557 should not update the ctime.
1561 Bugzilla : 15713/16362
1562 Description: Assertion in iopen_connect_dentry in 1.6.3
1563 Details : looking up an inode via iopen with the wrong generation number can
1564 populate the dcache with a disconneced dentry while the inode
1565 number is in the process of being reallocated. This causes an
1566 assertion failure in iopen since the inode's dentry list contains
1567 both a connected and disconnected dentry.
1571 Description: assertion failure in ldlm_handle2lock()
1572 Details : fix a race between class_handle_unhash() and class_handle2object()
1573 introduced in lustre 1.6.5 by bug 13622.
1578 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1579 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1584 Description: oops in sd_iostats_seq_show()
1585 Details : unloading/reloading the scsi low level driver triggers a kernel
1586 bug when trying to access the sd iostat file.
1591 Description: Kernel panics during QLogic driver reload
1592 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1593 patch, causing memory corruption.
1598 Description: journal_dev option does not work in b1_6
1599 Details : pass mount option during pre-mount.
1601 Severity : enhancement
1603 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1604 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1605 extent information of a file. It can be used to map logical blocks
1606 in a file to physical blocks in the block device.
1610 Description: LDLM soft lockups - improvement
1611 Details : It is be possible to send the lock handle along with each read
1612 or write request because the client is already doing a lock match
1613 itself so there isn't any reason the OST should have to re-do that
1617 Frequency : only on Cray X2
1619 Description: X2 build failures
1620 Details : fix build failures on Cray X2.
1624 Description: xid & resent requests
1625 Details : Initialize RPC XID from clock at startup (randomly if clock is
1628 Severity : enhancement
1630 Description: Add lustre_start utility to start or stop multiple Lustre servers
1635 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1636 Details : In case of memory pressure, list_del() can be called twice on
1637 req->rq_history_list, causing a kernel oops.
1641 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1642 Details : fix stack overflow in the distributed lock manager by defering
1643 export eviction after a failed AST to the elt thread instead of
1644 handling it in the dlm interpret routine.
1648 Description: Convert some comments to new format.
1649 Details : Update documenting comments to match doxygen conventions.
1653 Description: Grammar fixes.
1654 Details : A couple of trivial spelling fixes.
1658 Description: OSD_COUNTERS-mandatory
1659 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1660 to simplify the code.
1664 Description: simplify cmm_device freeing logic.
1665 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1669 Description: Add lockdep support to dt_object_operations locking interface.
1670 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1671 parameter indicating lock ordering. Update mdd code to use new
1676 Description: Introduce failloc constants for lockless IO tests.
1677 Details : Add two new failloc constants to test lockless IO. Only one of
1678 them in implemented---another is checked in yet to be landed
1683 Description: Add lockdep support for inode mutex.
1684 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1685 in the situations where Lustre has to lock more than one inode
1690 Description: Add optional invariants checking support.
1691 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1692 configure switch. This macro is to be used for consistency and
1693 sanity checks that are too expensive to be left in `production'
1698 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1699 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1704 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1705 Details : Use spin_lock_nested() in (the only) situation where more than
1706 one ldlm_lock is locked simultaneously. Also, fix possible
1707 dead-lock in ldlm_lock_change_resource() by enforcing particular
1712 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1713 Details : Instead of specifying each ldlm_lock call-back through separate
1714 parameter, wrap them into struct ldlm_callback_suite.
1718 Description: Kill join_lru obd method and its callers.
1719 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1720 in memory---a requirement implemented through ->o_join_lru() obd
1721 method in HEAD. Remove this method and its users.
1725 Description: Add asynchronous ldlm ENQUEUE completion handler.
1726 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1727 case that stock ldlm_completion_ast() cannot handle as it waits
1728 until lock is granted. Introduce new ldlm_completion_ast_async()
1729 for this. Also comment ldlm_completion_ast().
1733 Description: ldlm_error <-> errno conversion.
1734 Details : Add functions to map (rather arbitrary) between LDLM error codes
1735 and standard errno values. CLIO needs this to prevent LDLM specific
1736 constants from escaping ldlm and osc.
1740 Description: Kill unused ldlm_handle2lock_ns() function.
1741 Details : Kill unused ldlm_handle2lock_ns() function.
1745 Description: Add lu_ref support to ldlm_lock
1746 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1747 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1748 and ldlm_resource. LDLM interface has to be changed, because code that
1749 releases a reference on a lock, has to "know" what reference this is.
1750 In the most frequent case
1752 lock = ldlm_handle2lock(handle);
1754 LDLM_LOCK_PUT(lock);
1756 no changes are required. When any other reference (received _not_ from
1757 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1758 instead of LDLM_LOCK_PUT().
1760 Arguably, changes are pervasive, and interface requires some discipline
1761 for proper use. On the other hand, it was very instrumental in finding
1762 a few leaked lock references.
1766 Description: Add ldlm_lock_addref_try().
1767 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1768 attempts to addref a lock that might be being canceled
1773 Description: Add ldlm_weigh_callback().
1774 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1775 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1776 hard-coded `number of pages' logic.
1780 Description: Add lockdep annotations to llog code.
1781 Details : Use appropriately tagged _nested() locking calls in the places
1782 where llog takes more than one ->lgh_lock lock.
1786 Description: Add loi_kms_set().
1787 Details : Wrap kms updates into a helper function.
1791 Description: Constify instances of struct lsm_operations.
1792 Details : Constify instances of struct lsm_operations.
1796 Description: lu_conf support.
1797 Details : On a server, a file system object is uniquely identified
1798 by a fid, which is sufficient to locate and load all object
1799 state (inode). On a client, on the other hand, more data are
1800 necessary instantiate an object. Change lu_object_find() and
1801 friends to take additional `lu_conf' argument describing object.
1802 Typically this includes layout information.
1806 Description: lu_context fixes.
1807 Details : Introduce new lu_context functions that are needed on the client
1808 side, where some system threads (ptlrpcd) are shared by multiple
1809 modules, and so cannot be stopped during module shutdown.
1813 Description: Add start and stop methods to lu_device_type_operations.
1814 Details : Introduce two new methods in lu_device_type_operations, that are
1815 invoked when first instance of a given type is created and last one
1816 is destroyed respectively. This is need by CLIO.
1820 Description: Add lu_ref support to struct lu_device.
1821 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1822 track leaked references.
1826 Description: Introduce lu_kmem_descr.
1827 Details : lu_kmem_descr and its companion interface allow to create
1828 and destroy a number of kmem caches at once.
1832 Description: Fix lu_object finalization race.
1833 Details : Fix a race between lu_object_find() finding an object and its
1834 concurrent finalization. This race is (most likely) not possible
1835 on the server, but might happen on the client.
1839 Description: Introduce lu_ref interface.
1840 Details : lu_ref is a debugging module allowing to track references to
1841 a given object. It is quite cpu expensive, and has to be
1842 explicitly enabled with --enable-lu_ref. See usage description
1847 Description: Factor lu_site procfs stats into a separate function.
1848 Details : Separate lu_site stats printing code into a separate function
1849 that can be reused on a client.
1853 Description: Constify instances of struct {lu,dt,md}_device_operations.
1854 Details : Constify instances of struct {lu,dt,md}_device_operations.
1858 Description: Introduce struct md_site and move meta-data specific parts of
1859 struct lu_site here.
1860 Details : Move md-specific fields out of struct lu_site into special struct
1861 md_site, so that lu_site can be used on a client.
1865 Description: Kill mdd_lov_destroy().
1866 Details : Remove unused mdd code.
1870 Description: Add st_block checking to multistat.c.
1871 Details : Add st_block checking to multistat.c.
1875 Description: Add lu_ref support to struct obd_device.
1876 Details : Add lu_ref tracking to obd_device.
1880 Description: Kill obd_set_fail_loc().
1881 Details : Remove unused code.
1885 Description: Add special type for ptlrpc_request interpret functions.
1886 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1887 there. Actual usage will be in CLIO.
1891 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1892 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1893 doesn't work with lockdep.
1897 Description: Add rwv.c test program.
1898 Details : New testing program exercising readv(2) and writev(2) (Qian).
1902 Description: Add sendfile.c test program.
1903 Details : New testing program exercising sendfile(2) (Jay).
1907 Description: Ratelimit a message that can be very frequent.
1908 Details : Ratelimit a memory allocation failure message that can
1913 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1914 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1918 Description: Liblustre build fixes.
1919 Details : Liblustre build fixes.
1923 Description: libcfs: add cfs_{need,cond}_resched() interface.
1924 Details : libcfs: add cfs_{need,cond}_resched() definition and
1925 implementations for Linux, NT, and liblustre.
1927 Severity : enhancement
1929 Description: More exported tunables for mballoc
1930 Details : Add support for tunable preallocation window and new tunables for
1931 large/small requests
1935 Description: Detect corruption of block bitmap and checking for preallocations
1936 Details : Checks validity of on-disk block bitmap. Also it does better
1937 checking of number of applied preallocations. When corruption is
1938 found, it turns filesystem readonly to prevent further corruptions.
1942 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1943 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1944 window to 0. If it does hit cache miss, and read-pattern is still
1945 stride-io mode, does not reset the stride window, but also does not
1946 increase the stride window length in this case.
1950 Frequency : only for big-endian servers
1951 Description: Check if system is big-endian while mounting fs with extents feature
1952 Details : Mounting a filesystem with extents feature will fail on big-endian
1953 systems since ext3-based ldiskfs is not supported on big-endian
1954 systems. This can be over-riden with "bigendian_extents" mount option.
1956 Severity : enhancement
1958 Description: The root squash functionality
1959 Details : A security feature, which is to prevent users from being able
1960 to mount lustre on their desktop, run as root, and delete
1961 all of the files in the filesystem. The goal is accomplished by
1962 remapping user id (UID) and group id (GID) of the root user to
1963 a UID and GID specified by the system administartor via Lustre
1964 configuration management server (MGS). The functionality also
1965 allows to specify sets of clients for which the remapping does
1970 Description: Excessive recovery window
1971 Details : With AT enabled, the recovery window can be excessively long (6000+
1972 seconds). To address this problem, we no longer use
1973 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1974 timeout no longer depends on the service time, it is set to
1975 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1976 time via pb_service_time.
1980 Description: Watchdog triggered on MDS failover
1981 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1982 the OSTs know that the MDS "UUID" can be reused for the same export
1983 from a different NID, so we do not need to wait for the export to be
1987 Frequency : rare, only if using MMP with Linux RAID
1989 Description: MMP doesn't work with Linux RAID
1990 Details : While using HA for Lustre servers with Linux RAID, it is possible
1991 that MMP will not detect multiple mounts. To make this work we
1992 need to unplug the device queue in RAID when the MMP block is being
1993 written. Also while reading the MMP block, we should read it from
1994 disk and not the cached one.
1996 Severity : enhancement
1998 Description: open file using fid
1999 Details : A file can be opened using just its fid, like
2000 <mntpt>/.lustre/fid/SEQ:OID:VER - this is needed for HSM and replication
2002 --------------------------------------------------------------------------------
2004 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
2006 * Support for kernels:
2007 2.6.5-7.283 (SLES 9),
2008 2.6.9-55.EL (RHEL 4),
2009 2.6.16.46-0.14 (SLES 10),
2010 2.6.18.8 vanilla (kernel.org)
2011 * Client support for unpatched kernels:
2012 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
2013 2.6.16 - 2.6.22 vanilla (kernel.org)
2014 * Due to recently discovered recovery problems, we do not recommend
2015 using patchless RHEL 4 clients with this or any earlier release.
2016 * Recommended e2fsprogs version: 1.39.cfs8
2017 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
2018 * Starting with this release, the ldiskfs backing filesystem required
2019 by Lustre is now in its own package, lustre-ldiskfs. This package
2020 should be installed. It is versioned separately from Lustre and
2021 may be released separately in future.
2025 Description: block reactivating mgc import until all deactivates complete
2026 Details : Fix race when failing back MDT/MGS to itself (testing)
2029 Frequency : at statup only
2031 Description: mds_lov_synchronize race leads to various problems
2032 Details : simultaneous MDT->OST connections at startup can cause the
2033 sync to abort, leaving the OSC in a bad state.
2035 Severity : enhancement
2037 Description: add optional extra BUILD_VERSION info
2038 Details : add a new environment variable (namely LUSTRE_VERS) which allows
2039 to override the lustre version.
2042 Frequency : 2.6.18 servers only
2044 Description: ll_kern_mount() doesn't release the module reference
2045 Details : The ldiskfs module reference count never drops down to 0
2046 because ll_kern_mount() doesn't release the module reference.
2051 Description: server LBUG when using old ost_num_threads parameter
2052 Details : Accept the old ost_num_threads parameter but warn that it
2053 is deprecated, and fix an off-by-one error that caused an LBUG.
2058 Description: Transient SCSI error results in persistent IO issue
2059 Details : iobuf->dr_error is not reinitialized to 0 between two
2063 Frequency : sometimes when underlying device returns I/O errors
2065 Description: OSTs not going read-only during write failures
2066 Details : OSTs are not remounted read-only when the journal commit threads
2067 get I/O errors because fsfilt_ext3 calls journal_start/stop()
2068 instead of the ext3 wrappers.
2072 Description: poor connect scaling with increasing client count
2073 Details : Don't run filter_grant_sanity_check for more than 100 exports
2074 to improve scaling for large numbers of clients.
2077 Frequency : SLES10 only
2079 Description: sanity-quota.sh quotacheck failed: rc = -22
2080 Details : Quotas cannot be enabled on SLES10.
2083 Frequency : liblustre clients only
2085 Description: getdirentries does not give error when run on compute nodes
2086 Details : getdirentries does not fail when the size specified as an argument
2087 is too small to contain at least one entry
2089 Severity : enhancement
2091 Description: Add LNET router traceability for debug purposes
2092 Details : If a checksum failure occurs with a router as part of the
2093 IO path, the NID of the last router that forwarded the bulk data
2094 is printed so it can be identified.
2099 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2100 Details : Due to a race condition, liblustre clients were occasionally
2101 evicted incorrectly.
2103 Severity : enhancement
2105 Description: lfs setstripe use optional parameters instead of postional
2108 Severity : enhancement
2110 Description: Nanosecond timestamp support for ldiskfs
2111 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2112 resolution timestamps. There is not yet support for this at
2113 the Lustre filesystem level.
2116 Frequency : during server recovery
2118 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2119 Details : request with rq_no_resend flag not awake l_wait_event if they get
2123 Frequency : nfs export on patchless client
2125 Description: connectathon hang when test nfs export over patchless client
2126 Details : Disconnected dentry cannot be found with lookup, so we do not need
2127 to unhash it or make it invalid
2130 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2131 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2132 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2137 Description: open req refcounting wrong on reconnect
2138 Details : If reconnect happened between getting open reply from server and
2139 call to mdc_set_replay_data in ll_file_open, we will schedule
2140 replay for unreferenced request that we are about to free.
2141 Subsequent close will crash in variety of ways.
2142 Check that request is still eligible for replay in
2143 mdc_set_replay_data().
2148 Description: disable writes to filesystem when reading health_check file
2149 Details : the default for reading the health_check proc file has changed
2150 to NOT do a journal transaction and write to disk, because this
2151 can cause reads of the /proc file to hang and block HA state
2152 checking on a healthy but otherwise heavily loaded system. It
2153 is possible to return to the previous behaviour during configure
2154 with --enable-health-write.
2156 Severity : enhancement
2158 Description: 64-bit inode version
2159 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2160 to the inode. This will be required for version-based recovery.
2165 Description: MDS fails to start if a duplicate client export is detected
2166 Details : in some rare cases it was possible for a client to connect to
2167 an MDS multiple times. Upon recovery the MDS would detect this
2168 and fail during startup. Handle this more gracefully.
2170 Severity : enhancement
2172 Description: Add -o localflock option to simulate old noflock
2174 Details : This will achieve local-only flock/fcntl locks
2180 Description: log_commit_thread vs filter_destroy race leads to crash
2181 Details : Take import reference before releasing llog record semaphore
2186 Description: Wrong request locking in request set processing
2187 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2188 imp_delayed_list, in this place should be used imp_lock.
2191 Frequency : when reconnection
2193 Description: Grant Leak when osc reconnect to OST
2194 Details : When osc reconnect ost, OST(filter) should check whether it
2195 should grant more space to client by comparing fed_grant and
2196 cl_avail_grant, and return the granted space to client instead
2197 of "new granted" space, because client will call osc_init_grant
2198 to update the client grant space info.
2201 Frequency : when client reconnect to OST
2203 Description: Grant Leak when osc do resend and replay bulk write
2204 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2205 bulk write request, because the grant info while be sync between
2206 OSC and OST when reconnect, and we should ignore the grant info
2207 these of resend/replay write req.
2212 Description: Grant space more than avaiable left space sometimes.
2213 Details : When then OST is about to be full, if two bulk writing from
2214 different clients came to OST. Accord the avaliable space of the
2215 OST, the first req should be permitted, and the second one
2216 should be denied by ENOSPC. But if the seconde arrived before
2217 the first one is commited. The OST might wrongly permit second
2218 writing, which will cause grant space > avaiable space.
2221 Frequency : when client is evicted
2223 Description: Grant might be wrongly erased when osc is evicted by OST
2224 Details : when the import is evicted by server, it will fork another
2225 thread ptlrpc_invalidate_import_thread to invalidate the
2226 import, where the grant will be set to 0. While the original
2227 thread will update the grant it got when connecting. So if
2228 the former happened latter, the grant will be wrongly errased
2229 because of this race.
2234 Description: Checking Stale with correct fid
2235 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2236 to check whether it is stale, because sometimes, we want the
2237 enqueue happened anyway, and op_data.fid2 will not be initialized.
2239 Severity : enhancement
2241 Description: update patchless client
2242 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2245 Frequency : only with 2.4 kernel
2247 Description: random memory corruption
2248 Details : size of struct ll_inode_info is to big for union inode.u and this
2249 can be cause of random memory corruption.
2254 Description: Memory leak in recovery
2255 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2256 It should also check obd_fail before fsfilt_start, otherwise if
2257 fsfilt_start return -EROFS,(failover mds during mds recovery).
2258 then the req will return with repmsg->transno = 0 and rc = EROFS.
2259 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2260 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2261 be freed no matter whether fsfilt_commit success or not.
2264 Frequency : only with huge count clients
2266 Description: Prevents from taking the superblock lock in llap_from_page for
2268 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2269 for llap_from_page call in ll_removepage prevents from taking the
2270 superblock lock for a soon died page.
2275 Description: Not check open intent error before release open handle
2276 Details : in some rare cases, the open intent error is not checked before
2277 release open handle, which may cause
2278 ASSERTION(open_req->rq_transno != 0), because it tries to release
2279 the failed open handle.
2284 Description: Set cat log bitmap only after create log success.
2285 Details : in some rare cases, the cat log bitmap is set too early. and it
2286 should be set only after create log success.
2290 Description: Accessing a block bevice can re-enable I/O when Lustre is
2291 tearing down a device.
2292 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2297 Description: service threads may hog cpus when there are a lot of requests
2298 Details : Insert cond_resched to give other threads a chance to use some CPU
2303 Description: the cat log was not initialized in recovery
2304 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2305 unlink log on mds will not be initialized until mds post
2306 recovery. And also in mds post recovery, the unlink log will
2307 initialization will be done asynchronausly, so there will be race
2308 between add unlink log and unlink log initialization.
2312 Description: brw_stats were being printed incorrectly
2313 Details : brw_stats were being printed as log2 but all of them were not
2314 recorded as log2. Also remove some code duplication arising from
2315 filter_tally_{read,write}.
2319 Frequency : rare, only in recovery.
2320 Description: ASSERTION(req->rq_type != LI_POISON) failed
2321 Details : imp_lock should be held while iterating over imp_sending_list for
2322 prevent destroy request after get timeout in ptlrpc_queue_wait.
2326 Description: replay-single.sh test 52 fails
2327 Details : A lock's skiplist need to be cleanup when it being unlinked
2328 from its resource list.
2332 Description: Short directio read returns full requested size rather than
2334 Details : Direct I/O operations should return actual amount of bytes
2335 transferred rather than requested size.
2337 Severity : enhancement
2339 Description: metadata RPC reduction (e.g. for rm performance)
2340 Details : decrease the amount of synchronous RPC between clients and servers
2341 by canceling conflicing lock before the operation on the client
2342 and packing thier handles into the main operation RPC to server.
2344 Severity : enhancement
2346 Description: Async OSC create to avoid the blocking unnecessarily.
2347 Details : If a OST has no remain object, system will block on the creating
2348 when need to create a new object on this OST. Now, ways use
2349 pre-created objects when available, instead of blocking on an
2350 empty osc while others are not empty. If we must block, we block
2351 for the shortest possible period of time.
2355 Description: improve handling recoverable errors
2356 Details : if request processig with error which can be recoverable on server
2357 request should be resend, otherwise page released from cache and
2360 Severity : enhancement
2362 Description: refine locking for avoid write wrong info into lov_objid file
2363 Details : fix possible races with add new target and write/update data in
2366 --------------------------------------------------------------------------------
2368 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2373 Frequency : on some architectures
2375 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2376 Details : Uninitialized flags sometimes cause configuration commands to
2380 Frequency : patchless clients only
2382 Description: missing __iget() symbol export
2383 Details : The __iget() symbol export is missing. To avoid the need for
2384 this on patchless clients the deathrow inode reaper is turned
2385 off, and we depend on the VM to clean up old inodes. This
2386 dependency was during via the fix for bug 12181.
2388 --------------------------------------------------------------------------------
2390 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2392 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2393 INTEROPERATE with older versions automatically. In many cases a
2394 special upgrade step is needed. Please read the
2395 user documentation before upgrading any part of a live system.
2396 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2397 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2398 * WARNING: Lustre configuration and startup changes are required with
2399 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2403 Severity : enhancement
2405 Description: MountConf
2406 Details : Lustre configuration is now managed via mkfs and mount
2407 commands instead of lmc and lconf. New obd types (MGS, MGC)
2408 are added for dynamic configuration management. See
2409 https://mail.clusterfs.com/wikis/lustre/MountConf for
2412 Severity : enhancement
2414 Description: dynamic OST addition
2415 Details : OSTs can now be added to a live filesystem
2417 Severity : enhancement
2419 Description: startup order invariance
2420 Details : MDTs and OSTs can be started in any order. Clients only
2421 require the MDT to complete startup.
2423 Severity : enhancement
2425 Description: parallel, asynchronous orphan cleanup
2426 Details : orphan cleanup is now performed in separate threads for each
2427 OST, allowing parallel non-blocking operation.
2429 Severity : enhancement
2431 Description: optimized stripe assignment
2432 Details : stripe assignments are now made based on ost space available,
2433 ost previous usage, and OSS previous usage, in order to try
2434 to optimize storage space and networking resources.
2436 Severity : enhancement
2438 Description: Permanently set tunables
2439 Details : All writable /proc/fs/lustre tunables can now be permanently
2440 set on a per-server basis, at mkfs time or on a live system.
2442 Severity : enhancement
2444 Description: Lustre message v2
2445 Details : Add lustre message format v2.
2447 Severity : enhancement
2449 Description: client OST exclusion list
2450 Details : Clients can be started with a list of OSTs that should be
2451 declared "inactive" for known non-responsive OSTs.
2455 Description: SPEC SFS validation failure on NFS v2 over lustre.
2456 Details : Changes the blocksize for regular files to be 2x RPC size,
2457 and not depend on stripe size.
2459 Severity : enhancement
2461 Description: Multiple MD RPCs in flight.
2462 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2463 To avoid overly-overloading MDS, introduce a limit on number of
2464 MDS RPCs in flight for a single client and add /proc controls
2465 to adjust this limit.
2467 Severity : enhancement
2469 Description: client read/write statistics
2470 Details : Add client read/write call usage stats for performance
2471 analysis of user processes.
2472 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2473 file access. extents_stats shows chunk size distribution.
2474 extents_stats_per_process show chunk size distribution per
2477 Severity : enhancement
2479 Description: mds statistics
2480 Details : Add detailed mds operations statistics in
2481 /proc/fs/lustre/mds/*/stats.
2485 Description: Failure of copying files with lustre special EAs.
2486 Details : Client side always return success for setxattr call for lustre
2487 special xattr (currently only "trusted.lov").
2492 Description: Refcount LNET uuids
2493 Details : The global LNET uuid list grew linearly with every startup;
2494 refcount repeated list entries instead of always adding to
2497 Severity : enhancement
2499 Description: Dynamic service threads
2500 Details : Within a small range, start extra service threads
2501 automatically when the request queue builds up.
2504 Frequency : mixed-endian client/server environments
2506 Description: mixed-endian crashes
2507 Details : The new msg_v2 system had some failures in mixed-endian
2510 Severity : enhancement
2512 Description: Easy OST removal
2513 Details : OSTs can be permanently deactivated with e.g. 'lctl
2514 conf_param lustre-OST0001.osc.active=0'
2516 Severity : enhancement
2518 Description: MGS proc entries
2519 Details : Added basic proc entries for the MGS showing what filesystems
2522 Severity : enhancement
2524 Description: provide MGS failover
2525 Details : Added config lock reacquisition after MGS server failover.
2527 Severity : enhancement
2529 Description: add Linux 2.4 support
2530 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2534 Description: a large application tries to do I/O to the same resource and dies
2535 in the middle of it.
2536 Details : Check the req->rq_arrival time after the call to
2537 ost_brw_lock_get(), but before we do anything about
2538 processing it & sending the BULK transfer request. This
2539 should help move old stale pending locks off the queue as
2540 quickly as obd_timeout.
2543 Frequency : when an incorrect nid is specified during startup
2545 Description: ptlrpc connect to non-existant node causes kernel crash
2546 Details : LNET can't be re-entered from an event callback, which
2547 happened when we expire a message after the export has been
2548 cleaned up. Instead, hand the zombie cleanup off to another
2551 Severity : enhancement
2553 Description: plain/inodebits lock performance improvement
2554 Details : Grouping plain/inodebits in granted list by their request modes
2555 and bits policy, thus improving the performance of search through
2559 Frequency : only if OST filesystem is corrupted
2561 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2562 Details : for a short time RPCs with bulk IO are in the replay list,
2563 but replay of bulk IOs is unimplemented. If the OST filesystem
2564 is corrupted due to disk cache incoherency and then replay is
2565 started it is possible to trip an assertion. Avoid putting
2566 committed RPCs into the replay list at all to avoid this issue.
2569 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2571 Description: System hang on startup
2572 Details : This bug allowed the liblustre (e.g. catamount) client to
2573 return to the app before handling all startup RPCs. This
2574 could leave the node unresponsive to lustre network traffic
2575 and manifested as a server ptllnd timeout.
2577 Severity : enhancement
2579 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2580 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2581 boolean to control whether to print peer debug info when a
2582 client's RPC times out.
2585 Frequency : only for kernels with patches from Lustre below 1.4.3
2587 Description: Remove old rdonly API
2588 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2591 Frequency : only for devices with external journals
2593 Description: Set external device read-only also
2594 Details : During a commanded failover stop, we set the disk device
2595 read-only while the server shuts down. We now also set any
2596 external journal device read-only at the same time.
2599 Frequency : when upgrading from 1.4 while trying to change parameters
2601 Description: The wrong (new) MDC name was used when setting parameters for
2602 upgraded MDT's. Also allows changing of OSC (and MDC)
2603 parameters if --writeconf is specified at tunefs upgrade time.
2606 Frequency : when setting specific ost indicies
2608 Description: QOS code breaks on skipped indicies
2609 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2610 created with --index need not be sequential.
2612 Severity : enhancement
2614 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2615 Details : The uninit_groups feature works in conjunction with the kernel
2616 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2617 up the pass1 processing of e2fsck. This is a read-only feature
2618 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2619 mount filesystems that have had this feature enabled.
2621 Severity : enhancement
2623 Description: Improve multi-block allocation algorithm to avoid fragmentation
2624 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2625 allocation locality and avoid filesystem fragmentation.
2627 ------------------------------------------------------------------------------
2629 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2631 * Support for kernels:
2632 2.4.21-47.0.1.EL (RHEL 3)
2633 2.6.5-7.283 (SLES 9)
2634 2.6.9-42.0.10.EL (RHEL 4)
2635 2.6.12.6 vanilla (kernel.org)
2636 2.6.16.27-0.9 (SLES 10)
2637 * Recommended e2fsprogs version: 1.39.cfs5
2639 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2643 Frequency : occasional, depends on client load and configuration
2644 Bugzilla : 12181, 12203
2645 Description: data loss for recently-modified files
2647 Details : In some cases it is possible that recently written or created
2648 files may not be written to disk in a timely manner (this should
2649 normally be within 30s unless client IO load is very high).
2650 The problem appears as zero-length files or files that are a
2651 multiple of 1MB in size after a client crash or client eviction
2652 that are missing data at the end of the file.
2654 This problem is more likely to be hit on clients where files are
2655 repeatedly created and unlinked in the same directory, clients
2656 have a large amount of RAM, have many CPUs, the filesystem has
2657 many OSTs, the clients are rebooted frequently, and/or the files
2658 are not accessed by other nodes after being written.
2660 The presence of the problem can be detected by looking at
2661 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2662 smaller than the second (nr_unused) then dirty files will not
2663 be flushed automatically to disk. "sync; sleep 10" should be
2664 run several times on the node before unmounting it to update
2665 Lustre (this is also safe to run on nodes without this problem).
2667 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2668 that can cause this same problem, so customers using that kernel
2669 also need to update the kernel in addition to Lustre. In order
2670 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2672 It is normal that files written just before a client crash (less
2673 than 30s) may not yet have been flushed to disk, even for local
2677 Frequency : frequent on thin XT3 nodes
2679 Description: UUID collision on thin XT3 Linux nodes
2680 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2681 randomly, since we relied on an insufficiently-seeded PRNG.
2686 Description: OSS hangs after "All ost request buffers busy"
2687 Details : A deadlock between quota and journal operations caused OSS
2688 hangs after printing "All ost request buffers busy."
2691 Frequency : always on liblustre builds
2693 Description: Cleanup compiler warnings on liblustre
2696 Frequency : always on liblustre builds on XT3
2698 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2702 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2703 > 32000 subdirectories
2704 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2705 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2706 subdirectory count crosses 32000. This will aid e2fsck to
2707 correctly handle more than 32000 subdirectories.
2710 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2712 Description: System hang on startup
2713 Details : This bug allowed the liblustre (e.g. catamount) client to
2714 return to the app before handling all startup RPCs. This
2715 could leave the node unresponsive to lustre network traffic
2716 and manifested as a server ptllnd timeout.
2718 Severity : enhancement
2720 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2721 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2722 boolean to control whether to print peer debug info when a
2723 client's RPC times out.
2728 Description: make O_SYNC working on 2.6 kernels
2729 Details : 2.6 kernels use different method for mark pages for write,
2730 so need add a code to lustre for O_SYNC work.
2735 Description: Failure to close file and release space on NFS
2736 Details : Put inode details into lock acquired in ll_intent_file_open.
2737 Use mdc_intent_lock in ll_intent_open to properly
2738 detect all kind of errors unhandled by mdc_enqueue.
2743 Description: proc file read during shutdown sometimes raced obd removal,
2745 Details : Add lock to prevent obd access after proc file removal.
2748 Frequency : Only for files larger than 4GB on 32-bit clients.
2750 Description: improperly doing page alignment of locks
2751 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2752 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2757 Description: RPCs being resent when they shouldn't be
2758 Details : Some RPCs that should not be resent are being resent. This
2759 can cause inconsistencies in the RPC state machine. Do not
2760 resend such requests.
2763 Frequency : rare, only with NFS export
2765 Description: Crash on NFS re-export node
2766 Details : under very unusual load conditions an assertion is hit in
2767 ll_intent_file_open()
2770 Frequency : only if OST filesystem is corrupted
2772 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2773 Details : for a short time RPCs with bulk IO are in the replay list,
2774 but replay of bulk IOs is unimplemented. If the OST filesystem
2775 is corrupted due to disk cache incoherency and then replay is
2776 started it is possible to trip an assertion. Avoid putting
2777 committed RPCs into the replay list at all to avoid this issue.
2782 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2783 Details : Large single O_DIRECT read and write calls can fail to allocate
2784 a sufficiently large buffer to process the request. In case of
2785 allocation failure the allocation is retried with a smaller
2786 buffer and broken into smaller requests.
2788 Severity : enhancement
2790 Description: Add -o localflock option to simulate old noflock behaviour.
2791 Details : This will achieve local-only flock/fcntl locks coherentness.
2796 Description: versioning check is incomplete
2797 Details : Checking the version difference of client vs. server, report
2798 error if the gap is too big.
2803 Description: add support PG_writeback bit
2804 Details : add support for PG_writeback bit for Lustre, for more carefull
2805 work with page cache in 2.6 kernel. This also fix some deadlocks
2806 and remove hack for work O_SYNC with 2.6 kernel.
2808 Severity : enhancement
2810 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2811 Details : The uninit_groups feature works in conjunction with the kernel
2812 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2813 up the pass1 processing of e2fsck. This is a read-only feature
2814 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2815 mount filesystems that have had this feature enabled.
2817 Severity : enhancement
2819 Description: Improve multi-block allocation algorithm to avoid fragmentation
2820 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2821 allocation locality and avoid filesystem fragmentation.
2823 ------------------------------------------------------------------------------
2825 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2827 * Support for kernels:
2828 2.6.9-42.0.3.EL (RHEL 4)
2829 2.6.5-7.276 (SLES 9)
2830 2.4.21-47.0.1.EL (RHEL 3)
2831 2.6.12.6 vanilla (kernel.org)
2832 2.6.16.21-0.8 (SLES10)
2833 * Recommended e2fsprogs version: 1.39.cfs2-0
2835 * The backwards-compatible /proc/sys/portals symlink has been removed
2836 in this release. Before upgrading, please ensure that you change
2837 any configuration scripts or /etc/sysctl.conf files that access
2838 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2839 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2840 in advance of the upgrade on any system running Lustre 1.4.6 or
2841 newer, since /proc/sys/lnet was added in that version.
2842 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2846 Frequency : only when quota is used
2848 Description: avoid scanning export list for quota master
2849 Details : Change the algorithms to avoid scanning export list in order
2850 to improve the efficiency.
2853 Frequency : MDS failover only, very rarely
2855 Description: "went back in time" messages on mds failover
2856 Details : The greatest transno may be lost when the current operation
2857 finishes with an error (transno==0) and the client's last_rcvd
2858 record is over-written. Save the greatest transno in the
2859 mds_last_transno for this case.
2862 Frequency : always for specific kernels and striping counts
2864 Description: client may get "Matching packet too big" without ACL support
2865 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2866 when trying to access files in certain configurations. The
2867 clients should in fact be denied when mounting because they do
2868 not understand ACLs.
2871 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2873 Description: many clients connecting with IO in progress causes connect timeouts
2874 Details : Avoid synchronous journal commits to avoid delays caused by many
2875 clients connecting/disconnecting when bulk IO is in progress.
2876 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2877 OST_IO_PORTAL to avoid delays behind potentially many pending
2881 Frequency : occasionally with multiple writers to a single file
2883 Description: shared writes to file may result in wrong size reported by stat()
2884 Details : Allow growing of kms when extent lock is cancelled
2887 Frequency : always with random mmap IO to multi-striped file
2889 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2890 Details : Only if the hole is at the end of OST object so that kms is too
2891 small. Fix is to increase kms accordingly in ll_nopage.
2894 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2896 Description: writes to a missing object would leak memory on the OST
2897 Details : If there is an inconsistency between the MDS and OST filesystems,
2898 such that the MDS references an object that doesn't exist, writes
2899 to that object will leak memory due to incorrect cleanup in the
2900 error handling path, eventually running out of memory on the OST.
2905 Description: Creating too long symlink causes lustre errors
2906 Details : Check symlink and name lengths before sending requests to MDS.
2909 Frequency : only if flock is enabled (not on by default)
2911 Description: posix locks not released on fd closure on 2.6.9+
2912 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2913 kernels, this caused such locks not to be released on fd close and
2914 then assertions on fs unmount about still used locks.
2917 Frequency : MDS failover only, very rarely
2919 Description: clients may get ASSERTION(granted_lock != NULL)
2920 Details : When request was taking a long time, and a client was resending
2921 a getattr by name lock request. The were multiple lock requests
2922 with the same client lock handle and
2923 mds_getattr_name->fixup_handle_for_resent_request found one of the
2924 lock handles but later failed with ASSERTION(granted_lock != NULL).
2929 Description: handle->h_buffer_credits > 0, assertion failure
2930 Details : h_buffer_credits is zero after truncate, causing assertion
2931 failure. This patch extends the transaction or creates a new
2935 Frequency : NFS re-export or patchless client
2936 Bugzilla : 11179, 10796
2937 Description: Crash on NFS re-export node (__d_move)
2938 Details : We do not want to hash the dentry if we don't have a lock.
2939 But if this dentry is later used in d_move, we'd hit uninitialised
2940 list head d_hash, so we just do this to init d_hash field but
2941 leave dentry unhashed.
2944 Frequency : NFS re-export or patchless client
2946 Description: NFS exports has problem with symbolic link
2947 Details : lustre client didn't properly install dentry when re-exported
2948 to NFS or running patchless client.
2951 Frequency : NFS re-export or patchless client
2953 Description: Various nfs/patchless fixes.
2954 Details : fixes reuse disconected alias for lookup process - this fixes
2955 warning "find_exported_dentry: npd != pd",
2956 fix permission error with open files at nfs.
2957 fix apply umask when do revalidate.
2960 Frequency : occasional
2962 Description: Crash on NFS re-export node
2963 Details : calling clear_page() on the wrong pointer triggered oops in
2964 generic_mapping_read().
2967 Frequency : rarely, using O_DIRECT IO
2969 Description: unaligned directio crashes client with LASSERT
2970 Details : check for unaligned buffers before trying any requests.
2973 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2975 Description: stale data returned from RAID cache
2976 Details : If only a small amount of IO is done to the RAID device before
2977 reading it again it is possible to get stale data from the RAID
2978 cache instead of reading it from disk.
2981 Frequency : always for sles10 kernel
2983 Description: sles10 support
2984 Details : ll_follow_link: compile fixes and using of nd_set_link
2985 under newer kernels.
2988 Frequency : depends on arch, kernel and compiler version, always on sles10
2991 Description: recursive or deep enough symlinks cause stack overflow
2992 Details : getting rid of large stack-allocated variable in
2996 Frequency : depends on hardware
2998 Description: lustre write performance loss in the SLES10 kernel
2999 Details : the performance loss is caused by using of write barriers in the
3000 ext3 code. The SLES10 kernel turns barrier support on by
3001 default. The fix is to undo that change for ldiskfs.
3003 ------------------------------------------------------------------------------
3005 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
3007 * Support for kernels:
3008 2.6.9-42.0.3EL (RHEL 4)
3009 2.6.5-7.276 (SLES 9)
3010 2.4.21-47.0.1.EL (RHEL 3)
3011 2.6.12.6 vanilla (kernel.org)
3015 Frequency : quota enabled and large files being deleted
3017 Description: releasing more than 4GB of quota at once hangs OST
3018 Details : If a user deletes more than 4GB of files on a single OST it
3019 will cause the OST to spin in an infinite loop. Release
3020 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3025 Description: statfs data retrieved from /proc may be stale or zero
3026 Details : When reading per-device statfs data from /proc, in the
3027 {kbytes,files}_{total,free,avail} files, it may appear
3028 as zero or be out of date.
3031 Frequency : systems with MD RAID1 external journal devices
3033 Description: lconf's call to blkid is confused by RAID1 journal devices
3034 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3035 of returning all block devices that match the journal UUID.
3038 Frequency : always, for aggregate stripe size over 4GB
3040 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
3041 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
3042 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
3045 Frequency : always if "lfs find" used on a local file/directory
3047 Description: "lfs find" segfaults if used on a local file/directory
3048 Details : The case where a directory component was not specified wasn't
3049 handled correctly. Handle this properly.
3052 Frequency : always on ppc64
3054 Description: the write to an ext3 filesystem mounted with mballoc got stuck
3055 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3056 perform endianness conversion.
3059 Frequency : rarely (truncate to non-zero file size after write under load)
3060 Bugzilla : 10730, 10687
3061 Description: Files padded with zeros to next 4K multiple
3062 Details : With filesystems mounted using the "extents" option (2.6 kernels)
3063 it is possible that files that are truncated to a non-zero size
3064 immediately after being written are filled with zero bytes beyond
3065 the truncated size. No file data is lost.
3067 Severity : enhancement
3069 Description: Allow recovery/failover for liblustre clients.
3070 Details : liblustre clients were unaware of failover configurations until
3073 Severity : enhancement
3075 Description: user file locks should fail when not mounting with flock option
3076 Details : Set up an error-returning stub in ll_file_operations.lock field
3077 to prevent incorrect behaviour when client is mounted without
3078 flock option. Also, set up properly f_op->flock field for
3082 Frequency : always on ia64
3084 Description: "lfs df" loops on printing out MDS statfs information
3085 Details : The obd_ioctl_data was not initialized and in some systems
3086 this caused a failure during the ioctl that did not return
3087 an error. Initialize the struct and return an error on failure.
3090 Frequency : SLES 9 only
3092 Description: Error of copying files with lustre special EAs as root
3093 Details : Client side always return success for setxattr call for lustre
3094 special xattr (currently only "trusted.lov").
3097 Frequency : rarely on clusters with both ia64+i386 clients
3099 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3100 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3101 clients are doing concurrent writes to the same file it is
3102 possible that the ia64 clients may LASSERT because the OST
3103 extent locks are not PAGE_SIZE aligned. Ensure that grown
3104 locks are always aligned on the request boundary.
3107 Frequency : specific use, occasional
3109 Description: Overwriting in use executable truncates on-disk binary image
3110 Details : If one node attempts to overwrite an executable in use by
3111 another node, we now correctly return ETXTBSY instead of
3112 truncating the file.
3114 Severity : enhancement
3116 Description: Async OSC create to avoid the blocking unnecessarily.
3117 Details : If a OST has no remain object, system will block on the creating
3118 when need to create a new object on this OST. Now, ways use
3119 pre-created objects when available, instead of blocking on an
3120 empty osc while others are not empty. If we must block, we block
3121 for the shortest possible period of time.
3126 Description: chmod on Lustre root is propagated to other clients
3127 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3128 invalid by the follow_mount time.
3131 Frequency : liblustre clients only
3133 Description: Race in 'instant cancel' lock handling could lead to such locks
3134 never to be granted in case of SMP MDS
3135 Details : Do not destroy not yet granted but cbpending locks in
3139 Frequency : replay/resend of open
3141 Description: non null lock assetion failure in mds_intent_policy
3142 Details : Trying to replay/resend lockless open requests resulted in
3143 mds_open() returning 0 with no lock. Now it sets a flag if
3144 a lock is going to be returned.
3146 Severity : enhancement
3148 Description: Checksum enhancements
3149 Details : New checksum enhancements allow for resending RPCs that failed
3152 Severity : enhancement
3154 Description: Tunables on number of dirty pages in cacche
3155 Details : Allow to set limit on number of dirty pages cached.
3160 Description: client crash on unmount - lock still has references
3161 Details : In some error handling cases it was possible to leak a lock
3162 reference on a client while accessing a file. This was not
3163 harmful to the client during operation, but would cause the
3164 client to crash when the filesystem is unmounted.
3167 Frequency : specific case, rare
3169 Description: ETXTBSY on mds though file not in use by client
3170 Details : ETXTBSY is no longer incorrectly returned when attempting to
3171 chmod or chown a directory that the user previously tried to
3172 execute or a currently-executing binary.
3175 Frequency : extremely rare except on liblustre-based clients
3177 Description: Lustre space not freed when files are deleted
3178 Details : Clean up open-unlinked files after client eviction. Previously
3179 the unlink was skipped and the files remained as orphans.
3184 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3185 Details : In some cases it was possible to send a blocking callback to
3186 a client doing a glimpse, even though that client didn't get
3187 a lock granted. When the glimpse lock is cancelled on the OST
3188 the freed lock is left on the waiting list and corrupted the list.
3191 Frequency : all core dumps
3193 Description: Broke core dumps to lustre
3194 Details : Negative dentry may be unhashed if parent does not have UPDATE
3195 lock, but some callers, e.g. do_coredump, expect dentry to be
3196 hashed after successful create, hash it in ll_create_it.
3198 ------------------------------------------------------------------------------
3200 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3202 * Support for kernels:
3203 2.6.9-42.0.2.EL (RHEL 4)
3204 2.6.5-7.276 (SLES 9)
3205 2.4.21-40.EL (RHEL 3)
3206 2.6.12.6 vanilla (kernel.org)
3210 Frequency : always on RHEL 3
3212 Description: Number of open files grows over time
3213 Details : The number of open files grows over time, whether or not
3214 Lustre is started. This was due to a filp leak introduced
3215 by one of our kernel patches.
3217 ------------------------------------------------------------------------------
3219 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3221 * Support for kernels:
3222 2.6.9-42.EL (RHEL 4)
3223 2.6.5-7.276 (SLES 9)
3224 2.4.21-40.EL (RHEL 3)
3225 2.6.12.6 vanilla (kernel.org)
3230 Bugzilla : 5719, 9635, 9792, 9684
3231 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3232 Details : If a server is under heavy load and cannot reply to new
3233 connection requests before the client resends the (re)connect,
3234 the connection handling code can behave badly if two service
3235 threads are concurrently handing separate (re)connections from
3236 the same client. Add better locking to the connection handling
3237 code, and ensure that only a single connection will be processed
3238 for a given client UUID, even if the lock is dropped.
3240 Severity : enhancement
3242 Description: add TCP zero-copy support to kernel
3243 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3244 sends if the hardware supports scatter-gather and checksumming.
3245 This allows socklnd to do client-write and server-read more
3246 efficiently and reduce CPU utilization from skbuf copying.
3249 Frequency : only if NFS exporting from client
3251 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3252 Details : If setting attributes on a file created under NFS that had
3253 never been opened it would be possible to oops the client
3254 if the file had no objects.
3257 Frequency : always for liblustre
3259 Description: liblustre client does MDS+OSTs setattr RPC for each write
3260 Details : When doing a write from a liblustre client, the client
3261 incorrectly issued an RPC to the MDS and each OST the file was
3262 striped over in order to update the timestamps. When writing
3263 with small chunks and many clients this could overwhelm the MDS
3264 with RPCs. In all cases it would slow down the write because
3265 these RPCs are unnecessary.
3267 Severity : enhancement
3269 Description: allow number of MDS service threads to be changed at module load
3270 Details : It is now possible to change the number of MDS service threads
3271 running. Adding "options mds mds_num_threads={N}" to the MDS's
3272 /etc/modprobe.conf will set the number of threads for the next
3273 time Lustre is restarted (assuming the "mds" module is also
3274 reloaded at that time). The default number of threads will
3275 stay the same, 32 for most systems.
3280 Description: OST crash if filesystem is unformatted or corrupt
3281 Details : If an OST is started on a device that has never been formatted
3282 or if the filesystem is corrupt and cannot even mount then the
3283 error handling cleanup routines would dereference a NULL pointer.
3288 Description: NULL pointer deref in llap_from_page.
3289 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3290 incorrect) mapping if the page was truncated/reclaimed while it was
3291 searched for. Check for this condition and skip such pages when
3292 doing readahead. Introduce extra check to llap_from_page() to
3293 verify page->mapping->host is non-NULL (so page is not anonymous).
3296 Frequency : Sometimes when using sys_sendfile
3298 Description: "page not covered by a lock" warnings from ll_readpage
3299 Details : sendfile called ll_readpage without right page locks present.
3300 Now we introduced ll_file_sendfile that does necessary locking
3301 around call to generic_file_sendfile() much like we do in
3305 Frequency : with certain MDS communication failures at client mount time
3307 Description: NULL pointer deref after failed client mount
3308 Details : a client connection request may delayed by the network layer
3309 and not be sent until after the PTLRPC layer has timed out the
3310 request. If the client fails the mount immediately it will try
3311 to clean up before the network times out the request. Add a
3312 reference from the request import to the obd device and delay
3313 the cleanup until the network drops the request.
3316 Frequency : occasionally during client (re)connect
3318 Description: assertion failure during client (re)connect
3319 Details : processing a client connection request may be delayed by the
3320 client or server longer than the client connect timeout. This
3321 causes the client to resend the connection request. If the
3322 original connection request is replied in this interval, the
3323 client may trip an assertion failure in ptlrpc_connect_interpret()
3324 which thought it would be the only running connect process.
3327 Frequency : only with obd_echo servers and clients that are rebooted
3329 Description: kernel BUG accessing uninitialized data structure
3330 Details : When running an obd_echo server it did not start the ping_evictor
3331 thread, and when a client was evicted an uninitialized data
3332 structure was accessed. Start the ping_evictor in the RPC
3333 service startup instead of the OBD startup.
3335 Severity : enhancement
3336 Bugzilla : 10193 (patchless)
3337 Description: Remove dependency on various unexported kernel interfaces.
3338 Details : No longer need reparent_to_init, exit_mm, exit_files,
3339 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3342 Frequency : rare (only users of deprecated and unsupported LDAP config)
3344 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3345 Details : LDAP apparently contains 'lustreName' attributes instead of
3346 'name'. A simple remapping of the name is sufficient.
3349 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3351 Description: waiting_locks_callback trips kernel BUG if client is evicted
3352 Details : Running with the dump_on_timeout debug flag turned on makes
3353 it possible that the waiting_locks_callback() can try to dump
3354 the Lustre kernel debug logs from an interrupt handler. Defer
3355 this log dumping to the expired_lock_main() thread.
3357 Severity : enhancement
3359 Description: Support NFS exporting on 2.6 kernels.
3360 Details : Implement non-rawops metadata methods for NFS server to use without
3361 changing NFS server code.
3364 Frequency : very rare (synthetic metadata workload only)
3366 Description: two racing renames might cause an MDS thread to deadlock
3367 Details : Running the "racer" program may cause one MDS thread to rename
3368 a file from being the source of a rename to being the target of
3369 a rename at exactly the same time that another thread is doing
3370 so, and the second thread has already enqueued these locks after
3371 doing a lookup of the target and is trying to relock them in
3372 order. Ensure that we don't try to re-lock the same resource.
3375 Frequency : only very large systems with liblustre clients
3377 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3378 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3379 all OSTs in parallel. This allows the eviction of stale liblustre
3380 clients to proceed much faster than if they were done in series,
3381 and also offers similar improvements for other set_info RPCs.
3386 Description: excessive CPU usage during initial read phase on client
3387 Details : During the initial read phase on a client, it would agressively
3388 retry readahead on the file, consuming too much CPU and impacting
3389 performance (since 1.4.5.8). Improve the readahead algorithm
3390 to avoid this, and also improve some other common cases (read
3391 of small files in particular, where "small" is files smaller than
3392 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3397 Description: MDS crash when receiving packet with unknown intent.
3398 Details : Do not LBUG in unknown intent case, just return -EFAULT
3400 Severity : enhancement
3401 Bugzilla : 9293, 9385
3402 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3403 Details : Do not serialize getattr (non-intent version) and statfs.
3406 Frequency : occasional, when OST network is overloaded/intermittent
3408 Description: client evicted by OST after bulk IO timeout
3409 Details : If a client sends a bulk IO request (read or write) the OST
3410 may evict the client if it is unresposive to its data GET/PUT
3411 request. This is incorrect if the network is overloaded (takes
3412 too long to transfer the RPC data) or dropped the OST GET/PUT
3413 request. There is no need to evict the client at all, since
3414 the pinger and/or lock callbacks will handle this, and the
3415 client can restart the bulk request.
3418 Frequency : Always when mmapping file with no objects
3420 Description: client crashes when mmapping file with no objects
3421 Details : Check that we actually have objects in a file before doing any
3422 operations on objects in ll_vm_open, ll_vm_close and
3428 Description: Request leak when working with deleted CWD
3429 Details : Introduce advanced request refcount tracking for requests
3430 referenced from lustre intent.
3432 Severity : Enhancement
3434 Description: Cache open file handles on client.
3435 Details : MDS now will return special lock along with openhandle, if
3436 requested and client is allowed to hold openhandle, even if unused,
3437 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3438 closing files for every read/write openration.
3440 Severity : Enhancement
3442 Description: Cache open negative dentries on client when possible.
3443 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3444 negative dentries on lock revocation.
3449 Description: Remounting a client read-only wasn't possible with a zconf mount
3450 Details : It wasn't possible to remount a client read-only with llmount.
3452 Severity : enhancement
3453 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3454 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3455 MPICH is included to add Lustre-specific ADIO interfaces.
3456 This is based closely on the UFS ADIO layer and only differs
3457 in file creation, in order to allow the OST striping to be set.
3458 This is user-contributed code and not supported by CFS.
3463 Description: extended inode attributes (immutable, append-only) work improperly
3464 when 2.4 and 2.6 kernels are used on client/server or vice versa
3465 Details : Introduce kernel-independent values for these flags.
3467 Severity : enhancement
3470 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3471 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3472 granularity. Now, Fractional MB granularity is be supported,
3473 this is very useful for low memory system.
3475 Severity : enhancement
3477 Description: Getattr by fid
3478 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3479 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3480 have locks and avoids getattr rpc requests for it on every stat.
3483 Frequency : Always, for filesystems larger than 2TB
3485 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3486 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3487 filesystems larger than 2TB to be mounted with mballoc enabled.
3490 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3492 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3493 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3494 filesystems larger than 2TB is mounted then the kernel will
3495 silently corrupt the start of the filesystem. CONFIG_LBD is
3496 enabled for all CFS-supported kernels, but the possibility of
3497 this happening with a modified kernel config exists.
3499 Severity : enhancement
3501 Description: add client O_DIRECT support for 2.6 kernels
3502 Details : It is now possible to do O_DIRECT reads and writes to files
3503 in the Lustre client mountpoint on 2.6 kernel clients.
3505 Severity : enhancement
3507 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3508 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3509 parallel, not waiting for response from every OST before sending
3510 a rpc to the next OST.
3515 Description: setattr vs write race when updating file timestamps
3516 Details : Client processes that update a file timestamp into the past
3517 right after writing to the file (e.g. tar) it is possible that
3518 the updated file modification time can be reset to the current
3519 time due to a race between processing the setattr and write RPC.
3521 Severity : enhancement
3523 Description: Bring 'lfs find' closer in line with regular Linux find.
3524 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3525 -print0 options and obtains all the needed info through the lustre
3528 Severity : enhancement
3530 Description: support up to 1024 configured devices on one node
3531 Details : change obd_dev array from statically allocated to dynamically
3532 allocated structs as they are first used to reduce memory usage
3537 Description: Flush dirty partially truncated pages during truncate
3538 Details : Immediatelly flush partially truncated pages in filter_setattr,
3539 this way we completely avoid having any pages in page cache on OST
3540 and can retire ugly workarounds during writes to flush such pages.
3545 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3546 Details : protect inode from truncation within vfs_unlink() context
3547 just take a reference before calling vfs_unlink() and release it
3548 when parent's i_sem is free.
3553 Description: last_id value checked outside lock on OST caused LASSERT failure
3554 Details : If there were multiple MDS->OST object precreate requests in
3555 flight, it was possible that the OST's last object id was checked
3556 outside a lock and incorrectly tripped an assertion. Move checks
3557 inside locks, and discard old precreate requests.
3560 Frequency : always, if extents are used on OSTs
3562 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3563 Details : OSTs using the extents format would not zero the high 16 bits of
3564 the index physical block number. This is not a problem for any
3565 OST filesystems smaller than 16TB, and no kernels support ext3
3566 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3567 new/modified files) and can be fixed for existing filesystems
3568 with e2fsprogs-1.39-cfs1.
3573 Description: import connection selection may be incorrect if timer wraps
3574 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3575 import connections to be ignored if the 32-bit jiffies counter
3576 wraps. Use a 64-bit jiffies counter.
3579 Frequency : very large clusters immediately after boot
3581 Description: LNET request buffers exhausted under heavy short-term load
3582 Details : If a large number of client requests are generated on a service
3583 that has previously never seen so many requests it is possible
3584 that the request buffer growth cannot keep up with the spike in
3585 demand. Instead of dropping incoming requests, they are held in
3586 the LND until the RPC service can accept more requests.
3589 Frequency : Sometimes during replay
3591 Description: Assertion failure in ll_local_open after replay.
3592 Details : If replay happened on an open request reply before we were able
3593 to set replay handler, reply will become not swabbed tripping the
3594 assertion in ll_local_open. Now we set the handler right after
3595 recognising of open request
3598 Frequency : very rare
3600 Description: kernel reports "badness in vsnprintf"
3601 Details : Reading from the "recovery_status" /proc file in small chunks
3602 may cause a negative length in lprocfs_obd_rd_recovery_status()
3603 call to vsnprintf() (which is otherwise harmless). Exit early
3604 if there is no more space in the output buffer.
3606 Severity : enhancement
3608 Description: clear OBD RPC statistics by writing to them
3609 Details : It is now possible to clear the OBD RPC statistics by writing
3610 to the "stats" file.
3615 Description: Client mtime is not the same on different clients after utimes
3616 Details : In some cases, the client was using the utimes() syscall on
3617 a file cached on another node. The clients now validate the
3618 ctime from the MDS + OSTs to determine which one is right.
3623 Description: Inability to activate failout mode
3624 Details : lconf script incorrectly assumed that in pythong string's numeric
3625 value is used in comparisons.
3628 Frequency : always with multiple stripes per file
3630 Description: Inefficient object allocation for mutli-stripe files
3631 Details : When selecting which OSTs to stripe files over, for files with
3632 a stripe count that divides evenly into the number of OSTs,
3633 the MDS is always picking the same starting OST for each file.
3634 Return the OST selection heuristic to the original design.
3639 Description: mount failures may take full timeout to return an error
3640 Details : Under some heavy load conditions it is possible that a
3641 failed mount can wait for the full obd_timeout interval,
3642 possibly several minutes, before reporting an error.
3643 Instead return an error as soon as the status is known.
3645 Frequency : quota enabled and large files being deleted
3647 Description: releasing more than 4GB of quota at once hangs OST
3648 Details : If a user deletes more than 4GB of files on a single OST it
3649 will cause the OST to spin in an infinite loop. Release
3650 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3655 Description: statfs data retrieved from /proc may be stale or zero
3656 Details : When reading per-device statfs data from /proc, in the
3657 {kbytes,files}_{total,free,avail} files, it may appear
3658 as zero or be out of date.
3661 Frequency : systems with MD RAID1 external journal devices
3663 Description: lconf's call to blkid is confused by RAID1 journal devices
3664 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3665 of returning all block devices that match the journal UUID.
3668 Frequency : always, for aggregate stripe size over 4GB
3670 Description: assertion fails when trying to use 4GB stripe size
3671 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3672 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3675 Frequency : always on ppc64
3677 Description: the first write on an ext3 filesystem with mballoc got stuck
3678 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3679 perform endianness conversion.
3681 ------------------------------------------------------------------------------
3683 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3685 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3686 INTEROPERATE with older versions automatically. Please read the
3687 user documentation before upgrading any part of a live system.
3688 * WARNING: Lustre networking configuration changes are required with
3689 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3692 * Support for kernels:
3693 2.6.9-22.0.2.EL (RHEL 4)
3694 2.6.5-7.244 (SLES 9)
3695 2.6.12.6 vanilla (kernel.org)
3698 Severity : enhancement
3699 Bugzilla : 7981/8208
3700 Description: Introduced Lustre Networking (LNET)
3701 Details : LNET is new networking infrastructure for Lustre, it includes
3702 a reorganized network configuration mode (see the user
3703 documentation for full details) as well as support for routing
3704 between different network fabrics. Lustre Networking Devices
3705 (LNDs) for the supported network fabrics have also been
3706 created for this new infrastructure.
3708 Severity : enhancement
3709 Description: Introduced Access control lists
3710 Details : clients can set ACLs on files and directories in order to have
3711 more fine-grained permissions than the standard Unix UGO+RWX.
3712 The MDS must be started with the "-o acl" mount option.
3714 Severity : enhancement
3715 Description: Introduced filesystem quotas
3716 Details : Administrators may now establish per-user quotas on the
3719 Severity : enhancement
3721 Description: Configuration change for the XT3
3722 The PTLLND is now used to run Lustre over Portals on the XT3
3723 The configure option(s) --with-cray-portals are no longer used.
3724 Rather --with-portals=<path-to-portals-includes> is used to
3725 enable building on the XT3. In addition to enable XT3 specific
3726 features the option --enable-cray-xt3 must be used.
3731 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3732 Details : A race between allocating a new llcd and re-getting the llcd_lock
3733 allowed another thread to grab newly-allocated llcd.
3735 Severity : enhancement
3737 Description: 2.6 OST async journal commit and locking fix to improve performance
3738 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3739 2.6 kernels are now async as they already were in 2.4 kernels so
3740 that they can commit concurrently with the network bulk transfer.
3741 For block-allocated files the filter allocation semaphore is held
3742 to avoid filesystem fragmentation during allocation. BKL lock
3743 removed for 2.6 xattr operations where it is no longer needed.
3748 Description: lconf incorrectly determined whether two IP networks could talk
3749 Details : In some more complicated routing and multiple-network
3750 configurations, lconf will avoid trying to make a network
3751 connection to a disjoint part of the IP space. It was doing the
3752 math incorrectly for one set of cases.
3757 Description: Fix for potential infinite loop processing records in an llog.
3758 Details : If an llog record is corrupted/zeroed, it is possible to loop
3759 forever in llog_process(). Validate the llog record length
3760 and skip the remainder of the block on error.
3763 Frequency : occasional (liblustre only)
3765 Description: liblustre could not open files whose last component is a symlink
3766 Details : sysio_path_walk() would incorrectly pass the open intent to
3767 intermediate path components.
3770 Frequency : rare (liblustre only with non-standard tuning)
3771 Bugzilla : 7201 (7350)
3772 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3773 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3774 after it has been released, only do this for referenced locks
3776 Severity : enhancement
3778 Description: specifying an (invalid) directory default stripe_size of -1
3779 would reset the directory default striping
3780 Details : stripe_size -1 was used internally to signal directory stripe
3781 removal, now use "all default" to signal dir stripe removal
3782 as a directory striping of "all default" is not useful
3785 Frequency : common for large clusters running liblustre clients
3787 Description: doing an ls when liblustre clients are running is slow
3788 Details : sending a glimpse AST to a liblustre client waits for every AST
3789 to time out, as liblustre clients will not respond. Since they
3790 cannot cache data we refresh the OST lock LVB from disk instead.
3792 Severity : enhancement
3794 Description: doing an ls at the same time as file IO can be slow
3795 Details : enqueue and other "small" requests can be blocked behind many
3796 large IO requests. Create a new OST IO portal for non-IO
3797 requests so they can be processed faster.
3800 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3802 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3803 created file with mode 000
3804 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3805 allow an ftruncate() as a non-root user to a file with mode 000.
3806 Lustre now respects this flag to disable mode checks when
3807 truncating a file owned by the user
3810 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3813 Description: Revoking locks from clients that went dead or catatonic might take
3815 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3816 cancellation of such locks instant on servers without waiting for
3817 any reply from clients. Clients drops these locks when cancel
3818 notification from server is received without replying.
3821 Frequency : liblustre-only, when liblustre client dies or becomes busy
3823 Description: Doing ls on Linux clients can take a long time with active
3825 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3826 granting such locks to it in the first place if possible. Locks
3827 are taken by proxy on the OST during the read or write and
3828 dropped immediately afterward. Add connect flags handling, do
3829 not grant locks to liblustre clients for glimpse ASTs.
3831 Severity : enhancement
3833 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3834 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3835 sequential reads and behaves badly with applications doing random
3836 reads. Improve it by reading ahead at least read region, and
3837 avoiding excessive large RPC for small reads.
3839 Severity : enhancement
3841 Description: Creating more than 1000 files for a single job may cause a load
3842 imbalance on the OSTs if there are also a large number of OSTs.
3843 Details : qos_prep_create() uses an OST index reseed value that is an
3844 even multiple of the number of available OSTs so that if the
3845 reseed happens in the middle of the object allocation it will
3846 still utilize the OSTs as uniformly as possible.
3851 Description: OST or MDS may oops in ping_evictor_main()
3852 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3853 but doesn't restart at beginning of obd_exports_timed list
3856 Severity : enhancement
3858 Description: improve by-nid export eviction on the MDS and OST
3859 Details : allow multiple exports with the same NID to be evicted at one
3860 time without re-searching the exports list.
3863 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3865 Description: MDS may oops in groups_free()
3866 Details : in rare race conditions a newly allocated group_info struct is
3867 freed again, and this can be NULL. The 2.4 compatibility code
3868 for groups_free() checked for a NULL pointer, but 2.6 did not.
3871 Frequency : common for liblustre clients doing little filesystem IO
3872 Bugzilla : 9352, 7313
3873 Description: server may evict liblustre clients accessing contended locks
3874 Details : if a client is granted a lock or receives a completion AST
3875 with a blocking AST already set it would not reply to the AST
3876 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3879 Frequency : lfs setstripe, only systems with more than 160 OSTs
3881 Description: unable to set striping with a starting offset beyond OST 160
3882 Details : llapi_create_file() incorrectly limited the starting stripe
3883 index to the maximum single-file stripe count.
3886 Frequency : LDAP users only
3888 Description: lconf did not handle in-kernel recovery with LDAP properly
3889 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3891 Severity : enhancement
3893 Description: bind OST threads to NUMA nodes to improve performance
3894 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3895 node and do their allocations there to localize memory access
3897 Severity : enhancement
3899 Description: llmount can determine client NID directly from Myrinet (GM)
3900 Details : the client NID code from gmnalnid was moved directly into
3901 llmount, removing the need to use this or specifying the
3902 client NID explicitly when mounting GM clients with zeroconf
3905 Frequency : if client is started with down MDS
3907 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3908 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3909 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3910 ptlrpc_import_delayed_req() because we didn't check if the
3911 request was interrupted, and we also didn't break out of the
3912 event loop if there was no timeout
3917 Description: data loss during non-page-aligned writes to a single file from
3918 both multiple nodes and multiple threads on one node at same time
3919 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3920 inconsistency led to false short-reads, that were cached and later
3921 used by ->prepare_write() to fill in partially written page,
3922 leading to data loss.
3925 Frequency : always, if lconf --abort_recovery used
3927 Description: lconf --abort_recovery fails with 'Operation not supported'
3928 Details : lconf was attempting to abort recovery on the MDT device and not
3931 Severity : enhancement
3933 Description: remove cleanup logs
3934 Details : replace lconf-generated cleanup logs with lustre internal
3935 cleanup routines. Eliminates the need for client-cleanup and
3938 Severity : enhancement
3940 Description: add support for EAs (user and system) on lustre filesystems
3941 Details : it is now possible to store extended attributes in the Lustre
3942 client filesystem, and with the user_xattr mount option it
3943 is possible to allow users to store EAs on their files also
3945 Severity : enhancement
3947 Description: Add possibility (config option) to show minimal available OST free
3949 Details : When compiled with --enable-mindf configure option, statfs(2)
3950 (and so, df) will return least minimal free space available from
3951 all OSTs as amount of free space on FS, instead of summary of
3952 free spaces of all OSTs.
3954 Severity : enhancement
3956 Description: do not expand extent locks acquired on OST-side
3957 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3958 by server: they are not cached anyway.
3961 Frequency : when mmap is used/binaries executed from Lustre
3963 Description: Unmmap pages before throwing them away from read cache.
3964 Details : llap_shrink cache now attempts to unmap pages before discarding
3965 them (if unmapping failed - do not discard). SLES9 kernel has
3966 extra checks that trigger if this unmapping is not done first.
3971 Description: lconf didn't resolve symlinks before checking to see whether a
3972 given mountpoint was already in use
3975 Frequency : when migrating failover services
3976 Bugzilla : 6395, 9514
3977 Description: When migrating a subset of services from a node (e.g. failback
3978 from a failover service node) the remaining services would
3979 time out and evict clients.
3980 Details : lconf --force (implied by --failover) sets the global obd_timeout
3981 to 5 seconds in order to quickly disconnect, but this caused
3982 other RPCs to time out too quickly. Do not change the global
3983 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3985 Severity : enhancement
3986 Frequency : if MDS is started with down OST
3987 Bugzilla : 9439,5706
3988 Description: Allow startup/shutdown of an MDS without depending on the
3989 availability of the OSTs.
3990 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3991 Add appropriate locking and lov-osc refcounts for safe
3992 cleaning. Add osc abort_inflight calls in case the
3993 synchronize never started.
3996 Frequency : occasional (Cray XT3 only)
3998 Description: root not authorized to access files in CRAY_PORTALS environment
3999 Details : The client process capabilities were not honoured on the MDS in
4000 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
4001 been accessed by an authorized user then root was able to access
4002 the file on the local client also. The root user capabilities
4003 are now allowed on the MDS, as this environment has secure UID.
4006 Frequency : occasional
4008 Description: ldiskfs "too long searching" message happens too often
4009 Details : A debugging message (otherwise harmless) prints too often on
4010 the OST console. This has been reduced to only happen when
4011 there are fragmentation problems on the filesystem.
4016 Description: Division by zero in statfs when all OSCs are inactive
4017 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
4018 lov_statfs divides by value returned by lov_get_stripecnt().
4022 Bugzilla : 9489, 3273
4023 Description: First write from each client to each OST was only 4kB in size,
4024 to initialize client writeback cache, which caused sub-optimal
4025 RPCs and poor layout on disk for the first writen file.
4026 Details : Clients now request an initial cache grant at (re)connect time
4027 and so that they can start streaming writes to the cache right
4028 away and always do full-sized RPCs if there is enough data.
4029 If the OST is rebooted the client also re-establishes its grant
4030 so that client cached writes will be honoured under the grant.
4035 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
4036 Details : Now I/O RPCs go to different portal number and (presumably) fast
4037 lock requests (and glimses) and other RPCs get their own service
4038 threads pool that should be able to service those RPCs
4041 Severity : enhancement
4043 Description: Ability to exchange lustre version between client and servers and
4044 issue warnings at client side if client is too old. Also for
4045 liblustre clients there is ability to refuse connection of too old
4047 Details : New 'version' field is added to connect data structure that is
4048 filled with version info. That info is later checked by server and
4052 Frequency : rare, liblustre only.
4053 Bugzilla : 9296, 9581
4054 Description: Two simultaneous writes from liblustre at offset within same page
4055 might proceed at the same time overwriting eachother with stale
4057 Details : I/O lock withing llu_file_prwv was released too early, before data
4058 actually was hitting the wire. Extended lock-holding time until
4059 server acknowledges receiving data.
4062 Frequency : extremely rare. Never observed in practice.
4064 Description: avoid generating lustre_handle cookie of 0.
4065 Details : class_handle_hash() generates handle cookies by incrementing
4066 global counter, and can hit 0 occasionaly (this is unlikely, but
4067 not impossible, because initial value of cookie counter is
4068 selected randonly). Value of 0 is used as a sentinel meaning
4069 "unassigned handle" --- avoid it. Also coalesce two critical
4070 sections in this function into one.
4072 Severity : enhancement
4074 Description: allow liblustre clients to delegate truncate locking to OST
4075 Details : To avoid overhead of locking, liblustre client instructs OST to
4076 take extent lock in ost_punch() on client's behalf. New connection
4077 flag is added to handle backward compatibility.
4079 Severity : enhancement
4080 Bugzilla : 4928, 7341, 9758
4081 Description: allow number of OST service threads to be specified
4082 Details : a module parameter allows the number of OST service threads
4083 to be specified via "options ost ost_num_threads={N}" in the
4084 OSS's /etc/modules.conf or /etc/modprobe.conf.
4088 Bugzilla : 6146, 9635, 9895
4089 Description: servers crash with bad pointer in target_handle_connect()
4090 Details : In rare cases when a client is reconnecting it was possible that
4091 the connection request was the last reference for that export.
4092 We would temporarily drop the export reference and get a new
4093 one, but this may have been the last reference and the export
4094 was just destroyed. Get new reference before dropping old one.
4096 Severity : enhancement
4097 Frequency : if client is started with failover MDS
4099 Description: Allow multiple MDS hostnames in the mount command
4100 Details : Try to read the configuration from all specified MDS
4101 hostnames during a client mount in case the "primary"
4104 Severity : enhancement
4106 Description: Stop sending data to evicted clients as soon as possible.
4107 Details : Check if the client we are about to send or are sending data to
4108 was evicted already. (Check is done every second of waiting,
4109 for which l_wait_event interface was extended to allow checking
4110 of exit condition at specified intervals).
4113 Frequency : rare, normally only when NFS exporting is done from client
4115 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4117 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4120 Frequency : always, if srand() is called before liblustre initialization
4122 Description: Liblustre uses system PRNG disturbing its usage by user application
4123 Details : Introduce internal to lustre fast and high-quality PRNG for
4124 lustre usage and make liblustre and some other places in generic
4125 lustre code to use it.
4127 Severity : enhancement
4128 Bugzilla : 9477, 9557, 9870
4129 Description: Verify that the MDS configuration logs are updated when xml is
4130 Details : Check if the .xml configuration logs are newer than the config
4131 logs stored on the MDS and report an error if this is the case.
4132 Request --write-conf, or allow starting with --old_conf.
4134 Severity : enhancement
4136 Description: Handle symlinks in the path when checking if Lustre is mounted.
4137 Details : Resolve intermediate symlinks when checking if a client has
4138 mounted a filesystem to avoid duplicate client mounts.
4143 Description: lconf can hit an error exception but still return success.
4144 Details : The lconf command catches the Command error exception at the top
4145 level script context and will exit with the associated exit
4146 status, but doesn't ensure that this exit status is non-zero.
4151 Description: failure of ptlrpc thread startup can cause oops
4152 Details : Starting a ptlrpc service thread can fail if there are a large
4153 number of threads or the server memory is very fragmented.
4154 Handle this without oopsing.
4157 Frequency : always, only if liblustre and non-default acceptor port was used
4159 Description: liblustre cannot connect to servers with non-default acceptor port
4160 Details : tcpnal_set_default_params() was not called and was therefore
4161 ignoring the environment varaible TCPNAL_PORT, as well as other
4162 TCPNAL_ environment variables
4167 Description: two objects could be created on the same OST for a single file
4168 Details : If an OST is down, in some cases it was possible to create two
4169 objects on a single OST for a single file. No problems other
4170 than potential performance impact and spurious error messages.
4174 Bugzilla : 5681, 9562
4175 Description: Client may oops in ll_unhash_aliases
4176 Details : Client dcache may become inconsistent in race condition.
4177 In some cases "getcwd" can fail if the current directory is
4183 Description: Inode refcounting problems in NFS export code
4184 Details : link_raw functions used to call d_instantiate without obtaining
4185 extra inode reference first.
4189 Bugzilla : 9942, 9903
4190 Description: Referencing freed requests leading to crash, memleaks with NFS.
4191 Details : We used to require that call to ll_revalidate_it was always
4192 followed by ll_lookup_it. Also with revalidate_special() it is
4193 possible to call ll_revalidate_it() twice for the same dentry
4194 even if first occurence returned success. This fix changes semantic
4195 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
4196 reference on a request referred from the intent.
4197 ll_intent_release() then releases such a request.
4200 Frequency : rare, normally benchmark loads only
4202 Description: unlinked inodes were kept in memory on the client
4203 Details : If a client is repeatedly creating and unlinking files it
4204 can accumulate a lot of stale inodes in the inode slab cache.
4205 If there is no other client load running this can cause the
4206 client node to run out of memory. Instead flush old inodes
4207 from client cache that have the same inode number as a new inode.
4210 Frequency : SLES9 2.6.5 kernel and long filenames only
4211 Bugzilla : 9969, 10379
4212 Description: utime reports stale NFS file handle
4213 Details : SLES9 uses out-of-dentry names in some cases, which confused
4214 the lustre dentry revalidation. Change it to always use the
4218 Frequency : rare, unless heavy write-truncate concurrency is continuous
4219 Bugzilla : 4180, 6984, 7171, 9963, 9331
4220 Description: OST becomes very slow and/or deadlocked during object unlink
4221 Details : filter_destroy() was holding onto the parent directory lock
4222 while truncating+unlinking objects. For very large objects this
4223 may block other threads for a long time and slow overall OST
4224 responsiveness. It may also be possible to get a lock ordering
4225 deadlock in this case, or run out of journal credits because of
4226 the combined truncate+unlink. Solution is to do object truncate
4227 first in one transaction without parent lock, and then do the
4228 final unlink in a new transaction with the parent lock. This
4229 reduces the lock hold time dramatically.
4232 Frequency : rare, 2.4 kernels only
4234 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
4235 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
4236 before unmounting their filesystem in order to avoid deadlock.
4237 The kernel_locked() function in 2.4 kernels only checks whether
4238 the kernel lock is held, not whether it is this process that is
4239 holding it as 2.6 kernels do.
4244 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
4245 Details : The client ptlrpc code may be trying to reconnect to a down
4246 server before a previous connection attempt has timed out.
4247 Increase the reconnect interval to be longer than the connection
4248 timeout interval to avoid sending duplicate connections to
4252 Frequency : echo_client brw_test command
4254 Description: fix echo_client to work with OST preallocated code
4255 Details : OST preallocation code (5137) didn't take echo_client IO path
4256 into account: echo_client calls filter methods outside of any
4257 OST thread and, hence, there is no per-thread preallocated
4258 pages and buffers to use. Solution: hijack pga pages for IO. As
4259 a byproduct, this avoids unnecessary data copying.
4263 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
4264 Description: Client can oops in mdc_commit_close() after open replay
4265 Details : It was possible for the MDS to return an open request with no
4266 transaction number in mds_finish_transno() if the client was
4267 evicted, but without actually returning an error. Clients
4268 would later try to replay that open and may trip an assertion
4269 Simplify the client close codepath, and always return an error
4270 from the MDS in case the open is not successful.
4273 Frequency : rare, 2.6 OSTs only
4275 Description: OST may deadlock under high load on fragmented files
4276 Details : If there was a heavy load and highly-fragmented OST filesystems
4277 it was possible to have all the OST threads deadlock waiting on
4278 allocation of biovecs, because the biovecs were not released
4279 until the entire RPC IO was completed. Instead, release biovecs
4280 as soon as they are complete to ensure forward IO progress.
4282 Severity : enhancement
4284 Description: Support for specifying external journal device at mount
4285 Details : If an OST or MDS device is formatted with an external journal
4286 device, this device major/minor is stored in the ext3 superblock
4287 and may not be valid for failover. Allow detecting and
4288 specifying the external journal at mount time.
4293 Description: Mounting an MDS with pending unlinked files may cause oops
4294 Details : target_finish_recovery() calls mds_postrecov() which returned
4295 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4296 considers this an error and immediately begins cleaning up the
4297 lov, just after starting the mds_lov process
4299 Severity : enhancement
4301 Description: Implement 'lfs df' to report actual free space on per-OST basis
4302 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4303 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4304 usage of objects; '-h' to report in human readable format.
4306 ------------------------------------------------------------------------------
4308 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
4315 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4316 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4317 at startup during orphan unlinking. Instead, skip buddy update
4318 before setup, as it will be regenerated after recovery is complete.
4323 Description: If an OST is inactive, its locks might reference stale inodes.
4324 Details : lov_change_cbdata() must iterate over all namespaces, even if
4325 they are inactive to clear inode references from the lock.
4327 Severity : enhancement
4328 Frequency : occasional, if non-standard max_dirty_mb used
4330 Description: Client will block write RPCs if not enough grant
4331 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4332 then the client will block writes while waiting for another RPC
4333 to complete instead of consuming its dirty limit. With change
4334 we get improved performance when max_dirty_mb is small.
4336 Severity : enhancement
4337 Bugzilla : 3389, 6253
4338 Description: Add support for supplementary groups on the MDS.
4339 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4340 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4341 lookups for user supplementary groups into a cache.
4345 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4346 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4347 for liblustre O_EXCL is enforced by MDS.
4351 Description: readdir never returns NULL in liblustre.
4352 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4353 dentry in struct dirent.
4357 Frequency : liblustre only, depends on application IO pattern
4358 Description: liblustre clients evicted if not contacting servers
4359 Details : Don't put liblustre clients into the ping_evictor list, so
4360 they will not be evicted by the pinger ever.
4362 Severity : enhancement
4364 Description: Add ability to evict clients by NID from MDS.
4365 Details : By echoing "nid:$NID" string into
4366 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4367 $NID would be instantly evicted from this MDS and from all active
4368 OSTs connected to it.
4372 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4373 Details : lookup_it_finish() used to query file size from OSTs that was not
4378 Description: service threads change working directory to that of init
4379 Details : Starting lustre service threads may pin the working directory
4380 of the parent thread, making that filesystem busy. Threads
4381 now change to the working directory of init to avoid this.
4385 Frequency : during shutdown only
4386 Description: shutdown with a failed MDS or OST can cause unmount to hang
4387 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4392 Frequency : 2.6 only
4393 Description: chmod/chown may include an extra supplementary group
4394 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4395 supplementary group and if none is specified this is used.
4398 Bugzilla : 5479 (6816)
4400 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4401 Details : If lookup is in progress on a file that is unlinked we might try
4402 to revalidate the inode and fail in revalidate after lookup is
4403 complete and ll_file_open() enqueues the open again but
4404 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4407 Frequency : always, if lconf --abort_recovery used
4409 Description: lconf --abort_recovery fails with 'Operation not supported'
4410 Details : lconf was attempting to abort recovery on the MDT device and not
4413 ------------------------------------------------------------------------------
4415 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4420 Frequency : rare (only unsupported configurations with a node running as an
4422 Bugzilla : 6514, 5137
4423 Description: Mounting a Lustre file system on a node running as an OST could
4425 Details : OSTs now preallocates memory needed to write out data at
4426 startup, instead of when needed, to avoid having to
4427 allocate memory in possibly low memory situations.
4428 Specifically, if the file system is mounted on on OST,
4429 memory pressure could force it to try to write out data,
4430 which it needed to allocate memory to do. Due to the low
4431 memory, it would be unable to do so and the node would
4432 become unresponsive.
4434 Severity : enhancement
4436 Description: Addition of lconf --service command line option
4437 Details : lconf now accepts a '--service <arg>' option, which is
4438 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4440 Severity : enhancement
4442 Description: Failover mode is now the default for OSTs.
4443 Details : By default, OSTs will now run in failover mode. To return to
4444 the old behaviour, add '--failout' to the lmc line for OSTs.
4446 Severity : enhancement
4448 Description: Health checks are now provided for MDS and OSTs
4449 Details : Additional detailed health check information on MSD and OSTs
4450 is now provided through the procfs health_check value.
4453 Frequency : occasional, depends on IO load
4455 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4456 after numerous create/delete cycles
4457 Details : The ext3 inode allocation policy would not allocate new inodes
4458 very well on the OSTs because there are no new directories
4459 being created. Instead we look for groups with free space if
4460 the parent directories are nearly full.
4465 Description: Network or server problems during mount may cause partially
4466 mounted clients instead of returning an error.
4467 Details : The config llog parsing code may overwrite the error return
4468 code during mount error handling, returning success instead
4474 Description: MDS can fail to allocate large reply buffers
4475 Details : After long uptimes the MDS can fail to allocate large reply
4476 buffers (e.g. zconf client mount config records) due to memory
4477 fragmentation or consumption by the buffer cache. Preallocate
4478 some large reply buffers so that these replies can be sent even
4479 under memory pressure.
4483 Frequency : rare (liblustre)
4484 Description: fsx running with liblustre complained that using truncate() to
4485 extend the file doesn't work. This patch corrects that issue.
4486 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4487 ATTR_SIZE and lsm use in llu_setattr_raw.
4491 Frequency : rare, only 2.6 kernels
4492 Description: Unusual file access patterns on the MDS may result in inode
4493 data being lost in very rare circumstances.
4494 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4495 code in the 2.6 kernel caused the inode and/or EA data not to
4496 be read from disk, causing single-file corruption.
4500 Frequency : rare, only 2.6 filesystems using extents
4501 Description: Heavy concurrent write and delete load may cause data corruption.
4502 Details : It was possible under high-load situations to have an extent
4503 metadata block in the block device cache from a just-unlinked
4504 file overwrite a newly-allocated data block. We now unmap any
4505 metadata buffers that alias just-allocated data blocks.
4509 Frequency : filesystems with default stripe_count larger than 77
4510 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4511 Details : lconf specifies an inode size of 4096 bytes when the default
4512 stripe_count is larger than 77. This conflicts with the default
4513 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4514 this case to avoid pinning too much memory for large EAs.
4516 ------------------------------------------------------------------------------
4518 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4523 Frequency : rare (extremely heavy IO load with hundreds of clients)
4525 Description: Client is evicted, gets IO error writing to file
4526 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4527 caused clients to be evicted for AST timeouts. The fixes in
4528 bug 5192 mean we no longer need to have such short AST timeouts
4529 so ldlm_timeout has been increased.
4532 Frequency : occasional during --force or --failover shutdown under load
4533 Bugzilla : 5949, 4834
4534 Description: Server oops/LBUG if stopped with --force or --failover under load
4535 Details : a collection of import/export refcount and cleanup ordering
4536 issues fixed for safer force cleanup
4539 Frequency : only filesystems larger than 120 OSTs
4540 Bugzilla : 5990, 6223
4541 Description: lfs getstripe would oops on a very large filesystem
4542 Details : lov_getconfig used kfree on vmalloc'd memory
4545 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4546 Bugzilla : 6242, 6243
4547 Description: reading from files that had been truncated to a non-zero size
4548 but never opened returned no data
4549 Details : ll_file_read() reads zeros from no-object files to EOF
4554 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4555 Details : There's little harm in aborting MDS/OSS recovery and letting it
4556 try again, so I removed the LASSERT and return an error instead.
4558 Severity : enhancement
4560 Description: New debugging infrastructure for tracking down data corruption
4561 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4562 (b) cover more of the client-side code path, and (c) try to narrow
4563 down where problems occurred
4567 Bugzilla : 3819, 4364, 4397, 6313
4568 Description: Racing close and eviction MDS could cause assertion in mds_close
4569 Details : It was possible to get multiple mfd references during close and
4570 client eviction, leading to one thread referencing a freed mfd.
4572 Severity: : enhancement
4573 Bugzilla : 3262, 6359
4574 Description: Attempts to reconnect to servers are now more aggressive.
4575 Details : This builds on the enhanced upcall-less recovery that was added
4576 in 1.4.2. When trying to reconnect to servers, clients will
4577 now try each server in the failover group every 10 seconds. By
4578 default, clients would previously try one server every 25 seconds.
4583 Description: After recovery, certain operations trigger a failed
4584 assertion on a client.
4585 Details : Failing over an mds, using lconf -d --failover, while a
4586 client was doing a readdir() call would cause the client to
4587 LBUG after recovery completed and the readdir() was resent.
4589 Severity : enhancement
4591 Description: Default groups are now added by lconf
4592 Details : You can now run lconf --group <servicename> without having to
4593 manually add groups with lmc.
4596 Frequency : occasional
4598 Description: Nodes with an elan id of 0 trigger a failed assertion
4601 Frequency : always when accessing e.g. tty/console device nodes
4603 Description: tty and some other devices nodes cannot be used on lustre
4604 Details : file's private_data field is used by device data and lustre
4605 values in there got lost. New field was added to struct file to
4606 store fs-specific private data.
4609 Frequency : when exporting Lustre via NFS
4611 Description: NFSD failed occasionally when looking up a path component
4612 Details : NFSD is looking up ".." which was broken in ext3 directories
4613 that had grown large enough to become hashed.
4616 Frequency : Clusters with multiple interfaces not on the same subnet
4618 Description: Nodes will repeatedly try to reconnect to an interface which it
4619 cannot reach and report an error to the log.
4620 Details : Extra peer list entries will be created by lconf with some peers
4621 unreachable. lconf now validates the peer before adding it.
4624 Frequency : Only if a default stripe is set on the filesystem root.
4626 Description: Setting a default stripe on the filesystem root prevented the
4627 filesystem from being remounted.
4628 Details : The client was sending extra request flags in the root getattr
4629 request and did not allocate a reply buffer for the dir EA.
4632 Frequency : occasional, higher if lots of files are accessed by one client
4633 Bugzilla : 6159, 6097
4634 Description: Client trips assertion regarding lsm mismatch/magic
4635 Details : While revalidating inodes the VFS looks up inodes with ifind()
4636 and in rare cases can find an inode that is being freed.
4637 The ll_test_inode() code will free the lsm during ifind()
4638 when it finds an existing inode and then the VFS later attaches
4639 this free lsm to a new inode.
4643 Bugzilla : 6422, 7030
4644 Description: MDS deadlock between mkdir and client eviction
4645 Details : Creating a new file via mkdir or mknod (starting a transaction
4646 and getting the ns lock) can deadlock with client eviction
4647 (gets ns lock and trying to finish a synchronous transaction).
4650 Frequency : occasional
4651 Description: While starting a server, the fsfilt_ext3 module could not be
4653 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4654 kernels. Previously, lconf would still use the ext3 name
4655 when trying to load modules. Now, it will correctly use
4656 ext3 on 2.4 and ldiskfs on 2.6.
4658 Severity : enhancement
4659 Description: The default stripe count has been changed to 1
4660 Details : The interpretation of the default stripe count (0, to lfs
4661 or lmc) has been changed to mean striping across a single
4662 OST, rather than all available. For general usage we have
4663 found a stripe count of 1 or 2 works best.
4665 Severity : enhancement
4666 Description: Add support for compiling against Cray portals.
4667 Details : Conditional compiling for some areas that are different
4671 Frequency : occasional
4672 Bugzilla : 6409, 6834
4673 Description: Creating files with an explicit stripe count may lead to
4674 a failed assertion on the MDS
4675 Details : If some OSTs are full or unavailable, creating files may
4676 trigger a failed assertion on the MDS. Now, Lustre will
4677 try to use other servers or return an error to the
4681 Frequency : occasional
4683 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4684 caused lock timeout during readahead (since 1.4.2).
4685 Details : Processes doing readahead might match a lock that hasn't been
4686 granted yet if there are overlapping and conflicting lock
4687 requests. The readahead process waits on ungranted lock
4688 (original lock is CBPENDING), while OST waits for that process
4689 to cancel CBPENDING read lock and eventually evicts client.
4691 Severity : enhancement
4693 Description: Initial enabling of flock support for clients
4694 Details : Implements fcntl advisory locking and file status functions.
4695 This feature is provided as an optional mount flag (default
4696 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4697 locking are implemented yet, and those that are are not guaranteed
4698 to be completely correct in production environments.
4699 mount -t lustre -o [flock|noflock] ...
4702 Frequency : occasional
4704 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4705 assertion in the ext3 JBD (journaling) layer.
4706 Details : The b_committed_data struct is protected by the big kernel lock
4707 in 2.4 kernels, serializing journal_commit_transaction() and
4708 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4709 access to this struct. In 2.6 kernels there is finer grained
4710 locking to improve SMP performance of the JBD layer.
4714 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4719 Description: merge and cleanup kernel patches.
4722 -----------------------------------------------------------------------------
4724 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4726 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4727 versions of Lustre. Please update all servers and clients to
4728 version 1.4.2 or later at the same time. You must also run
4729 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4730 to update the configuration logs.
4732 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4733 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4734 - swab llog records of type '0' so we get proper header size/idx (5861)
4735 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4736 - fix rename of one directory over another leaking an inode (5953)
4737 - avoid SetPageDirty on 2.6 (5981)
4738 - don't re-add just-being-destroyed locks to the waiting list (5653)
4739 - when creating new directories, inherit the parent's custom
4740 striping settings if present parent (3048)
4741 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4742 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4743 - don't hold client locks on temporary worklist from l_lru (5666)
4744 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4745 - automatically evict dead clients (5921)
4746 - Update file size properly in create+truncate+fstat case (6196)
4747 - Do not unhash mountpoint dentries, do not allow removal of
4749 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4750 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4751 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4752 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4753 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4754 - Correctly handle reads of files with no objects (6243)
4755 - lctl recover will also mark a device active if deactivate used (5933)
4757 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4758 - allow --write-conf on an MDS with different nettype than client (5619)
4759 - don't write config llogs to MDS for mounts not from that MDS (5617)
4760 - lconf should create multiple TCP connections from a client (5201)
4761 - init scripts are now turned off by default; run chkconfig --on
4762 lustre and chkconfig --on lustrefs to use them
4763 - upcalls are no longer needed for clients to recover to failover
4765 - add --abort-recovery option to lconf to abort recovery on device
4767 - add support for an arbitrary number of OSTs (3026)
4768 - Quota support protocol changes.
4769 - forward compatibility changes to wire structs (6007)
4770 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4771 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4772 - improved "lustre status" script
4773 - initialize blocksize for non-regular files (6062)
4774 - added --disable-server and --disable-client configure options (5782)
4775 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4776 - Vanilla 2.4.29 support
4777 - increase maximum number of obd devices to 520 (6242)
4778 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4779 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4780 SLES 9 2.6.5 kernels
4781 - limit stripes per file to 160 (the maximum EA size) (6093)
4783 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4786 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4787 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4788 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4789 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4790 - small ext3 extents cleanups and fixes (5733)
4791 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4792 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4793 - avoid needless client->OST connect, fix handle mismatch (5317)
4794 - fix DLM error path that led to out-of-sync client, long delays (5779)
4795 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4796 - fix several locking issues related to i_size (5492,5624,5654,5672)
4797 - don't move pending lock onto export if it is already evicted (5683)
4798 - fix kernel oops when creating .foo in unlinked directory (5548)
4799 - fix deadlock in obdfilter statistics vs. object create (5811)
4800 - use time_{before,after} to avoid timer jiffies wrap (5882)
4801 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4802 - Do not leak request if server was not able to process it (5154)
4803 - If mds_open unable to find parent dir, make that negative lookup(5154)
4804 - don't create new directories with extent-mapping (5909, 5936)
4806 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4807 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4808 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4809 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4810 - support for RHEL4 kernel on i686 (5773)
4811 - provide error messages when incompatible logs are encountered (5898)
4813 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4814 * version 1.4.0.10 (1.4.1 release candidate 1)
4816 - don't keep a lock reference when lock is not granted (4238)
4817 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4818 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4819 - drop import inflight refcount on signal_completed_replay error (5255)
4820 - unlock page after async write error during send (3677)
4821 - handle missing objects in filter_preprw_read properly (5265)
4822 - no transno return for symlink open, don't save no-trasno open (3440)
4823 - don't try to complete elan receive that already failed (4012)
4824 - free RPC server reply state on error (5406)
4825 - clean up thread from ptlrpc_start_thread() on error (5160)
4826 - readahead could read extra page into cache that wasn't ejected (5388)
4827 - prevent races in class_attach/setup/cleanup/detach (5260)
4828 - don't dereference de->d_inode after l_dput of de (5458)
4829 - use "int" for stripe value returned from lock_to_stripe (5544)
4830 - mballoc allocation and error-checking fixes in 2.6 (5504)
4831 - block device patches to fix I/O request sizes in 2.6 (5482)
4832 - look up hostnames for IB nals (5602)
4833 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4834 - don't start multiple acceptors for the same port (5277)
4835 - fix incorrect LASSERT in mds_getattr_name (5635)
4836 - export a proc file for general "ping" checking (5628)
4837 - fix "lfs check" to not block when the MDS is down (5628)
4839 - service request history (4965)
4840 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4841 - create an "evict_client" /proc entry on OSTs, like the MDS has
4842 - fix mount usage message, return errors per mount(8) (5168)
4843 - change grep [] to grep "[]" in tests so they work in more UMLs
4844 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4845 - remove ancient LOV_MAGIC_V0 EA support (5047)
4846 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4847 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4848 - allow readahead allocations to fail when low on memory (5383)
4849 - mmap locking landed again, after considerable improvement (2828)
4850 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4852 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4855 - send OST transaction number in read/write reply to free req (4966)
4856 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4857 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4858 - the watchdog thread now runs as interruptible (5246)
4859 - flock/lockf fixes (but it's still disabled, pending 5135)
4860 - don't use EXT3 constants in llite code (5094)
4861 - memory shortage at startup could cause assertion (5176)
4863 - reorganization of lov code
4864 - single portals codebase
4866 - add extents/mballoc support (5025)
4867 - direct I/O reads in the obdfilter (4048)
4868 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4870 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4873 - send OST transaction number in read/write reply to free req (4966)
4874 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4875 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4876 - the watchdog thread now runs as interruptible (5246)
4877 - handle missing objects in filter_preprw_read properly (5265)
4878 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4879 - drop import inflight refcount on signal_completed_replay error (5255)
4880 - unlock page after async write error during send (3677)
4881 - return original error code on reconstructed replies (3761)
4882 - no transno return for symlink open, don't save no-trasno open (3440)
4884 - add pid to ldlm debugging output (4922)
4885 - bump the watchdog timeouts -- we can't handle 30sec yet
4886 - extra debugging for orphan dentry/inode bug (5259)
4888 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4891 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4892 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4893 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4894 - fix NULL dereference in /proc/sys/portals/routes (4827)
4895 - allow failed mdc_close() operations to be interrupted (4561)
4896 - stop precreate on OST before MDS would time out on it (4778)
4897 - don't send partial-page writes before EOF from client (4410)
4898 - discard client grant for sub-page writes on large-page clients (4520)
4899 - don't free dentries not owned by NFS code, check generation (4806)
4900 - fix lsm leak if mds_create_objects() fails (4801)
4901 - limit debug_daemon file size, always print CERROR messages (4789)
4902 - use transno after validating reply (3892)
4903 - process timed out requests if import state changes (3754)
4904 - update mtime on OST during writes, return in glimpse (4829)
4905 - add mkfsoptions to LDAP (4679)
4906 - use ->max_readahead method instead of zapping global ra (5039)
4907 - don't interrupt __l_wait_event() during strace
4909 - add software watchdogs to catch hung threads quickly (4941)
4910 - make lustrefs init script start after nfs is mounted
4911 - fix CWARN/ERROR duplication (4930)
4912 - return async write errors to application if possible (2248)
4913 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4914 - print NAL number in %x format (4645)
4915 - update barely-supported suse-2.4.21-171 series (4842)
4916 - support for sles 9 %post scripts
4917 - support for building 2.6 kernel-source packages
4918 - support for sles km_* packages
4920 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4923 - ignore -ENOENT errors in osc_destroy (3639)
4924 - notify osc create thread that OSC is being cleaned up (4600)
4925 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4926 - reconstruct ost_handle() like mds_handle() (4657)
4927 - create a new thread to do import eviction to avoid deadlock (3969)
4928 - let lconf resolve symlinked-to devices (4629)
4929 - don't unlink "objects" from directory with default EA (4554)
4930 - hold socknal file ref over connect in case target is down (4394)
4931 - allow more than 32000 subdirectories in a single directory (3244)
4932 - fix blocks count for O_DIRECT writes (3751)
4933 - OST returns ENOSPC from object create when no space left (4539)
4934 - don't send truncate RPC if file size isn't changing (4410)
4935 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4936 - bind to privileged port in socknal and tcpnal (3689)
4938 - rate limit CERROR/CWARN console message to avoid overload (4519)
4939 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4940 - basic mmap support (3918)
4941 - kernel patch series update from b1_4 (4711)
4943 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4946 - avoid crash during MDS cleanup with OST shut down (2775)
4947 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4948 - don't use bad inodes on the MDS (3744)
4949 - dynamic object preallocation to improve recovery speed (4236)
4950 - don't hold spinlock over lock dumping or change debug flags (4401)
4951 - don't zero obd_dev when it is force cleaned (3651)
4952 - print grants to console if they go negative (4431)
4953 - "lctl deactivate" will stop automatic recovery attempts (3406)
4954 - look for existing locks in ldlm_handle_enqueue() (3764)
4955 - don't resolve lock handle twice in recovery avoiding race (4401)
4956 - revalidate should check working dir is a directory (4134)
4958 - don't always mark "slow" obdfilter messages as errors (4418)
4960 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4963 - don't close LustreDB during write_conf until it is done (3860)
4964 - fix typo in lconf for_each_profile (3821)
4965 - allow dumping logs from multiple threads at one time (3820)
4966 - don't allow multiple threads in OSC recovery (3812)
4967 - fix debug_size parameters (3864)
4968 - fix mds_postrecov to initialize import for llog ctxt (3121)
4969 - replace config semaphore with spinlock (3306)
4970 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4971 - don't allow enqueue to complete on a destroyed export (3822)
4972 - down write_lock before checking llog header bitmap (3825)
4973 - recover from lock replay timeout (3764)
4974 - up llog sem before sending rpc (3652)
4975 - reduce ns lock hold times when setting kms (3267)
4976 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4977 - fix NULL deref and obd_dev leak on setup error (3312)
4978 - replace some LBUG about llog ops with error handling (3841)
4979 - don't match INVALID dentries from d_lookup and spin (3784)
4980 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4981 - fix invalid assertion in ptlrpc_set_wait (3880)
4983 - add libwrap support for the TCP acceptor (3996)
4984 - add /proc/sys/portals/routes for non-root route listing (3994)
4985 - allow setting MDS UUID in .xml (2580)
4986 - print the stack of a process that LBUGs (4228)
4988 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4991 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4992 - make sure to unset replay flag from failed open requests (3440)
4993 - if default stripe count is 0, use OST count for inode size (3636)
4994 - update parent mtime/ctime on client for create/unlink (2611)
4995 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4996 - free recovery state on server during a forced cleanup (3571)
4997 - unregister_reply for resent reqs (3063)
4998 - loop back devices mounting and status check on 2.6 (3563)
4999 - fix resource-creation race that can provoke i_size == 0 (3513)
5000 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
5001 - more debugging for page-accounting assertion (3746)
5002 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
5003 - avoid deadlock after precreation failure (3758)
5004 - fix race and lock order deadlock in orphan handling (3450, 3750)
5005 - add validity checks when grabbing inodes from l_ast_data (3599)
5007 - add /proc/.../recovery_status to obdfilter (3428)
5008 - lightweight CDEBUG infrastructure, debug daemon (3668)
5009 - change default OSC RPC parameters to be better on small clusters
5010 - turn off OST read cache for files smaller than 32MB
5011 - install man pages and include them in rpms (3100)
5012 - add new init script for (un)mounting lustre filesystems (2593)
5013 - run chkconfig in %post for init scripts (3701)
5014 - drop scimac NAL (unmaintained)
5016 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
5019 - clean kiobufs before and after use (3485)
5020 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
5021 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
5022 - change init script to fail more clearly if not run as root (1528)
5023 - allow clients to reconnect during replay (1742)
5024 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
5025 - don't do DNS lookups on NIDs too small for IP addresses (3442)
5026 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
5027 - fix cond_resched (3554)
5028 - only evict unfinished clients after recovery (3515)
5029 - allow bulk resend, prevent data loss (3570)
5030 - dynamic ptlrpc request buffer allocation (2102)
5031 - don't allow unlinking open directory if it isn't empty (2904)
5032 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
5033 - remove extraneous obd dereference causing LASSERT failure (3334)
5034 - don't use get_cycles() when creating temp. files on the mds (3156)
5035 - hold i_sem when setting i_size in ll_extent_lock() (3564)
5036 - handle EEXIST for set-stripe, set proper directory name (3336)
5038 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
5039 - fix ksocknal_fmb_callback() error messages (2918)
5041 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
5044 - don't copy lvb into (possibly NULL) reply on error (2983)
5045 - don't deref dentry after dput, don't free lvb on error (2922)
5046 - use the kms to determine writeback rpc length (2947)
5047 - increment oti_logcookies when osc is inactive (2948)
5048 - update client's i_blocks count via lvb messages (2543)
5049 - handle intent open/close of special files properly (1557)
5050 - mount MDS with errors=remount-ro, like obdfilter (2009)
5051 - initialize lock handle to avoid ASSERT on error cleanup (3057)
5052 - don't use cancelling-locks' kms values (2947)
5053 - use highest lock extent for kms, not last one (2925)
5054 - don't dereference ERR_PTR() dentry in error handling path (3107)
5055 - fix thread race in portals_debug_dumplog() (3122)
5056 - create lprocfs device entries at setup instead of at attach (1519)
5057 - common AST error handler, don't evict client on completion race (3145)
5058 - zero nameidata in detach_mnt in 2.6 (3118)
5059 - verify d_inode after revalidate_special is valid in 2.6 (3116)
5060 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
5061 - initialize RPC timeout timer earlier for 2.6 (3219)
5062 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
5063 - print nal/nid for unknown nid (3258)
5064 - additional checks for oscc recovery before doing precreate (3284)
5065 - fix ll_extent_lock() error return code for 64-bit systems (3043)
5066 - don't crash in mdc_close for bad permissions on open (3285)
5067 - zero i_rdev for non-device files (3147)
5068 - clear page->private before handing to FS, better assertion (3119)
5069 - tune the read pipeline (3236)
5070 - fix incorrect decref of invalidated dentry (2350)
5071 - provide read-ahead stats and refine rpc in flight stats (3328)
5072 - don't hold journal transaction open across create RPC (3313)
5073 - update atime on MDS at close time (3265)
5074 - close LDAP connection when recovering to avoid server load (3315)
5075 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
5076 - don't leak open file on MDS after open resend (3325)
5077 - serialize filter_precreate and filter_destroy_precreated (3329)
5078 - loop device shouldn't call sync_dev() for nul device (3092)
5079 - clear page cache after eviction (2766)
5080 - resynchronize MDS->OST in background (2824)
5081 - refuse to mount the same filesystem twice on same mountpoint (3394)
5082 - allow llmount to create routes for mounting behind routers (3320)
5083 - push lock cancellation to blocking thread for glimpse ASTs (3409)
5084 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
5085 - fix rare problem with rename on htree directories (3417)
5087 - allow default OST striping configuration per directory (1414)
5088 - fix compilation for qswnal for 2.6 kernels (3125)
5089 - increase maximum number of MDS request buffers for large systems
5090 - change liblustreapi to be useful for external progs like lfsck (3098)
5091 - increase local configuration timeout for slow disks (3353)
5092 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
5094 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
5097 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
5098 - don't overwrite extent policy data in reply if lock was blocked (2901)
5099 - drop filter export grants atomically with removal from device (2663)
5100 - del obd_self_export from work_list in class_disconnect_exports (2908)
5101 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
5102 - swab reply message in mdc_close, other PPC fixes (2464)
5103 - fix destroying of named logs (2325)
5104 - overwrite old logs when running lconf --write_conf (2264)
5105 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
5106 - fix race in target_handle_connect (2898)
5107 - mds_reint_create() should take same inode create lock (2926)
5108 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
5109 - don't close files for self_export to avoid uninitialized obd (2936)
5110 - allow MDS with the same name as client node (2939)
5111 - hold dentry reference for closed log files for unlink (2325)
5112 - reserve space for all logs during transactions (2059)
5113 - don't evict page beyond end of stripe extent (2925)
5114 - don't oops on a deleted current working directory (2399)
5115 - handle hard links to targets without a parent properly (2517)
5116 - don't dereference NULL lock when racing during eviction (2867)
5117 - don't grow lock extents when lots of conflicting locks (2919)
5119 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
5122 - account for cache space usage on clients to avoid data loss (974)
5123 - lfsck support in lustre kernel code (2349)
5124 - reduce journal credits needed for BRW writes (2370)
5125 - orphan handling to avoid losing space on client/server crashes
5126 - ptlrpcd can be blocked, stopping ALL progress (2477)
5127 - use lock value blocks to assist in proper KMS, faster stat (1021)
5128 - takes i_sem instead of DLM locks internally on obdfilter (2720)
5129 - recovery for initial connections (2355)
5130 - fixes for mds_cleanup_orphans (1934)
5131 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
5132 - block all file creations until orphan recovery completes (1901)
5133 - client remove rq_connection from request struct (2423)
5134 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
5135 - recovery timer race (2670)
5136 - mdc_close recovey bug (2532)
5137 - ptlrpc cleanup bug (2710)
5138 - mds timeout on local locks (2588)
5139 - namespace lock held during RPCs (2431)
5140 - handle interrupted sync write properly (2503)
5141 - don't try to handle a message that hasn't been replied to (2699)
5142 - client assert failure during cleanup after abort recovery (2701)
5143 - leak mdc device after failed mount (2712)
5144 - ptlrpc_check_set allows timedout requests to complete (2714)
5145 - wait for inflight reqs when ptlrpcd finishes (2710)
5146 - make sure unregistered services are removed from the srv_list
5147 - reset bulk XID's when resending them (caught by 1138 test)
5148 - unregister_bulk after timeout
5149 - fix lconf error (2694)
5150 - handle write after unfinished setstripe, stripe-only getstripe (2388)
5151 - readahead locks pages, leaves pending causing memory pressure (2673)
5152 - increase OST request buffers to 4096 on large machines (2729)
5153 - fix up permission of existing directories in simple_mkdir (2661)
5154 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
5155 - don't assign transno to errored transactions (2742)
5156 - don't delete objects on OST if given a bogus objid from MDS (2751)
5157 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
5158 - if rq_no_resend, then timeout request after recovery (2432)
5159 - fix MDS llog_logid record size, 64-bit array alignment (2733)
5160 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
5161 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
5162 - MDS recovery shouldn't delete objects that it has given out (2730)
5163 - if enqueue arrives after completion, don't clobber LVB (2819)
5164 - don't unlock pages twice when trigger_group_io returns error (2814)
5165 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
5166 - don't write pages to disk if there was an error (1450)
5167 - don't ping imports that have recovery disabled (2676)
5168 - take buffered bytes into account when balancing socknal conn (2817)
5169 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
5170 - reconnect unlink llog connection after MDS reconnects to OST (2816)
5171 - remove little-endian swabbing of llog records (1987)
5172 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
5173 - retry reposting request buffers if they fail (1191)
5174 - grow extent at grant time to avoid granting a revoked lock (2809)
5175 - lock revoke doesn't evict page if covered by a second lock (2765)
5176 - disable VM readahead to avoid reading outside lock extents (2805)
5178 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
5179 - updated kernel patches for hp-2.4.20 kernel (2681)
5181 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
5184 - fix truncated write corruption (2366)
5185 - fix for failed assertion in iopen_connect_dentry (1792,2517)
5187 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
5188 - ensure len(uuid) < 37 in lmc (1171)
5189 - fix ia64 OOPS in llog_test (2255)
5190 - zero end of page at obdfilter for partial page writes (2648)
5191 - don't leave stale dentries around after renames (bug 2428)
5192 - fix timeouts when evicting a client with a single lock held (2642)
5193 - set deadline for the initial HELLO message to drain (2634)
5194 - print out dotted-quad IP addresses in the socknal (2302)
5196 - additional debugging for MDS client eviction problem (2443)
5197 - fix mkfsoptions support for osts (2603, 2604)
5199 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
5202 - add series for the vanilla 2.6.0 kernel
5203 - add series for the vanilla 2.4.24 kernel
5204 - add series for a cray x86/64 UL kernel drop
5205 - fix xattr patches for the vanilla 2.4.19 series
5207 - generate true UUIDs in lmc (1171)
5208 - have portals stack dumping break in UML (2466)
5209 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
5210 - allocate with _NOFS in ldlm to avoid deadlock (1933)
5211 - wake callback waiting threads on client eviction (2460)
5212 - Add --ptldebug and --subsystem to lmc (1719)
5213 - update assertion to allow safe interrupt allocation
5214 - set rq_no_resend for cancel requests (2432)
5215 - recalculate ptlrpcd timeout after resend (2494)
5216 - call vfs_rmdir when removing pending directories (2368)
5217 - fix renaming a file to itself (2429)
5218 - lmc creates a default one-stripe lov (2454)
5219 - expand procfs space to handle large clusters (2326)
5220 - increase UML stack to avoid overflow
5221 - update lconf's list of debug and subsystem masks
5222 - fix lfs find --obd (2510)
5223 - /proc tunable for disabling filter read caching (2591)
5224 - stop rpm packages from altering slapd.conf (2301)
5225 - disable nagle in the socknal under 0conf (2578)
5226 - choose mds inode size based on stripe count (2572)
5227 - fix kernel-source rpm problems (2516)
5229 - add --disable-doc to avoid pdf generation (2421)
5230 - update documentation, tests, type-os, comments
5231 - avoid format warnings on ia64
5232 - remove the TOE NAL
5233 - tiny code cleanups by removing unused fields
5235 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
5238 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
5239 - remove the most blatant lies from BUILDING file (2371)
5240 - change default debug level to reasonable production setting
5241 - reduce client side cache size to reduce cache flush time
5242 - reduce max RPCs in flight to avoid unnecessary file fragmentation
5243 - make TCP zerocopy and pinger support enabled by default (2476)
5244 - sync writes completed after process exits caused crashes (2319)
5245 - maintain correct mount count on the MDS (2356)
5246 - backout 1557, because 2316 wasn't really fixed
5247 - better file I/O statistics gathering in /proc
5248 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
5249 - another kernel patch to fix zero-copy TCP function export
5250 - don't take duplicate lock when processing re-sent getattr (2420)
5251 - lctl uses obd_self_export instead of creating new conn (2353)
5252 - MDS/OST recovery case which requires object creation asserted (2425)
5253 - move lfs from /usr/sbin to /usr/bin in packages
5254 - fix race between mds_client_add and mds_client_free (2417)
5255 - use kmalloc instead of slabs in portals (2430)
5256 - don't create duplicate records when a failover MDS is present (2442)
5257 - remove unnecessary mount age check (2332)
5258 - don't remove directory inodes from locks prematurely (2451)
5259 - don't break if MDS service name is the same as hostname (2103)
5260 - fix races in client write RPC generation when cache full (2482)
5262 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
5265 - remove now-unused request->rq_obd (278)
5266 - if an allocation fails, print out how much memory we've used (1933)
5267 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
5268 - add the "configurable stack size" patch to most series files (1256)
5269 - ability to write large log records, for 100+ OST configs (2306)
5270 - fix NULL deref when filter_prep fails (2314)
5271 - fix operator precedence error in filter_sync
5272 - dynamic allocation of socknal TX descriptors (2315)
5273 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
5274 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
5275 - ensure that CWARN messages in Portals always get to the syslog
5276 - __init/__exit are not for prototype decls (ldlm_init/exit)
5277 - x86-64 compile warning fixes
5278 - fix gateway LMC keyword conflict (2318)
5279 - fix MDS lock inversions in getattr/reint paths (1844)
5280 - fix a rare lock re-ordering bug, which caused deadlock (2322)
5281 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
5282 - DLM race condition prevented some lock evictions (2328)
5283 - ENOMEM detection and retry on socknal sends (2230)
5284 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
5285 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
5286 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
5287 - better error messages when a client is rejected during recovery (1505)
5288 - avoid cancelling locks which were never granted, after failure (2330)
5289 - fix i_sem/journal inversion in mds_client_add (2333)
5290 - fix truncate/getattr lock cycle deadlock (2334)
5291 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
5292 - fix two rare exit paths which could leak an l_lock() ref (2321)
5293 - fencepost error in MDS/OST orphan recovery (2226)
5294 - make log record alignment 8 bytes (1988)
5295 - lstripe now fails when requested offset > ost_count (2237)
5296 - ensure that all kernel series have a complete list.h (1607)
5297 - fix crashes in special-file operations (2316)
5298 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5300 - allow configurable automake binary, for testing new versions
5301 - small update to the lfs documentation
5303 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5305 * fix negative export reference count in fsfilt_sync (2312)
5307 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5308 * release candidate 0.9.1
5310 - orphans are moved into the PENDING directory for possible recovery
5311 - replayed opens now open by fid for orphan/rename safety (1042)
5312 - last close of an orphan inode generates a transno (683)
5313 - chdir() and mount() now pin the directory entry (1020)
5314 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5315 - discard very old requests without processing them (1502)
5316 - remove some common, well-understood CERRORs (1505)
5317 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5318 - clear "grant" flags in OST replies until OST grant code lands (1644)
5319 - fix read performance by not clobbering i_blksize on client (1598)
5320 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5321 - make LRU size a /proc tunable, clears locks when reduced (707)
5322 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5323 - server locks take references on exports now (1558)
5324 - build fixes for 2.4.20-rh trees (1663)
5325 - return an error from lov_create if all OSCs are inactive (1751)
5326 - fix import levels when a reconnect happens without a timeout (1597)
5327 - exit early from mds_open if we get a lookup error (1749)
5328 - partial page read at EOF wouldn't wait for disk before sending (1642)
5329 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5330 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5331 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5332 - write/truncate lock inversion (1639)
5333 - Don't auto-load obdclass, portals modules during cleanup (1495)
5334 - fix timestamps from jumping to "now" (1763)
5335 - extra journal assertions (1648)
5336 - add an extra multiunlink test (1771)
5337 - fix read_record/write_record API (1776)
5338 - fix leak of offset_extent, possible incorrect i_size later (1772)
5339 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5340 - Debugging for the kqswnal_get_idle_tx problems (1820)
5341 - Allow recovery to be attempted multiple times (1536)
5342 - Write out MDS last_rcvd file after it is first created (1600)
5343 - Fix tx_descriptor leak in failed transmit situations (1827)
5344 - ext3 journaling fixes for assertion failure after IO error (1871)
5345 - class_export_put() on freed export after completion AST error (1896)
5346 - Fix revalidate looping in VFS (1322)
5347 - Don't access a freed export during MDS_REINT timeout (1521)
5348 - Add open-unlink recovery support on the MDS (1673,1764)
5349 - Return an error if no MDS data was read from last_rcvd (1946)
5350 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5351 - Don't setup a disk device that doesn't match exported UUID (317)
5352 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5353 - avoid committing NULL handle in force close
5354 - local.sh is now a one-stripe LOV configuration
5355 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5356 - fix ext3 htree duplicate directory entry corruption (1516)
5357 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5358 - update to new LOV EA format (2097)
5359 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5361 2003-06-15 Phil Schwan <phil@clusterfs.com>
5364 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5365 - per-import/export recovery handling (958, 931, 959)
5366 - multiple last-rcvd slots, for serving multiple FSes (949)
5367 - connections are again shared between multiple imp/exports (963, 964)
5368 - "umount -f" would hang if any requests needed to be sent (393, 978)
5369 - avoid pinning large req buffer by copying for queued messages (989)
5370 - add "uuid" to "lctl device" command to help upcalls (991)
5371 - "open" RPCs with transnos would confuse recovery counters (1037)
5372 - do proper endian conversion of all wire messages (288, 340, 891)
5373 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5374 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5375 - fix ptlrpc_connection leak in target_handle_connect (1174)
5376 - fix import refcounting bug in OST and MDS cleanup (1134)
5377 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5378 - fix very unlikely obd_types race condition (501)
5379 - remove osc_open hack for echo_client (1187)
5380 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5381 - a failure in read_inode2 leads to deadlock (1139)
5382 - cancel ack-locks as soon as transaction is committed (1072)
5383 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5384 - make sure to commitrw after any preprw to avoid deadlock (1162)
5385 - failing to execute a file in a lustre FS would lock inode (1203)
5386 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5387 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5388 - obd_uuid2tgt would walk off the end of the list (1255)
5389 - on IA64 the osc would give portals incorrect bulk size (1258)
5390 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5391 - fix lock inversion caused by new llite matching code (1282)
5392 - limit the number of dirty pages on a client to 10MB (1286)
5393 - timed out locks were not being corrected cancelled (1289)
5394 - fix O_DIRECT above 4GB on IA-32 (1292)
5395 * major user-visible changes
5396 - fail out/fail over policy now controlled by the upcall (993)
5398 - add OBD_PING to check server availability and failure (954)
5399 - lustre messages are now sent in sending host order (288, 340, 891)
5400 - add eadatalen to MDS getattr reply (340)
5401 - OST read replies may contain second buffer, with per-page status (593)
5403 2003-03-11 Phil Schwan <phil@clusterfs.com>
5406 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5407 - failed open()s could cause deadlock; fixed (867, 869)
5408 - stop cancelling OST locks when files are closed (481)
5409 - overlapping XID spaces caused network corruption (851, 853)
5410 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5411 - setattr_raw added, to do single-RPC, server-side setattrs
5412 - lmc/lconf syntax change for OST UUIDs
5413 - fix crashy race condition between ptlrpc_free_req and osc_close
5414 - don't use request in mdc_enqueue if we hit a timeout (889)
5415 - don't set the inode i_size for regular files from the MDS (896)
5416 - handle out of order completion AST (842)
5417 - don't LBUG if a lock request times out after receiving AST (913)
5418 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5419 - if a bad lock AST arrives, send an error instead of dropping entirely
5420 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5421 - fix leak in bulk IO when only partially completed (899, 900, 926)
5422 - fix O_DIRECT for ia64 (55)
5423 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5424 - C-z after timeout could hang a process forever; fixed (977)
5426 - client-side I/O cache (678, 924, 929, 941, 970)
5428 - READPAGE and SETATTRs which don't take server-side locks get
5431 2003-02-11 Phil Schwan <phil@clusterfs.com>
5434 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5435 - Fix fsfilt_extN_readpage() to read a full page of directory
5436 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5437 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5438 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5439 - Most intent operations were removed, in favour of a new RPC mode
5440 that does a single RPC to the server and bypasses most of the VFS
5441 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5442 - Aggressively cancel local locks on DLM servers
5443 - mds_reint_unlink sends EA to the client if it's the last nlink.
5444 client uses that EA to unlink OST objects.
5445 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5446 - recursive symlinks were fixed (439)
5447 - fixed NULL deref in DEBUG_REQ
5448 - filter_update_lastobjid no longer calls sync, which annoyed extN
5449 - fixed multi-client small-writes to a single file problem (445)
5450 - fixed mtime updates during file writes (607)
5451 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5452 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5453 - fixed Linux OST size reporting problem (444, 656)
5454 - OST now updates object mtime with writes or setattr (607, 619)
5455 - client verifies file size before zeroing page past EOF (445)
5456 - OST now writes last allocated objid to disk with allocation (108)
5457 - LOV on echo now works (409)
5459 - mds_reint_unlink sends a new buffer, with the EA included. this
5460 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5461 set if a regular file was being unlinked, and it was the last link
5462 - use PtlGet from the target for bulk writes (315)
5463 - OST now updates object mtime with writes or setattr (607, 619)
5464 - LDLM now has a grant-time callback to revalidate locked items, if
5466 - Many MDS operations were reorganized to combat race conditions
5468 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5469 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5471 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5474 - Fully reactivate OST imports after reconnection (512, others)
5475 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5476 - More graceful error handling for truncating on dead OST (515)
5477 - Don't error out unless we're actually accessing dead stripes (474)
5478 - Fix garbage sizes when stripes are missing (410)
5479 - LRU counters were broken, causing constant lock purge (433, 432)
5480 - garbage on read from stripes with failed OSTs (441)
5481 - mark OSCs as active before reconnecting during recovery (438)
5482 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5483 - lfind did not preserve OST order in output (443)
5484 - symlinks cause hung clients, incorrect data (439)
5485 - stop dereferencing request after dropping refcount (457)
5486 - don't LASSERT(spin_is_locked) on non-SMP (455)
5487 - fixes for many rename() bugs
5488 - fstat didn't correctly synchronize attributes (399)
5489 - server must handle lock cancellation during blocking AST prep (487)
5490 - bulk descriptors were free()d too soon (511)
5491 - fix paths in lconf, which would load incorrect modules (451, 507)
5492 - fix confusing lconf 'host not found' error message (386)
5493 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5494 - fix race condition in mdc_blocking_ast() for inode access (526)
5495 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5496 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5497 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5498 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5499 - fix lstripe in several ways (kernel side) (527)
5500 - fix request leak in ldlm_cli_enqueue (262)
5501 - incorrect OSC was marked inactive after OST failure
5502 - call mds_fs_cleanup before unmounting filesystem (524)
5503 - fix races between taking ns_lock and ldlm_lock_change_resource
5504 - fix races updating LOV export open file list
5505 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5506 - fix recovery NULL deref in ldlm_cli_cancel_unused
5507 - fix some DLM races by using new hash table for lock handles (419)
5508 - permit the client to specify desired inodes, at replay
5509 - duplicate requests when we queue them for replay reintegration
5510 - fix last_rcvd offset calculation
5511 - sync after each recovered transaction, so we always make progress
5512 - never, not always, ERESTART requests without transnos
5513 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5515 - skip replay if the MDS says that the client is already connected
5516 - don't check for a recovery-enabled export to match lctl's UUID
5517 - don't INC_USE_COUNT for phantom exports
5518 - don't crash when cleaning up phantom exports (567)
5519 - don't double-finish or set replay data for errored mdc_open requests
5520 - abort requests when they time out, so we don't get old replies
5521 - send/receive replies for AST messages again
5522 - if the client says that it doesn't have the lock, cancel it on the
5524 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5525 mark it as destroyed, it will all work out when decref is called
5526 - fix module use counts (22, 581)
5528 - ASTs now expect a reply (server cancels lock on error reply)
5530 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5533 - fix many simultaneous client startup (392)
5534 - fix dentry->d_it clobbering
5535 - credentials weren't being shipped for readdir/getattr operations
5536 - remove invalid assertions triggered during some concurrent MD
5538 - proper Lustre versions added (336, 389)
5539 - fix memory leak for create error case (398)
5540 - fix LOV locking bug that would get cli/srv out of sync
5541 - fix echo client over LOV (409)
5542 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5543 - fix double-O_EXCL intent crash (424)
5544 - avoid sending multiple lock CANCELs (352)
5546 - MDS can do multi-client recovery (modulo bugs in new code)
5548 - many updates, edits, cleanups
5550 2002-11-18 Phil Schwan <phil@clusterfs.com>
5553 - fix null d_it dereference (346)
5554 - fix full OST/dbench hang (333)
5555 - fix permission problem with file removal (286)
5556 - fix removal of OSCs from LOV when they fail
5557 - fix NULL deref during bulk timeout (214)
5558 - fix problems related to multiple filesystems on one MDS (241)
5559 - fixed serious subtle metadata locking bugs
5560 - free locks on clients when inodes are removed due to memory
5562 - fix inode pointer in lock data (285)
5563 - partial support for multiple MDS on a single host (241)
5564 - data locks weren't cancelled at clear_inode time (290, 311)
5565 - intent locks could lead to unbounded lock growth (205)
5566 - added a maximum lock count, an LRU list, and a flusher
5567 - fix multiple rename (365)
5568 - properly abstracted the echo client
5569 - OSC locked 1 byte too many; fixed
5570 - rewrote brw callback code:
5571 - fixed recovery bugs related to LOVs (306)
5572 - fixed too-many-pages-in-one-write crash (191)
5573 - fixed (again) crash in sync_io_timeout (214)
5574 - probably fixed callback-related race (385)
5576 - Add capability to MDS protocol
5577 - LDLM cancellations and callbacks on different portals
5579 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5582 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5583 - defer open object destruction to close time (601981, 138)
5584 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5585 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5586 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5587 - don't call obd_setattr in ll_file_release for destroyed objects
5588 * protocol change to lustre_msg: move |version| and add |flags|
5589 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5590 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5591 * added replay of create, unlink, link and rename operations during
5592 MDS failover; recovery should be much more robust now
5593 * remove failed OSCs from LOVs (only lov_create uses this so far)
5594 * the lustre-HOWTO was brought (more) up to date (582544)
5596 2002-10-23 Phil Schwan <phil@clusterfs.com>
5599 - in-use dentries weren't being reused properly (617851)
5600 - prevent multiple LDLM setup (599178)
5601 - fix LOV size calculations for truncate (617853)
5602 - fix client handling of MDS intent errors (POSIX)
5603 - fix permission bug in lovstripe.c test (624321)
5604 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5605 - truncate past end of file could corrupt data
5606 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5607 - a race in recovery could return ETIMEDOUT to apps (623947)
5608 - building outside the source directory was fixed
5609 * the lustre-HOWTO was brought (more) up to date (582544)
5610 * major progress was made on recovery functionality
5612 2002-10-10 Phil Schwan <phil@clusterfs.com>
5615 - recovery deadlock fix
5616 - rm -rf causes LBUG fix (617817)
5617 - file open by multiple tasks fix (618962)
5618 - directory permissions bugs (602707 and 620007)
5619 - journal_stop fixed with locking (611313)
5620 - O_APPEND failures resolved (618273, perhaps 614459)
5621 - lconf PATH fix (619770)
5622 - IA64 build fix (621450)
5623 - RPC buffer sizes scale with amount of memory
5625 2002-10-01 Phil Schwan <phil@clusterfs.com>
5628 - locks would be cancelled without throwing away data pages,
5629 resulting in inconsistent data (605627)
5630 - inode attributes were not always being refreshed (605627, 612449)
5631 - lconf now continues to cleanup after lctl reports an error
5632 - MDS now enforces user permissions (602707)
5633 - lprocfs cleanup fixed, but not yet enabled (614157)
5634 - fixed infinite server hang, should a client not respond to an AST
5635 - avoid going into recovery if user calls readlink() with a buffer
5636 that's too small (613941)
5637 - AST RPCs no longer require replies (614867) -- this may be changed
5638 - don't crash server if client sends an IOV that's too big (611336)
5639 - fixed lock conversion deadlock (611892)
5640 - fixed the following of symlinks (614622)
5641 * recovery: the server can remove locks from a client that dies, other
5642 clients can make progress
5643 * more extN patch fixes
5644 * compile-time configurable ptlrpc buffer allocations
5646 - collaborative read cache document
5647 - Lustre Lite Performance CDR document-in-progress
5649 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5652 - fix typo in patch-2.4.18
5654 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5657 - clear ptlrpc request each time in handle_incoming_request()
5658 - unlink of files now destroys the object on the OST
5660 2002-09-19 Peter Braam <braam@clusterfs.com>
5662 * add hard link support
5663 * change obdfile creation method
5664 * kernel patch changed
5666 2002-09-19 Peter Braam <braam@clusterfs.com>
5669 - stack overflow bug in extN fixed
5671 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5673 * documentation updates
5674 - add man pages for config tools
5675 - update tests/README to describe testing with new config tools
5676 - finish metadata API descriptions
5677 * bug fixes and cleanups
5678 - statfs workaround for 16TB limit
5679 - LOV stripe allocation improved, can stripe on subset of OSTs
5680 - LOV file size/IO offset was wrong for files > 4GB in size
5681 - object EA data was being dropped, caused files to be unreadable
5682 - memory overflow with non-LOV OST caused memory corruption
5683 - fixed regression tests to work with new config tools, obdfilter
5684 - fixed bug when directory size became larger than 1 block
5685 - fixed bug (for single client case) when PWD was deleted
5686 - invalidate local directory pages when doing intent-based ops
5687 - avoid LDLM oops when lock callback contained bad data
5689 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5691 * documentation updates
5692 * bug fixes and cleanups
5693 - configuration tools
5696 - 64-bit compile warnings
5697 - 64-bit internal statfs data
5699 * test_brw on persistent OST devices
5701 * lprocfs (disabled)
5703 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5705 * documentation updates
5706 * bug fixes and cleanups
5707 * configuration tools
5709 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5712 * many small fixes to 0_5_4
5713 * io/network handling
5714 * thinkos in MDS operations
5716 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5719 * crucial basic fixes to 0.5.3
5720 * IOR, Iozone work over Elan
5723 2002-08-07 Phil Schwan <phil@clusterfs.com>
5724 * version 0_5_3, our first alpha
5725 * we use the new Portals iovs
5726 * documentation updates
5727 * bug fixes and cleanups
5728 * small changes in the DLM wire protocol
5730 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5731 * version 0_5_1 with some initial stability,
5732 * locking on MD and file I/O.
5733 * documentation updates
5734 * several bug fixes since 0.5.0
5735 * small changes in wire protocol
5737 2002-07-18 Phil Schwan <phil@clusterfs.com>
5739 * delivered as Lustre Light Alpha
5740 * fixed a crash after handling invalid MDS requests
5741 * fixed directory pages for architectures with non-4k pages sizes
5743 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5744 * release version v0_4_4
5745 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5747 * Fixes a number of interruption problems with OST operations.
5748 * Update documentation for portals header changes
5749 * Move all wire protocol structs/defines to lustre_idl.h
5750 * Fixes symlink length bug.
5751 * Add tcpdump to repository.
5753 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5754 * release version v0_4_3
5755 * Fixes statfs for inodes on extN.
5756 * Fixes bug in runtests which would delete /etc/hosts.
5757 * Use 64-bit object IDs wherever possible (not into VFS though)
5758 Remove ost_get_info, which is unused by lustre, and out of date.
5760 2002-07-03 Peter Braam <braam@clusterfs.com>
5761 * release version v0_4_2 Fixes a lookup error (type not passed)
5762 * move forward to head of Portals
5763 * move forward to latest Lustre kernel
5765 2002-06-25 Peter Braam <braam@clusterfs.com>
5766 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob