1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.31 (SLES 10),
5 2.6.18-92.1.22.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.
16 Severity : enhancement
18 Description: Update to RHEL5 kernel-2.6.18-92.1.22.el5.
21 Frequency : start MDS on uncleanly shutdowned MDS device
23 Descriptoin: aborting recovery hang on MDS
24 Details : don't throttle destroy RPCs for the MDT.
27 Frequency : on remount
29 Description: external journal device not working after the remount
30 Details : clear dev_rdonly flag for external journal devices in
34 Frequency : race on file read and write
36 Description: Lustre doesn't delete files
37 Details : Clients drop lock reference and release openhandle when they find
43 Description: shutdown vs evict race
44 Details : client_disconnect_export vs connect request race.
45 if client will evicted at this time - we start invalidate
46 thread without referece to import and import can be freed
50 Frequency : rare, need acl's on inode.
52 Description: client can't handle ost additional correctly
53 Details : if ost was added after client connected to mds client can have
54 hit lnet_try_match_md ... to big messages to wide striped files.
55 in this case need teach client to handle config events about add
56 lov target and update client max ea size at that event.
58 Severity : enhancement
60 Description: Changelogs
61 Details : Changelogs are a lightweight mechanism to track filesystem
62 metadata and namespace changes. The changelog is recorded
63 permanently on the MDTs, and is periodically "consumed" / purged
64 when records are no longer needed.
66 Severity : enhancement
68 Description: compact fld format with extents
69 Details : Store range of seq rather than every seq in FLD. Seq
70 controller update FLD rather than clients. In Case of CMD, mdt0
71 has FLD, all other metadata server act as non persistent proxy
72 for FLD queries and cache fld entries in fld cache.
77 Description: don't skip ost target if they assigned to file
78 Details : Drop slow OSCs if we can, but not for requested start idx.
79 This means "if OSC is slow and it is not the requested
80 start OST, then it can be skipped, otherwise skip it only
81 if it is inactive/recovering/out-of-space.
85 Description: more cleanup in mds_lov
86 Details : not send LOV EA under replay, we can't know about they size at this
87 time. Don't allow client connect to mds before any ost connected,
88 for avoid problems with LOV EA size and returning EIO to client.
90 Severity : enhancement
92 Description: Interoperability at server side (Disk interoperability)
94 Severity : enhancement
96 Description: Update to RHEL5 kernel-2.6.18-92.1.17.el5.
98 Severity : enhancement
100 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.31.
102 Severity : enhancement
104 Description: New client IO stack (CLIO).
106 Severity : enhancement
108 Description: Commit on sharing. Eliminate inter-client dependencies between
109 uncommitted transactions by doing transaction commits.
110 Thereby clients may recovery independently.
113 Frequency : Create a symlink file with a very long name
115 Description: ldlm_cancel_pack()) ASSERTION(max >= dlm->lock_count + count)
116 Details : If there is no extra space in the request for early cancels,
117 ldlm_req_handles_avail() returns 0 instead of a negative value.
119 Severity : enhancement
121 Description: Add /proc entry for import status
122 Details : The mdc, osc, and mgc import directories now have
123 an import directory that contains useful import data for debugging
126 Severity : enhancement
128 Description: Re-disable certain /proc logging
129 Details : Enable and disable client's offset_stats, extents_stats and
130 extents_stats_per_process stats logging on the fly.
133 Frequency : Only on FC kernels 2.6.22+
135 Description: oops in statahead
136 Details : Do not drop reference count for the dentry from VFS when lookup,
137 VFS will do that by itself.
139 Severity : enhancement
141 Description: Generic /proc file permissions
142 Details : Set /Proc file permissions in a more generic way to enable non-
143 root users operate on some /proc files.
147 Description: Hitting mdc_commit_close() ASSERTION
148 Details : Properly handle request reference release in
149 ll_release_openhandle().
153 Description: quota recovery deadlock during mds failover
154 Details : This patch includes att18982, att18236, att18237 in bz14840.
156 1. fix osts hang when mds does failover with quotaon
157 2. prevent watchdog storm when osts threads wait for the
162 Frequency : only patchless client
163 Description: add workaround for race between add/remove dentry from hash
165 Severity : enhancement
167 Description: Allow OST glimpses to return PW locks
171 Description: LBUG when llog conf file is full
172 Details : When llog bitmap is full, ENOSPC should be returned for plain log.
176 Description: Prevent import from entering FULL state when server in recovery
180 Description: service mount cannot take device name with ":"
181 Details : Only when device name contains ":/" will mount treat it as
187 Description: replace ptlrpcd with the statahead thread to interpret the async
188 statahead RPC callback
192 Frequency : on recovery
193 Description: I/O failures after umount during fail back
194 Details : if client reconnected to restarted server we need join to recovery
195 instead of find server handler is changed and process self
196 eviction with cancel all locks.
198 Severity : enhancement
200 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
202 Severity : enhancement
204 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
206 Severity : enhancement
208 Description: Upcall on Lustre log has been dumped
209 Details : Allow for a user mode script to be called once a Lustre log has
210 been dumped. It passes the filename of the dumped log to the
211 script, the location of the script can be specified via
212 /proc/sys/lnet/debug_log_upcall.
217 Description: avoid idr_remove called for id which is not allocated.
218 Details : Move assigment s_dev for clustered nfs to end of initialization,
219 to avoid problem with error handling.
224 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
225 Details : When connection is reused this not moved from CONN_UNUSED_HASH
226 into CONN_USED_HASH and this prodice warning when put connection
227 again in unused hash.
230 Severity : enhancement
232 Description: File striping can now be set to use an arbitrary pool of OSTs.
234 Severity : enhancement
236 Description: Export bytes_read/bytes_write count on OSC/OST.
240 Description: Early reply size mismatch, MGC loses connection
241 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
242 the connect flags are properly negotiated.
247 Description: quotas are not honored with O_DIRECT
248 Details : all writes with the flag O_DIRECT will use grants which leads to
249 this problem. Now using OBD_BRW_SYNC to guard this.
253 Description: add quota statistics
254 Details : 1. sort out quota proc entries and proc code.
255 2. add quota statistics
257 Severity : enhancement
259 Description: enable quota support for HEAD.
263 Description: Properly propagate oinfo flags from lov to osc for statfs
264 Details : restore missing copy oi_flags to lov requests.
266 Severity : enhancement
268 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
269 : l_getgroups(8), lst(8), routerstat(8)
270 Details : included man pages for llobdstat(8), llstat(8),
271 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
273 Severity : enhancement
275 Description: configure's --enable-quota should check the
276 : kernel .config for CONFIG_QUOTA
277 Details : configure is terminated if --enable-quota is passed but
278 : no quota support is in kernel
282 Description: Remove portals compatibility
283 Details : Remove portals compatibility, not interoperable with releases
288 Description: Resolve device initialization race
289 Details : Prevent proc handler from accessing devices added to the
290 obd_devs array but yet be intialized.
292 Severity : enhancement
294 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
296 Severity : enhancement
298 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
303 Description: Using wrong pointer in osc_brw_prep_request
304 Details : Access to array[-1] can produce panic if kernel compiled with
305 CONFIG_PAGE_ALLOC enabled
309 Description: Client runs out of low memory
310 Details : Consider only lowmem when counting initial number of llap pages
314 Description: *optional* service tags registration
315 Details : if the "service tags" package is installed on a Lustre node
316 When the filesystem is mounted, a local-node service tag will
317 be created. See http://inventory.sun.com/ for more information
318 about the Service Tags asset management system.
322 Description: Kernel BUG tries to release flock
323 Details : Lustre does not destroy flock lock before last reference goes
324 away. So always drop flock locks when client is evicted and
325 perform unlock regardless of successfulness of speaking to MDS.
329 Description: add refcount protection for osc callbacks, avoid panic on shutdown
333 Description: sanity test 65a fails if stripecount of -1 is set
334 Details : handle -1 striping on filesystem in ll_dirstripe_verify
339 Description: ASSERTION(CheckWriteback(page,cmd)) failed
340 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
343 Severity : enhancement
345 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
349 Description: do not process already freed flock
350 Details : flock can possibly be freed by another thread before it reaches
351 to ldlm_flock_completion_ast.
355 Description: LBUG during stress test
356 Details : Need properly lock accesses the flock deadlock detection list.
360 Description: oops in page fault handler
361 Details : kernel page fault handler can return two special 'pages' in error
362 case, don't try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
366 Description: timeout with invalidate import.
367 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which
368 should be handled by ptlrpcd. This produce long age waiting and
369 -ETIMEOUT ptlrpc_invalidate_import and as result LASSERT.
371 Severity : enhancement
373 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
377 Description: enable MGS and MDT services start separately
378 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
379 a co-located MGS without starting the MGS, which is a complement
380 to 'nosvc' mount option.
384 Frequency : after recovery
385 Description: precreate to many object's after del orphan.
386 Details : del orphan st in oscc last_id == next_id and this triger growing
387 count of precreated objects. Set flag LOW to skip increase count
388 of precreated objects.
392 Frequency : rare, on clear nid stats
393 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
394 Details : when clean nid stats sometimes try destroy live entry,
395 and this produce panic in free.
399 Description: Stack overflow during MDS log replay
400 ease stack pressure by using a thread dealing llog_process.
404 Description: wait until IO finished before start new when do lock cancel.
405 Details : VM protocol want old IO finished before start new, in this case
406 need wait until PG_writeback is cleared until check dirty flag and
407 call writepages in lock cancel callback.
409 Severity : enhancement
411 Description: using special macro for print time and cleanup in includes.
415 Description: mds_mfd_close() ASSERTION(rc == 0)
416 Details : In mds_mfd_close(), we need protect inode's writecount change
417 within its orphan write semaphore to prevent possible races.
421 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
425 Frequency : rare, on shutdown ost
426 Description: don't hit live lock with umount ost.
427 Details : shrink_dcache_parent can be in long loop with destroy dentries,
428 use shrink_dcache_sb instead.
432 Description: don't panic with use echo client
433 Details : echo client pass NULL as client nid pointer and this produce null
438 Description: fix build on ppc32
439 Details : compile code with -m64 flag produce wrong object file for ppc32.
443 Description: add message levels for liblustreapi
447 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
448 Details : Move the 'good_osts' check before the 'total_bavail' check. This
449 will result in an -EAGAIN and in the exit call path we call
450 alloc_rr() which will with increasing aggressiveness attempt to
451 aquire precreated objects on the minimum number of required OSCs.
455 Description: Use old size assignment to avoid deadlock
456 Details : Reverts the changes in bugs 2369 and bug 14138 that introduced
457 the scheduling while holding a spinlock. We do not need locking
458 for size in ll_update_inode() because size is only updated from
459 the MDS for directories or files without objects, so there is no
460 other place to do the update, and concurrent access to such inodes
461 are protected by the inode lock.
465 Description: resolve "_IOWR redefined" build error on SLES10
469 Description: dump the memory debugging after all modules are unloaded to
470 suppress false negative in conf_sanity test 39
472 Severity : enhancement
474 Description: build kernel-ib packages for OFED 1.3 in our release cycle
479 Description: fix SLES kernel versioning
480 Details : the kernel version for our SLES 10 kernel did not include a "-"
481 before the "smp" at the end. while this was not a problem in
482 general, it did mean that software trying to use the kernel
483 version to try to detect a vendor specific kernel would fail.
484 this was most evident by the OFED build scripts.
488 Description: Don't update lov_desc members until making sure they are valid
489 Details : When updating lov_desc members via proc fs, need fix their
490 validities before doing the real update.
494 Description: don't put request into delay list while invalidate in flight.
495 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
496 import in flight. this produce timeout for invalidate and sometimes
497 can cause stale data.
501 Frequency : on ppc only
502 Description: not convert ost objects for directory because it's not exist.
503 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
507 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
509 Severity : enhancement
511 Description: Update to support 2.6.22.14 vanilla kernel.
515 Frequency : rare, on recovery
516 Description: read procfs can produce deadlock in some situation
517 Details : Holding lprocfs lock which send rpc can produce block for destroy
518 obd objects and this also block reconnect with -EALREADY. This
519 isn't fix all lprocfs bugs - but make it rare.
521 Severity : enhancement
523 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
526 Frequency : frequent on X2 node
528 Description: mdc_set_open_replay_data LBUG
529 Details : Set replay data for requests that are eligible for replay.
533 Description: lustre_mgs: operation 101 on unconnected MGS
534 Details : When MGC is disconnected from MGS long enough, MGS will evict the
535 MGC, and late on MGC cannot successfully connect to MGS and a lot
536 of the error messages complaining that MGS is not connected.
540 Frequency : on network error
541 Description: panic with double free request if network error
542 Details : mdc_finish_enqueue is finish request if any network error ocuring,
543 but it's true only for synchronus enqueue, for async enqueue
544 (via ptlrpcd) this incorrect and ptlrpcd want finish request
547 Severity : enhancement
549 Description: client-side metadata stat-ahead during readdir(directory readahead)
550 Details : perform client-side metadata stat-ahead when the client detects
551 readdir and sequential stat of dir entries therein
554 Frequency : on start mds
556 Description: Implement get_info(last_id) in obdfilter.
559 Frequency : occasional
561 Description: Correctly check stale fid, not start epoch if ost not support SOM
562 Details : open with flag O_CREATE need set old fid in op_fid3 because
563 op_fid2 was overwritten with new generated fid, but mds can answer
564 with one of these two fids and both is not stale. Setattr
565 incorrectly started an epoch and assume will be called
566 done_writing, but without SOM done_writing ever being called.
569 Frequency : rare, depends on device drivers and load
571 Description: MDS or OSS nodes crash due to stack overflow
572 Details : Code changes in 1.8.0 increased the stack usage of some functions.
573 In some cases, in conjunction with device drivers that use a lot
574 of stack the MDS (or possibly OSS) service threads could overflow
575 the stack. One change which was identified to consume additional
576 stack has been reworked to avoid the extra stack usage.
579 Frequency : occasional
581 Description: Do not fail import if osc_interpret_create gets -EAGAIN
582 Details : If osc_interpret_create got -EAGAIN it immediately exits and
583 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
584 oscc_has_objects and see OSC has no objests and call
585 oscc_internal_create to resend create request.
587 Severity : enhancement
589 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
591 Severity : enhancement
593 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
596 Frequency : very rare
598 Description: Fix replay if there is an un-replied request and open
599 Details : In some cases, older replay request will revert the
600 mcd->mcd_last_xid on MDS which is used to record the client's
603 Severity : enhancement
605 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
607 Severity : enhancement
609 Description: Add rhel5 support to HEAD.
611 Serverity : enhancement
613 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
618 Description: Don't allow skipping OSTs if index has been specified.
619 Details : Don't allow skipping OSTs if index has been specified, make
620 locking in internal create lots better.
624 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
625 Details : access to bitfield in structure is always rounded to long
626 and this produce problem with not atomic change any bit.
630 Description: Lustre make rpms failed.
631 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
632 builds Lustre rpms from tarball.
635 Frequency : rare on shutdown ost
637 Description: If llog cancel was not send before clean_exports phase, this can
638 produce deadlock in llog code.
639 Details : If llog thread has last reference to obd and call class_import_put
640 this produce deadlock because llog_cleanup_commit_master wait when
641 last llog_commit_thread exited, but this never success because was
642 called from llog_commit_thread.
646 Description: allow userland application know is lost one of stripes.
647 Details : fill lvb_blocks with error code on ost and return it to
648 application if error flag found.
652 Description: NULL lov_tgts causing MDS oops
653 Details : more safe checks for NULL lov_tgts for avoid oops.
655 Severity : enhancement
657 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
661 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
662 Details : If an OST is down the MDS will hang indefinitely in
663 obd_statfs_rqset() waiting for the statfs data. While for
664 MDS QOS usage of statfs, it should not stuck in waiting.
666 Severity : enhancement
668 Description: remote_acl support
669 Details : Support ACL-based permission check for remote user.
670 Support setfacl/getfacl for remote user with the utils
671 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
672 the system "{s,g}etfacl" utils.
674 Severity : enhancement
676 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
678 Severity : enhancement
680 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
684 Description: make mgs_setparam() handle fsname containing dash
685 Details : fsname containing a dash does not work with lctl conf_param
687 Severity : enhancement
689 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
691 Severity : enhancement
693 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
697 Frequency : rare, at shutdown
698 Description: access already free / zero obd_namespace.
699 Details : if client_disconnect_export was called without force flag set,
700 and exist connect request in flight, this can produce access to
701 NULL pointer (or already free pointer) when connect_interpret
702 store ocd flags in obd_namespace.
706 Frequency : only at startup
707 Description: not alloc memory with spinlock held.
708 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
709 if any spinlock held.
711 Severity : enhancement
713 Description: make lustre randomly fail allocating memory
714 Details : Make lustre randomly failed allocating memory for testing purpose.
716 Severity : enhancement
718 Description: lost problems with lov objid file
719 Details : Fixes some scability and access to not inited memory problems
720 in work with lov objdid file.
725 Description: lfs find does not continue on file error
726 Details : Continue other files processing when a file/dir is absent.
730 Description: Inconsistent usage of lustre_pack_reply()
731 Details : Standardize the usage of lustre_pack_reply() such that it
732 always generate a CERROR on failure.
735 Frequency : occasional
737 Description: MDS hang or stay in waiting lock
738 Details : If client receive lock with CBPENDING flag ldlm need send lock
739 cancel as separate rpc, to avoid situation when cancel request
740 can't processed due all i/o threads stay in wait lock.
744 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
748 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
752 Description: Update to SLES9 kernel-2.6.5-7.287.3.
756 Description: Update to RHEL4 latest kernel.
758 Severity : enhancement
760 Description: Build SLES10 patchless client fails
761 Details : The configure was broken by run ./configure with
762 --with-linux-obj=.... argument for patchless client. When the
763 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
764 can't search header adequately. Use absolute path such as
765 -I($LINUX)/include instead.
769 Description: interrupt oig_wait produce painc on resend.
770 Details : brw_redo_request can be used for resend requests from ptlrpcd and
771 private set, and this produce situation when rq_ptlrpcd_data not
772 copyed to new allocated request and triggered LBUG on assert
773 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
774 ptlrpcd set if request is changed and can be safety changed to use
779 Description: LASSERT_{REQ,REP}SWAB macros are buggy
780 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
781 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
782 these macros and replace them with their difinitions instead.
786 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
787 Details : Update which_patch & target file for SLES10 latest kernel.
790 Description: Remove client patches from SLES 10 kernel.
791 Details : This causes SLES 10 clients to behave as patchless clients
792 even on a Lustre-patched (server) kernel.
794 Severity : enhancement
796 Description: self-adjustable client's lru lists
797 Details : use adaptive algorithm for managing client cached locks lru
798 lists according to current server load, other client's work
799 pattern, memory activities, etc. Both, server and client
800 side namespaces provide number of proc tunables for controlling
803 Severity : enhancement
805 Description: light-weight GSS support
806 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
807 and reduce performance overhead.
809 Severity : enhancement
811 Description: Linux keyring support
812 Details : Support using service of Linux keyring for Lustre GSS internal
813 context refresh/cache mechanism.
817 Description: Fix errors in lfs documentation
818 Details : Fixes man pages
822 Description: don't use GFP_* in generic Lustre code.
823 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
827 Description: obdclass is limited by single OBD_ALLOC(idarray)
828 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
832 Description: warnings with build patchless client with vanila 2.6.19 and up
833 Details : change old ctl_table style and replace ctl_table/ctl_table_header
834 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
838 Frequency : Only for SLES
839 Description: sanity_quota fail test_1
840 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
841 file, which makes TSTID[2] inunique.
845 Frequency : Only for SLES9
846 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
847 Details : Update target/ChangeLog/which_patch .
851 Frequency : Only for test_10 in sanity-quota.sh
852 Description: a bug in quota test code
853 Details : checking a test flag in a wrong place
857 Frequency : Only for test_13 in sanity-quota.sh
858 Description: a bug in quota test script
859 Details : shouldn't check overfull allocated quota
863 Description: jbd statistics
864 Details : Port older jbd statistics patch for sles10
866 Severity : enhancement
868 Description: Add printing inode info into message about error in writepage.
873 Description: reply_lock_interpret crash due to race with it and lock cancel.
874 Details : Do not replay locks that are being cancelled. Do not reference
875 locks by their address during replay, just by their handle.
879 Frequency : When flocks are used.
880 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
881 Details : Flock locks might destroy just granted lock if it could be merged
882 with another existing flock, this is done in completion handler,
883 so teach ldlm_cli_enquque_fini that this is a valid case for
889 Description: Oops in read and write path when failing to allocate lock.
890 Details : Check if lock allocation failed and return error back.
894 Description: lstripe command fails for valid OST index
895 Details : The stripe offset is compared to lov->desc.ld_tgt_count
896 instead of lov->desc.ld_active_tgt_count.
900 Description: sanity.sh failed test 103
901 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
906 Description: df doesn't work properly if diskfs blocksize != 4K
907 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
911 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
912 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
914 Severity : enhancement
916 Description: Adaptive timeouts
917 Details : RPC timeouts adapt to changing server load and network
918 conditions to reduce resend attempts and improve recovery time.
922 Description: llapi_file_create() does not allow some changes
923 Details : add llapi_file_open() that allows specifying the mode and
924 open flags, and also returns an open file handle.
928 Description: merge and cleanup kernel patches.
929 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
933 Description: Add journal checksum support.(Kernel part)
934 Details : The journal checksum feature adds two new flags i.e
935 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
936 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
937 indicates that the commit block contains the checksum for
938 the blocks described by the descriptor blocks. Now commit
939 record can be sent to disk without waiting for descriptor
940 blocks to be written to disk. This behavior is controlled
941 using JBD2_FEATURE_ASYNC_COMMIT flag.
945 Description: OSS needs mutliple precreate threads
946 Details : Add ability to start more than one create thread per OSS.
950 Description: Sanity.sh test_65k failed: setstripe should have succeeded
951 Details : execute lfs setstripe on client
955 Description: mds_obd_create error creating tmp object
956 Details : When the user sets quota on root, llog will be affected and can't
957 create files and write files.
960 Frequency : Always on ia64 patchless client, and possibly others.
962 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
963 Details : This allows the patchless client to be loaded on architectures
968 Description: RedHat Update kernel for RHEL5
969 Details : Add the kernel config file for RHEL5.
973 Description: RedHat Update kernel for RHEL5
974 Details : Modify the kernel config file more closer RHEL5.
978 Description: Build failure against Centos5 (RHEL5)
979 Details : Define PAGE_SIZE when it isn't present.
983 Description: test_7 on sanity-quota.sh can't be used on separate machines
984 Details : add facet to handle it
988 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
989 Details : with NFS, the anon dentry's parent was set to itself in
990 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
991 corresponding dentry other than use rec->ur_name.
993 Severity : enhancement
995 Description: lfs setstripe enhancement
996 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1000 Description: enable data checksumming by default
1001 Details : enable checksum by default, allow --disable-checksum
1002 configure option and "-o nochecksum" mount option. Checksums
1003 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1007 Description: lustre support for RHEL5
1008 Details : Add support for RHEL5.
1012 Description: Client eviction due to failover config
1013 Details : after a connection loss, the lustre client should attempt to
1014 reconnect to the last active server first before trying the
1015 other potential connections.
1019 Description: when mds and osts use different quota unit(32bit and 64bit),
1020 quota will be released repeatly.
1021 Details : void sending multiple quota reqs to mds, which will keep the
1022 status between the reqs.
1026 Description: osts not allocated evenly to files
1027 Details : change the condition to increase offset_idx
1030 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1032 Description: Data corruption for OSTs that are formatted larger than 2TB
1034 Details : When generating the bio request for lustre file writes the
1035 sector number would overflow a temporary variable before being
1036 used for the IO. The data reads correctly from Lustre (which
1037 will overflow in a similar manner) but other file data or
1038 filesystem metadata may be corrupted in some cases.
1042 Description: Tune the kernel for good SCSI performance.
1043 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1044 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1049 Description: rewrite ext2-derived code in obdclass/uuid.c
1050 Details : rewrite inherited code (uuid parsing code from ext2 utils)
1051 from scratch preserving functionality.
1055 Description: "lfs find -obd UUID" prints directories
1056 Details : "lfs find -obd UUID" will return all directory names instead
1057 of just file names. It is incorrect because the directories
1058 do not reside on the OSTs.
1062 Description: use i_size_read and i_size_write in 2.6 port
1063 Details : replace inode->i_size access with i_size_read/write()
1066 Frequency : when removing large files
1068 Description: scheduling issue during removal of large Lustre files
1069 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
1070 It causes scheduling issues when removing large files (17TB in the
1074 Frequency : only with liblustre clients on XT3
1076 Description: evictions taking too long
1077 Details : allow llrd to evict clients directly on OSTs
1080 Frequency : only on ppc
1082 Description: /proc/fs/lustre/devices broken on ppc
1083 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
1084 We should make sure the type of 'index' is loff_t and then cast
1085 explicitly as needed below. Do not assign an explicitly cast
1089 Frequency : only for rhel5
1091 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
1092 Details : Modify the target file & which_kernel.
1094 Severity : enhancement
1096 Description: omit set fsid for export NFS
1097 Details : fix set/restore device id for avoid EMFILE error and mark lustre
1098 fs as FS_REQUIRES_DEV for avoid problems with generate fsid.
1102 Frequency : Always, for kernels after 2.6.16
1103 Description: Fix warning idr_remove called for id=.. which is not allocated.
1104 Details : Last kernels save old s_dev before kill super and not allow
1105 to restore from callback - restore it before call kill_anon_super.
1109 Description: buffer overruns could theoretically occur
1110 Details : llapi_semantic_traverse() modifies the "path" argument by
1111 appending values to the end of the origin string, and a
1112 overrun may occur. Adding buffer overrun check in liblustreapi.
1116 Description: Fix error on 'ls .' at the top of the Lustre mount.
1117 Details : Don't revalidate dentry if it is a root dentry.
1121 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
1122 Details : Modify vm-tunables-rhel4.patch.
1126 Description: Kernel config for 2.6.18-vanilla.
1127 Details : Modify targets/2.6-vanilla.target.in.
1128 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
1129 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
1130 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
1131 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
1135 Description: improve handling recoverable errors
1136 Details : if request processig with error which can be recoverable on server
1137 request should be resend, otherwise page released from cache and
1142 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1143 Details : Modify target file & which_patch.
1144 A flaw was found in the IA32 system call emulation provided
1145 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1146 value could be stored in the %RAX register, which could trigger an
1147 out-of-bounds system call table access. An untrusted local user
1148 could exploit this flaw to run code in the kernel
1149 (ie a root privilege escalation). (CVE-2007-4573).
1153 Description: O_DIRECT bypasses client statistics.
1154 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1155 read_ahead_stats were updated. copied the stats section from
1156 osc_send_oap_rpc() into async_internal().
1160 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1164 Description: change order of libsysio includes
1165 Details : '#include sysio.h' should always come before '#include xtio.h'
1169 Description: handle "serious error: objid * already exists" more gracefully
1170 Details : If LAST_ID value on disk is smaller than the objects existing in
1171 the O/0/d* directories, it indicates disk corruption and causes an
1172 LBUG(). If the object is 0-length, then we should use the existing
1173 object. This will help to avoid a full fsck in most cases.
1175 Severity : enhancement
1177 Description: adapt the lustre_config script to support the upgrade case
1178 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1179 server targets to 1.6 in parallel.
1183 Description: To avoid grant space > avaible space when the disk is almost
1184 full. Without this patch you might see the error "grant XXXX >
1185 available" or some LBUG about grant, when the disk is almost
1187 Details : In filter_check_grant, for non_grant cache write, we should
1188 check the left space by if (*left > ungranted + bytes), instead
1189 of (*left > ungranted), because only we are sure the left space
1190 is enough for another "bytes", then the ungrant space should be
1191 increase. In client, we should update cl_avail_grant only there
1192 is OBD_MD_FLGRANT in the reply.
1196 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1197 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1198 described in CVE-2007-4573. This problem could allow untrusted
1199 local users to gain root access.
1202 Frequency : when using O_DIRECT and quotas
1204 Description: Incorrect file ownership on O_DIRECT output files
1205 Details : block usage reported by 'lfs quota' does not take into account
1206 files that have been written with O_DIRECT.
1211 Description: touch file failed when fs is not full
1212 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1213 otherwise we can get ENOSP while fs is not full.
1217 Description: parallel lock callbacks
1218 Details : Instead of sending blocking and completion callbacks as separated
1219 requests, adding them to a set and sending in parallel.
1222 Frequency : only for Cray XT3
1223 Bugzilla : 12829/13455
1224 Description: Changing primary group doesn't change the group lustre assigns to
1226 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1227 overridden with the primary group provided by the group upcall,
1228 whereas the supplied fsgid can be trusted if it is in the list of
1229 supplementary groups returned by the group upcall.
1231 Severity : enhancement
1233 Description: Allow masking D_WARNING, D_ERROR messages from console
1234 Details : Console messages can now be disabled via lnet.printk.
1238 Description: User code with malformed file open parameter crashes client node
1239 Details : Before packing join_file req, all the related reference should be
1240 checked carefully in case some malformed flags cause fake
1241 join_file req on client.
1245 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1246 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1247 mode is equal to the granted mode. However, it should also check
1248 that LDLM_FL_CP_REQD is not set.
1252 Description: Heavy nfs access might result in deadlocks
1253 Details : After ELC code landed, it is now improper to enqueue any mds
1254 locks under och_sem, because enqueue might want to decide to
1255 cancel open locks for same inode we are holding och_sem for.
1259 Description: Client eviction while running blogbench
1260 Details : A lot of unlink operations with concurrent I/O can lead to a
1261 deadlock causing evictions. To address the problem, the number of
1262 oustanding OST_DESTROY requests is now throttled to
1263 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1264 callbacks are processed in priority.
1268 Description: enable ACLs on MDS by default
1269 Details : ACLs must be enabled on MDS by default.
1272 Frequency : PPC/PPC64 only
1274 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1275 Details : fix duplicated definitions between asm-ppc64/types.h and
1276 lustre_types.h on PPC.
1279 Frequency : PPC/PPC64 only
1281 Description: asm-ppc/segment.h does not exist
1282 Details : fix compile issue on PPC.
1286 Description: better handle error messages in extents code
1289 Frequency : RHEL4 only
1291 Description: mkfs is very slow on IA64/RHEL4
1292 Details : A performance regression has been discovered in the MPT Fusion
1293 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1294 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1295 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1297 Severity : enhancement
1299 Description: SNMP support enhancement
1300 Details : Adding total number of sampled request for an MDS node in snmp
1303 Severity : enhancement
1305 Description: Optimize ldlm waiting list processing for PR extent locks
1306 Details : When processing waiting list for read extent lock and meeting read
1307 lock that is same or wider to it that is not contended, skip
1308 processing rest of the list and immediatelly return current
1309 status of conflictness, since we are guaranteed there are no
1310 conflicting locks in the rest of the list.
1314 Description: Time out and refuse to reconnect
1315 Details : When the failover node is the primary node, it is possible
1316 to have two identical connections in imp_conn_list. We must
1317 compare not conn's pointers but NIDs, otherwise we can defeat
1318 connection throttling.
1322 Description: port llog fixes from b1_6 into HEAD
1323 Details : Port llog reference couting and some llog cleanups from b1_6
1324 (bug 10800) into HEAD, for protect from panic and access to already
1325 free llog structures.
1329 Description: Detect stride IO mode in read-ahead
1330 Details : When a client does stride read, read-ahead should detect that and
1331 read-ahead pages according to the detected stride pattern.
1335 Description: data checksumming impacts single node performance
1336 Details : add support for several checksum algorithm. Currently, only CRC32
1337 and Adler-32 are supported. The checksum type can be changed on
1338 the fly via /proc/fs/lustre/osc/*/checksum_type.
1342 Description: use adler32 for page checksums
1343 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1348 Description: build for x2 fails
1349 Details : fix compile issue on Cray systems.
1353 Description: Properly match for duplicate locks
1354 Details : Due to different lock order from skiplists code, we need to
1355 traverse entire list for now
1358 Frequency : only on PPC/SLES10
1360 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1361 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1362 be defined. Add a hack in mkfs_lustre.c to work around this compile
1367 Description: LASSERT on MDS when client holding flock lock dies
1368 Details : ldlm pool logic depends on number of granted locks equal to
1369 number of released locks which is not true for flock locks, so
1370 just exclude such locks from consideration.
1374 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1375 Details : Use fsfilt_sync() for both the whole filesystem sync and
1376 individual file sync to eliminate dangerous inode locking
1377 with I_LOCK that can lead to a deadlock.
1381 Description: performance in 1.6.3
1382 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1383 sure that 1MB requests are not fragmented by the block layer.
1385 Severity : enhancement
1387 Description: organize the server-side client stats on per-nid basis
1388 Details : Change the structure of stats under obdfilter and mds to
1396 The "uuid"s file would list the uuids of _active_ exports.
1397 And the clear entry is to clear all stats and stale nids.
1399 Severity : enhancement
1401 Description: eliminate client locks in face of contention
1402 Details : file contention detection and lockless i/o implementation
1403 for contended files.
1407 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1410 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1412 Description: MDS may lose file striping (and hence file data) in some cases
1413 Details : If there are additional extended attributes stored on the MDS,
1414 in particular ACLs, SELinux, or user attributes (if user_xattr
1415 is specified for the client mount options) then there is a risk
1416 of attribute loss. Additionally, the Lustre file striping
1417 needs to be larger than default (e.g. striped over all OSTs),
1418 and an additional attribute must be stored initially in the
1419 inode and then increase in size enough to be moved to the
1420 external attribute block (e.g. ACL growing in size) for file
1425 Description: skiplist implementation simplification
1426 Details : skiplists are used to group compatible locks on granted list
1427 that was implemented as tracking first and last lock of each
1428 lock group the patch changes that to using doubly linked lists
1432 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1433 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1434 because LASSERT on both the data supplied by a client, and the
1435 data on disk is dangerous and incorrect.
1437 Severity : enhancement
1439 Description: Slow truncate/writes to huge files at high offsets.
1440 Details : Directly associate cached pages to lock that protect those pages,
1441 this allows us to quickly find what pages to write and remove
1442 once lock callback is received.
1446 Description: more ldlm soft lockups
1447 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1448 starve other threads from the resource lock for a long time in
1449 case of long waiting queue, so change the debug level from
1450 D_OTHER to the less frequently used D_INFO.
1452 Severity : enhancement
1454 Description: add -gid, -group, -uid, -user options to lfs find
1458 Description: Hung threads in invalidate_inode_pages2_range
1459 Details : The direct IO path doesn't call check_rpcs to submit a new RPC
1460 once one is completed. As a result, some RPCs are stuck in the
1461 queue and are never sent.
1465 Description: filter threads hungs on waiting journal commit
1466 Details : Cleanup filter group llog code, then only filter group llog will
1467 be only created in the MDS/OST syncing process.
1471 Description: Procfs and llog threads access destoryed import sometimes.
1472 Details : Sync the import destoryed process with procfs and llog threads by
1473 the import refcount and semaphore.
1475 Severity : enhancement
1477 Description: openlock cache of b1_6 port to HEAD
1482 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1483 Details : If insertion of an extent fails, then discard the inode
1484 preallocation and free data blocks else it can lead to duplicate
1489 Description: don't always update ctime in ext3_xattr_set_handle()
1490 Details : Current xattr code updates inode ctime in ext3_xattr_set_handle.
1491 In some cases the ctime should not be updated, for example for
1492 2.0->1.8 compatibility it is necessary to delete an xattr and it
1493 should not update the ctime.
1497 Bugzilla : 15713/16362
1498 Description: Assertion in iopen_connect_dentry in 1.6.3
1499 Details : looking up an inode via iopen with the wrong generation number can
1500 populate the dcache with a disconneced dentry while the inode
1501 number is in the process of being reallocated. This causes an
1502 assertion failure in iopen since the inode's dentry list contains
1503 both a connected and disconnected dentry.
1507 Description: assertion failure in ldlm_handle2lock()
1508 Details : fix a race between class_handle_unhash() and class_handle2object()
1509 introduced in lustre 1.6.5 by bug 13622.
1514 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1515 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1520 Description: oops in sd_iostats_seq_show()
1521 Details : unloading/reloading the scsi low level driver triggers a kernel
1522 bug when trying to access the sd iostat file.
1527 Description: Kernel panics during QLogic driver reload
1528 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1529 patch, causing memory corruption.
1534 Description: journal_dev option does not work in b1_6
1535 Details : pass mount option during pre-mount.
1537 Severity : enhancement
1539 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1540 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1541 extent information of a file. It can be used to map logical blocks
1542 in a file to physical blocks in the block device.
1546 Description: LDLM soft lockups - improvement
1547 Details : It is be possible to send the lock handle along with each read
1548 or write request because the client is already doing a lock match
1549 itself so there isn't any reason the OST should have to re-do that
1553 Frequency : only on Cray X2
1555 Description: X2 build failures
1556 Details : fix build failures on Cray X2.
1560 Description: xid & resent requests
1561 Details : Initialize RPC XID from clock at startup (randomly if clock is
1564 Severity : enhancement
1566 Description: Add lustre_start utility to start or stop multiple Lustre servers
1571 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1572 Details : In case of memory pressure, list_del() can be called twice on
1573 req->rq_history_list, causing a kernel oops.
1577 Description: kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1578 Details : fix stack overflow in the distributed lock manager by defering
1579 export eviction after a failed AST to the elt thread instead of
1580 handling it in the dlm interpret routine.
1584 Description: Convert some comments to new format.
1585 Details : Update documenting comments to match doxygen conventions.
1589 Description: Grammar fixes.
1590 Details : A couple of trivial spelling fixes.
1594 Description: OSD_COUNTERS-mandatory
1595 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1596 to simplify the code.
1600 Description: simplify cmm_device freeing logic.
1601 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1605 Description: Add lockdep support to dt_object_operations locking interface.
1606 Details : Augment ->do_{read,write}_lock() prototypes with a `role'
1607 parameter indicating lock ordering. Update mdd code to use new
1612 Description: Introduce failloc constants for lockless IO tests.
1613 Details : Add two new failloc constants to test lockless IO. Only one of
1614 them in implemented---another is checked in yet to be landed
1619 Description: Add lockdep support for inode mutex.
1620 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1621 in the situations where Lustre has to lock more than one inode
1626 Description: Add optional invariants checking support.
1627 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1628 configure switch. This macro is to be used for consistency and
1629 sanity checks that are too expensive to be left in `production'
1634 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1635 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1640 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1641 Details : Use spin_lock_nested() in (the only) situation where more than
1642 one ldlm_lock is locked simultaneously. Also, fix possible
1643 dead-lock in ldlm_lock_change_resource() by enforcing particular
1648 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1649 Details : Instead of specifying each ldlm_lock call-back through separate
1650 parameter, wrap them into struct ldlm_callback_suite.
1654 Description: Kill join_lru obd method and its callers.
1655 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1656 in memory---a requirement implemented through ->o_join_lru() obd
1657 method in HEAD. Remove this method and its users.
1661 Description: Add asynchronous ldlm ENQUEUE completion handler.
1662 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1663 case that stock ldlm_completion_ast() cannot handle as it waits
1664 until lock is granted. Introduce new ldlm_completion_ast_async()
1665 for this. Also comment ldlm_completion_ast().
1669 Description: ldlm_error <-> errno conversion.
1670 Details : Add functions to map (rather arbitrary) between LDLM error codes
1671 and standard errno values. CLIO needs this to prevent LDLM specific
1672 constants from escaping ldlm and osc.
1676 Description: Kill unused ldlm_handle2lock_ns() function.
1677 Details : Kill unused ldlm_handle2lock_ns() function.
1681 Description: Add lu_ref support to ldlm_lock
1682 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1683 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1684 and ldlm_resource. LDLM interface has to be changed, because code that
1685 releases a reference on a lock, has to "know" what reference this is.
1686 In the most frequent case
1688 lock = ldlm_handle2lock(handle);
1690 LDLM_LOCK_PUT(lock);
1692 no changes are required. When any other reference (received _not_ from
1693 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1694 instead of LDLM_LOCK_PUT().
1696 Arguably, changes are pervasive, and interface requires some discipline
1697 for proper use. On the other hand, it was very instrumental in finding
1698 a few leaked lock references.
1702 Description: Add ldlm_lock_addref_try().
1703 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1704 attempts to addref a lock that might be being canceled
1709 Description: Add ldlm_weigh_callback().
1710 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1711 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1712 hard-coded `number of pages' logic.
1716 Description: Add lockdep annotations to llog code.
1717 Details : Use appropriately tagged _nested() locking calls in the places
1718 where llog takes more than one ->lgh_lock lock.
1722 Description: Add loi_kms_set().
1723 Details : Wrap kms updates into a helper function.
1727 Description: Constify instances of struct lsm_operations.
1728 Details : Constify instances of struct lsm_operations.
1732 Description: lu_conf support.
1733 Details : On a server, a file system object is uniquely identified
1734 by a fid, which is sufficient to locate and load all object
1735 state (inode). On a client, on the other hand, more data are
1736 necessary instantiate an object. Change lu_object_find() and
1737 friends to take additional `lu_conf' argument describing object.
1738 Typically this includes layout information.
1742 Description: lu_context fixes.
1743 Details : Introduce new lu_context functions that are needed on the client
1744 side, where some system threads (ptlrpcd) are shared by multiple
1745 modules, and so cannot be stopped during module shutdown.
1749 Description: Add start and stop methods to lu_device_type_operations.
1750 Details : Introduce two new methods in lu_device_type_operations, that are
1751 invoked when first instance of a given type is created and last one
1752 is destroyed respectively. This is need by CLIO.
1756 Description: Add lu_ref support to struct lu_device.
1757 Details : Add lu_ref support to lu_object and lu_device. lu_ref is used to
1758 track leaked references.
1762 Description: Introduce lu_kmem_descr.
1763 Details : lu_kmem_descr and its companion interface allow to create
1764 and destroy a number of kmem caches at once.
1768 Description: Fix lu_object finalization race.
1769 Details : Fix a race between lu_object_find() finding an object and its
1770 concurrent finalization. This race is (most likely) not possible
1771 on the server, but might happen on the client.
1775 Description: Introduce lu_ref interface.
1776 Details : lu_ref is a debugging module allowing to track references to
1777 a given object. It is quite cpu expensive, and has to be
1778 explicitly enabled with --enable-lu_ref. See usage description
1783 Description: Factor lu_site procfs stats into a separate function.
1784 Details : Separate lu_site stats printing code into a separate function
1785 that can be reused on a client.
1789 Description: Constify instances of struct {lu,dt,md}_device_operations.
1790 Details : Constify instances of struct {lu,dt,md}_device_operations.
1794 Description: Introduce struct md_site and move meta-data specific parts of
1795 struct lu_site here.
1796 Details : Move md-specific fields out of struct lu_site into special struct
1797 md_site, so that lu_site can be used on a client.
1801 Description: Kill mdd_lov_destroy().
1802 Details : Remove unused mdd code.
1806 Description: Add st_block checking to multistat.c.
1807 Details : Add st_block checking to multistat.c.
1811 Description: Add lu_ref support to struct obd_device.
1812 Details : Add lu_ref tracking to obd_device.
1816 Description: Kill obd_set_fail_loc().
1817 Details : Remove unused code.
1821 Description: Add special type for ptlrpc_request interpret functions.
1822 Details : Add lu_env parameter to ->rq_interpreter call-back. NULL is passed
1823 there. Actual usage will be in CLIO.
1827 Description: Replace RW_LOCK_UNLOCKED() macro with rwlock_init().
1828 Details : Replace RW_LOCK_UNLOCKED() with rwlock_init() as the former
1829 doesn't work with lockdep.
1833 Description: Add rwv.c test program.
1834 Details : New testing program exercising readv(2) and writev(2) (Qian).
1838 Description: Add sendfile.c test program.
1839 Details : New testing program exercising sendfile(2) (Jay).
1843 Description: Ratelimit a message that can be very frequent.
1844 Details : Ratelimit a memory allocation failure message that can
1849 Description: Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1850 Details : Use cdebug_show() in CDEBUG-style macros defined outside of libcfs.
1854 Description: Liblustre build fixes.
1855 Details : Liblustre build fixes.
1859 Description: libcfs: add cfs_{need,cond}_resched() interface.
1860 Details : libcfs: add cfs_{need,cond}_resched() definition and
1861 implementations for Linux, NT, and liblustre.
1863 Severity : enhancement
1865 Description: More exported tunables for mballoc
1866 Details : Add support for tunable preallocation window and new tunables for
1867 large/small requests
1871 Description: Detect corruption of block bitmap and checking for preallocations
1872 Details : Checks validity of on-disk block bitmap. Also it does better
1873 checking of number of applied preallocations. When corruption is
1874 found, it turns filesystem readonly to prevent further corruptions.
1878 Description: (rw.c:1323:ll_read_ahead_pages()) ASSERTION(page_idx > ria->ria_stoff) failed
1879 Details : Once the unmatched stride IO mode is detected, shrink the stride-ahead
1880 window to 0. If it does hit cache miss, and read-pattern is still
1881 stride-io mode, does not reset the stride window, but also does not
1882 increase the stride window length in this case.
1886 Frequency : only for big-endian servers
1887 Description: Check if system is big-endian while mounting fs with extents feature
1888 Details : Mounting a filesystem with extents feature will fail on big-endian
1889 systems since ext3-based ldiskfs is not supported on big-endian
1890 systems. This can be over-riden with "bigendian_extents" mount option.
1892 Severity : enhancement
1894 Description: The root squash functionality
1895 Details : A security feature, which is to prevent users from being able
1896 to mount lustre on their desktop, run as root, and delete
1897 all of the files in the filesystem. The goal is accomplished by
1898 remapping user id (UID) and group id (GID) of the root user to
1899 a UID and GID specified by the system administartor via Lustre
1900 configuration management server (MGS). The functionality also
1901 allows to specify sets of clients for which the remapping does
1906 Description: Excessive recovery window
1907 Details : With AT enabled, the recovery window can be excessively long (6000+
1908 seconds). To address this problem, we no longer use
1909 OBD_RECOVERY_FACTOR when extending the recovery window (the connect
1910 timeout no longer depends on the service time, it is set to
1911 INITIAL_CONNECT_TIMEOUT now) and clients report the old service
1912 time via pb_service_time.
1916 Description: Watchdog triggered on MDS failover
1917 Details : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
1918 the OSTs know that the MDS "UUID" can be reused for the same export
1919 from a different NID, so we do not need to wait for the export to be
1922 --------------------------------------------------------------------------------
1924 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1926 * Support for kernels:
1927 2.6.5-7.283 (SLES 9),
1928 2.6.9-55.EL (RHEL 4),
1929 2.6.16.46-0.14 (SLES 10),
1930 2.6.18.8 vanilla (kernel.org)
1931 * Client support for unpatched kernels:
1932 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1933 2.6.16 - 2.6.22 vanilla (kernel.org)
1934 * Due to recently discovered recovery problems, we do not recommend
1935 using patchless RHEL 4 clients with this or any earlier release.
1936 * Recommended e2fsprogs version: 1.39.cfs8
1937 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1938 * Starting with this release, the ldiskfs backing filesystem required
1939 by Lustre is now in its own package, lustre-ldiskfs. This package
1940 should be installed. It is versioned separately from Lustre and
1941 may be released separately in future.
1945 Description: block reactivating mgc import until all deactivates complete
1946 Details : Fix race when failing back MDT/MGS to itself (testing)
1949 Frequency : at statup only
1951 Description: mds_lov_synchronize race leads to various problems
1952 Details : simultaneous MDT->OST connections at startup can cause the
1953 sync to abort, leaving the OSC in a bad state.
1955 Severity : enhancement
1957 Description: add optional extra BUILD_VERSION info
1958 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1959 to override the lustre version.
1962 Frequency : 2.6.18 servers only
1964 Description: ll_kern_mount() doesn't release the module reference
1965 Details : The ldiskfs module reference count never drops down to 0
1966 because ll_kern_mount() doesn't release the module reference.
1971 Description: server LBUG when using old ost_num_threads parameter
1972 Details : Accept the old ost_num_threads parameter but warn that it
1973 is deprecated, and fix an off-by-one error that caused an LBUG.
1978 Description: Transient SCSI error results in persistent IO issue
1979 Details : iobuf->dr_error is not reinitialized to 0 between two
1983 Frequency : sometimes when underlying device returns I/O errors
1985 Description: OSTs not going read-only during write failures
1986 Details : OSTs are not remounted read-only when the journal commit threads
1987 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1988 instead of the ext3 wrappers.
1992 Description: poor connect scaling with increasing client count
1993 Details : Don't run filter_grant_sanity_check for more than 100 exports
1994 to improve scaling for large numbers of clients.
1997 Frequency : SLES10 only
1999 Description: sanity-quota.sh quotacheck failed: rc = -22
2000 Details : Quotas cannot be enabled on SLES10.
2003 Frequency : liblustre clients only
2005 Description: getdirentries does not give error when run on compute nodes
2006 Details : getdirentries does not fail when the size specified as an argument
2007 is too small to contain at least one entry
2009 Severity : enhancement
2011 Description: Add LNET router traceability for debug purposes
2012 Details : If a checksum failure occurs with a router as part of the
2013 IO path, the NID of the last router that forwarded the bulk data
2014 is printed so it can be identified.
2019 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
2020 Details : Due to a race condition, liblustre clients were occasionally
2021 evicted incorrectly.
2023 Severity : enhancement
2025 Description: lfs setstripe use optional parameters instead of postional
2028 Severity : enhancement
2030 Description: Nanosecond timestamp support for ldiskfs
2031 Details : The on-disk ldiskfs filesystem has added support for nanosecond
2032 resolution timestamps. There is not yet support for this at
2033 the Lustre filesystem level.
2036 Frequency : during server recovery
2038 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
2039 Details : request with rq_no_resend flag not awake l_wait_event if they get
2043 Frequency : nfs export on patchless client
2045 Description: connectathon hang when test nfs export over patchless client
2046 Details : Disconnected dentry cannot be found with lookup, so we do not need
2047 to unhash it or make it invalid
2050 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
2051 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
2052 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
2057 Description: open req refcounting wrong on reconnect
2058 Details : If reconnect happened between getting open reply from server and
2059 call to mdc_set_replay_data in ll_file_open, we will schedule
2060 replay for unreferenced request that we are about to free.
2061 Subsequent close will crash in variety of ways.
2062 Check that request is still eligible for replay in
2063 mdc_set_replay_data().
2068 Description: disable writes to filesystem when reading health_check file
2069 Details : the default for reading the health_check proc file has changed
2070 to NOT do a journal transaction and write to disk, because this
2071 can cause reads of the /proc file to hang and block HA state
2072 checking on a healthy but otherwise heavily loaded system. It
2073 is possible to return to the previous behaviour during configure
2074 with --enable-health-write.
2076 Severity : enhancement
2078 Description: 64-bit inode version
2079 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
2080 to the inode. This will be required for version-based recovery.
2085 Description: MDS fails to start if a duplicate client export is detected
2086 Details : in some rare cases it was possible for a client to connect to
2087 an MDS multiple times. Upon recovery the MDS would detect this
2088 and fail during startup. Handle this more gracefully.
2090 Severity : enhancement
2092 Description: Add -o localflock option to simulate old noflock
2094 Details : This will achieve local-only flock/fcntl locks
2100 Description: log_commit_thread vs filter_destroy race leads to crash
2101 Details : Take import reference before releasing llog record semaphore
2106 Description: Wrong request locking in request set processing
2107 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
2108 imp_delayed_list, in this place should be used imp_lock.
2111 Frequency : when reconnection
2113 Description: Grant Leak when osc reconnect to OST
2114 Details : When osc reconnect ost, OST(filter) should check whether it
2115 should grant more space to client by comparing fed_grant and
2116 cl_avail_grant, and return the granted space to client instead
2117 of "new granted" space, because client will call osc_init_grant
2118 to update the client grant space info.
2121 Frequency : when client reconnect to OST
2123 Description: Grant Leak when osc do resend and replay bulk write
2124 Details : When osc reconnect to OST, OST(filter)should clear grant info of
2125 bulk write request, because the grant info while be sync between
2126 OSC and OST when reconnect, and we should ignore the grant info
2127 these of resend/replay write req.
2132 Description: Grant space more than avaiable left space sometimes.
2133 Details : When then OST is about to be full, if two bulk writing from
2134 different clients came to OST. Accord the avaliable space of the
2135 OST, the first req should be permitted, and the second one
2136 should be denied by ENOSPC. But if the seconde arrived before
2137 the first one is commited. The OST might wrongly permit second
2138 writing, which will cause grant space > avaiable space.
2141 Frequency : when client is evicted
2143 Description: Grant might be wrongly erased when osc is evicted by OST
2144 Details : when the import is evicted by server, it will fork another
2145 thread ptlrpc_invalidate_import_thread to invalidate the
2146 import, where the grant will be set to 0. While the original
2147 thread will update the grant it got when connecting. So if
2148 the former happened latter, the grant will be wrongly errased
2149 because of this race.
2154 Description: Checking Stale with correct fid
2155 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
2156 to check whether it is stale, because sometimes, we want the
2157 enqueue happened anyway, and op_data.fid2 will not be initialized.
2159 Severity : enhancement
2161 Description: update patchless client
2162 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
2165 Frequency : only with 2.4 kernel
2167 Description: random memory corruption
2168 Details : size of struct ll_inode_info is to big for union inode.u and this
2169 can be cause of random memory corruption.
2174 Description: Memory leak in recovery
2175 Details : Lov_mds_md was not free in an error handler in mds_create_object.
2176 It should also check obd_fail before fsfilt_start, otherwise if
2177 fsfilt_start return -EROFS,(failover mds during mds recovery).
2178 then the req will return with repmsg->transno = 0 and rc = EROFS.
2179 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
2180 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
2181 be freed no matter whether fsfilt_commit success or not.
2184 Frequency : only with huge count clients
2186 Description: Prevents from taking the superblock lock in llap_from_page for
2188 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
2189 for llap_from_page call in ll_removepage prevents from taking the
2190 superblock lock for a soon died page.
2195 Description: Not check open intent error before release open handle
2196 Details : in some rare cases, the open intent error is not checked before
2197 release open handle, which may cause
2198 ASSERTION(open_req->rq_transno != 0), because it tries to release
2199 the failed open handle.
2204 Description: Set cat log bitmap only after create log success.
2205 Details : in some rare cases, the cat log bitmap is set too early. and it
2206 should be set only after create log success.
2210 Description: Accessing a block bevice can re-enable I/O when Lustre is
2211 tearing down a device.
2212 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
2217 Description: service threads may hog cpus when there are a lot of requests
2218 Details : Insert cond_resched to give other threads a chance to use some CPU
2223 Description: the cat log was not initialized in recovery
2224 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
2225 unlink log on mds will not be initialized until mds post
2226 recovery. And also in mds post recovery, the unlink log will
2227 initialization will be done asynchronausly, so there will be race
2228 between add unlink log and unlink log initialization.
2232 Description: brw_stats were being printed incorrectly
2233 Details : brw_stats were being printed as log2 but all of them were not
2234 recorded as log2. Also remove some code duplication arising from
2235 filter_tally_{read,write}.
2239 Frequency : rare, only in recovery.
2240 Description: ASSERTION(req->rq_type != LI_POISON) failed
2241 Details : imp_lock should be held while iterating over imp_sending_list for
2242 prevent destroy request after get timeout in ptlrpc_queue_wait.
2246 Description: replay-single.sh test 52 fails
2247 Details : A lock's skiplist need to be cleanup when it being unlinked
2248 from its resource list.
2252 Description: Short directio read returns full requested size rather than
2254 Details : Direct I/O operations should return actual amount of bytes
2255 transferred rather than requested size.
2257 Severity : enhancement
2259 Description: metadata RPC reduction (e.g. for rm performance)
2260 Details : decrease the amount of synchronous RPC between clients and servers
2261 by canceling conflicing lock before the operation on the client
2262 and packing thier handles into the main operation RPC to server.
2264 Severity : enhancement
2266 Description: Async OSC create to avoid the blocking unnecessarily.
2267 Details : If a OST has no remain object, system will block on the creating
2268 when need to create a new object on this OST. Now, ways use
2269 pre-created objects when available, instead of blocking on an
2270 empty osc while others are not empty. If we must block, we block
2271 for the shortest possible period of time.
2275 Description: improve handling recoverable errors
2276 Details : if request processig with error which can be recoverable on server
2277 request should be resend, otherwise page released from cache and
2280 Severity : enhancement
2282 Description: refine locking for avoid write wrong info into lov_objid file
2283 Details : fix possible races with add new target and write/update data in
2286 --------------------------------------------------------------------------------
2288 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2293 Frequency : on some architectures
2295 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2296 Details : Uninitialized flags sometimes cause configuration commands to
2300 Frequency : patchless clients only
2302 Description: missing __iget() symbol export
2303 Details : The __iget() symbol export is missing. To avoid the need for
2304 this on patchless clients the deathrow inode reaper is turned
2305 off, and we depend on the VM to clean up old inodes. This
2306 dependency was during via the fix for bug 12181.
2308 --------------------------------------------------------------------------------
2310 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2312 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2313 INTEROPERATE with older versions automatically. In many cases a
2314 special upgrade step is needed. Please read the
2315 user documentation before upgrading any part of a live system.
2316 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
2317 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2318 * WARNING: Lustre configuration and startup changes are required with
2319 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2323 Severity : enhancement
2325 Description: MountConf
2326 Details : Lustre configuration is now managed via mkfs and mount
2327 commands instead of lmc and lconf. New obd types (MGS, MGC)
2328 are added for dynamic configuration management. See
2329 https://mail.clusterfs.com/wikis/lustre/MountConf for
2332 Severity : enhancement
2334 Description: dynamic OST addition
2335 Details : OSTs can now be added to a live filesystem
2337 Severity : enhancement
2339 Description: startup order invariance
2340 Details : MDTs and OSTs can be started in any order. Clients only
2341 require the MDT to complete startup.
2343 Severity : enhancement
2345 Description: parallel, asynchronous orphan cleanup
2346 Details : orphan cleanup is now performed in separate threads for each
2347 OST, allowing parallel non-blocking operation.
2349 Severity : enhancement
2351 Description: optimized stripe assignment
2352 Details : stripe assignments are now made based on ost space available,
2353 ost previous usage, and OSS previous usage, in order to try
2354 to optimize storage space and networking resources.
2356 Severity : enhancement
2358 Description: Permanently set tunables
2359 Details : All writable /proc/fs/lustre tunables can now be permanently
2360 set on a per-server basis, at mkfs time or on a live system.
2362 Severity : enhancement
2364 Description: Lustre message v2
2365 Details : Add lustre message format v2.
2367 Severity : enhancement
2369 Description: client OST exclusion list
2370 Details : Clients can be started with a list of OSTs that should be
2371 declared "inactive" for known non-responsive OSTs.
2375 Description: SPEC SFS validation failure on NFS v2 over lustre.
2376 Details : Changes the blocksize for regular files to be 2x RPC size,
2377 and not depend on stripe size.
2379 Severity : enhancement
2381 Description: Multiple MD RPCs in flight.
2382 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2383 To avoid overly-overloading MDS, introduce a limit on number of
2384 MDS RPCs in flight for a single client and add /proc controls
2385 to adjust this limit.
2387 Severity : enhancement
2389 Description: client read/write statistics
2390 Details : Add client read/write call usage stats for performance
2391 analysis of user processes.
2392 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2393 file access. extents_stats shows chunk size distribution.
2394 extents_stats_per_process show chunk size distribution per
2397 Severity : enhancement
2399 Description: mds statistics
2400 Details : Add detailed mds operations statistics in
2401 /proc/fs/lustre/mds/*/stats.
2405 Description: Failure of copying files with lustre special EAs.
2406 Details : Client side always return success for setxattr call for lustre
2407 special xattr (currently only "trusted.lov").
2412 Description: Refcount LNET uuids
2413 Details : The global LNET uuid list grew linearly with every startup;
2414 refcount repeated list entries instead of always adding to
2417 Severity : enhancement
2419 Description: Dynamic service threads
2420 Details : Within a small range, start extra service threads
2421 automatically when the request queue builds up.
2424 Frequency : mixed-endian client/server environments
2426 Description: mixed-endian crashes
2427 Details : The new msg_v2 system had some failures in mixed-endian
2430 Severity : enhancement
2432 Description: Easy OST removal
2433 Details : OSTs can be permanently deactivated with e.g. 'lctl
2434 conf_param lustre-OST0001.osc.active=0'
2436 Severity : enhancement
2438 Description: MGS proc entries
2439 Details : Added basic proc entries for the MGS showing what filesystems
2442 Severity : enhancement
2444 Description: provide MGS failover
2445 Details : Added config lock reacquisition after MGS server failover.
2447 Severity : enhancement
2449 Description: add Linux 2.4 support
2450 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2454 Description: a large application tries to do I/O to the same resource and dies
2455 in the middle of it.
2456 Details : Check the req->rq_arrival time after the call to
2457 ost_brw_lock_get(), but before we do anything about
2458 processing it & sending the BULK transfer request. This
2459 should help move old stale pending locks off the queue as
2460 quickly as obd_timeout.
2463 Frequency : when an incorrect nid is specified during startup
2465 Description: ptlrpc connect to non-existant node causes kernel crash
2466 Details : LNET can't be re-entered from an event callback, which
2467 happened when we expire a message after the export has been
2468 cleaned up. Instead, hand the zombie cleanup off to another
2471 Severity : enhancement
2473 Description: plain/inodebits lock performance improvement
2474 Details : Grouping plain/inodebits in granted list by their request modes
2475 and bits policy, thus improving the performance of search through
2479 Frequency : only if OST filesystem is corrupted
2481 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2482 Details : for a short time RPCs with bulk IO are in the replay list,
2483 but replay of bulk IOs is unimplemented. If the OST filesystem
2484 is corrupted due to disk cache incoherency and then replay is
2485 started it is possible to trip an assertion. Avoid putting
2486 committed RPCs into the replay list at all to avoid this issue.
2489 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2491 Description: System hang on startup
2492 Details : This bug allowed the liblustre (e.g. catamount) client to
2493 return to the app before handling all startup RPCs. This
2494 could leave the node unresponsive to lustre network traffic
2495 and manifested as a server ptllnd timeout.
2497 Severity : enhancement
2499 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2500 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2501 boolean to control whether to print peer debug info when a
2502 client's RPC times out.
2505 Frequency : only for kernels with patches from Lustre below 1.4.3
2507 Description: Remove old rdonly API
2508 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2511 Frequency : only for devices with external journals
2513 Description: Set external device read-only also
2514 Details : During a commanded failover stop, we set the disk device
2515 read-only while the server shuts down. We now also set any
2516 external journal device read-only at the same time.
2519 Frequency : when upgrading from 1.4 while trying to change parameters
2521 Description: The wrong (new) MDC name was used when setting parameters for
2522 upgraded MDT's. Also allows changing of OSC (and MDC)
2523 parameters if --writeconf is specified at tunefs upgrade time.
2526 Frequency : when setting specific ost indicies
2528 Description: QOS code breaks on skipped indicies
2529 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2530 created with --index need not be sequential.
2532 Severity : enhancement
2534 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2535 Details : The uninit_groups feature works in conjunction with the kernel
2536 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2537 up the pass1 processing of e2fsck. This is a read-only feature
2538 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2539 mount filesystems that have had this feature enabled.
2541 Severity : enhancement
2543 Description: Improve multi-block allocation algorithm to avoid fragmentation
2544 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2545 allocation locality and avoid filesystem fragmentation.
2547 ------------------------------------------------------------------------------
2549 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2551 * Support for kernels:
2552 2.4.21-47.0.1.EL (RHEL 3)
2553 2.6.5-7.283 (SLES 9)
2554 2.6.9-42.0.10.EL (RHEL 4)
2555 2.6.12.6 vanilla (kernel.org)
2556 2.6.16.27-0.9 (SLES 10)
2557 * Recommended e2fsprogs version: 1.39.cfs5
2559 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2563 Frequency : occasional, depends on client load and configuration
2564 Bugzilla : 12181, 12203
2565 Description: data loss for recently-modified files
2567 Details : In some cases it is possible that recently written or created
2568 files may not be written to disk in a timely manner (this should
2569 normally be within 30s unless client IO load is very high).
2570 The problem appears as zero-length files or files that are a
2571 multiple of 1MB in size after a client crash or client eviction
2572 that are missing data at the end of the file.
2574 This problem is more likely to be hit on clients where files are
2575 repeatedly created and unlinked in the same directory, clients
2576 have a large amount of RAM, have many CPUs, the filesystem has
2577 many OSTs, the clients are rebooted frequently, and/or the files
2578 are not accessed by other nodes after being written.
2580 The presence of the problem can be detected by looking at
2581 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2582 smaller than the second (nr_unused) then dirty files will not
2583 be flushed automatically to disk. "sync; sleep 10" should be
2584 run several times on the node before unmounting it to update
2585 Lustre (this is also safe to run on nodes without this problem).
2587 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2588 that can cause this same problem, so customers using that kernel
2589 also need to update the kernel in addition to Lustre. In order
2590 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2592 It is normal that files written just before a client crash (less
2593 than 30s) may not yet have been flushed to disk, even for local
2597 Frequency : frequent on thin XT3 nodes
2599 Description: UUID collision on thin XT3 Linux nodes
2600 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2601 randomly, since we relied on an insufficiently-seeded PRNG.
2606 Description: OSS hangs after "All ost request buffers busy"
2607 Details : A deadlock between quota and journal operations caused OSS
2608 hangs after printing "All ost request buffers busy."
2611 Frequency : always on liblustre builds
2613 Description: Cleanup compiler warnings on liblustre
2616 Frequency : always on liblustre builds on XT3
2618 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2622 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2623 > 32000 subdirectories
2624 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2625 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2626 subdirectory count crosses 32000. This will aid e2fsck to
2627 correctly handle more than 32000 subdirectories.
2630 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2632 Description: System hang on startup
2633 Details : This bug allowed the liblustre (e.g. catamount) client to
2634 return to the app before handling all startup RPCs. This
2635 could leave the node unresponsive to lustre network traffic
2636 and manifested as a server ptllnd timeout.
2638 Severity : enhancement
2640 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2641 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2642 boolean to control whether to print peer debug info when a
2643 client's RPC times out.
2648 Description: make O_SYNC working on 2.6 kernels
2649 Details : 2.6 kernels use different method for mark pages for write,
2650 so need add a code to lustre for O_SYNC work.
2655 Description: Failure to close file and release space on NFS
2656 Details : Put inode details into lock acquired in ll_intent_file_open.
2657 Use mdc_intent_lock in ll_intent_open to properly
2658 detect all kind of errors unhandled by mdc_enqueue.
2663 Description: proc file read during shutdown sometimes raced obd removal,
2665 Details : Add lock to prevent obd access after proc file removal.
2668 Frequency : Only for files larger than 4GB on 32-bit clients.
2670 Description: improperly doing page alignment of locks
2671 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2672 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2677 Description: RPCs being resent when they shouldn't be
2678 Details : Some RPCs that should not be resent are being resent. This
2679 can cause inconsistencies in the RPC state machine. Do not
2680 resend such requests.
2683 Frequency : rare, only with NFS export
2685 Description: Crash on NFS re-export node
2686 Details : under very unusual load conditions an assertion is hit in
2687 ll_intent_file_open()
2690 Frequency : only if OST filesystem is corrupted
2692 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2693 Details : for a short time RPCs with bulk IO are in the replay list,
2694 but replay of bulk IOs is unimplemented. If the OST filesystem
2695 is corrupted due to disk cache incoherency and then replay is
2696 started it is possible to trip an assertion. Avoid putting
2697 committed RPCs into the replay list at all to avoid this issue.
2702 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2703 Details : Large single O_DIRECT read and write calls can fail to allocate
2704 a sufficiently large buffer to process the request. In case of
2705 allocation failure the allocation is retried with a smaller
2706 buffer and broken into smaller requests.
2708 Severity : enhancement
2710 Description: Add -o localflock option to simulate old noflock behaviour.
2711 Details : This will achieve local-only flock/fcntl locks coherentness.
2716 Description: versioning check is incomplete
2717 Details : Checking the version difference of client vs. server, report
2718 error if the gap is too big.
2723 Description: add support PG_writeback bit
2724 Details : add support for PG_writeback bit for Lustre, for more carefull
2725 work with page cache in 2.6 kernel. This also fix some deadlocks
2726 and remove hack for work O_SYNC with 2.6 kernel.
2728 Severity : enhancement
2730 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2731 Details : The uninit_groups feature works in conjunction with the kernel
2732 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2733 up the pass1 processing of e2fsck. This is a read-only feature
2734 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2735 mount filesystems that have had this feature enabled.
2737 Severity : enhancement
2739 Description: Improve multi-block allocation algorithm to avoid fragmentation
2740 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2741 allocation locality and avoid filesystem fragmentation.
2743 ------------------------------------------------------------------------------
2745 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2747 * Support for kernels:
2748 2.6.9-42.0.3.EL (RHEL 4)
2749 2.6.5-7.276 (SLES 9)
2750 2.4.21-47.0.1.EL (RHEL 3)
2751 2.6.12.6 vanilla (kernel.org)
2752 2.6.16.21-0.8 (SLES10)
2753 * Recommended e2fsprogs version: 1.39.cfs2-0
2755 * The backwards-compatible /proc/sys/portals symlink has been removed
2756 in this release. Before upgrading, please ensure that you change
2757 any configuration scripts or /etc/sysctl.conf files that access
2758 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2759 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2760 in advance of the upgrade on any system running Lustre 1.4.6 or
2761 newer, since /proc/sys/lnet was added in that version.
2762 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2766 Frequency : only when quota is used
2768 Description: avoid scanning export list for quota master
2769 Details : Change the algorithms to avoid scanning export list in order
2770 to improve the efficiency.
2773 Frequency : MDS failover only, very rarely
2775 Description: "went back in time" messages on mds failover
2776 Details : The greatest transno may be lost when the current operation
2777 finishes with an error (transno==0) and the client's last_rcvd
2778 record is over-written. Save the greatest transno in the
2779 mds_last_transno for this case.
2782 Frequency : always for specific kernels and striping counts
2784 Description: client may get "Matching packet too big" without ACL support
2785 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2786 when trying to access files in certain configurations. The
2787 clients should in fact be denied when mounting because they do
2788 not understand ACLs.
2791 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2793 Description: many clients connecting with IO in progress causes connect timeouts
2794 Details : Avoid synchronous journal commits to avoid delays caused by many
2795 clients connecting/disconnecting when bulk IO is in progress.
2796 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2797 OST_IO_PORTAL to avoid delays behind potentially many pending
2801 Frequency : occasionally with multiple writers to a single file
2803 Description: shared writes to file may result in wrong size reported by stat()
2804 Details : Allow growing of kms when extent lock is cancelled
2807 Frequency : always with random mmap IO to multi-striped file
2809 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2810 Details : Only if the hole is at the end of OST object so that kms is too
2811 small. Fix is to increase kms accordingly in ll_nopage.
2814 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2816 Description: writes to a missing object would leak memory on the OST
2817 Details : If there is an inconsistency between the MDS and OST filesystems,
2818 such that the MDS references an object that doesn't exist, writes
2819 to that object will leak memory due to incorrect cleanup in the
2820 error handling path, eventually running out of memory on the OST.
2825 Description: Creating too long symlink causes lustre errors
2826 Details : Check symlink and name lengths before sending requests to MDS.
2829 Frequency : only if flock is enabled (not on by default)
2831 Description: posix locks not released on fd closure on 2.6.9+
2832 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2833 kernels, this caused such locks not to be released on fd close and
2834 then assertions on fs unmount about still used locks.
2837 Frequency : MDS failover only, very rarely
2839 Description: clients may get ASSERTION(granted_lock != NULL)
2840 Details : When request was taking a long time, and a client was resending
2841 a getattr by name lock request. The were multiple lock requests
2842 with the same client lock handle and
2843 mds_getattr_name->fixup_handle_for_resent_request found one of the
2844 lock handles but later failed with ASSERTION(granted_lock != NULL).
2849 Description: handle->h_buffer_credits > 0, assertion failure
2850 Details : h_buffer_credits is zero after truncate, causing assertion
2851 failure. This patch extends the transaction or creates a new
2855 Frequency : NFS re-export or patchless client
2856 Bugzilla : 11179, 10796
2857 Description: Crash on NFS re-export node (__d_move)
2858 Details : We do not want to hash the dentry if we don't have a lock.
2859 But if this dentry is later used in d_move, we'd hit uninitialised
2860 list head d_hash, so we just do this to init d_hash field but
2861 leave dentry unhashed.
2864 Frequency : NFS re-export or patchless client
2866 Description: NFS exports has problem with symbolic link
2867 Details : lustre client didn't properly install dentry when re-exported
2868 to NFS or running patchless client.
2871 Frequency : NFS re-export or patchless client
2873 Description: Various nfs/patchless fixes.
2874 Details : fixes reuse disconected alias for lookup process - this fixes
2875 warning "find_exported_dentry: npd != pd",
2876 fix permission error with open files at nfs.
2877 fix apply umask when do revalidate.
2880 Frequency : occasional
2882 Description: Crash on NFS re-export node
2883 Details : calling clear_page() on the wrong pointer triggered oops in
2884 generic_mapping_read().
2887 Frequency : rarely, using O_DIRECT IO
2889 Description: unaligned directio crashes client with LASSERT
2890 Details : check for unaligned buffers before trying any requests.
2893 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2895 Description: stale data returned from RAID cache
2896 Details : If only a small amount of IO is done to the RAID device before
2897 reading it again it is possible to get stale data from the RAID
2898 cache instead of reading it from disk.
2901 Frequency : always for sles10 kernel
2903 Description: sles10 support
2904 Details : ll_follow_link: compile fixes and using of nd_set_link
2905 under newer kernels.
2908 Frequency : depends on arch, kernel and compiler version, always on sles10
2911 Description: recursive or deep enough symlinks cause stack overflow
2912 Details : getting rid of large stack-allocated variable in
2916 Frequency : depends on hardware
2918 Description: lustre write performance loss in the SLES10 kernel
2919 Details : the performance loss is caused by using of write barriers in the
2920 ext3 code. The SLES10 kernel turns barrier support on by
2921 default. The fix is to undo that change for ldiskfs.
2923 ------------------------------------------------------------------------------
2925 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2927 * Support for kernels:
2928 2.6.9-42.0.3EL (RHEL 4)
2929 2.6.5-7.276 (SLES 9)
2930 2.4.21-47.0.1.EL (RHEL 3)
2931 2.6.12.6 vanilla (kernel.org)
2935 Frequency : quota enabled and large files being deleted
2937 Description: releasing more than 4GB of quota at once hangs OST
2938 Details : If a user deletes more than 4GB of files on a single OST it
2939 will cause the OST to spin in an infinite loop. Release
2940 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2945 Description: statfs data retrieved from /proc may be stale or zero
2946 Details : When reading per-device statfs data from /proc, in the
2947 {kbytes,files}_{total,free,avail} files, it may appear
2948 as zero or be out of date.
2951 Frequency : systems with MD RAID1 external journal devices
2953 Description: lconf's call to blkid is confused by RAID1 journal devices
2954 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2955 of returning all block devices that match the journal UUID.
2958 Frequency : always, for aggregate stripe size over 4GB
2960 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2961 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2962 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2965 Frequency : always if "lfs find" used on a local file/directory
2967 Description: "lfs find" segfaults if used on a local file/directory
2968 Details : The case where a directory component was not specified wasn't
2969 handled correctly. Handle this properly.
2972 Frequency : always on ppc64
2974 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2975 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2976 perform endianness conversion.
2979 Frequency : rarely (truncate to non-zero file size after write under load)
2980 Bugzilla : 10730, 10687
2981 Description: Files padded with zeros to next 4K multiple
2982 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2983 it is possible that files that are truncated to a non-zero size
2984 immediately after being written are filled with zero bytes beyond
2985 the truncated size. No file data is lost.
2987 Severity : enhancement
2989 Description: Allow recovery/failover for liblustre clients.
2990 Details : liblustre clients were unaware of failover configurations until
2993 Severity : enhancement
2995 Description: user file locks should fail when not mounting with flock option
2996 Details : Set up an error-returning stub in ll_file_operations.lock field
2997 to prevent incorrect behaviour when client is mounted without
2998 flock option. Also, set up properly f_op->flock field for
3002 Frequency : always on ia64
3004 Description: "lfs df" loops on printing out MDS statfs information
3005 Details : The obd_ioctl_data was not initialized and in some systems
3006 this caused a failure during the ioctl that did not return
3007 an error. Initialize the struct and return an error on failure.
3010 Frequency : SLES 9 only
3012 Description: Error of copying files with lustre special EAs as root
3013 Details : Client side always return success for setxattr call for lustre
3014 special xattr (currently only "trusted.lov").
3017 Frequency : rarely on clusters with both ia64+i386 clients
3019 Description: ia64+i686 clients doing shared IO on the same file may LBUG
3020 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
3021 clients are doing concurrent writes to the same file it is
3022 possible that the ia64 clients may LASSERT because the OST
3023 extent locks are not PAGE_SIZE aligned. Ensure that grown
3024 locks are always aligned on the request boundary.
3027 Frequency : specific use, occasional
3029 Description: Overwriting in use executable truncates on-disk binary image
3030 Details : If one node attempts to overwrite an executable in use by
3031 another node, we now correctly return ETXTBSY instead of
3032 truncating the file.
3034 Severity : enhancement
3036 Description: Async OSC create to avoid the blocking unnecessarily.
3037 Details : If a OST has no remain object, system will block on the creating
3038 when need to create a new object on this OST. Now, ways use
3039 pre-created objects when available, instead of blocking on an
3040 empty osc while others are not empty. If we must block, we block
3041 for the shortest possible period of time.
3046 Description: chmod on Lustre root is propagated to other clients
3047 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
3048 invalid by the follow_mount time.
3051 Frequency : liblustre clients only
3053 Description: Race in 'instant cancel' lock handling could lead to such locks
3054 never to be granted in case of SMP MDS
3055 Details : Do not destroy not yet granted but cbpending locks in
3059 Frequency : replay/resend of open
3061 Description: non null lock assetion failure in mds_intent_policy
3062 Details : Trying to replay/resend lockless open requests resulted in
3063 mds_open() returning 0 with no lock. Now it sets a flag if
3064 a lock is going to be returned.
3066 Severity : enhancement
3068 Description: Checksum enhancements
3069 Details : New checksum enhancements allow for resending RPCs that failed
3072 Severity : enhancement
3074 Description: Tunables on number of dirty pages in cacche
3075 Details : Allow to set limit on number of dirty pages cached.
3080 Description: client crash on unmount - lock still has references
3081 Details : In some error handling cases it was possible to leak a lock
3082 reference on a client while accessing a file. This was not
3083 harmful to the client during operation, but would cause the
3084 client to crash when the filesystem is unmounted.
3087 Frequency : specific case, rare
3089 Description: ETXTBSY on mds though file not in use by client
3090 Details : ETXTBSY is no longer incorrectly returned when attempting to
3091 chmod or chown a directory that the user previously tried to
3092 execute or a currently-executing binary.
3095 Frequency : extremely rare except on liblustre-based clients
3097 Description: Lustre space not freed when files are deleted
3098 Details : Clean up open-unlinked files after client eviction. Previously
3099 the unlink was skipped and the files remained as orphans.
3104 Description: OST failure "would be an LBUG" in waiting_locks_callback()
3105 Details : In some cases it was possible to send a blocking callback to
3106 a client doing a glimpse, even though that client didn't get
3107 a lock granted. When the glimpse lock is cancelled on the OST
3108 the freed lock is left on the waiting list and corrupted the list.
3111 Frequency : all core dumps
3113 Description: Broke core dumps to lustre
3114 Details : Negative dentry may be unhashed if parent does not have UPDATE
3115 lock, but some callers, e.g. do_coredump, expect dentry to be
3116 hashed after successful create, hash it in ll_create_it.
3118 ------------------------------------------------------------------------------
3120 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
3122 * Support for kernels:
3123 2.6.9-42.0.2.EL (RHEL 4)
3124 2.6.5-7.276 (SLES 9)
3125 2.4.21-40.EL (RHEL 3)
3126 2.6.12.6 vanilla (kernel.org)
3130 Frequency : always on RHEL 3
3132 Description: Number of open files grows over time
3133 Details : The number of open files grows over time, whether or not
3134 Lustre is started. This was due to a filp leak introduced
3135 by one of our kernel patches.
3137 ------------------------------------------------------------------------------
3139 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3141 * Support for kernels:
3142 2.6.9-42.EL (RHEL 4)
3143 2.6.5-7.276 (SLES 9)
3144 2.4.21-40.EL (RHEL 3)
3145 2.6.12.6 vanilla (kernel.org)
3150 Bugzilla : 5719, 9635, 9792, 9684
3151 Description: OST (or MDS) trips assertions in (re)connection under heavy load
3152 Details : If a server is under heavy load and cannot reply to new
3153 connection requests before the client resends the (re)connect,
3154 the connection handling code can behave badly if two service
3155 threads are concurrently handing separate (re)connections from
3156 the same client. Add better locking to the connection handling
3157 code, and ensure that only a single connection will be processed
3158 for a given client UUID, even if the lock is dropped.
3160 Severity : enhancement
3162 Description: add TCP zero-copy support to kernel
3163 Details : Add support to the kernel TCP stack to allow zero-copy bulk
3164 sends if the hardware supports scatter-gather and checksumming.
3165 This allows socklnd to do client-write and server-read more
3166 efficiently and reduce CPU utilization from skbuf copying.
3169 Frequency : only if NFS exporting from client
3171 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
3172 Details : If setting attributes on a file created under NFS that had
3173 never been opened it would be possible to oops the client
3174 if the file had no objects.
3177 Frequency : always for liblustre
3179 Description: liblustre client does MDS+OSTs setattr RPC for each write
3180 Details : When doing a write from a liblustre client, the client
3181 incorrectly issued an RPC to the MDS and each OST the file was
3182 striped over in order to update the timestamps. When writing
3183 with small chunks and many clients this could overwhelm the MDS
3184 with RPCs. In all cases it would slow down the write because
3185 these RPCs are unnecessary.
3187 Severity : enhancement
3189 Description: allow number of MDS service threads to be changed at module load
3190 Details : It is now possible to change the number of MDS service threads
3191 running. Adding "options mds mds_num_threads={N}" to the MDS's
3192 /etc/modprobe.conf will set the number of threads for the next
3193 time Lustre is restarted (assuming the "mds" module is also
3194 reloaded at that time). The default number of threads will
3195 stay the same, 32 for most systems.
3200 Description: OST crash if filesystem is unformatted or corrupt
3201 Details : If an OST is started on a device that has never been formatted
3202 or if the filesystem is corrupt and cannot even mount then the
3203 error handling cleanup routines would dereference a NULL pointer.
3208 Description: NULL pointer deref in llap_from_page.
3209 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3210 incorrect) mapping if the page was truncated/reclaimed while it was
3211 searched for. Check for this condition and skip such pages when
3212 doing readahead. Introduce extra check to llap_from_page() to
3213 verify page->mapping->host is non-NULL (so page is not anonymous).
3216 Frequency : Sometimes when using sys_sendfile
3218 Description: "page not covered by a lock" warnings from ll_readpage
3219 Details : sendfile called ll_readpage without right page locks present.
3220 Now we introduced ll_file_sendfile that does necessary locking
3221 around call to generic_file_sendfile() much like we do in
3225 Frequency : with certain MDS communication failures at client mount time
3227 Description: NULL pointer deref after failed client mount
3228 Details : a client connection request may delayed by the network layer
3229 and not be sent until after the PTLRPC layer has timed out the
3230 request. If the client fails the mount immediately it will try
3231 to clean up before the network times out the request. Add a
3232 reference from the request import to the obd device and delay
3233 the cleanup until the network drops the request.
3236 Frequency : occasionally during client (re)connect
3238 Description: assertion failure during client (re)connect
3239 Details : processing a client connection request may be delayed by the
3240 client or server longer than the client connect timeout. This
3241 causes the client to resend the connection request. If the
3242 original connection request is replied in this interval, the
3243 client may trip an assertion failure in ptlrpc_connect_interpret()
3244 which thought it would be the only running connect process.
3247 Frequency : only with obd_echo servers and clients that are rebooted
3249 Description: kernel BUG accessing uninitialized data structure
3250 Details : When running an obd_echo server it did not start the ping_evictor
3251 thread, and when a client was evicted an uninitialized data
3252 structure was accessed. Start the ping_evictor in the RPC
3253 service startup instead of the OBD startup.
3255 Severity : enhancement
3256 Bugzilla : 10193 (patchless)
3257 Description: Remove dependency on various unexported kernel interfaces.
3258 Details : No longer need reparent_to_init, exit_mm, exit_files,
3259 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3262 Frequency : rare (only users of deprecated and unsupported LDAP config)
3264 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3265 Details : LDAP apparently contains 'lustreName' attributes instead of
3266 'name'. A simple remapping of the name is sufficient.
3269 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3271 Description: waiting_locks_callback trips kernel BUG if client is evicted
3272 Details : Running with the dump_on_timeout debug flag turned on makes
3273 it possible that the waiting_locks_callback() can try to dump
3274 the Lustre kernel debug logs from an interrupt handler. Defer
3275 this log dumping to the expired_lock_main() thread.
3277 Severity : enhancement
3279 Description: Support NFS exporting on 2.6 kernels.
3280 Details : Implement non-rawops metadata methods for NFS server to use without
3281 changing NFS server code.
3284 Frequency : very rare (synthetic metadata workload only)
3286 Description: two racing renames might cause an MDS thread to deadlock
3287 Details : Running the "racer" program may cause one MDS thread to rename
3288 a file from being the source of a rename to being the target of
3289 a rename at exactly the same time that another thread is doing
3290 so, and the second thread has already enqueued these locks after
3291 doing a lookup of the target and is trying to relock them in
3292 order. Ensure that we don't try to re-lock the same resource.
3295 Frequency : only very large systems with liblustre clients
3297 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3298 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3299 all OSTs in parallel. This allows the eviction of stale liblustre
3300 clients to proceed much faster than if they were done in series,
3301 and also offers similar improvements for other set_info RPCs.
3306 Description: excessive CPU usage during initial read phase on client
3307 Details : During the initial read phase on a client, it would agressively
3308 retry readahead on the file, consuming too much CPU and impacting
3309 performance (since 1.4.5.8). Improve the readahead algorithm
3310 to avoid this, and also improve some other common cases (read
3311 of small files in particular, where "small" is files smaller than
3312 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3317 Description: MDS crash when receiving packet with unknown intent.
3318 Details : Do not LBUG in unknown intent case, just return -EFAULT
3320 Severity : enhancement
3321 Bugzilla : 9293, 9385
3322 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3323 Details : Do not serialize getattr (non-intent version) and statfs.
3326 Frequency : occasional, when OST network is overloaded/intermittent
3328 Description: client evicted by OST after bulk IO timeout
3329 Details : If a client sends a bulk IO request (read or write) the OST
3330 may evict the client if it is unresposive to its data GET/PUT
3331 request. This is incorrect if the network is overloaded (takes
3332 too long to transfer the RPC data) or dropped the OST GET/PUT
3333 request. There is no need to evict the client at all, since
3334 the pinger and/or lock callbacks will handle this, and the
3335 client can restart the bulk request.
3338 Frequency : Always when mmapping file with no objects
3340 Description: client crashes when mmapping file with no objects
3341 Details : Check that we actually have objects in a file before doing any
3342 operations on objects in ll_vm_open, ll_vm_close and
3348 Description: Request leak when working with deleted CWD
3349 Details : Introduce advanced request refcount tracking for requests
3350 referenced from lustre intent.
3352 Severity : Enhancement
3354 Description: Cache open file handles on client.
3355 Details : MDS now will return special lock along with openhandle, if
3356 requested and client is allowed to hold openhandle, even if unused,
3357 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3358 closing files for every read/write openration.
3360 Severity : Enhancement
3362 Description: Cache open negative dentries on client when possible.
3363 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3364 negative dentries on lock revocation.
3369 Description: Remounting a client read-only wasn't possible with a zconf mount
3370 Details : It wasn't possible to remount a client read-only with llmount.
3372 Severity : enhancement
3373 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3374 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3375 MPICH is included to add Lustre-specific ADIO interfaces.
3376 This is based closely on the UFS ADIO layer and only differs
3377 in file creation, in order to allow the OST striping to be set.
3378 This is user-contributed code and not supported by CFS.
3383 Description: extended inode attributes (immutable, append-only) work improperly
3384 when 2.4 and 2.6 kernels are used on client/server or vice versa
3385 Details : Introduce kernel-independent values for these flags.
3387 Severity : enhancement
3390 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3391 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3392 granularity. Now, Fractional MB granularity is be supported,
3393 this is very useful for low memory system.
3395 Severity : enhancement
3397 Description: Getattr by fid
3398 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3399 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3400 have locks and avoids getattr rpc requests for it on every stat.
3403 Frequency : Always, for filesystems larger than 2TB
3405 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3406 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3407 filesystems larger than 2TB to be mounted with mballoc enabled.
3410 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3412 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3413 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3414 filesystems larger than 2TB is mounted then the kernel will
3415 silently corrupt the start of the filesystem. CONFIG_LBD is
3416 enabled for all CFS-supported kernels, but the possibility of
3417 this happening with a modified kernel config exists.
3419 Severity : enhancement
3421 Description: add client O_DIRECT support for 2.6 kernels
3422 Details : It is now possible to do O_DIRECT reads and writes to files
3423 in the Lustre client mountpoint on 2.6 kernel clients.
3425 Severity : enhancement
3427 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3428 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3429 parallel, not waiting for response from every OST before sending
3430 a rpc to the next OST.
3435 Description: setattr vs write race when updating file timestamps
3436 Details : Client processes that update a file timestamp into the past
3437 right after writing to the file (e.g. tar) it is possible that
3438 the updated file modification time can be reset to the current
3439 time due to a race between processing the setattr and write RPC.
3441 Severity : enhancement
3443 Description: Bring 'lfs find' closer in line with regular Linux find.
3444 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3445 -print0 options and obtains all the needed info through the lustre
3448 Severity : enhancement
3450 Description: support up to 1024 configured devices on one node
3451 Details : change obd_dev array from statically allocated to dynamically
3452 allocated structs as they are first used to reduce memory usage
3457 Description: Flush dirty partially truncated pages during truncate
3458 Details : Immediatelly flush partially truncated pages in filter_setattr,
3459 this way we completely avoid having any pages in page cache on OST
3460 and can retire ugly workarounds during writes to flush such pages.
3465 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3466 Details : protect inode from truncation within vfs_unlink() context
3467 just take a reference before calling vfs_unlink() and release it
3468 when parent's i_sem is free.
3473 Description: last_id value checked outside lock on OST caused LASSERT failure
3474 Details : If there were multiple MDS->OST object precreate requests in
3475 flight, it was possible that the OST's last object id was checked
3476 outside a lock and incorrectly tripped an assertion. Move checks
3477 inside locks, and discard old precreate requests.
3480 Frequency : always, if extents are used on OSTs
3482 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3483 Details : OSTs using the extents format would not zero the high 16 bits of
3484 the index physical block number. This is not a problem for any
3485 OST filesystems smaller than 16TB, and no kernels support ext3
3486 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3487 new/modified files) and can be fixed for existing filesystems
3488 with e2fsprogs-1.39-cfs1.
3493 Description: import connection selection may be incorrect if timer wraps
3494 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3495 import connections to be ignored if the 32-bit jiffies counter
3496 wraps. Use a 64-bit jiffies counter.
3499 Frequency : very large clusters immediately after boot
3501 Description: LNET request buffers exhausted under heavy short-term load
3502 Details : If a large number of client requests are generated on a service
3503 that has previously never seen so many requests it is possible
3504 that the request buffer growth cannot keep up with the spike in
3505 demand. Instead of dropping incoming requests, they are held in
3506 the LND until the RPC service can accept more requests.
3509 Frequency : Sometimes during replay
3511 Description: Assertion failure in ll_local_open after replay.
3512 Details : If replay happened on an open request reply before we were able
3513 to set replay handler, reply will become not swabbed tripping the
3514 assertion in ll_local_open. Now we set the handler right after
3515 recognising of open request
3518 Frequency : very rare
3520 Description: kernel reports "badness in vsnprintf"
3521 Details : Reading from the "recovery_status" /proc file in small chunks
3522 may cause a negative length in lprocfs_obd_rd_recovery_status()
3523 call to vsnprintf() (which is otherwise harmless). Exit early
3524 if there is no more space in the output buffer.
3526 Severity : enhancement
3528 Description: clear OBD RPC statistics by writing to them
3529 Details : It is now possible to clear the OBD RPC statistics by writing
3530 to the "stats" file.
3535 Description: Client mtime is not the same on different clients after utimes
3536 Details : In some cases, the client was using the utimes() syscall on
3537 a file cached on another node. The clients now validate the
3538 ctime from the MDS + OSTs to determine which one is right.
3543 Description: Inability to activate failout mode
3544 Details : lconf script incorrectly assumed that in pythong string's numeric
3545 value is used in comparisons.
3548 Frequency : always with multiple stripes per file
3550 Description: Inefficient object allocation for mutli-stripe files
3551 Details : When selecting which OSTs to stripe files over, for files with
3552 a stripe count that divides evenly into the number of OSTs,
3553 the MDS is always picking the same starting OST for each file.
3554 Return the OST selection heuristic to the original design.
3559 Description: mount failures may take full timeout to return an error
3560 Details : Under some heavy load conditions it is possible that a
3561 failed mount can wait for the full obd_timeout interval,
3562 possibly several minutes, before reporting an error.
3563 Instead return an error as soon as the status is known.
3565 Frequency : quota enabled and large files being deleted
3567 Description: releasing more than 4GB of quota at once hangs OST
3568 Details : If a user deletes more than 4GB of files on a single OST it
3569 will cause the OST to spin in an infinite loop. Release
3570 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3575 Description: statfs data retrieved from /proc may be stale or zero
3576 Details : When reading per-device statfs data from /proc, in the
3577 {kbytes,files}_{total,free,avail} files, it may appear
3578 as zero or be out of date.
3581 Frequency : systems with MD RAID1 external journal devices
3583 Description: lconf's call to blkid is confused by RAID1 journal devices
3584 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3585 of returning all block devices that match the journal UUID.
3588 Frequency : always, for aggregate stripe size over 4GB
3590 Description: assertion fails when trying to use 4GB stripe size
3591 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3592 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3595 Frequency : always on ppc64
3597 Description: the first write on an ext3 filesystem with mballoc got stuck
3598 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3599 perform endianness conversion.
3601 ------------------------------------------------------------------------------
3603 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3605 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3606 INTEROPERATE with older versions automatically. Please read the
3607 user documentation before upgrading any part of a live system.
3608 * WARNING: Lustre networking configuration changes are required with
3609 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3612 * Support for kernels:
3613 2.6.9-22.0.2.EL (RHEL 4)
3614 2.6.5-7.244 (SLES 9)
3615 2.6.12.6 vanilla (kernel.org)
3618 Severity : enhancement
3619 Bugzilla : 7981/8208
3620 Description: Introduced Lustre Networking (LNET)
3621 Details : LNET is new networking infrastructure for Lustre, it includes
3622 a reorganized network configuration mode (see the user
3623 documentation for full details) as well as support for routing
3624 between different network fabrics. Lustre Networking Devices
3625 (LNDs) for the supported network fabrics have also been
3626 created for this new infrastructure.
3628 Severity : enhancement
3629 Description: Introduced Access control lists
3630 Details : clients can set ACLs on files and directories in order to have
3631 more fine-grained permissions than the standard Unix UGO+RWX.
3632 The MDS must be started with the "-o acl" mount option.
3634 Severity : enhancement
3635 Description: Introduced filesystem quotas
3636 Details : Administrators may now establish per-user quotas on the
3639 Severity : enhancement
3641 Description: Configuration change for the XT3
3642 The PTLLND is now used to run Lustre over Portals on the XT3
3643 The configure option(s) --with-cray-portals are no longer used.
3644 Rather --with-portals=<path-to-portals-includes> is used to
3645 enable building on the XT3. In addition to enable XT3 specific
3646 features the option --enable-cray-xt3 must be used.
3651 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3652 Details : A race between allocating a new llcd and re-getting the llcd_lock
3653 allowed another thread to grab newly-allocated llcd.
3655 Severity : enhancement
3657 Description: 2.6 OST async journal commit and locking fix to improve performance
3658 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3659 2.6 kernels are now async as they already were in 2.4 kernels so
3660 that they can commit concurrently with the network bulk transfer.
3661 For block-allocated files the filter allocation semaphore is held
3662 to avoid filesystem fragmentation during allocation. BKL lock
3663 removed for 2.6 xattr operations where it is no longer needed.
3668 Description: lconf incorrectly determined whether two IP networks could talk
3669 Details : In some more complicated routing and multiple-network
3670 configurations, lconf will avoid trying to make a network
3671 connection to a disjoint part of the IP space. It was doing the
3672 math incorrectly for one set of cases.
3677 Description: Fix for potential infinite loop processing records in an llog.
3678 Details : If an llog record is corrupted/zeroed, it is possible to loop
3679 forever in llog_process(). Validate the llog record length
3680 and skip the remainder of the block on error.
3683 Frequency : occasional (liblustre only)
3685 Description: liblustre could not open files whose last component is a symlink
3686 Details : sysio_path_walk() would incorrectly pass the open intent to
3687 intermediate path components.
3690 Frequency : rare (liblustre only with non-standard tuning)
3691 Bugzilla : 7201 (7350)
3692 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3693 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3694 after it has been released, only do this for referenced locks
3696 Severity : enhancement
3698 Description: specifying an (invalid) directory default stripe_size of -1
3699 would reset the directory default striping
3700 Details : stripe_size -1 was used internally to signal directory stripe
3701 removal, now use "all default" to signal dir stripe removal
3702 as a directory striping of "all default" is not useful
3705 Frequency : common for large clusters running liblustre clients
3707 Description: doing an ls when liblustre clients are running is slow
3708 Details : sending a glimpse AST to a liblustre client waits for every AST
3709 to time out, as liblustre clients will not respond. Since they
3710 cannot cache data we refresh the OST lock LVB from disk instead.
3712 Severity : enhancement
3714 Description: doing an ls at the same time as file IO can be slow
3715 Details : enqueue and other "small" requests can be blocked behind many
3716 large IO requests. Create a new OST IO portal for non-IO
3717 requests so they can be processed faster.
3720 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3722 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3723 created file with mode 000
3724 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3725 allow an ftruncate() as a non-root user to a file with mode 000.
3726 Lustre now respects this flag to disable mode checks when
3727 truncating a file owned by the user
3730 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3733 Description: Revoking locks from clients that went dead or catatonic might take
3735 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3736 cancellation of such locks instant on servers without waiting for
3737 any reply from clients. Clients drops these locks when cancel
3738 notification from server is received without replying.
3741 Frequency : liblustre-only, when liblustre client dies or becomes busy
3743 Description: Doing ls on Linux clients can take a long time with active
3745 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3746 granting such locks to it in the first place if possible. Locks
3747 are taken by proxy on the OST during the read or write and
3748 dropped immediately afterward. Add connect flags handling, do
3749 not grant locks to liblustre clients for glimpse ASTs.
3751 Severity : enhancement
3753 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3754 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3755 sequential reads and behaves badly with applications doing random
3756 reads. Improve it by reading ahead at least read region, and
3757 avoiding excessive large RPC for small reads.
3759 Severity : enhancement
3761 Description: Creating more than 1000 files for a single job may cause a load
3762 imbalance on the OSTs if there are also a large number of OSTs.
3763 Details : qos_prep_create() uses an OST index reseed value that is an
3764 even multiple of the number of available OSTs so that if the
3765 reseed happens in the middle of the object allocation it will
3766 still utilize the OSTs as uniformly as possible.
3771 Description: OST or MDS may oops in ping_evictor_main()
3772 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3773 but doesn't restart at beginning of obd_exports_timed list
3776 Severity : enhancement
3778 Description: improve by-nid export eviction on the MDS and OST
3779 Details : allow multiple exports with the same NID to be evicted at one
3780 time without re-searching the exports list.
3783 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3785 Description: MDS may oops in groups_free()
3786 Details : in rare race conditions a newly allocated group_info struct is
3787 freed again, and this can be NULL. The 2.4 compatibility code
3788 for groups_free() checked for a NULL pointer, but 2.6 did not.
3791 Frequency : common for liblustre clients doing little filesystem IO
3792 Bugzilla : 9352, 7313
3793 Description: server may evict liblustre clients accessing contended locks
3794 Details : if a client is granted a lock or receives a completion AST
3795 with a blocking AST already set it would not reply to the AST
3796 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3799 Frequency : lfs setstripe, only systems with more than 160 OSTs
3801 Description: unable to set striping with a starting offset beyond OST 160
3802 Details : llapi_create_file() incorrectly limited the starting stripe
3803 index to the maximum single-file stripe count.
3806 Frequency : LDAP users only
3808 Description: lconf did not handle in-kernel recovery with LDAP properly
3809 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3811 Severity : enhancement
3813 Description: bind OST threads to NUMA nodes to improve performance
3814 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3815 node and do their allocations there to localize memory access
3817 Severity : enhancement
3819 Description: llmount can determine client NID directly from Myrinet (GM)
3820 Details : the client NID code from gmnalnid was moved directly into
3821 llmount, removing the need to use this or specifying the
3822 client NID explicitly when mounting GM clients with zeroconf
3825 Frequency : if client is started with down MDS
3827 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3828 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3829 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3830 ptlrpc_import_delayed_req() because we didn't check if the
3831 request was interrupted, and we also didn't break out of the
3832 event loop if there was no timeout
3837 Description: data loss during non-page-aligned writes to a single file from
3838 both multiple nodes and multiple threads on one node at same time
3839 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3840 inconsistency led to false short-reads, that were cached and later
3841 used by ->prepare_write() to fill in partially written page,
3842 leading to data loss.
3845 Frequency : always, if lconf --abort_recovery used
3847 Description: lconf --abort_recovery fails with 'Operation not supported'
3848 Details : lconf was attempting to abort recovery on the MDT device and not
3851 Severity : enhancement
3853 Description: remove cleanup logs
3854 Details : replace lconf-generated cleanup logs with lustre internal
3855 cleanup routines. Eliminates the need for client-cleanup and
3858 Severity : enhancement
3860 Description: add support for EAs (user and system) on lustre filesystems
3861 Details : it is now possible to store extended attributes in the Lustre
3862 client filesystem, and with the user_xattr mount option it
3863 is possible to allow users to store EAs on their files also
3865 Severity : enhancement
3867 Description: Add possibility (config option) to show minimal available OST free
3869 Details : When compiled with --enable-mindf configure option, statfs(2)
3870 (and so, df) will return least minimal free space available from
3871 all OSTs as amount of free space on FS, instead of summary of
3872 free spaces of all OSTs.
3874 Severity : enhancement
3876 Description: do not expand extent locks acquired on OST-side
3877 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3878 by server: they are not cached anyway.
3881 Frequency : when mmap is used/binaries executed from Lustre
3883 Description: Unmmap pages before throwing them away from read cache.
3884 Details : llap_shrink cache now attempts to unmap pages before discarding
3885 them (if unmapping failed - do not discard). SLES9 kernel has
3886 extra checks that trigger if this unmapping is not done first.
3891 Description: lconf didn't resolve symlinks before checking to see whether a
3892 given mountpoint was already in use
3895 Frequency : when migrating failover services
3896 Bugzilla : 6395, 9514
3897 Description: When migrating a subset of services from a node (e.g. failback
3898 from a failover service node) the remaining services would
3899 time out and evict clients.
3900 Details : lconf --force (implied by --failover) sets the global obd_timeout
3901 to 5 seconds in order to quickly disconnect, but this caused
3902 other RPCs to time out too quickly. Do not change the global
3903 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3905 Severity : enhancement
3906 Frequency : if MDS is started with down OST
3907 Bugzilla : 9439,5706
3908 Description: Allow startup/shutdown of an MDS without depending on the
3909 availability of the OSTs.
3910 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3911 Add appropriate locking and lov-osc refcounts for safe
3912 cleaning. Add osc abort_inflight calls in case the
3913 synchronize never started.
3916 Frequency : occasional (Cray XT3 only)
3918 Description: root not authorized to access files in CRAY_PORTALS environment
3919 Details : The client process capabilities were not honoured on the MDS in
3920 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3921 been accessed by an authorized user then root was able to access
3922 the file on the local client also. The root user capabilities
3923 are now allowed on the MDS, as this environment has secure UID.
3926 Frequency : occasional
3928 Description: ldiskfs "too long searching" message happens too often
3929 Details : A debugging message (otherwise harmless) prints too often on
3930 the OST console. This has been reduced to only happen when
3931 there are fragmentation problems on the filesystem.
3936 Description: Division by zero in statfs when all OSCs are inactive
3937 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3938 lov_statfs divides by value returned by lov_get_stripecnt().
3942 Bugzilla : 9489, 3273
3943 Description: First write from each client to each OST was only 4kB in size,
3944 to initialize client writeback cache, which caused sub-optimal
3945 RPCs and poor layout on disk for the first writen file.
3946 Details : Clients now request an initial cache grant at (re)connect time
3947 and so that they can start streaming writes to the cache right
3948 away and always do full-sized RPCs if there is enough data.
3949 If the OST is rebooted the client also re-establishes its grant
3950 so that client cached writes will be honoured under the grant.
3955 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3956 Details : Now I/O RPCs go to different portal number and (presumably) fast
3957 lock requests (and glimses) and other RPCs get their own service
3958 threads pool that should be able to service those RPCs
3961 Severity : enhancement
3963 Description: Ability to exchange lustre version between client and servers and
3964 issue warnings at client side if client is too old. Also for
3965 liblustre clients there is ability to refuse connection of too old
3967 Details : New 'version' field is added to connect data structure that is
3968 filled with version info. That info is later checked by server and
3972 Frequency : rare, liblustre only.
3973 Bugzilla : 9296, 9581
3974 Description: Two simultaneous writes from liblustre at offset within same page
3975 might proceed at the same time overwriting eachother with stale
3977 Details : I/O lock withing llu_file_prwv was released too early, before data
3978 actually was hitting the wire. Extended lock-holding time until
3979 server acknowledges receiving data.
3982 Frequency : extremely rare. Never observed in practice.
3984 Description: avoid generating lustre_handle cookie of 0.
3985 Details : class_handle_hash() generates handle cookies by incrementing
3986 global counter, and can hit 0 occasionaly (this is unlikely, but
3987 not impossible, because initial value of cookie counter is
3988 selected randonly). Value of 0 is used as a sentinel meaning
3989 "unassigned handle" --- avoid it. Also coalesce two critical
3990 sections in this function into one.
3992 Severity : enhancement
3994 Description: allow liblustre clients to delegate truncate locking to OST
3995 Details : To avoid overhead of locking, liblustre client instructs OST to
3996 take extent lock in ost_punch() on client's behalf. New connection
3997 flag is added to handle backward compatibility.
3999 Severity : enhancement
4000 Bugzilla : 4928, 7341, 9758
4001 Description: allow number of OST service threads to be specified
4002 Details : a module parameter allows the number of OST service threads
4003 to be specified via "options ost ost_num_threads={N}" in the
4004 OSS's /etc/modules.conf or /etc/modprobe.conf.
4008 Bugzilla : 6146, 9635, 9895
4009 Description: servers crash with bad pointer in target_handle_connect()
4010 Details : In rare cases when a client is reconnecting it was possible that
4011 the connection request was the last reference for that export.
4012 We would temporarily drop the export reference and get a new
4013 one, but this may have been the last reference and the export
4014 was just destroyed. Get new reference before dropping old one.
4016 Severity : enhancement
4017 Frequency : if client is started with failover MDS
4019 Description: Allow multiple MDS hostnames in the mount command
4020 Details : Try to read the configuration from all specified MDS
4021 hostnames during a client mount in case the "primary"
4024 Severity : enhancement
4026 Description: Stop sending data to evicted clients as soon as possible.
4027 Details : Check if the client we are about to send or are sending data to
4028 was evicted already. (Check is done every second of waiting,
4029 for which l_wait_event interface was extended to allow checking
4030 of exit condition at specified intervals).
4033 Frequency : rare, normally only when NFS exporting is done from client
4035 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
4037 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
4040 Frequency : always, if srand() is called before liblustre initialization
4042 Description: Liblustre uses system PRNG disturbing its usage by user application
4043 Details : Introduce internal to lustre fast and high-quality PRNG for
4044 lustre usage and make liblustre and some other places in generic
4045 lustre code to use it.
4047 Severity : enhancement
4048 Bugzilla : 9477, 9557, 9870
4049 Description: Verify that the MDS configuration logs are updated when xml is
4050 Details : Check if the .xml configuration logs are newer than the config
4051 logs stored on the MDS and report an error if this is the case.
4052 Request --write-conf, or allow starting with --old_conf.
4054 Severity : enhancement
4056 Description: Handle symlinks in the path when checking if Lustre is mounted.
4057 Details : Resolve intermediate symlinks when checking if a client has
4058 mounted a filesystem to avoid duplicate client mounts.
4063 Description: lconf can hit an error exception but still return success.
4064 Details : The lconf command catches the Command error exception at the top
4065 level script context and will exit with the associated exit
4066 status, but doesn't ensure that this exit status is non-zero.
4071 Description: failure of ptlrpc thread startup can cause oops
4072 Details : Starting a ptlrpc service thread can fail if there are a large
4073 number of threads or the server memory is very fragmented.
4074 Handle this without oopsing.
4077 Frequency : always, only if liblustre and non-default acceptor port was used
4079 Description: liblustre cannot connect to servers with non-default acceptor port
4080 Details : tcpnal_set_default_params() was not called and was therefore
4081 ignoring the environment varaible TCPNAL_PORT, as well as other
4082 TCPNAL_ environment variables
4087 Description: two objects could be created on the same OST for a single file