1 tbd Sun Microsystems, Inc.
4 2.6.16.60-0.27 (SLES 10),
5 2.6.18-92.1.10.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: Re-disable certain /proc logging
19 Details : Enable and disable client's offset_stats, extents_stats and
20 extents_stats_per_process stats logging on the fly.
23 Frequency : Only on FC kernels 2.6.22+
25 Description: oops in statahead
26 Details : Do not drop reference count for the dentry from VFS when lookup,
27 VFS will do that by itself.
29 Severity : enhancement
31 Description: Generic /proc file permissions
32 Details : Set /Proc file permissions in a more generic way to enable non-
33 root users operate on some /proc files.
37 Description: Hitting mdc_commit_close() ASSERTION
38 Details : Properly handle request reference release in
39 ll_release_openhandle().
43 Frequency : only patchless client
44 Description: add workaround for race between add/remove dentry from hash
46 Severity : enhancement
48 Description: Allow OST glimpses to return PW locks
52 Description: LBUG when llog conf file is full
53 Details : When llog bitmap is full, ENOSPC should be returned for plain
58 Description: Prevent import from entering FULL state when server in recovery
62 Description: service mount cannot take device name with ":"
63 Details : Only when device name contains ":/" will mount treat it as
69 Description: replace ptlrpcd with the statahead thread to interpret the async
70 statahead RPC callback
74 Frequency : on recovery
75 Description: I/O failures after umount during fail back
76 Details : if client reconnected to restarted server we need join to recovery
77 instead of find server handler is changed and process self eviction
78 with cancel all locks.
80 Severity : enhancement
82 Description: Update to RHEL5 kernel-2.6.18-92.1.10.el5.
84 Severity : enhancement
86 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.27.
88 Severity : enhancement
90 Description: Upcall on Lustre log has been dumped
91 Details : Allow for a user mode script to be called once a Lustre log has
92 been dumped. It passes the filename of the dumped log to the
93 script, the location of the script can be specified via
94 /proc/sys/lnet/debug_log_upcall.
99 Description: avoid idr_remove called for id which is not allocated.
100 Details : Move assigment s_dev for clustered nfs to end of initialization,
101 to avoid problem with error handling.
106 Description: avoid Already found the key in hash [CONN_UNUSED_HASH] messages
107 Details : When connection is reused this not moved from CONN_UNUSED_HASH
108 into CONN_USED_HASH and this prodice warning when put connection
109 again in unused hash.
111 Severity : enhancement
113 Description: File striping can now be set to use an arbitrary pool of OSTs.
115 Severity : enhancement
117 Description: Export bytes_read/bytes_write count on OSC/OST.
121 Description: Early reply size mismatch, MGC loses connection
122 Details : Apply the MGS_CONNECT_SUPPORTED mask at reconnect time so
123 the connect flags are properly negotiated.
127 Description: Properly propagate oinfo flags from lov to osc for statfs
128 Details : restore missing copy oi_flags to lov requests.
130 Severity : enhancement
132 Description: Add man pages for llobdstat(8), llstat(8), plot-llstat(8),
133 : l_getgroups(8), lst(8), routerstat(8)
134 Details : included man pages for llobdstat(8), llstat(8),
135 : plot-llstat(8), l_getgroups(8), lst(8), routerstat(8)
137 Severity : enhancement
139 Description: configure's --enable-quota should check the
140 : kernel .config for CONFIG_QUOTA
141 Details : configure is terminated if --enable-quota is passed but
142 : no quota support is in kernel
146 Description: Remove portals compatibility
147 Details : Remove portals compatibility, not interoperable with releases
152 Description: Resolve device initialization race
153 Details : Prevent proc handler from accessing devices added to the
154 obd_devs array but yet be intialized.
156 Severity : enhancement
158 Description: Update to SLES10 SP2 kernel-2.6.16.60-0.23.
160 Severity : enhancement
162 Description: Update to RHEL5 kernel-2.6.18-92.1.6.el5.
167 Description: Using wrong pointer in osc_brw_prep_request
168 Details : Access to array[-1] can produce panic if kernel compiled with
169 CONFIG_PAGE_ALLOC enabled
173 Description: Client runs out of low memory
174 Details : Consider only lowmem when counting initial number of llap pages
178 Description: *optional* service tags registration
179 Details : if the "service tags" package is installed on a Lustre node
180 When the filesystem is mounted, a local-node service tag will
181 be created. See http://inventory.sun.com/ for more information
182 about the Service Tags asset management system.
186 Description: Kernel BUG tries to release flock
187 Details : Lustre does not destroy flock lock before last reference goes
188 away. So always drop flock locks when client is evicted and
189 perform unlock regardless of successfulness of speaking to MDS.
193 Description: add recount protection for osc callbacks, so avoid panic on shutdown
197 Description: sanity test 65a fails if stripecount of -1 is set
198 Details : handle -1 striping on filesystem in ll_dirstripe_verify
203 Description: ASSERTION(CheckWriteback(page,cmd)) failed
204 Details : badly clear PG_Writeback bit in ll_ap_completion can produce false
207 Severity : enhancement
209 Description: Update to RHEL5 kernel-2.6.18-53.1.21.el5.
213 Description: do not process already freed flock
214 Details : flock can possibly be freed by another thread before it reaches
215 to ldlm_flock_completion_ast.
219 Description: LBUG during stress test
220 Details : Need properly lock accesses the flock deadlock detection list.
224 Description: oops in page fault handler
225 Details : kernel page fault handler can return two special 'pages' in error case, don't
226 try dereference NOPAGE_SIGBUS and NOPAGE_OMM.
230 Description: timeout with invalidate import.
231 Details : ptlrpcd_check call obd_zombie_impexp_cull and wait request which should be
232 handled by ptlrpcd. This produce long age waiting and -ETIMEOUT
233 ptlrpc_invalidate_import and as result LASSERT.
235 Severity : enhancement
237 Description: Update to RHEL5 kernel-2.6.18-53.1.19.el5.
241 Description: enable MGS and MDT services start separately
242 Details : add a 'nomgs' option in mount.lustre to enable start a MDT with
243 a co-located MGS without starting the MGS, which is a complement
244 to 'nosvc' mount option.
248 Frequency : after recovery
249 Description: precreate to many object's after del orphan.
250 Details : del orphan st in oscc last_id == next_id and this triger growing
251 count of precreated objects. Set flag LOW to skip increase count
252 of precreated objects.
256 Frequency : rare, on clear nid stats
257 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
258 Details : when clean nid stats sometimes try destroy live entry,
259 and this produce panic in free.
263 Description: Stack overflow during MDS log replay
264 ease stack pressure by using a thread dealing llog_process.
268 Description: wait until IO finished before start new when do lock cancel.
269 Details : VM protocol want old IO finished before start new, in this case
270 need wait until PG_writeback is cleared until check dirty flag and
271 call writepages in lock cancel callback.
273 Severity : enhancement
275 Description: using special macro for print time and cleanup in includes.
279 Description: mds_mfd_close() ASSERTION(rc == 0)
280 Details : In mds_mfd_close(), we need protect inode's writecount change
281 within its orphan write semaphore to prevent possible races.
285 Description: Obsolete CURRENT_SECONDS and use cfs_time_current_sec() instead.
289 Frequency : rare, on shutdown ost
290 Description: don't hit live lock with umount ost.
291 Details : shrink_dcache_parent can be in long loop with destroy dentries,
292 use shrink_dcache_sb instead.
296 Description: don't panic with use echo client
297 Details : echo client pass NULL as client nid pointer and this produce null
302 Description: fix build on ppc32
303 Details : compile code with -m64 flag produce wrong object file for ppc32.
307 Description: add message levels for liblustreapi
311 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
312 Details : Move the 'good_osts' check before the 'total_bavail' check. This
313 will result in an -EAGAIN and in the exit call path we call
314 alloc_rr() which will with increasing aggressiveness attempt to
315 aquire precreated objects on the minimum number of required OSCs.
319 Description: Use old size assignment to avoid deadlock
320 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
321 the scheduling while holding a spinlock. We do not need locking
322 for size in ll_update_inode() because size is only updated from
323 the MDS for directories or files without objects, so there is no
324 other place to do the update, and concurrent access to such inodes
325 are protected by the inode lock.
329 Description: resolve "_IOWR redefined" build error on SLES10
333 Description: dump the memory debugging after all modules are unloaded to
334 suppress false negative in conf_sanity test 39
336 Severity : enhancement
338 Description: build kernel-ib packages for OFED 1.3 in our release cycle
343 Description: fix SLES kernel versioning
344 Details : the kernel version for our SLES 10 kernel did not include a "-"
345 before the "smp" at the end. while this was not a problem in
346 general, it did mean that software trying to use the kernel
347 version to try to detect a vendor specific kernel would fail.
348 this was most evident by the OFED build scripts.
352 Description: Don't update lov_desc members until making sure they are valid
353 Details : When updating lov_desc members via proc fs, need fix their
354 validities before doing the real update.
358 Description: don't put request into delay list while invalidate in flight.
359 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
360 import in flight. this produce timeout for invalidate and sometimes
361 can cause stale data.
365 Frequency : on ppc only
366 Description: not convert ost objects for directory because it's not exist.
367 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
371 Description: Add FMODE_EXEC file flag for SLES10 SP1 kernel.
373 Severity : enhancement
375 Description: Update to support 2.6.22.14 vanilla kernel.
379 Frequency : rare, on recovery
380 Description: read procfs can produce deadlock in some situation
381 Details : Holding lprocfs lock which send rpc can produce block for destroy
382 obd objects and this also block reconnect with -EALREADY. This isn't
383 fix all lprocfs bugs - but make it rare.
385 Severity : enhancement
387 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
390 Frequency : frequent on X2 node
392 Description: mdc_set_open_replay_data LBUG
393 Details : Set replay data for requests that are eligible for replay.
397 Description: lustre_mgs: operation 101 on unconnected MGS
398 Details : When MGC is disconnected from MGS long enough, MGS will evict the
399 MGC, and late on MGC cannot successfully connect to MGS and a lot
400 of the error messages complaining that MGS is not connected.
404 Frequency : on network error
405 Description: panic with double free request if network error
406 Details : mdc_finish_enqueue is finish request if any network error ocuring,
407 but it's true only for synchronus enqueue, for async enqueue
408 (via ptlrpcd) this incorrect and ptlrpcd want finish request
411 Severity : enhancement
413 Description: client-side metadata stat-ahead during readdir(directory readahead)
414 Details : perform client-side metadata stat-ahead when the client detects
415 readdir and sequential stat of dir entries therein
418 Frequency : on start mds
420 Description: Implement get_info(last_id) in obdfilter.
423 Frequency : occasional
425 Description: Correctly check stale fid, not start epoch if ost not support SOM
426 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
427 overwrited with new generated fid, but mds can anwer with one of these
428 two fids and both is not stale. setattr incorectly start epoch and
429 assume will be called done_writeting, but without SOM done_writing
433 Frequency : rare, depends on device drivers and load
435 Description: MDS or OSS nodes crash due to stack overflow
436 Details : Code changes in 1.8.0 increased the stack usage of some functions.
437 In some cases, in conjunction with device drivers that use a lot
438 of stack the MDS (or possibly OSS) service threads could overflow
439 the stack. One change which was identified to consume additional
440 stack has been reworked to avoid the extra stack usage.
443 Frequency : occasional
445 Description: Do not fail import if osc_interpret_create gets -EAGAIN
446 Details : If osc_interpret_create got -EAGAIN it immediately exits and
447 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
448 oscc_has_objects and see OSC has no objests and call
449 oscc_internal_create to resend create request.
451 Severity : enhancement
453 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
455 Severity : enhancement
457 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
460 Frequency : very rare
462 Description: Fix replay if there is an un-replied request and open
463 Details : In some cases, older replay request will revert the
464 mcd->mcd_last_xid on MDS which is used to record the client's
467 Severity : enhancement
469 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
471 Severity : enhancement
473 Description: Add rhel5 support to HEAD.
475 Serverity : enhancement
477 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
482 Description: Don't allow skipping OSTs if index has been specified.
483 Details : Don't allow skipping OSTs if index has been specified, make locking
484 in internal create lots better.
488 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
489 Details : access to bitfield in structure is always rounded to long
490 and this produce problem with not atomic change any bit.
494 Description: Lustre make rpms failed.
495 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
496 builds Lustre rpms from tarball.
499 Frequency : rare on shutdown ost
501 Description: If llog cancel was not send before clean_exports phase, this can
502 produce deadlock in llog code.
503 Details : If llog thread has last reference to obd and call class_import_put
504 this produce deadlock because llog_cleanup_commit_master wait when
505 last llog_commit_thread exited, but this never success because was
506 called from llog_commit_thread.
510 Description: allow userland application know is lost one of stripes.
511 Details : fill lvb_blocks with error code on ost and return it to
512 application if error flag found.
516 Description: NULL lov_tgts causing MDS oops
517 Details : more safe checks for NULL lov_tgts for avoid oops.
519 Severity : enhancement
521 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
525 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
526 Details : If an OST is down the MDS will hang indefinitely in
527 obd_statfs_rqset() waiting for the statfs data. While for
528 MDS QOS usage of statfs, it should not stuck in waiting.
530 Severity : enhancement
532 Description: remote_acl support
533 Details : Support ACL-based permission check for remote user.
534 Support setfacl/getfacl for remote user with the utils
535 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
536 the system "{s,g}etfacl" utils.
538 Severity : enhancement
540 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
542 Severity : enhancement
544 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
548 Description: make mgs_setparam() handle fsname containing dash
549 Details : fsname containing a dash does not work with lctl conf_param
551 Severity : enhancement
553 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
555 Severity : enhancement
557 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
561 Frequency : rare, at shutdown
562 Description: access already free / zero obd_namespace.
563 Details : if client_disconnect_export was called without force flag set,
564 and exist connect request in flight, this can produce access to
565 NULL pointer (or already free pointer) when connect_interpret
566 store ocd flags in obd_namespace.
570 Frequency : only at startup
571 Description: not alloc memory with spinlock held.
572 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
573 if any spinlock held.
575 Severity : enhancement
577 Description: make lustre randomly fail allocating memory
578 Details : Make lustre randomly failed allocating memory for testing purpose.
580 Severity : enhancement
582 Description: lost problems with lov objid file
583 Details : Fixes some scability and access to not inited memory problems
584 in work with lov objdid file.
589 Description: lfs find does not continue on file error
590 Details : Continue other files processing when a file/dir is absent.
594 Description: Inconsistent usage of lustre_pack_reply()
595 Details : Standardize the usage of lustre_pack_reply() such that it
596 always generate a CERROR on failure.
599 Frequency : occasional
601 Description: MDS hang or stay in waiting lock
602 Details : If client receive lock with CBPENDING flag ldlm need send lock
603 cancel as separate rpc, to avoid situation when cancel request
604 can't processed due all i/o threads stay in wait lock.
608 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
612 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
616 Description: Update to SLES9 kernel-2.6.5-7.287.3.
620 Description: Update to RHEL4 latest kernel.
622 Severity : enhancement
624 Description: Build SLES10 patchless client fails
625 Details : The configure was broken by run ./configure with
626 --with-linux-obj=.... argument for patchless client. When the
627 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
628 can't search header adequately. Use absolute path such as
629 -I($LINUX)/include instead.
633 Description: interrupt oig_wait produce painc on resend.
634 Details : brw_redo_request can be used for resend requests from ptlrpcd and
635 private set, and this produce situation when rq_ptlrpcd_data not
636 copyed to new allocated request and triggered LBUG on assert
637 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
638 ptlrpcd set if request is changed and can be safety changed to use
643 Description: LASSERT_{REQ,REP}SWAB macros are buggy
644 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
645 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
646 these macros and replace them with their difinitions instead.
650 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
651 Details : Update which_patch & target file for SLES10 latest kernel.
654 Description: Remove client patches from SLES 10 kernel.
655 Details : This causes SLES 10 clients to behave as patchless clients
656 even on a Lustre-patched (server) kernel.
658 Severity : enhancement
660 Description: self-adjustable client's lru lists
661 Details : use adaptive algorithm for managing client cached locks lru
662 lists according to current server load, other client's work
663 pattern, memory activities, etc. Both, server and client
664 side namespaces provide number of proc tunables for controlling
667 Severity : enhancement
669 Description: light-weight GSS support
670 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
671 and reduce performance overhead.
673 Severity : enhancement
675 Description: Linux keyring support
676 Details : Support using service of Linux keyring for Lustre GSS internal
677 context refresh/cache mechanism.
681 Description: Fix errors in lfs documentation
682 Details : Fixes man pages
686 Description: don't use GFP_* in generic Lustre code.
687 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
691 Description: obdclass is limited by single OBD_ALLOC(idarray)
692 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
696 Description: warnings with build patchless client with vanila 2.6.19 and up
697 Details : change old ctl_table style and replace ctl_table/ctl_table_header
698 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
702 Frequency : Only for SLES
703 Description: sanity_quota fail test_1
704 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
705 file, which makes TSTID[2] inunique.
709 Frequency : Only for SLES9
710 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
711 Details : Update target/ChangeLog/which_patch .
715 Frequency : Only for test_10 in sanity-quota.sh
716 Description: a bug in quota test code
717 Details : checking a test flag in a wrong place
721 Frequency : Only for test_13 in sanity-quota.sh
722 Description: a bug in quota test script
723 Details : shouldn't check overfull allocated quota
727 Description: jbd statistics
728 Details : Port older jbd statistics patch for sles10
730 Severity : enhancement
732 Description: Add printing inode info into message about error in writepage.
737 Description: reply_lock_interpret crash due to race with it and lock cancel.
738 Details : Do not replay locks that are being cancelled. Do not reference
739 locks by their address during replay, just by their handle.
743 Frequency : When flocks are used.
744 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
745 Details : Flock locks might destroy just granted lock if it could be merged
746 with another existing flock, this is done in completion handler,
747 so teach ldlm_cli_enquque_fini that this is a valid case for
753 Description: Oops in read and write path when failing to allocate lock.
754 Details : Check if lock allocation failed and return error back.
758 Description: lstripe command fails for valid OST index
759 Details : The stripe offset is compared to lov->desc.ld_tgt_count
760 instead of lov->desc.ld_active_tgt_count.
764 Description: sanity.sh failed test 103
765 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
770 Description: df doesn't work properly if diskfs blocksize != 4K
771 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
775 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
776 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
778 Severity : enhancement
780 Description: Adaptive timeouts
781 Details : RPC timeouts adapt to changing server load and network
782 conditions to reduce resend attempts and improve recovery time.
786 Description: llapi_file_create() does not allow some changes
787 Details : add llapi_file_open() that allows specifying the mode and
788 open flags, and also returns an open file handle.
792 Description: merge and cleanup kernel patches.
793 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
797 Description: Add journal checksum support.(Kernel part)
798 Details : The journal checksum feature adds two new flags i.e
799 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
800 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
801 indicates that the commit block contains the checksum for
802 the blocks described by the descriptor blocks. Now commit
803 record can be sent to disk without waiting for descriptor
804 blocks to be written to disk. This behavior is controlled
805 using JBD2_FEATURE_ASYNC_COMMIT flag.
809 Description: OSS needs mutliple precreate threads
810 Details : Add ability to start more than one create thread per OSS.
814 Description: Sanity.sh test_65k failed: setstripe should have succeeded
815 Details : execute lfs setstripe on client
819 Description: mds_obd_create error creating tmp object
820 Details : When the user sets quota on root, llog will be affected and can't
821 create files and write files.
824 Frequency : Always on ia64 patchless client, and possibly others.
826 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
827 Details : This allows the patchless client to be loaded on architectures
832 Description: RedHat Update kernel for RHEL5
833 Details : Add the kernel config file for RHEL5.
837 Description: RedHat Update kernel for RHEL5
838 Details : Modify the kernel config file more closer RHEL5.
842 Description: Build failure against Centos5 (RHEL5)
843 Details : Define PAGE_SIZE when it isn't present.
847 Description: test_7 on sanity-quota.sh can't be used on separate machines
848 Details : add facet to handle it
852 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
853 Details : with NFS, the anon dentry's parent was set to itself in
854 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
855 corresponding dentry other than use rec->ur_name.
857 Severity : enhancement
859 Description: lfs setstripe enhancement
860 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
864 Description: enable data checksumming by default
865 Details : enable checksum by default, allow --disable-checksum
866 configure option and "-o nochecksum" mount option. Checksums
867 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
871 Description: lustre support for RHEL5
872 Details : Add support for RHEL5.
876 Description: Client eviction due to failover config
877 Details : after a connection loss, the lustre client should attempt to
878 reconnect to the last active server first before trying the
879 other potential connections.
883 Description: when mds and osts use different quota unit(32bit and 64bit),
884 quota will be released repeatly.
885 Details : void sending multiple quota reqs to mds, which will keep the
886 status between the reqs.
890 Description: osts not allocated evenly to files
891 Details : change the condition to increase offset_idx
894 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
896 Description: Data corruption for OSTs that are formatted larger than 2TB
898 Details : When generating the bio request for lustre file writes the
899 sector number would overflow a temporary variable before being
900 used for the IO. The data reads correctly from Lustre (which
901 will overflow in a similar manner) but other file data or
902 filesystem metadata may be corrupted in some cases.
906 Description: Tune the kernel for good SCSI performance.
907 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
908 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
913 Description: rewrite ext2-derived code in obdclass/uuid.c
914 Details : rewrite inherited code (uuid parsing code from ext2 utils)
915 from scratch preserving functionality.
919 Description: "lfs find -obd UUID" prints directories
920 Details : "lfs find -obd UUID" will return all directory names instead
921 of just file names. It is incorrect because the directories
922 do not reside on the OSTs.
926 Description: use i_size_read and i_size_write in 2.6 port
927 Details : replace inode->i_size access with i_size_read/write()
930 Frequency : when removing large files
932 Description: scheduling issue during removal of large Lustre files
933 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
934 It causes scheduling issues when removing large files (17TB in the
938 Frequency : only with liblustre clients on XT3
940 Description: evictions taking too long
941 Details : allow llrd to evict clients directly on OSTs
944 Frequency : only on ppc
946 Description: /proc/fs/lustre/devices broken on ppc
947 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
948 We should make sure the type of 'index' is loff_t and then cast
949 explicitly as needed below. Do not assign an explicitly cast
953 Frequency : only for rhel5
955 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
956 Details : Modify the target file & which_kernel.
958 Severity : enhancement
960 Description: omit set fsid for export NFS
961 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
962 as FS_REQUIRES_DEV for avoid problems with generate fsid.
966 Frequency : Always, for kernels after 2.6.16
967 Description: Fix warning idr_remove called for id=.. which is not allocated.
968 Details : Last kernels save old s_dev before kill super and not allow
969 to restore from callback - restore it before call kill_anon_super.
973 Description: buffer overruns could theoretically occur
974 Details : llapi_semantic_traverse() modifies the "path" argument by
975 appending values to the end of the origin string, and a
976 overrun may occur. Adding buffer overrun check in liblustreapi.
980 Description: Fix error on 'ls .' at the top of the Lustre mount.
981 Details : Don't revalidate dentry if it is a root dentry.
985 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
986 Details : Modify vm-tunables-rhel4.patch.
990 Description: Kernel config for 2.6.18-vanilla.
991 Details : Modify targets/2.6-vanilla.target.in.
992 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
993 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
994 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
995 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
999 Description: improve handling recoverable errors
1000 Details : if request processig with error which can be recoverable on server
1001 request should be resend, otherwise page released from cache and
1006 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
1007 Details : Modify target file & which_patch.
1008 A flaw was found in the IA32 system call emulation provided
1009 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
1010 value could be stored in the %RAX register, which could trigger an
1011 out-of-bounds system call table access. An untrusted local user
1012 could exploit this flaw to run code in the kernel
1013 (ie a root privilege escalation). (CVE-2007-4573).
1017 Description: O_DIRECT bypasses client statistics.
1018 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1019 read_ahead_stats were updated. copied the stats section from
1020 osc_send_oap_rpc() into async_internal().
1024 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
1028 Description: change order of libsysio includes
1029 Details : '#include sysio.h' should always come before '#include xtio.h'
1033 Description: handle "serious error: objid * already exists" more gracefully
1034 Details : If LAST_ID value on disk is smaller than the objects existing in
1035 the O/0/d* directories, it indicates disk corruption and causes an
1036 LBUG(). If the object is 0-length, then we should use the existing
1037 object. This will help to avoid a full fsck in most cases.
1039 Severity : enhancement
1041 Description: adapt the lustre_config script to support the upgrade case
1042 Details : Add "-u" option for lustre_config script to support upgrading 1.4
1043 server targets to 1.6 in parallel.
1047 Description: To avoid grant space > avaible space when the disk is almost
1048 full. Without this patch you might see the error "grant XXXX >
1049 available" or some LBUG about grant, when the disk is almost
1051 Details : In filter_check_grant, for non_grant cache write, we should
1052 check the left space by if (*left > ungranted + bytes), instead
1053 of (*left > ungranted), because only we are sure the left space
1054 is enough for another "bytes", then the ungrant space should be
1055 increase. In client, we should update cl_avail_grant only there
1056 is OBD_MD_FLGRANT in the reply.
1060 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1061 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1062 described in CVE-2007-4573. This problem could allow untrusted
1063 local users to gain root access.
1066 Frequency : when using O_DIRECT and quotas
1068 Description: Incorrect file ownership on O_DIRECT output files
1069 Details : block usage reported by 'lfs quota' does not take into account
1070 files that have been written with O_DIRECT.
1075 Description: touch file failed when fs is not full
1076 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
1077 otherwise we can get ENOSP while fs is not full.
1081 Description: parallel lock callbacks
1082 Details : Instead of sending blocking and completion callbacks as separated
1083 requests, adding them to a set and sending in parallel.
1086 Frequency : only for Cray XT3
1087 Bugzilla : 12829/13455
1088 Description: Changing primary group doesn't change the group lustre assigns to
1090 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
1091 overridden with the primary group provided by the group upcall,
1092 whereas the supplied fsgid can be trusted if it is in the list of
1093 supplementary groups returned by the group upcall.
1095 Severity : enhancement
1097 Description: Allow masking D_WARNING, D_ERROR messages from console
1098 Details : Console messages can now be disabled via lnet.printk.
1102 Description: User code with malformed file open parameter crashes client node
1103 Details : Before packing join_file req, all the related reference should be
1104 checked carefully in case some malformed flags cause fake
1105 join_file req on client.
1109 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
1110 Details : ldlm_completion_ast() assumes that a lock is granted when the req
1111 mode is equal to the granted mode. However, it should also check
1112 that LDLM_FL_CP_REQD is not set.
1116 Description: Heavy nfs access might result in deadlocks
1117 Details : After ELC code landed, it is now improper to enqueue any mds
1118 locks under och_sem, because enqueue might want to decide to
1119 cancel open locks for same inode we are holding och_sem for.
1123 Description: Client eviction while running blogbench
1124 Details : A lot of unlink operations with concurrent I/O can lead to a
1125 deadlock causing evictions. To address the problem, the number of
1126 oustanding OST_DESTROY requests is now throttled to
1127 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
1128 callbacks are processed in priority.
1132 Description: enable ACLs on MDS by default
1133 Details : ACLs must be enabled on MDS by default.
1136 Frequency : PPC/PPC64 only
1138 Description: conflicts between asm-ppc64/types.h and lustre_types.h
1139 Details : fix duplicated definitions between asm-ppc64/types.h and
1140 lustre_types.h on PPC.
1143 Frequency : PPC/PPC64 only
1145 Description: asm-ppc/segment.h does not exist
1146 Details : fix compile issue on PPC.
1150 Description: better handle error messages in extents code
1153 Frequency : RHEL4 only
1155 Description: mkfs is very slow on IA64/RHEL4
1156 Details : A performance regression has been discovered in the MPT Fusion
1157 driver between versions 3.02.73rh and 3.02.99.00rh. As a
1158 consequence, we have downgraded the MPT Fusion driver in the RHEL4
1159 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
1161 Severity : enhancement
1163 Description: SNMP support enhancement
1164 Details : Adding total number of sampled request for an MDS node in snmp
1167 Severity : enhancement
1169 Description: Optimize ldlm waiting list processing for PR extent locks
1170 Details : When processing waiting list for read extent lock and meeting read
1171 lock that is same or wider to it that is not contended, skip
1172 processing rest of the list and immediatelly return current
1173 status of conflictness, since we are guaranteed there are no
1174 conflicting locks in the rest of the list.
1178 Description: Time out and refuse to reconnect
1179 Details : When the failover node is the primary node, it is possible
1180 to have two identical connections in imp_conn_list. We must
1181 compare not conn's pointers but NIDs, otherwise we can defeat
1182 connection throttling.
1186 Description: port llog fixes from b1_6 into HEAD
1187 Details : Port llog reference couting and some llog cleanups from b1_6
1188 (bug 10800) into HEAD, for protect from panic and access to already
1189 free llog structures.
1193 Description: Detect stride IO mode in read-ahead
1194 Details : When a client does stride read, read-ahead should detect that and
1195 read-ahead pages according to the detected stride pattern.
1199 Description: data checksumming impacts single node performance
1200 Details : add support for several checksum algorithm. Currently, only CRC32
1201 and Adler-32 are supported. The checksum type can be changed on
1202 the fly via /proc/fs/lustre/osc/*/checksum_type.
1206 Description: use adler32 for page checksums
1207 Details : when available, use the Adler-32 algorithm instead of CRC32 for
1212 Description: build for x2 fails
1213 Details : fix compile issue on Cray systems.
1217 Description: Properly match for duplicate locks
1218 Details : Due to different lock order from skiplists code, we need to
1219 traverse entire list for now
1222 Frequency : only on PPC/SLES10
1224 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
1225 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
1226 be defined. Add a hack in mkfs_lustre.c to work around this compile
1231 Description: LASSERT on MDS when client holding flock lock dies
1232 Details : ldlm pool logic depends on number of granted locks equal to
1233 number of released locks which is not true for flock locks, so
1234 just exclude such locks from consideration.
1238 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
1239 Details : Use fsfilt_sync() for both the whole filesystem sync and
1240 individual file sync to eliminate dangerous inode locking
1241 with I_LOCK that can lead to a deadlock.
1245 Description: performance in 1.6.3
1246 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
1247 sure that 1MB requests are not fragmented by the block layer.
1249 Severity : enhancement
1251 Description: organize the server-side client stats on per-nid basis
1252 Details : Change the structure of stats under obdfilter and mds to
1260 The "uuid"s file would list the uuids of _active_ exports.
1261 And the clear entry is to clear all stats and stale nids.
1263 Severity : enhancement
1265 Description: eliminate client locks in face of contention
1266 Details : file contention detection and lockless i/o implementation
1267 for contended files.
1271 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
1274 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
1276 Description: MDS may lose file striping (and hence file data) in some cases
1277 Details : If there are additional extended attributes stored on the MDS,
1278 in particular ACLs, SELinux, or user attributes (if user_xattr
1279 is specified for the client mount options) then there is a risk
1280 of attribute loss. Additionally, the Lustre file striping
1281 needs to be larger than default (e.g. striped over all OSTs),
1282 and an additional attribute must be stored initially in the
1283 inode and then increase in size enough to be moved to the
1284 external attribute block (e.g. ACL growing in size) for file
1289 Description: skiplist implementation simplification
1290 Details : skiplists are used to group compatible locks on granted list
1291 that was implemented as tracking first and last lock of each
1292 lock group the patch changes that to using doubly linked lists
1296 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
1297 Details : Change LASSERTs to client eviction (i.e. abort client's recovery)
1298 because LASSERT on both the data supplied by a client, and the
1299 data on disk is dangerous and incorrect.
1301 Severity : enhancement
1303 Description: Slow truncate/writes to huge files at high offsets.
1304 Details : Directly associate cached pages to lock that protect those pages,
1305 this allows us to quickly find what pages to write and remove
1306 once lock callback is received.
1310 Description: more ldlm soft lockups
1311 Details : In ldlm_resource_add_lock(), call to ldlm_resource_dump()
1312 starve other threads from the resource lock for a long time in
1313 case of long waiting queue, so change the debug level from
1314 D_OTHER to the less frequently used D_INFO.
1316 Severity : enhancement
1318 Description: add -gid, -group, -uid, -user options to lfs find
1322 Description: Hung threads in invalidate_inode_pages2_range
1323 Details : The direct IO path doesn't call check_rpcs to submit a new RPC once
1324 one is completed. As a result, some RPCs are stuck in the queue
1329 Description: filter threads hungs on waiting journal commit
1330 Details : Cleanup filter group llog code, then only filter group llog will
1331 be only created in the MDS/OST syncing process.
1335 Description: Procfs and llog threads access destoryed import sometimes.
1336 Details : Sync the import destoryed process with procfs and llog threads by
1337 the import refcount and semaphore.
1339 Severity : enhancement
1341 Description: openlock cache of b1_6 port to HEAD
1346 Description: kernel BUG at ldiskfs2_ext_new_extent_cb
1347 Details : If insertion of an extent fails, then discard the inode
1348 preallocation and free data blocks else it can lead to duplicate
1353 Description: don't always update ctime in ext3_xattr_set_handle()
1354 Details : Current xattr code updates the inode ctime in ext3_xattr_set_handle.
1355 In some cases the ctime should not be updated, for example for
1356 2.0->1.8 compatibility it is necessary to delete an xattr and it
1357 should not update the ctime.
1361 Bugzilla : 15713/16362
1362 Description: Assertion in iopen_connect_dentry in 1.6.3
1363 Details : looking up an inode via iopen with the wrong generation number can
1364 populate the dcache with a disconneced dentry while the inode
1365 number is in the process of being reallocated. This causes an
1366 assertion failure in iopen since the inode's dentry list contains
1367 both a connected and disconnected dentry.
1371 Description: assertion failure in ldlm_handle2lock()
1372 Details : fix a race between class_handle_unhash() and class_handle2object()
1373 introduced in lustre 1.6.5 by bug 13622.
1378 Description: Kernel BUG: sd_iostats_bump: unexpected disk index
1379 Details : remove the limit of 256 scsi disks in the sd_iostat patch
1384 Description: oops in sd_iostats_seq_show()
1385 Details : unloading/reloading the scsi low level driver triggers a kernel
1386 bug when trying to access the sd iostat file.
1391 Description: Kernel panics during QLogic driver reload
1392 Details : REQ_BLOCK_PC requests are not handled properly in the sd iostat
1393 patch, causing memory corruption.
1398 Description: journal_dev option does not work in b1_6
1399 Details : pass mount option during pre-mount.
1401 Severity : enhancement
1403 Description: Add a FIEMAP(FIle Extent MAP) ioctl
1404 Details : FIEMAP ioctl will allow an application to efficiently fetch the
1405 extent information of a file. It can be used to map logical blocks
1406 in a file to physical blocks in the block device.
1410 Description: LDLM soft lockups - improvement
1411 Details : It is be possible to send the lock handle along with each read
1412 or write request because the client is already doing a lock match
1413 itself so there isn't any reason the OST should have to re-do that
1417 Frequency : only on Cray X2
1419 Description: X2 build failures
1420 Details : fix build failures on Cray X2.
1424 Description: xid & resent requests
1425 Details : Initialize RPC XID from clock at startup (randomly if clock is
1428 Severity : enhancement
1430 Description: Add lustre_start utility to start or stop multiple Lustre servers
1435 Description: Lustre GPF in {:ptlrpc:ptlrpc_server_free_request+373}
1436 Details : In case of memory pressure, list_del() can be called twice on
1437 req->rq_history_list, causing a kernel oops.
1441 Description: (ptllnd_peer.c:557:kptllnd_peer_check_sends()) ASSERTION(!in_interrupt()) failed
1442 Details : fix stack overflow in the distributed lock manager by defering export
1443 eviction after a failed ast to the elt thread instead of handling
1444 it in the dlm interpret routine.
1448 Description: Convert some comments to new format.
1449 Details : Update documenting comments to match doxygen conventions.
1453 Description: Grammar fixes.
1454 Details : A couple of trivial spelling fixes.
1458 Description: OSD_COUNTERS-mandatory
1459 Details : Make previously optional ->oti_{w,r}_locks sanity checks mandatory
1460 to simplify the code.
1464 Description: simplify cmm_device freeing logic.
1465 Details : Call cmm_device_free() in the failure path of cmm_device_alloc().
1469 Description: Add lockdep support to dt_object_operations locking interface.
1470 Details : Augment ->do_{read,write}_lock() prototypes with a `role' parameter
1471 indicating lock ordering. Update mdd code to use new locking
1476 Description: Introduce failloc constants for lockless IO tests.
1477 Details : Add two new failloc constants to test lockless IO. Only one of
1478 them in implemented---another is checked in yet to be landed
1483 Description: Add lockdep support for inode mutex.
1484 Details : Introduce and use new LOCK_INODE_MUTEX_PARENT() macro to be used
1485 in the situations where Lustre has to lock more than one inode
1490 Description: Add optional invariants checking support.
1491 Details : Add new LINVRNT() macro, optional on new --enable-invariants
1492 configure switch. This macro is to be used for consistency and
1493 sanity checks that are too expensive to be left in `production'
1498 Description: Zap lock->l_granted_mode with explicit LCK_MINMODE.
1499 Details : Use LCK_MINMODE rather than 0 to reset lock->l_granted_mode to
1504 Description: Add lockdep support for ldlm_lock and ldlm_resource.
1505 Details : Use spin_lock_nested() in (the only) situation where more than
1506 one ldlm_lock is locked simultaneously. Also, fix possible
1507 dead-lock in ldlm_lock_change_resource() by enforcing particular
1512 Description: Use struct ldlm_callback_suite in ldlm_lock_create().
1513 Details : Instead of specifying each ldlm_lock call-back through separate
1514 parameter, wrap them into struct ldlm_callback_suite.
1518 Description: Kill join_lru obd method and its callers.
1519 Details : CLIO uses lock weighting policy to keep locks over mmapped regions
1520 in memory---a requirement implemented through ->o_join_lru() obd
1521 method in HEAD. Remove this method and its users.
1525 Description: Add asynchronous ldlm ENQUEUE completion handler.
1526 Details : CLIO posts ENQUEUE requests asynchronously through ptlrpcd---a
1527 case that stock ldlm_completion_ast() cannot handle as it waits
1528 until lock is granted. Introduce new ldlm_completion_ast_async()
1529 for this. Also comment ldlm_completion_ast().
1533 Description: ldlm_error <-> errno conversion.
1534 Details : Add functions to map (rather arbitrary) between LDLM error codes
1535 and standard errno values. CLIO needs this to prevent LDLM specific
1536 constants from escaping ldlm and osc.
1540 Description: Kill unused ldlm_handle2lock_ns() function.
1541 Details : Kill unused ldlm_handle2lock_ns() function.
1545 Description: Add lu_ref support to ldlm_lock
1546 Details : lu_ref support for ldlm_lock and ldlm_resource. See lu_ref patch.
1547 lu_ref fields ->l_reference and ->lr_reference are added to ldlm_lock
1548 and ldlm_resource. LDLM interface has to be changed, because code that
1549 releases a reference on a lock, has to "know" what reference this is.
1550 In the most frequent case
1552 lock = ldlm_handle2lock(handle);
1554 LDLM_LOCK_PUT(lock);
1556 no changes are required. When any other reference (received _not_ from
1557 ldlm_handle2lock()) is released, LDLM_LOCK_RELEASE() has to be called
1558 instead of LDLM_LOCK_PUT().
1560 Arguably, changes are pervasive, and interface requires some discipline
1561 for proper use. On the other hand, it was very instrumental in finding
1562 a few leaked lock references.
1566 Description: Add ldlm_lock_addref_try().
1567 Details : Introduce ldlm_lock_addref_try() function (used by CLIO) that
1568 attempts to addref a lock that might be being canceled
1573 Description: Add ldlm_weigh_callback().
1574 Details : Add new ->l_weigh_ast() call-back to ldlm_lock. It is called
1575 by ldlm_cancel_shrink_policy() to estimate lock "value", instead of
1576 hard-coded `number of pages' logic.
1580 Description: Add lockdep annotations to llog code.
1581 Details : Use appropriately tagged _nested() locking calls in the places
1582 where llog takes more than one ->lgh_lock lock.
1586 Description: Add loi_kms_set().
1587 Details : Wrap kms updates into a helper function.
1591 Description: Constify instances of struct lsm_operations.
1592 Details : Constify instances of struct lsm_operations.
1596 Description: lu_conf support.
1597 Details : On a server, a file system object is uniquely identified
1598 by a fid, which is sufficient to locate and load all object
1599 state (inode). On a client, on the other hand, more data are
1600 necessary instantiate an object. Change lu_object_find() and
1601 friends to take additional `lu_conf' argument describing object.
1602 Typically this includes layout information.
1604 --------------------------------------------------------------------------------
1606 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
1608 * Support for kernels:
1609 2.6.5-7.283 (SLES 9),
1610 2.6.9-55.EL (RHEL 4),
1611 2.6.16.46-0.14 (SLES 10),
1612 2.6.18.8 vanilla (kernel.org)
1613 * Client support for unpatched kernels:
1614 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1615 2.6.16 - 2.6.22 vanilla (kernel.org)
1616 * Due to recently discovered recovery problems, we do not recommend
1617 using patchless RHEL 4 clients with this or any earlier release.
1618 * Recommended e2fsprogs version: 1.39.cfs8
1619 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1620 * Starting with this release, the ldiskfs backing filesystem required
1621 by Lustre is now in its own package, lustre-ldiskfs. This package
1622 should be installed. It is versioned separately from Lustre and
1623 may be released separately in future.
1627 Description: block reactivating mgc import until all deactivates complete
1628 Details : Fix race when failing back MDT/MGS to itself (testing)
1631 Frequency : at statup only
1633 Description: mds_lov_synchronize race leads to various problems
1634 Details : simultaneous MDT->OST connections at startup can cause the
1635 sync to abort, leaving the OSC in a bad state.
1637 Severity : enhancement
1639 Description: add optional extra BUILD_VERSION info
1640 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1641 to override the lustre version.
1644 Frequency : 2.6.18 servers only
1646 Description: ll_kern_mount() doesn't release the module reference
1647 Details : The ldiskfs module reference count never drops down to 0
1648 because ll_kern_mount() doesn't release the module reference.
1653 Description: server LBUG when using old ost_num_threads parameter
1654 Details : Accept the old ost_num_threads parameter but warn that it
1655 is deprecated, and fix an off-by-one error that caused an LBUG.
1660 Description: Transient SCSI error results in persistent IO issue
1661 Details : iobuf->dr_error is not reinitialized to 0 between two
1665 Frequency : sometimes when underlying device returns I/O errors
1667 Description: OSTs not going read-only during write failures
1668 Details : OSTs are not remounted read-only when the journal commit threads
1669 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1670 instead of the ext3 wrappers.
1674 Description: poor connect scaling with increasing client count
1675 Details : Don't run filter_grant_sanity_check for more than 100 exports
1676 to improve scaling for large numbers of clients.
1679 Frequency : SLES10 only
1681 Description: sanity-quota.sh quotacheck failed: rc = -22
1682 Details : Quotas cannot be enabled on SLES10.
1685 Frequency : liblustre clients only
1687 Description: getdirentries does not give error when run on compute nodes
1688 Details : getdirentries does not fail when the size specified as an argument
1689 is too small to contain at least one entry
1691 Severity : enhancement
1693 Description: Add LNET router traceability for debug purposes
1694 Details : If a checksum failure occurs with a router as part of the
1695 IO path, the NID of the last router that forwarded the bulk data
1696 is printed so it can be identified.
1701 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1702 Details : Due to a race condition, liblustre clients were occasionally
1703 evicted incorrectly.
1705 Severity : enhancement
1707 Description: lfs setstripe use optional parameters instead of postional
1710 Severity : enhancement
1712 Description: Nanosecond timestamp support for ldiskfs
1713 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1714 resolution timestamps. There is not yet support for this at
1715 the Lustre filesystem level.
1718 Frequency : during server recovery
1720 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1721 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1725 Frequency : nfs export on patchless client
1727 Description: connectathon hang when test nfs export over patchless client
1728 Details : Disconnected dentry cannot be found with lookup, so we do not need
1729 to unhash it or make it invalid
1732 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1733 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1734 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1739 Description: open req refcounting wrong on reconnect
1740 Details : If reconnect happened between getting open reply from server and
1741 call to mdc_set_replay_data in ll_file_open, we will schedule
1742 replay for unreferenced request that we are about to free.
1743 Subsequent close will crash in variety of ways.
1744 Check that request is still eligible for replay in
1745 mdc_set_replay_data().
1750 Description: disable writes to filesystem when reading health_check file
1751 Details : the default for reading the health_check proc file has changed
1752 to NOT do a journal transaction and write to disk, because this
1753 can cause reads of the /proc file to hang and block HA state
1754 checking on a healthy but otherwise heavily loaded system. It
1755 is possible to return to the previous behaviour during configure
1756 with --enable-health-write.
1758 Severity : enhancement
1760 Description: 64-bit inode version
1761 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1762 to the inode. This will be required for version-based recovery.
1767 Description: MDS fails to start if a duplicate client export is detected
1768 Details : in some rare cases it was possible for a client to connect to
1769 an MDS multiple times. Upon recovery the MDS would detect this
1770 and fail during startup. Handle this more gracefully.
1772 Severity : enhancement
1774 Description: Add -o localflock option to simulate old noflock
1776 Details : This will achieve local-only flock/fcntl locks
1782 Description: log_commit_thread vs filter_destroy race leads to crash
1783 Details : Take import reference before releasing llog record semaphore
1788 Description: Wrong request locking in request set processing
1789 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1790 imp_delayed_list, in this place should be used imp_lock.
1793 Frequency : when reconnection
1795 Description: Grant Leak when osc reconnect to OST
1796 Details : When osc reconnect ost, OST(filter) should check whether it
1797 should grant more space to client by comparing fed_grant and
1798 cl_avail_grant, and return the granted space to client instead
1799 of "new granted" space, because client will call osc_init_grant
1800 to update the client grant space info.
1803 Frequency : when client reconnect to OST
1805 Description: Grant Leak when osc do resend and replay bulk write
1806 Details : When osc reconnect to OST, OST(filter)should clear grant info of
1807 bulk write request, because the grant info while be sync between
1808 OSC and OST when reconnect, and we should ignore the grant info
1809 these of resend/replay write req.
1814 Description: Grant space more than avaiable left space sometimes.
1815 Details : When then OST is about to be full, if two bulk writing from
1816 different clients came to OST. Accord the avaliable space of the
1817 OST, the first req should be permitted, and the second one
1818 should be denied by ENOSPC. But if the seconde arrived before
1819 the first one is commited. The OST might wrongly permit second
1820 writing, which will cause grant space > avaiable space.
1823 Frequency : when client is evicted
1825 Description: Grant might be wrongly erased when osc is evicted by OST
1826 Details : when the import is evicted by server, it will fork another
1827 thread ptlrpc_invalidate_import_thread to invalidate the
1828 import, where the grant will be set to 0. While the original
1829 thread will update the grant it got when connecting. So if
1830 the former happened latter, the grant will be wrongly errased
1831 because of this race.
1836 Description: Checking Stale with correct fid
1837 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1838 to check whether it is stale, because sometimes, we want the
1839 enqueue happened anyway, and op_data.fid2 will not be initialized.
1841 Severity : enhancement
1843 Description: update patchless client
1844 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1847 Frequency : only with 2.4 kernel
1849 Description: random memory corruption
1850 Details : size of struct ll_inode_info is to big for union inode.u and this
1851 can be cause of random memory corruption.
1856 Description: Memory leak in recovery
1857 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1858 It should also check obd_fail before fsfilt_start, otherwise if
1859 fsfilt_start return -EROFS,(failover mds during mds recovery).
1860 then the req will return with repmsg->transno = 0 and rc = EROFS.
1861 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1862 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1863 be freed no matter whether fsfilt_commit success or not.
1866 Frequency : only with huge count clients
1868 Description: Prevents from taking the superblock lock in llap_from_page for
1870 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1871 for llap_from_page call in ll_removepage prevents from taking the
1872 superblock lock for a soon died page.
1877 Description: Not check open intent error before release open handle
1878 Details : in some rare cases, the open intent error is not checked before
1879 release open handle, which may cause
1880 ASSERTION(open_req->rq_transno != 0), because it tries to release
1881 the failed open handle.
1886 Description: Set cat log bitmap only after create log success.
1887 Details : in some rare cases, the cat log bitmap is set too early. and it
1888 should be set only after create log success.
1892 Description: Accessing a block bevice can re-enable I/O when Lustre is
1893 tearing down a device.
1894 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1899 Description: service threads may hog cpus when there are a lot of requests
1900 Details : Insert cond_resched to give other threads a chance to use some CPU
1905 Description: the cat log was not initialized in recovery
1906 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1907 unlink log on mds will not be initialized until mds post
1908 recovery. And also in mds post recovery, the unlink log will
1909 initialization will be done asynchronausly, so there will be race
1910 between add unlink log and unlink log initialization.
1914 Description: brw_stats were being printed incorrectly
1915 Details : brw_stats were being printed as log2 but all of them were not
1916 recorded as log2. Also remove some code duplication arising from
1917 filter_tally_{read,write}.
1921 Frequency : rare, only in recovery.
1922 Description: ASSERTION(req->rq_type != LI_POISON) failed
1923 Details : imp_lock should be held while iterating over imp_sending_list for
1924 prevent destroy request after get timeout in ptlrpc_queue_wait.
1928 Description: replay-single.sh test 52 fails
1929 Details : A lock's skiplist need to be cleanup when it being unlinked
1930 from its resource list.
1934 Description: Short directio read returns full requested size rather than
1936 Details : Direct I/O operations should return actual amount of bytes
1937 transferred rather than requested size.
1939 Severity : enhancement
1941 Description: metadata RPC reduction (e.g. for rm performance)
1942 Details : decrease the amount of synchronous RPC between clients and servers
1943 by canceling conflicing lock before the operation on the client
1944 and packing thier handles into the main operation RPC to server.
1946 Severity : enhancement
1948 Description: Async OSC create to avoid the blocking unnecessarily.
1949 Details : If a OST has no remain object, system will block on the creating
1950 when need to create a new object on this OST. Now, ways use
1951 pre-created objects when available, instead of blocking on an
1952 empty osc while others are not empty. If we must block, we block
1953 for the shortest possible period of time.
1957 Description: improve handling recoverable errors
1958 Details : if request processig with error which can be recoverable on server
1959 request should be resend, otherwise page released from cache and
1962 Severity : enhancement
1964 Description: refine locking for avoid write wrong info into lov_objid file
1965 Details : fix possible races with add new target and write/update data in
1968 --------------------------------------------------------------------------------
1970 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1975 Frequency : on some architectures
1977 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1978 Details : Uninitialized flags sometimes cause configuration commands to
1982 Frequency : patchless clients only
1984 Description: missing __iget() symbol export
1985 Details : The __iget() symbol export is missing. To avoid the need for
1986 this on patchless clients the deathrow inode reaper is turned
1987 off, and we depend on the VM to clean up old inodes. This
1988 dependency was during via the fix for bug 12181.
1990 --------------------------------------------------------------------------------
1992 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1994 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1995 INTEROPERATE with older versions automatically. In many cases a
1996 special upgrade step is needed. Please read the
1997 user documentation before upgrading any part of a live system.
1998 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1999 version will not interoperate with 1.6 betas before beta5 (1.5.95).
2000 * WARNING: Lustre configuration and startup changes are required with
2001 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2005 Severity : enhancement
2007 Description: MountConf
2008 Details : Lustre configuration is now managed via mkfs and mount
2009 commands instead of lmc and lconf. New obd types (MGS, MGC)
2010 are added for dynamic configuration management. See
2011 https://mail.clusterfs.com/wikis/lustre/MountConf for
2014 Severity : enhancement
2016 Description: dynamic OST addition
2017 Details : OSTs can now be added to a live filesystem
2019 Severity : enhancement
2021 Description: startup order invariance
2022 Details : MDTs and OSTs can be started in any order. Clients only
2023 require the MDT to complete startup.
2025 Severity : enhancement
2027 Description: parallel, asynchronous orphan cleanup
2028 Details : orphan cleanup is now performed in separate threads for each
2029 OST, allowing parallel non-blocking operation.
2031 Severity : enhancement
2033 Description: optimized stripe assignment
2034 Details : stripe assignments are now made based on ost space available,
2035 ost previous usage, and OSS previous usage, in order to try
2036 to optimize storage space and networking resources.
2038 Severity : enhancement
2040 Description: Permanently set tunables
2041 Details : All writable /proc/fs/lustre tunables can now be permanently
2042 set on a per-server basis, at mkfs time or on a live system.
2044 Severity : enhancement
2046 Description: Lustre message v2
2047 Details : Add lustre message format v2.
2049 Severity : enhancement
2051 Description: client OST exclusion list
2052 Details : Clients can be started with a list of OSTs that should be
2053 declared "inactive" for known non-responsive OSTs.
2057 Description: SPEC SFS validation failure on NFS v2 over lustre.
2058 Details : Changes the blocksize for regular files to be 2x RPC size,
2059 and not depend on stripe size.
2061 Severity : enhancement
2063 Description: Multiple MD RPCs in flight.
2064 Details : Further unserialise some read-only MDS RPCs - learn about intents.
2065 To avoid overly-overloading MDS, introduce a limit on number of
2066 MDS RPCs in flight for a single client and add /proc controls
2067 to adjust this limit.
2069 Severity : enhancement
2071 Description: client read/write statistics
2072 Details : Add client read/write call usage stats for performance
2073 analysis of user processes.
2074 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2075 file access. extents_stats shows chunk size distribution.
2076 extents_stats_per_process show chunk size distribution per
2079 Severity : enhancement
2081 Description: mds statistics
2082 Details : Add detailed mds operations statistics in
2083 /proc/fs/lustre/mds/*/stats.
2087 Description: Failure of copying files with lustre special EAs.
2088 Details : Client side always return success for setxattr call for lustre
2089 special xattr (currently only "trusted.lov").
2094 Description: Refcount LNET uuids
2095 Details : The global LNET uuid list grew linearly with every startup;
2096 refcount repeated list entries instead of always adding to
2099 Severity : enhancement
2101 Description: Dynamic service threads
2102 Details : Within a small range, start extra service threads
2103 automatically when the request queue builds up.
2106 Frequency : mixed-endian client/server environments
2108 Description: mixed-endian crashes
2109 Details : The new msg_v2 system had some failures in mixed-endian
2112 Severity : enhancement
2114 Description: Easy OST removal
2115 Details : OSTs can be permanently deactivated with e.g. 'lctl
2116 conf_param lustre-OST0001.osc.active=0'
2118 Severity : enhancement
2120 Description: MGS proc entries
2121 Details : Added basic proc entries for the MGS showing what filesystems
2124 Severity : enhancement
2126 Description: provide MGS failover
2127 Details : Added config lock reacquisition after MGS server failover.
2129 Severity : enhancement
2131 Description: add Linux 2.4 support
2132 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2136 Description: a large application tries to do I/O to the same resource and dies
2137 in the middle of it.
2138 Details : Check the req->rq_arrival time after the call to
2139 ost_brw_lock_get(), but before we do anything about
2140 processing it & sending the BULK transfer request. This
2141 should help move old stale pending locks off the queue as
2142 quickly as obd_timeout.
2145 Frequency : when an incorrect nid is specified during startup
2147 Description: ptlrpc connect to non-existant node causes kernel crash
2148 Details : LNET can't be re-entered from an event callback, which
2149 happened when we expire a message after the export has been
2150 cleaned up. Instead, hand the zombie cleanup off to another
2153 Severity : enhancement
2155 Description: plain/inodebits lock performance improvement
2156 Details : Grouping plain/inodebits in granted list by their request modes
2157 and bits policy, thus improving the performance of search through
2161 Frequency : only if OST filesystem is corrupted
2163 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2164 Details : for a short time RPCs with bulk IO are in the replay list,
2165 but replay of bulk IOs is unimplemented. If the OST filesystem
2166 is corrupted due to disk cache incoherency and then replay is
2167 started it is possible to trip an assertion. Avoid putting
2168 committed RPCs into the replay list at all to avoid this issue.
2171 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2173 Description: System hang on startup
2174 Details : This bug allowed the liblustre (e.g. catamount) client to
2175 return to the app before handling all startup RPCs. This
2176 could leave the node unresponsive to lustre network traffic
2177 and manifested as a server ptllnd timeout.
2179 Severity : enhancement
2181 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2182 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2183 boolean to control whether to print peer debug info when a
2184 client's RPC times out.
2187 Frequency : only for kernels with patches from Lustre below 1.4.3
2189 Description: Remove old rdonly API
2190 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2193 Frequency : only for devices with external journals
2195 Description: Set external device read-only also
2196 Details : During a commanded failover stop, we set the disk device
2197 read-only while the server shuts down. We now also set any
2198 external journal device read-only at the same time.
2201 Frequency : when upgrading from 1.4 while trying to change parameters
2203 Description: The wrong (new) MDC name was used when setting parameters for
2204 upgraded MDT's. Also allows changing of OSC (and MDC)
2205 parameters if --writeconf is specified at tunefs upgrade time.
2208 Frequency : when setting specific ost indicies
2210 Description: QOS code breaks on skipped indicies
2211 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2212 created with --index need not be sequential.
2214 Severity : enhancement
2216 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2217 Details : The uninit_groups feature works in conjunction with the kernel
2218 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2219 up the pass1 processing of e2fsck. This is a read-only feature
2220 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2221 mount filesystems that have had this feature enabled.
2223 Severity : enhancement
2225 Description: Improve multi-block allocation algorithm to avoid fragmentation
2226 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2227 allocation locality and avoid filesystem fragmentation.
2229 ------------------------------------------------------------------------------
2231 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2233 * Support for kernels:
2234 2.4.21-47.0.1.EL (RHEL 3)
2235 2.6.5-7.283 (SLES 9)
2236 2.6.9-42.0.10.EL (RHEL 4)
2237 2.6.12.6 vanilla (kernel.org)
2238 2.6.16.27-0.9 (SLES 10)
2239 * Recommended e2fsprogs version: 1.39.cfs5
2241 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2245 Frequency : occasional, depends on client load and configuration
2246 Bugzilla : 12181, 12203
2247 Description: data loss for recently-modified files
2249 Details : In some cases it is possible that recently written or created
2250 files may not be written to disk in a timely manner (this should
2251 normally be within 30s unless client IO load is very high).
2252 The problem appears as zero-length files or files that are a
2253 multiple of 1MB in size after a client crash or client eviction
2254 that are missing data at the end of the file.
2256 This problem is more likely to be hit on clients where files are
2257 repeatedly created and unlinked in the same directory, clients
2258 have a large amount of RAM, have many CPUs, the filesystem has
2259 many OSTs, the clients are rebooted frequently, and/or the files
2260 are not accessed by other nodes after being written.
2262 The presence of the problem can be detected by looking at
2263 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2264 smaller than the second (nr_unused) then dirty files will not
2265 be flushed automatically to disk. "sync; sleep 10" should be
2266 run several times on the node before unmounting it to update
2267 Lustre (this is also safe to run on nodes without this problem).
2269 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2270 that can cause this same problem, so customers using that kernel
2271 also need to update the kernel in addition to Lustre. In order
2272 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2274 It is normal that files written just before a client crash (less
2275 than 30s) may not yet have been flushed to disk, even for local
2279 Frequency : frequent on thin XT3 nodes
2281 Description: UUID collision on thin XT3 Linux nodes
2282 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2283 randomly, since we relied on an insufficiently-seeded PRNG.
2288 Description: OSS hangs after "All ost request buffers busy"
2289 Details : A deadlock between quota and journal operations caused OSS
2290 hangs after printing "All ost request buffers busy."
2293 Frequency : always on liblustre builds
2295 Description: Cleanup compiler warnings on liblustre
2298 Frequency : always on liblustre builds on XT3
2300 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2304 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2305 > 32000 subdirectories
2306 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2307 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2308 subdirectory count crosses 32000. This will aid e2fsck to
2309 correctly handle more than 32000 subdirectories.
2312 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2314 Description: System hang on startup
2315 Details : This bug allowed the liblustre (e.g. catamount) client to
2316 return to the app before handling all startup RPCs. This
2317 could leave the node unresponsive to lustre network traffic
2318 and manifested as a server ptllnd timeout.
2320 Severity : enhancement
2322 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2323 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2324 boolean to control whether to print peer debug info when a
2325 client's RPC times out.
2330 Description: make O_SYNC working on 2.6 kernels
2331 Details : 2.6 kernels use different method for mark pages for write,
2332 so need add a code to lustre for O_SYNC work.
2337 Description: Failure to close file and release space on NFS
2338 Details : Put inode details into lock acquired in ll_intent_file_open.
2339 Use mdc_intent_lock in ll_intent_open to properly
2340 detect all kind of errors unhandled by mdc_enqueue.
2345 Description: proc file read during shutdown sometimes raced obd removal,
2347 Details : Add lock to prevent obd access after proc file removal.
2350 Frequency : Only for files larger than 4GB on 32-bit clients.
2352 Description: improperly doing page alignment of locks
2353 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2354 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2359 Description: RPCs being resent when they shouldn't be
2360 Details : Some RPCs that should not be resent are being resent. This
2361 can cause inconsistencies in the RPC state machine. Do not
2362 resend such requests.
2365 Frequency : rare, only with NFS export
2367 Description: Crash on NFS re-export node
2368 Details : under very unusual load conditions an assertion is hit in
2369 ll_intent_file_open()
2372 Frequency : only if OST filesystem is corrupted
2374 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2375 Details : for a short time RPCs with bulk IO are in the replay list,
2376 but replay of bulk IOs is unimplemented. If the OST filesystem
2377 is corrupted due to disk cache incoherency and then replay is
2378 started it is possible to trip an assertion. Avoid putting
2379 committed RPCs into the replay list at all to avoid this issue.
2384 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2385 Details : Large single O_DIRECT read and write calls can fail to allocate
2386 a sufficiently large buffer to process the request. In case of
2387 allocation failure the allocation is retried with a smaller
2388 buffer and broken into smaller requests.
2390 Severity : enhancement
2392 Description: Add -o localflock option to simulate old noflock behaviour.
2393 Details : This will achieve local-only flock/fcntl locks coherentness.
2398 Description: versioning check is incomplete
2399 Details : Checking the version difference of client vs. server, report
2400 error if the gap is too big.
2405 Description: add support PG_writeback bit
2406 Details : add support for PG_writeback bit for Lustre, for more carefull
2407 work with page cache in 2.6 kernel. This also fix some deadlocks
2408 and remove hack for work O_SYNC with 2.6 kernel.
2410 Severity : enhancement
2412 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2413 Details : The uninit_groups feature works in conjunction with the kernel
2414 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2415 up the pass1 processing of e2fsck. This is a read-only feature
2416 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2417 mount filesystems that have had this feature enabled.
2419 Severity : enhancement
2421 Description: Improve multi-block allocation algorithm to avoid fragmentation
2422 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2423 allocation locality and avoid filesystem fragmentation.
2425 ------------------------------------------------------------------------------
2427 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2429 * Support for kernels:
2430 2.6.9-42.0.3.EL (RHEL 4)
2431 2.6.5-7.276 (SLES 9)
2432 2.4.21-47.0.1.EL (RHEL 3)
2433 2.6.12.6 vanilla (kernel.org)
2434 2.6.16.21-0.8 (SLES10)
2435 * Recommended e2fsprogs version: 1.39.cfs2-0
2437 * The backwards-compatible /proc/sys/portals symlink has been removed
2438 in this release. Before upgrading, please ensure that you change
2439 any configuration scripts or /etc/sysctl.conf files that access
2440 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2441 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2442 in advance of the upgrade on any system running Lustre 1.4.6 or
2443 newer, since /proc/sys/lnet was added in that version.
2444 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2448 Frequency : only when quota is used
2450 Description: avoid scanning export list for quota master
2451 Details : Change the algorithms to avoid scanning export list in order
2452 to improve the efficiency.
2455 Frequency : MDS failover only, very rarely
2457 Description: "went back in time" messages on mds failover
2458 Details : The greatest transno may be lost when the current operation
2459 finishes with an error (transno==0) and the client's last_rcvd
2460 record is over-written. Save the greatest transno in the
2461 mds_last_transno for this case.
2464 Frequency : always for specific kernels and striping counts
2466 Description: client may get "Matching packet too big" without ACL support
2467 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2468 when trying to access files in certain configurations. The
2469 clients should in fact be denied when mounting because they do
2470 not understand ACLs.
2473 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2475 Description: many clients connecting with IO in progress causes connect timeouts
2476 Details : Avoid synchronous journal commits to avoid delays caused by many
2477 clients connecting/disconnecting when bulk IO is in progress.
2478 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2479 OST_IO_PORTAL to avoid delays behind potentially many pending
2483 Frequency : occasionally with multiple writers to a single file
2485 Description: shared writes to file may result in wrong size reported by stat()
2486 Details : Allow growing of kms when extent lock is cancelled
2489 Frequency : always with random mmap IO to multi-striped file
2491 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2492 Details : Only if the hole is at the end of OST object so that kms is too
2493 small. Fix is to increase kms accordingly in ll_nopage.
2496 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2498 Description: writes to a missing object would leak memory on the OST
2499 Details : If there is an inconsistency between the MDS and OST filesystems,
2500 such that the MDS references an object that doesn't exist, writes
2501 to that object will leak memory due to incorrect cleanup in the
2502 error handling path, eventually running out of memory on the OST.
2507 Description: Creating too long symlink causes lustre errors
2508 Details : Check symlink and name lengths before sending requests to MDS.
2511 Frequency : only if flock is enabled (not on by default)
2513 Description: posix locks not released on fd closure on 2.6.9+
2514 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2515 kernels, this caused such locks not to be released on fd close and
2516 then assertions on fs unmount about still used locks.
2519 Frequency : MDS failover only, very rarely
2521 Description: clients may get ASSERTION(granted_lock != NULL)
2522 Details : When request was taking a long time, and a client was resending
2523 a getattr by name lock request. The were multiple lock requests
2524 with the same client lock handle and
2525 mds_getattr_name->fixup_handle_for_resent_request found one of the
2526 lock handles but later failed with ASSERTION(granted_lock != NULL).
2531 Description: handle->h_buffer_credits > 0, assertion failure
2532 Details : h_buffer_credits is zero after truncate, causing assertion
2533 failure. This patch extends the transaction or creates a new
2537 Frequency : NFS re-export or patchless client
2538 Bugzilla : 11179, 10796
2539 Description: Crash on NFS re-export node (__d_move)
2540 Details : We do not want to hash the dentry if we don't have a lock.
2541 But if this dentry is later used in d_move, we'd hit uninitialised
2542 list head d_hash, so we just do this to init d_hash field but
2543 leave dentry unhashed.
2546 Frequency : NFS re-export or patchless client
2548 Description: NFS exports has problem with symbolic link
2549 Details : lustre client didn't properly install dentry when re-exported
2550 to NFS or running patchless client.
2553 Frequency : NFS re-export or patchless client
2555 Description: Various nfs/patchless fixes.
2556 Details : fixes reuse disconected alias for lookup process - this fixes
2557 warning "find_exported_dentry: npd != pd",
2558 fix permission error with open files at nfs.
2559 fix apply umask when do revalidate.
2562 Frequency : occasional
2564 Description: Crash on NFS re-export node
2565 Details : calling clear_page() on the wrong pointer triggered oops in
2566 generic_mapping_read().
2569 Frequency : rarely, using O_DIRECT IO
2571 Description: unaligned directio crashes client with LASSERT
2572 Details : check for unaligned buffers before trying any requests.
2575 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2577 Description: stale data returned from RAID cache
2578 Details : If only a small amount of IO is done to the RAID device before
2579 reading it again it is possible to get stale data from the RAID
2580 cache instead of reading it from disk.
2583 Frequency : always for sles10 kernel
2585 Description: sles10 support
2586 Details : ll_follow_link: compile fixes and using of nd_set_link
2587 under newer kernels.
2590 Frequency : depends on arch, kernel and compiler version, always on sles10
2593 Description: recursive or deep enough symlinks cause stack overflow
2594 Details : getting rid of large stack-allocated variable in
2598 Frequency : depends on hardware
2600 Description: lustre write performance loss in the SLES10 kernel
2601 Details : the performance loss is caused by using of write barriers in the
2602 ext3 code. The SLES10 kernel turns barrier support on by
2603 default. The fix is to undo that change for ldiskfs.
2605 ------------------------------------------------------------------------------
2607 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2609 * Support for kernels:
2610 2.6.9-42.0.3EL (RHEL 4)
2611 2.6.5-7.276 (SLES 9)
2612 2.4.21-47.0.1.EL (RHEL 3)
2613 2.6.12.6 vanilla (kernel.org)
2617 Frequency : quota enabled and large files being deleted
2619 Description: releasing more than 4GB of quota at once hangs OST
2620 Details : If a user deletes more than 4GB of files on a single OST it
2621 will cause the OST to spin in an infinite loop. Release
2622 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2627 Description: statfs data retrieved from /proc may be stale or zero
2628 Details : When reading per-device statfs data from /proc, in the
2629 {kbytes,files}_{total,free,avail} files, it may appear
2630 as zero or be out of date.
2633 Frequency : systems with MD RAID1 external journal devices
2635 Description: lconf's call to blkid is confused by RAID1 journal devices
2636 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2637 of returning all block devices that match the journal UUID.
2640 Frequency : always, for aggregate stripe size over 4GB
2642 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2643 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2644 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2647 Frequency : always if "lfs find" used on a local file/directory
2649 Description: "lfs find" segfaults if used on a local file/directory
2650 Details : The case where a directory component was not specified wasn't
2651 handled correctly. Handle this properly.
2654 Frequency : always on ppc64
2656 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2657 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2658 perform endianness conversion.
2661 Frequency : rarely (truncate to non-zero file size after write under load)
2662 Bugzilla : 10730, 10687
2663 Description: Files padded with zeros to next 4K multiple
2664 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2665 it is possible that files that are truncated to a non-zero size
2666 immediately after being written are filled with zero bytes beyond
2667 the truncated size. No file data is lost.
2669 Severity : enhancement
2671 Description: Allow recovery/failover for liblustre clients.
2672 Details : liblustre clients were unaware of failover configurations until
2675 Severity : enhancement
2677 Description: user file locks should fail when not mounting with flock option
2678 Details : Set up an error-returning stub in ll_file_operations.lock field
2679 to prevent incorrect behaviour when client is mounted without
2680 flock option. Also, set up properly f_op->flock field for
2684 Frequency : always on ia64
2686 Description: "lfs df" loops on printing out MDS statfs information
2687 Details : The obd_ioctl_data was not initialized and in some systems
2688 this caused a failure during the ioctl that did not return
2689 an error. Initialize the struct and return an error on failure.
2692 Frequency : SLES 9 only
2694 Description: Error of copying files with lustre special EAs as root
2695 Details : Client side always return success for setxattr call for lustre
2696 special xattr (currently only "trusted.lov").
2699 Frequency : rarely on clusters with both ia64+i386 clients
2701 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2702 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2703 clients are doing concurrent writes to the same file it is
2704 possible that the ia64 clients may LASSERT because the OST
2705 extent locks are not PAGE_SIZE aligned. Ensure that grown
2706 locks are always aligned on the request boundary.
2709 Frequency : specific use, occasional
2711 Description: Overwriting in use executable truncates on-disk binary image
2712 Details : If one node attempts to overwrite an executable in use by
2713 another node, we now correctly return ETXTBSY instead of
2714 truncating the file.
2716 Severity : enhancement
2718 Description: Async OSC create to avoid the blocking unnecessarily.
2719 Details : If a OST has no remain object, system will block on the creating
2720 when need to create a new object on this OST. Now, ways use
2721 pre-created objects when available, instead of blocking on an
2722 empty osc while others are not empty. If we must block, we block
2723 for the shortest possible period of time.
2728 Description: chmod on Lustre root is propagated to other clients
2729 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2730 invalid by the follow_mount time.
2733 Frequency : liblustre clients only
2735 Description: Race in 'instant cancel' lock handling could lead to such locks
2736 never to be granted in case of SMP MDS
2737 Details : Do not destroy not yet granted but cbpending locks in
2741 Frequency : replay/resend of open
2743 Description: non null lock assetion failure in mds_intent_policy
2744 Details : Trying to replay/resend lockless open requests resulted in
2745 mds_open() returning 0 with no lock. Now it sets a flag if
2746 a lock is going to be returned.
2748 Severity : enhancement
2750 Description: Checksum enhancements
2751 Details : New checksum enhancements allow for resending RPCs that failed
2754 Severity : enhancement
2756 Description: Tunables on number of dirty pages in cacche
2757 Details : Allow to set limit on number of dirty pages cached.
2762 Description: client crash on unmount - lock still has references
2763 Details : In some error handling cases it was possible to leak a lock
2764 reference on a client while accessing a file. This was not
2765 harmful to the client during operation, but would cause the
2766 client to crash when the filesystem is unmounted.
2769 Frequency : specific case, rare
2771 Description: ETXTBSY on mds though file not in use by client
2772 Details : ETXTBSY is no longer incorrectly returned when attempting to
2773 chmod or chown a directory that the user previously tried to
2774 execute or a currently-executing binary.
2777 Frequency : extremely rare except on liblustre-based clients
2779 Description: Lustre space not freed when files are deleted
2780 Details : Clean up open-unlinked files after client eviction. Previously
2781 the unlink was skipped and the files remained as orphans.
2786 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2787 Details : In some cases it was possible to send a blocking callback to
2788 a client doing a glimpse, even though that client didn't get
2789 a lock granted. When the glimpse lock is cancelled on the OST
2790 the freed lock is left on the waiting list and corrupted the list.
2793 Frequency : all core dumps
2795 Description: Broke core dumps to lustre
2796 Details : Negative dentry may be unhashed if parent does not have UPDATE
2797 lock, but some callers, e.g. do_coredump, expect dentry to be
2798 hashed after successful create, hash it in ll_create_it.
2800 ------------------------------------------------------------------------------
2802 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2804 * Support for kernels:
2805 2.6.9-42.0.2.EL (RHEL 4)
2806 2.6.5-7.276 (SLES 9)
2807 2.4.21-40.EL (RHEL 3)
2808 2.6.12.6 vanilla (kernel.org)
2812 Frequency : always on RHEL 3
2814 Description: Number of open files grows over time
2815 Details : The number of open files grows over time, whether or not
2816 Lustre is started. This was due to a filp leak introduced
2817 by one of our kernel patches.
2819 ------------------------------------------------------------------------------
2821 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2823 * Support for kernels:
2824 2.6.9-42.EL (RHEL 4)
2825 2.6.5-7.276 (SLES 9)
2826 2.4.21-40.EL (RHEL 3)
2827 2.6.12.6 vanilla (kernel.org)
2832 Bugzilla : 5719, 9635, 9792, 9684
2833 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2834 Details : If a server is under heavy load and cannot reply to new
2835 connection requests before the client resends the (re)connect,
2836 the connection handling code can behave badly if two service
2837 threads are concurrently handing separate (re)connections from
2838 the same client. Add better locking to the connection handling
2839 code, and ensure that only a single connection will be processed
2840 for a given client UUID, even if the lock is dropped.
2842 Severity : enhancement
2844 Description: add TCP zero-copy support to kernel
2845 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2846 sends if the hardware supports scatter-gather and checksumming.
2847 This allows socklnd to do client-write and server-read more
2848 efficiently and reduce CPU utilization from skbuf copying.
2851 Frequency : only if NFS exporting from client
2853 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2854 Details : If setting attributes on a file created under NFS that had
2855 never been opened it would be possible to oops the client
2856 if the file had no objects.
2859 Frequency : always for liblustre
2861 Description: liblustre client does MDS+OSTs setattr RPC for each write
2862 Details : When doing a write from a liblustre client, the client
2863 incorrectly issued an RPC to the MDS and each OST the file was
2864 striped over in order to update the timestamps. When writing
2865 with small chunks and many clients this could overwhelm the MDS
2866 with RPCs. In all cases it would slow down the write because
2867 these RPCs are unnecessary.
2869 Severity : enhancement
2871 Description: allow number of MDS service threads to be changed at module load
2872 Details : It is now possible to change the number of MDS service threads
2873 running. Adding "options mds mds_num_threads={N}" to the MDS's
2874 /etc/modprobe.conf will set the number of threads for the next
2875 time Lustre is restarted (assuming the "mds" module is also
2876 reloaded at that time). The default number of threads will
2877 stay the same, 32 for most systems.
2882 Description: OST crash if filesystem is unformatted or corrupt
2883 Details : If an OST is started on a device that has never been formatted
2884 or if the filesystem is corrupt and cannot even mount then the
2885 error handling cleanup routines would dereference a NULL pointer.
2890 Description: NULL pointer deref in llap_from_page.
2891 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2892 incorrect) mapping if the page was truncated/reclaimed while it was
2893 searched for. Check for this condition and skip such pages when
2894 doing readahead. Introduce extra check to llap_from_page() to
2895 verify page->mapping->host is non-NULL (so page is not anonymous).
2898 Frequency : Sometimes when using sys_sendfile
2900 Description: "page not covered by a lock" warnings from ll_readpage
2901 Details : sendfile called ll_readpage without right page locks present.
2902 Now we introduced ll_file_sendfile that does necessary locking
2903 around call to generic_file_sendfile() much like we do in
2907 Frequency : with certain MDS communication failures at client mount time
2909 Description: NULL pointer deref after failed client mount
2910 Details : a client connection request may delayed by the network layer
2911 and not be sent until after the PTLRPC layer has timed out the
2912 request. If the client fails the mount immediately it will try
2913 to clean up before the network times out the request. Add a
2914 reference from the request import to the obd device and delay
2915 the cleanup until the network drops the request.
2918 Frequency : occasionally during client (re)connect
2920 Description: assertion failure during client (re)connect
2921 Details : processing a client connection request may be delayed by the
2922 client or server longer than the client connect timeout. This
2923 causes the client to resend the connection request. If the
2924 original connection request is replied in this interval, the
2925 client may trip an assertion failure in ptlrpc_connect_interpret()
2926 which thought it would be the only running connect process.
2929 Frequency : only with obd_echo servers and clients that are rebooted
2931 Description: kernel BUG accessing uninitialized data structure
2932 Details : When running an obd_echo server it did not start the ping_evictor
2933 thread, and when a client was evicted an uninitialized data
2934 structure was accessed. Start the ping_evictor in the RPC
2935 service startup instead of the OBD startup.
2937 Severity : enhancement
2938 Bugzilla : 10193 (patchless)
2939 Description: Remove dependency on various unexported kernel interfaces.
2940 Details : No longer need reparent_to_init, exit_mm, exit_files,
2941 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2944 Frequency : rare (only users of deprecated and unsupported LDAP config)
2946 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2947 Details : LDAP apparently contains 'lustreName' attributes instead of
2948 'name'. A simple remapping of the name is sufficient.
2951 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2953 Description: waiting_locks_callback trips kernel BUG if client is evicted
2954 Details : Running with the dump_on_timeout debug flag turned on makes
2955 it possible that the waiting_locks_callback() can try to dump
2956 the Lustre kernel debug logs from an interrupt handler. Defer
2957 this log dumping to the expired_lock_main() thread.
2959 Severity : enhancement
2961 Description: Support NFS exporting on 2.6 kernels.
2962 Details : Implement non-rawops metadata methods for NFS server to use without
2963 changing NFS server code.
2966 Frequency : very rare (synthetic metadata workload only)
2968 Description: two racing renames might cause an MDS thread to deadlock
2969 Details : Running the "racer" program may cause one MDS thread to rename
2970 a file from being the source of a rename to being the target of
2971 a rename at exactly the same time that another thread is doing
2972 so, and the second thread has already enqueued these locks after
2973 doing a lookup of the target and is trying to relock them in
2974 order. Ensure that we don't try to re-lock the same resource.
2977 Frequency : only very large systems with liblustre clients
2979 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2980 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2981 all OSTs in parallel. This allows the eviction of stale liblustre
2982 clients to proceed much faster than if they were done in series,
2983 and also offers similar improvements for other set_info RPCs.
2988 Description: excessive CPU usage during initial read phase on client
2989 Details : During the initial read phase on a client, it would agressively
2990 retry readahead on the file, consuming too much CPU and impacting
2991 performance (since 1.4.5.8). Improve the readahead algorithm
2992 to avoid this, and also improve some other common cases (read
2993 of small files in particular, where "small" is files smaller than
2994 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2999 Description: MDS crash when receiving packet with unknown intent.
3000 Details : Do not LBUG in unknown intent case, just return -EFAULT
3002 Severity : enhancement
3003 Bugzilla : 9293, 9385
3004 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3005 Details : Do not serialize getattr (non-intent version) and statfs.
3008 Frequency : occasional, when OST network is overloaded/intermittent
3010 Description: client evicted by OST after bulk IO timeout
3011 Details : If a client sends a bulk IO request (read or write) the OST
3012 may evict the client if it is unresposive to its data GET/PUT
3013 request. This is incorrect if the network is overloaded (takes
3014 too long to transfer the RPC data) or dropped the OST GET/PUT
3015 request. There is no need to evict the client at all, since
3016 the pinger and/or lock callbacks will handle this, and the
3017 client can restart the bulk request.
3020 Frequency : Always when mmapping file with no objects
3022 Description: client crashes when mmapping file with no objects
3023 Details : Check that we actually have objects in a file before doing any
3024 operations on objects in ll_vm_open, ll_vm_close and
3030 Description: Request leak when working with deleted CWD
3031 Details : Introduce advanced request refcount tracking for requests
3032 referenced from lustre intent.
3034 Severity : Enhancement
3036 Description: Cache open file handles on client.
3037 Details : MDS now will return special lock along with openhandle, if
3038 requested and client is allowed to hold openhandle, even if unused,
3039 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3040 closing files for every read/write openration.
3042 Severity : Enhancement
3044 Description: Cache open negative dentries on client when possible.
3045 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3046 negative dentries on lock revocation.
3051 Description: Remounting a client read-only wasn't possible with a zconf mount
3052 Details : It wasn't possible to remount a client read-only with llmount.
3054 Severity : enhancement
3055 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3056 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3057 MPICH is included to add Lustre-specific ADIO interfaces.
3058 This is based closely on the UFS ADIO layer and only differs
3059 in file creation, in order to allow the OST striping to be set.
3060 This is user-contributed code and not supported by CFS.
3065 Description: extended inode attributes (immutable, append-only) work improperly
3066 when 2.4 and 2.6 kernels are used on client/server or vice versa
3067 Details : Introduce kernel-independent values for these flags.
3069 Severity : enhancement
3072 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3073 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3074 granularity. Now, Fractional MB granularity is be supported,
3075 this is very useful for low memory system.
3077 Severity : enhancement
3079 Description: Getattr by fid
3080 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3081 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3082 have locks and avoids getattr rpc requests for it on every stat.
3085 Frequency : Always, for filesystems larger than 2TB
3087 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3088 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3089 filesystems larger than 2TB to be mounted with mballoc enabled.
3092 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3094 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3095 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3096 filesystems larger than 2TB is mounted then the kernel will
3097 silently corrupt the start of the filesystem. CONFIG_LBD is
3098 enabled for all CFS-supported kernels, but the possibility of
3099 this happening with a modified kernel config exists.
3101 Severity : enhancement
3103 Description: add client O_DIRECT support for 2.6 kernels
3104 Details : It is now possible to do O_DIRECT reads and writes to files
3105 in the Lustre client mountpoint on 2.6 kernel clients.
3107 Severity : enhancement
3109 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3110 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3111 parallel, not waiting for response from every OST before sending
3112 a rpc to the next OST.
3117 Description: setattr vs write race when updating file timestamps
3118 Details : Client processes that update a file timestamp into the past
3119 right after writing to the file (e.g. tar) it is possible that
3120 the updated file modification time can be reset to the current
3121 time due to a race between processing the setattr and write RPC.
3123 Severity : enhancement
3125 Description: Bring 'lfs find' closer in line with regular Linux find.
3126 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3127 -print0 options and obtains all the needed info through the lustre
3130 Severity : enhancement
3132 Description: support up to 1024 configured devices on one node
3133 Details : change obd_dev array from statically allocated to dynamically
3134 allocated structs as they are first used to reduce memory usage
3139 Description: Flush dirty partially truncated pages during truncate
3140 Details : Immediatelly flush partially truncated pages in filter_setattr,
3141 this way we completely avoid having any pages in page cache on OST
3142 and can retire ugly workarounds during writes to flush such pages.
3147 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3148 Details : protect inode from truncation within vfs_unlink() context
3149 just take a reference before calling vfs_unlink() and release it
3150 when parent's i_sem is free.
3155 Description: last_id value checked outside lock on OST caused LASSERT failure
3156 Details : If there were multiple MDS->OST object precreate requests in
3157 flight, it was possible that the OST's last object id was checked
3158 outside a lock and incorrectly tripped an assertion. Move checks
3159 inside locks, and discard old precreate requests.
3162 Frequency : always, if extents are used on OSTs
3164 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3165 Details : OSTs using the extents format would not zero the high 16 bits of
3166 the index physical block number. This is not a problem for any
3167 OST filesystems smaller than 16TB, and no kernels support ext3
3168 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3169 new/modified files) and can be fixed for existing filesystems
3170 with e2fsprogs-1.39-cfs1.
3175 Description: import connection selection may be incorrect if timer wraps
3176 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3177 import connections to be ignored if the 32-bit jiffies counter
3178 wraps. Use a 64-bit jiffies counter.
3181 Frequency : very large clusters immediately after boot
3183 Description: LNET request buffers exhausted under heavy short-term load
3184 Details : If a large number of client requests are generated on a service
3185 that has previously never seen so many requests it is possible
3186 that the request buffer growth cannot keep up with the spike in
3187 demand. Instead of dropping incoming requests, they are held in
3188 the LND until the RPC service can accept more requests.
3191 Frequency : Sometimes during replay
3193 Description: Assertion failure in ll_local_open after replay.
3194 Details : If replay happened on an open request reply before we were able
3195 to set replay handler, reply will become not swabbed tripping the
3196 assertion in ll_local_open. Now we set the handler right after
3197 recognising of open request
3200 Frequency : very rare
3202 Description: kernel reports "badness in vsnprintf"
3203 Details : Reading from the "recovery_status" /proc file in small chunks
3204 may cause a negative length in lprocfs_obd_rd_recovery_status()
3205 call to vsnprintf() (which is otherwise harmless). Exit early
3206 if there is no more space in the output buffer.
3208 Severity : enhancement
3210 Description: clear OBD RPC statistics by writing to them
3211 Details : It is now possible to clear the OBD RPC statistics by writing
3212 to the "stats" file.
3217 Description: Client mtime is not the same on different clients after utimes
3218 Details : In some cases, the client was using the utimes() syscall on
3219 a file cached on another node. The clients now validate the
3220 ctime from the MDS + OSTs to determine which one is right.
3225 Description: Inability to activate failout mode
3226 Details : lconf script incorrectly assumed that in pythong string's numeric
3227 value is used in comparisons.
3230 Frequency : always with multiple stripes per file
3232 Description: Inefficient object allocation for mutli-stripe files
3233 Details : When selecting which OSTs to stripe files over, for files with
3234 a stripe count that divides evenly into the number of OSTs,
3235 the MDS is always picking the same starting OST for each file.
3236 Return the OST selection heuristic to the original design.
3241 Description: mount failures may take full timeout to return an error
3242 Details : Under some heavy load conditions it is possible that a
3243 failed mount can wait for the full obd_timeout interval,
3244 possibly several minutes, before reporting an error.
3245 Instead return an error as soon as the status is known.
3247 Frequency : quota enabled and large files being deleted
3249 Description: releasing more than 4GB of quota at once hangs OST
3250 Details : If a user deletes more than 4GB of files on a single OST it
3251 will cause the OST to spin in an infinite loop. Release
3252 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
3257 Description: statfs data retrieved from /proc may be stale or zero
3258 Details : When reading per-device statfs data from /proc, in the
3259 {kbytes,files}_{total,free,avail} files, it may appear
3260 as zero or be out of date.
3263 Frequency : systems with MD RAID1 external journal devices
3265 Description: lconf's call to blkid is confused by RAID1 journal devices
3266 Details : Use the "blkid -l" flag to locate the MD RAID device instead
3267 of returning all block devices that match the journal UUID.
3270 Frequency : always, for aggregate stripe size over 4GB
3272 Description: assertion fails when trying to use 4GB stripe size
3273 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
3274 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
3277 Frequency : always on ppc64
3279 Description: the first write on an ext3 filesystem with mballoc got stuck
3280 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
3281 perform endianness conversion.
3283 ------------------------------------------------------------------------------
3285 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
3287 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3288 INTEROPERATE with older versions automatically. Please read the
3289 user documentation before upgrading any part of a live system.
3290 * WARNING: Lustre networking configuration changes are required with
3291 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3294 * Support for kernels:
3295 2.6.9-22.0.2.EL (RHEL 4)
3296 2.6.5-7.244 (SLES 9)
3297 2.6.12.6 vanilla (kernel.org)
3300 Severity : enhancement
3301 Bugzilla : 7981/8208
3302 Description: Introduced Lustre Networking (LNET)
3303 Details : LNET is new networking infrastructure for Lustre, it includes
3304 a reorganized network configuration mode (see the user
3305 documentation for full details) as well as support for routing
3306 between different network fabrics. Lustre Networking Devices
3307 (LNDs) for the supported network fabrics have also been
3308 created for this new infrastructure.
3310 Severity : enhancement
3311 Description: Introduced Access control lists
3312 Details : clients can set ACLs on files and directories in order to have
3313 more fine-grained permissions than the standard Unix UGO+RWX.
3314 The MDS must be started with the "-o acl" mount option.
3316 Severity : enhancement
3317 Description: Introduced filesystem quotas
3318 Details : Administrators may now establish per-user quotas on the
3321 Severity : enhancement
3323 Description: Configuration change for the XT3
3324 The PTLLND is now used to run Lustre over Portals on the XT3
3325 The configure option(s) --with-cray-portals are no longer used.
3326 Rather --with-portals=<path-to-portals-includes> is used to
3327 enable building on the XT3. In addition to enable XT3 specific
3328 features the option --enable-cray-xt3 must be used.
3333 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3334 Details : A race between allocating a new llcd and re-getting the llcd_lock
3335 allowed another thread to grab newly-allocated llcd.
3337 Severity : enhancement
3339 Description: 2.6 OST async journal commit and locking fix to improve performance
3340 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3341 2.6 kernels are now async as they already were in 2.4 kernels so
3342 that they can commit concurrently with the network bulk transfer.
3343 For block-allocated files the filter allocation semaphore is held
3344 to avoid filesystem fragmentation during allocation. BKL lock
3345 removed for 2.6 xattr operations where it is no longer needed.
3350 Description: lconf incorrectly determined whether two IP networks could talk
3351 Details : In some more complicated routing and multiple-network
3352 configurations, lconf will avoid trying to make a network
3353 connection to a disjoint part of the IP space. It was doing the
3354 math incorrectly for one set of cases.
3359 Description: Fix for potential infinite loop processing records in an llog.
3360 Details : If an llog record is corrupted/zeroed, it is possible to loop
3361 forever in llog_process(). Validate the llog record length
3362 and skip the remainder of the block on error.
3365 Frequency : occasional (liblustre only)
3367 Description: liblustre could not open files whose last component is a symlink
3368 Details : sysio_path_walk() would incorrectly pass the open intent to
3369 intermediate path components.
3372 Frequency : rare (liblustre only with non-standard tuning)
3373 Bugzilla : 7201 (7350)
3374 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3375 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3376 after it has been released, only do this for referenced locks
3378 Severity : enhancement
3380 Description: specifying an (invalid) directory default stripe_size of -1
3381 would reset the directory default striping
3382 Details : stripe_size -1 was used internally to signal directory stripe
3383 removal, now use "all default" to signal dir stripe removal
3384 as a directory striping of "all default" is not useful
3387 Frequency : common for large clusters running liblustre clients
3389 Description: doing an ls when liblustre clients are running is slow
3390 Details : sending a glimpse AST to a liblustre client waits for every AST
3391 to time out, as liblustre clients will not respond. Since they
3392 cannot cache data we refresh the OST lock LVB from disk instead.
3394 Severity : enhancement
3396 Description: doing an ls at the same time as file IO can be slow
3397 Details : enqueue and other "small" requests can be blocked behind many
3398 large IO requests. Create a new OST IO portal for non-IO
3399 requests so they can be processed faster.
3402 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3404 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3405 created file with mode 000
3406 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3407 allow an ftruncate() as a non-root user to a file with mode 000.
3408 Lustre now respects this flag to disable mode checks when
3409 truncating a file owned by the user
3412 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3415 Description: Revoking locks from clients that went dead or catatonic might take
3417 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3418 cancellation of such locks instant on servers without waiting for
3419 any reply from clients. Clients drops these locks when cancel
3420 notification from server is received without replying.
3423 Frequency : liblustre-only, when liblustre client dies or becomes busy
3425 Description: Doing ls on Linux clients can take a long time with active
3427 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3428 granting such locks to it in the first place if possible. Locks
3429 are taken by proxy on the OST during the read or write and
3430 dropped immediately afterward. Add connect flags handling, do
3431 not grant locks to liblustre clients for glimpse ASTs.
3433 Severity : enhancement
3435 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3436 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3437 sequential reads and behaves badly with applications doing random
3438 reads. Improve it by reading ahead at least read region, and
3439 avoiding excessive large RPC for small reads.
3441 Severity : enhancement
3443 Description: Creating more than 1000 files for a single job may cause a load
3444 imbalance on the OSTs if there are also a large number of OSTs.
3445 Details : qos_prep_create() uses an OST index reseed value that is an
3446 even multiple of the number of available OSTs so that if the
3447 reseed happens in the middle of the object allocation it will
3448 still utilize the OSTs as uniformly as possible.
3453 Description: OST or MDS may oops in ping_evictor_main()
3454 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3455 but doesn't restart at beginning of obd_exports_timed list
3458 Severity : enhancement
3460 Description: improve by-nid export eviction on the MDS and OST
3461 Details : allow multiple exports with the same NID to be evicted at one
3462 time without re-searching the exports list.
3465 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3467 Description: MDS may oops in groups_free()
3468 Details : in rare race conditions a newly allocated group_info struct is
3469 freed again, and this can be NULL. The 2.4 compatibility code
3470 for groups_free() checked for a NULL pointer, but 2.6 did not.
3473 Frequency : common for liblustre clients doing little filesystem IO
3474 Bugzilla : 9352, 7313
3475 Description: server may evict liblustre clients accessing contended locks
3476 Details : if a client is granted a lock or receives a completion AST
3477 with a blocking AST already set it would not reply to the AST
3478 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3481 Frequency : lfs setstripe, only systems with more than 160 OSTs
3483 Description: unable to set striping with a starting offset beyond OST 160
3484 Details : llapi_create_file() incorrectly limited the starting stripe
3485 index to the maximum single-file stripe count.
3488 Frequency : LDAP users only
3490 Description: lconf did not handle in-kernel recovery with LDAP properly
3491 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3493 Severity : enhancement
3495 Description: bind OST threads to NUMA nodes to improve performance
3496 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3497 node and do their allocations there to localize memory access
3499 Severity : enhancement
3501 Description: llmount can determine client NID directly from Myrinet (GM)
3502 Details : the client NID code from gmnalnid was moved directly into
3503 llmount, removing the need to use this or specifying the
3504 client NID explicitly when mounting GM clients with zeroconf
3507 Frequency : if client is started with down MDS
3509 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3510 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3511 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3512 ptlrpc_import_delayed_req() because we didn't check if the
3513 request was interrupted, and we also didn't break out of the
3514 event loop if there was no timeout
3519 Description: data loss during non-page-aligned writes to a single file from
3520 both multiple nodes and multiple threads on one node at same time
3521 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3522 inconsistency led to false short-reads, that were cached and later
3523 used by ->prepare_write() to fill in partially written page,
3524 leading to data loss.
3527 Frequency : always, if lconf --abort_recovery used
3529 Description: lconf --abort_recovery fails with 'Operation not supported'
3530 Details : lconf was attempting to abort recovery on the MDT device and not
3533 Severity : enhancement
3535 Description: remove cleanup logs
3536 Details : replace lconf-generated cleanup logs with lustre internal
3537 cleanup routines. Eliminates the need for client-cleanup and
3540 Severity : enhancement
3542 Description: add support for EAs (user and system) on lustre filesystems
3543 Details : it is now possible to store extended attributes in the Lustre
3544 client filesystem, and with the user_xattr mount option it
3545 is possible to allow users to store EAs on their files also
3547 Severity : enhancement
3549 Description: Add possibility (config option) to show minimal available OST free
3551 Details : When compiled with --enable-mindf configure option, statfs(2)
3552 (and so, df) will return least minimal free space available from
3553 all OSTs as amount of free space on FS, instead of summary of
3554 free spaces of all OSTs.
3556 Severity : enhancement
3558 Description: do not expand extent locks acquired on OST-side
3559 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3560 by server: they are not cached anyway.
3563 Frequency : when mmap is used/binaries executed from Lustre
3565 Description: Unmmap pages before throwing them away from read cache.
3566 Details : llap_shrink cache now attempts to unmap pages before discarding
3567 them (if unmapping failed - do not discard). SLES9 kernel has
3568 extra checks that trigger if this unmapping is not done first.
3573 Description: lconf didn't resolve symlinks before checking to see whether a
3574 given mountpoint was already in use
3577 Frequency : when migrating failover services
3578 Bugzilla : 6395, 9514
3579 Description: When migrating a subset of services from a node (e.g. failback
3580 from a failover service node) the remaining services would
3581 time out and evict clients.
3582 Details : lconf --force (implied by --failover) sets the global obd_timeout
3583 to 5 seconds in order to quickly disconnect, but this caused
3584 other RPCs to time out too quickly. Do not change the global
3585 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3587 Severity : enhancement
3588 Frequency : if MDS is started with down OST
3589 Bugzilla : 9439,5706
3590 Description: Allow startup/shutdown of an MDS without depending on the
3591 availability of the OSTs.
3592 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3593 Add appropriate locking and lov-osc refcounts for safe
3594 cleaning. Add osc abort_inflight calls in case the
3595 synchronize never started.
3598 Frequency : occasional (Cray XT3 only)
3600 Description: root not authorized to access files in CRAY_PORTALS environment
3601 Details : The client process capabilities were not honoured on the MDS in
3602 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3603 been accessed by an authorized user then root was able to access
3604 the file on the local client also. The root user capabilities
3605 are now allowed on the MDS, as this environment has secure UID.
3608 Frequency : occasional
3610 Description: ldiskfs "too long searching" message happens too often
3611 Details : A debugging message (otherwise harmless) prints too often on
3612 the OST console. This has been reduced to only happen when
3613 there are fragmentation problems on the filesystem.
3618 Description: Division by zero in statfs when all OSCs are inactive
3619 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3620 lov_statfs divides by value returned by lov_get_stripecnt().
3624 Bugzilla : 9489, 3273
3625 Description: First write from each client to each OST was only 4kB in size,
3626 to initialize client writeback cache, which caused sub-optimal
3627 RPCs and poor layout on disk for the first writen file.
3628 Details : Clients now request an initial cache grant at (re)connect time
3629 and so that they can start streaming writes to the cache right
3630 away and always do full-sized RPCs if there is enough data.
3631 If the OST is rebooted the client also re-establishes its grant
3632 so that client cached writes will be honoured under the grant.
3637 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3638 Details : Now I/O RPCs go to different portal number and (presumably) fast
3639 lock requests (and glimses) and other RPCs get their own service
3640 threads pool that should be able to service those RPCs
3643 Severity : enhancement
3645 Description: Ability to exchange lustre version between client and servers and
3646 issue warnings at client side if client is too old. Also for
3647 liblustre clients there is ability to refuse connection of too old
3649 Details : New 'version' field is added to connect data structure that is
3650 filled with version info. That info is later checked by server and
3654 Frequency : rare, liblustre only.
3655 Bugzilla : 9296, 9581
3656 Description: Two simultaneous writes from liblustre at offset within same page
3657 might proceed at the same time overwriting eachother with stale
3659 Details : I/O lock withing llu_file_prwv was released too early, before data
3660 actually was hitting the wire. Extended lock-holding time until
3661 server acknowledges receiving data.
3664 Frequency : extremely rare. Never observed in practice.
3666 Description: avoid generating lustre_handle cookie of 0.
3667 Details : class_handle_hash() generates handle cookies by incrementing
3668 global counter, and can hit 0 occasionaly (this is unlikely, but
3669 not impossible, because initial value of cookie counter is
3670 selected randonly). Value of 0 is used as a sentinel meaning
3671 "unassigned handle" --- avoid it. Also coalesce two critical
3672 sections in this function into one.
3674 Severity : enhancement
3676 Description: allow liblustre clients to delegate truncate locking to OST
3677 Details : To avoid overhead of locking, liblustre client instructs OST to
3678 take extent lock in ost_punch() on client's behalf. New connection
3679 flag is added to handle backward compatibility.
3681 Severity : enhancement
3682 Bugzilla : 4928, 7341, 9758
3683 Description: allow number of OST service threads to be specified
3684 Details : a module parameter allows the number of OST service threads
3685 to be specified via "options ost ost_num_threads={N}" in the
3686 OSS's /etc/modules.conf or /etc/modprobe.conf.
3690 Bugzilla : 6146, 9635, 9895
3691 Description: servers crash with bad pointer in target_handle_connect()
3692 Details : In rare cases when a client is reconnecting it was possible that
3693 the connection request was the last reference for that export.
3694 We would temporarily drop the export reference and get a new
3695 one, but this may have been the last reference and the export
3696 was just destroyed. Get new reference before dropping old one.
3698 Severity : enhancement
3699 Frequency : if client is started with failover MDS
3701 Description: Allow multiple MDS hostnames in the mount command
3702 Details : Try to read the configuration from all specified MDS
3703 hostnames during a client mount in case the "primary"
3706 Severity : enhancement
3708 Description: Stop sending data to evicted clients as soon as possible.
3709 Details : Check if the client we are about to send or are sending data to
3710 was evicted already. (Check is done every second of waiting,
3711 for which l_wait_event interface was extended to allow checking
3712 of exit condition at specified intervals).
3715 Frequency : rare, normally only when NFS exporting is done from client
3717 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3719 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3722 Frequency : always, if srand() is called before liblustre initialization
3724 Description: Liblustre uses system PRNG disturbing its usage by user application
3725 Details : Introduce internal to lustre fast and high-quality PRNG for
3726 lustre usage and make liblustre and some other places in generic
3727 lustre code to use it.
3729 Severity : enhancement
3730 Bugzilla : 9477, 9557, 9870
3731 Description: Verify that the MDS configuration logs are updated when xml is
3732 Details : Check if the .xml configuration logs are newer than the config
3733 logs stored on the MDS and report an error if this is the case.
3734 Request --write-conf, or allow starting with --old_conf.
3736 Severity : enhancement
3738 Description: Handle symlinks in the path when checking if Lustre is mounted.
3739 Details : Resolve intermediate symlinks when checking if a client has
3740 mounted a filesystem to avoid duplicate client mounts.
3745 Description: lconf can hit an error exception but still return success.
3746 Details : The lconf command catches the Command error exception at the top
3747 level script context and will exit with the associated exit
3748 status, but doesn't ensure that this exit status is non-zero.
3753 Description: failure of ptlrpc thread startup can cause oops
3754 Details : Starting a ptlrpc service thread can fail if there are a large
3755 number of threads or the server memory is very fragmented.
3756 Handle this without oopsing.
3759 Frequency : always, only if liblustre and non-default acceptor port was used
3761 Description: liblustre cannot connect to servers with non-default acceptor port
3762 Details : tcpnal_set_default_params() was not called and was therefore
3763 ignoring the environment varaible TCPNAL_PORT, as well as other
3764 TCPNAL_ environment variables
3769 Description: two objects could be created on the same OST for a single file
3770 Details : If an OST is down, in some cases it was possible to create two
3771 objects on a single OST for a single file. No problems other
3772 than potential performance impact and spurious error messages.
3776 Bugzilla : 5681, 9562
3777 Description: Client may oops in ll_unhash_aliases
3778 Details : Client dcache may become inconsistent in race condition.
3779 In some cases "getcwd" can fail if the current directory is
3785 Description: Inode refcounting problems in NFS export code
3786 Details : link_raw functions used to call d_instantiate without obtaining
3787 extra inode reference first.
3791 Bugzilla : 9942, 9903
3792 Description: Referencing freed requests leading to crash, memleaks with NFS.
3793 Details : We used to require that call to ll_revalidate_it was always
3794 followed by ll_lookup_it. Also with revalidate_special() it is
3795 possible to call ll_revalidate_it() twice for the same dentry
3796 even if first occurence returned success. This fix changes semantic
3797 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3798 reference on a request referred from the intent.
3799 ll_intent_release() then releases such a request.
3802 Frequency : rare, normally benchmark loads only
3804 Description: unlinked inodes were kept in memory on the client
3805 Details : If a client is repeatedly creating and unlinking files it
3806 can accumulate a lot of stale inodes in the inode slab cache.
3807 If there is no other client load running this can cause the
3808 client node to run out of memory. Instead flush old inodes
3809 from client cache that have the same inode number as a new inode.
3812 Frequency : SLES9 2.6.5 kernel and long filenames only
3813 Bugzilla : 9969, 10379
3814 Description: utime reports stale NFS file handle
3815 Details : SLES9 uses out-of-dentry names in some cases, which confused
3816 the lustre dentry revalidation. Change it to always use the
3820 Frequency : rare, unless heavy write-truncate concurrency is continuous
3821 Bugzilla : 4180, 6984, 7171, 9963, 9331
3822 Description: OST becomes very slow and/or deadlocked during object unlink
3823 Details : filter_destroy() was holding onto the parent directory lock
3824 while truncating+unlinking objects. For very large objects this
3825 may block other threads for a long time and slow overall OST
3826 responsiveness. It may also be possible to get a lock ordering
3827 deadlock in this case, or run out of journal credits because of
3828 the combined truncate+unlink. Solution is to do object truncate
3829 first in one transaction without parent lock, and then do the
3830 final unlink in a new transaction with the parent lock. This
3831 reduces the lock hold time dramatically.
3834 Frequency : rare, 2.4 kernels only
3836 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3837 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3838 before unmounting their filesystem in order to avoid deadlock.
3839 The kernel_locked() function in 2.4 kernels only checks whether
3840 the kernel lock is held, not whether it is this process that is
3841 holding it as 2.6 kernels do.
3846 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3847 Details : The client ptlrpc code may be trying to reconnect to a down
3848 server before a previous connection attempt has timed out.
3849 Increase the reconnect interval to be longer than the connection
3850 timeout interval to avoid sending duplicate connections to
3854 Frequency : echo_client brw_test command
3856 Description: fix echo_client to work with OST preallocated code
3857 Details : OST preallocation code (5137) didn't take echo_client IO path
3858 into account: echo_client calls filter methods outside of any
3859 OST thread and, hence, there is no per-thread preallocated
3860 pages and buffers to use. Solution: hijack pga pages for IO. As
3861 a byproduct, this avoids unnecessary data copying.
3865 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3866 Description: Client can oops in mdc_commit_close() after open replay
3867 Details : It was possible for the MDS to return an open request with no
3868 transaction number in mds_finish_transno() if the client was
3869 evicted, but without actually returning an error. Clients
3870 would later try to replay that open and may trip an assertion
3871 Simplify the client close codepath, and always return an error
3872 from the MDS in case the open is not successful.
3875 Frequency : rare, 2.6 OSTs only
3877 Description: OST may deadlock under high load on fragmented files
3878 Details : If there was a heavy load and highly-fragmented OST filesystems
3879 it was possible to have all the OST threads deadlock waiting on
3880 allocation of biovecs, because the biovecs were not released
3881 until the entire RPC IO was completed. Instead, release biovecs
3882 as soon as they are complete to ensure forward IO progress.
3884 Severity : enhancement
3886 Description: Support for specifying external journal device at mount
3887 Details : If an OST or MDS device is formatted with an external journal
3888 device, this device major/minor is stored in the ext3 superblock
3889 and may not be valid for failover. Allow detecting and
3890 specifying the external journal at mount time.
3895 Description: Mounting an MDS with pending unlinked files may cause oops
3896 Details : target_finish_recovery() calls mds_postrecov() which returned
3897 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3898 considers this an error and immediately begins cleaning up the
3899 lov, just after starting the mds_lov process
3901 Severity : enhancement
3903 Description: Implement 'lfs df' to report actual free space on per-OST basis
3904 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3905 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3906 usage of objects; '-h' to report in human readable format.
3908 ------------------------------------------------------------------------------
3910 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3917 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3918 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3919 at startup during orphan unlinking. Instead, skip buddy update
3920 before setup, as it will be regenerated after recovery is complete.
3925 Description: If an OST is inactive, its locks might reference stale inodes.
3926 Details : lov_change_cbdata() must iterate over all namespaces, even if
3927 they are inactive to clear inode references from the lock.
3929 Severity : enhancement
3930 Frequency : occasional, if non-standard max_dirty_mb used
3932 Description: Client will block write RPCs if not enough grant
3933 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3934 then the client will block writes while waiting for another RPC
3935 to complete instead of consuming its dirty limit. With change
3936 we get improved performance when max_dirty_mb is small.
3938 Severity : enhancement
3939 Bugzilla : 3389, 6253
3940 Description: Add support for supplementary groups on the MDS.
3941 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3942 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3943 lookups for user supplementary groups into a cache.
3947 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3948 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3949 for liblustre O_EXCL is enforced by MDS.
3953 Description: readdir never returns NULL in liblustre.
3954 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3955 dentry in struct dirent.
3959 Frequency : liblustre only, depends on application IO pattern
3960 Description: liblustre clients evicted if not contacting servers
3961 Details : Don't put liblustre clients into the ping_evictor list, so
3962 they will not be evicted by the pinger ever.
3964 Severity : enhancement
3966 Description: Add ability to evict clients by NID from MDS.
3967 Details : By echoing "nid:$NID" string into
3968 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3969 $NID would be instantly evicted from this MDS and from all active
3970 OSTs connected to it.
3974 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3975 Details : lookup_it_finish() used to query file size from OSTs that was not
3980 Description: service threads change working directory to that of init
3981 Details : Starting lustre service threads may pin the working directory
3982 of the parent thread, making that filesystem busy. Threads
3983 now change to the working directory of init to avoid this.
3987 Frequency : during shutdown only
3988 Description: shutdown with a failed MDS or OST can cause unmount to hang
3989 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3994 Frequency : 2.6 only
3995 Description: chmod/chown may include an extra supplementary group
3996 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3997 supplementary group and if none is specified this is used.
4000 Bugzilla : 5479 (6816)
4002 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4003 Details : If lookup is in progress on a file that is unlinked we might try
4004 to revalidate the inode and fail in revalidate after lookup is
4005 complete and ll_file_open() enqueues the open again but
4006 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4009 Frequency : always, if lconf --abort_recovery used
4011 Description: lconf --abort_recovery fails with 'Operation not supported'
4012 Details : lconf was attempting to abort recovery on the MDT device and not
4015 ------------------------------------------------------------------------------
4017 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4022 Frequency : rare (only unsupported configurations with a node running as an
4024 Bugzilla : 6514, 5137
4025 Description: Mounting a Lustre file system on a node running as an OST could
4027 Details : OSTs now preallocates memory needed to write out data at
4028 startup, instead of when needed, to avoid having to
4029 allocate memory in possibly low memory situations.
4030 Specifically, if the file system is mounted on on OST,
4031 memory pressure could force it to try to write out data,
4032 which it needed to allocate memory to do. Due to the low
4033 memory, it would be unable to do so and the node would
4034 become unresponsive.
4036 Severity : enhancement
4038 Description: Addition of lconf --service command line option
4039 Details : lconf now accepts a '--service <arg>' option, which is
4040 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4042 Severity : enhancement
4044 Description: Failover mode is now the default for OSTs.
4045 Details : By default, OSTs will now run in failover mode. To return to
4046 the old behaviour, add '--failout' to the lmc line for OSTs.
4048 Severity : enhancement
4050 Description: Health checks are now provided for MDS and OSTs
4051 Details : Additional detailed health check information on MSD and OSTs
4052 is now provided through the procfs health_check value.
4055 Frequency : occasional, depends on IO load
4057 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4058 after numerous create/delete cycles
4059 Details : The ext3 inode allocation policy would not allocate new inodes
4060 very well on the OSTs because there are no new directories
4061 being created. Instead we look for groups with free space if
4062 the parent directories are nearly full.
4067 Description: Network or server problems during mount may cause partially
4068 mounted clients instead of returning an error.
4069 Details : The config llog parsing code may overwrite the error return
4070 code during mount error handling, returning success instead
4076 Description: MDS can fail to allocate large reply buffers
4077 Details : After long uptimes the MDS can fail to allocate large reply
4078 buffers (e.g. zconf client mount config records) due to memory
4079 fragmentation or consumption by the buffer cache. Preallocate
4080 some large reply buffers so that these replies can be sent even
4081 under memory pressure.
4085 Frequency : rare (liblustre)
4086 Description: fsx running with liblustre complained that using truncate() to
4087 extend the file doesn't work. This patch corrects that issue.
4088 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4089 ATTR_SIZE and lsm use in llu_setattr_raw.
4093 Frequency : rare, only 2.6 kernels
4094 Description: Unusual file access patterns on the MDS may result in inode
4095 data being lost in very rare circumstances.
4096 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4097 code in the 2.6 kernel caused the inode and/or EA data not to
4098 be read from disk, causing single-file corruption.
4102 Frequency : rare, only 2.6 filesystems using extents
4103 Description: Heavy concurrent write and delete load may cause data corruption.
4104 Details : It was possible under high-load situations to have an extent
4105 metadata block in the block device cache from a just-unlinked
4106 file overwrite a newly-allocated data block. We now unmap any
4107 metadata buffers that alias just-allocated data blocks.
4111 Frequency : filesystems with default stripe_count larger than 77
4112 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4113 Details : lconf specifies an inode size of 4096 bytes when the default
4114 stripe_count is larger than 77. This conflicts with the default
4115 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4116 this case to avoid pinning too much memory for large EAs.
4118 ------------------------------------------------------------------------------
4120 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4125 Frequency : rare (extremely heavy IO load with hundreds of clients)
4127 Description: Client is evicted, gets IO error writing to file
4128 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4129 caused clients to be evicted for AST timeouts. The fixes in
4130 bug 5192 mean we no longer need to have such short AST timeouts
4131 so ldlm_timeout has been increased.
4134 Frequency : occasional during --force or --failover shutdown under load
4135 Bugzilla : 5949, 4834
4136 Description: Server oops/LBUG if stopped with --force or --failover under load
4137 Details : a collection of import/export refcount and cleanup ordering
4138 issues fixed for safer force cleanup
4141 Frequency : only filesystems larger than 120 OSTs
4142 Bugzilla : 5990, 6223
4143 Description: lfs getstripe would oops on a very large filesystem
4144 Details : lov_getconfig used kfree on vmalloc'd memory
4147 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4148 Bugzilla : 6242, 6243
4149 Description: reading from files that had been truncated to a non-zero size
4150 but never opened returned no data
4151 Details : ll_file_read() reads zeros from no-object files to EOF
4156 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4157 Details : There's little harm in aborting MDS/OSS recovery and letting it
4158 try again, so I removed the LASSERT and return an error instead.
4160 Severity : enhancement
4162 Description: New debugging infrastructure for tracking down data corruption
4163 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4164 (b) cover more of the client-side code path, and (c) try to narrow
4165 down where problems occurred
4169 Bugzilla : 3819, 4364, 4397, 6313
4170 Description: Racing close and eviction MDS could cause assertion in mds_close
4171 Details : It was possible to get multiple mfd references during close and
4172 client eviction, leading to one thread referencing a freed mfd.
4174 Severity: : enhancement
4175 Bugzilla : 3262, 6359
4176 Description: Attempts to reconnect to servers are now more aggressive.
4177 Details : This builds on the enhanced upcall-less recovery that was added
4178 in 1.4.2. When trying to reconnect to servers, clients will
4179 now try each server in the failover group every 10 seconds. By
4180 default, clients would previously try one server every 25 seconds.
4185 Description: After recovery, certain operations trigger a failed
4186 assertion on a client.
4187 Details : Failing over an mds, using lconf -d --failover, while a
4188 client was doing a readdir() call would cause the client to
4189 LBUG after recovery completed and the readdir() was resent.
4191 Severity : enhancement
4193 Description: Default groups are now added by lconf
4194 Details : You can now run lconf --group <servicename> without having to
4195 manually add groups with lmc.
4198 Frequency : occasional
4200 Description: Nodes with an elan id of 0 trigger a failed assertion
4203 Frequency : always when accessing e.g. tty/console device nodes
4205 Description: tty and some other devices nodes cannot be used on lustre
4206 Details : file's private_data field is used by device data and lustre
4207 values in there got lost. New field was added to struct file to
4208 store fs-specific private data.
4211 Frequency : when exporting Lustre via NFS
4213 Description: NFSD failed occasionally when looking up a path component
4214 Details : NFSD is looking up ".." which was broken in ext3 directories
4215 that had grown large enough to become hashed.
4218 Frequency : Clusters with multiple interfaces not on the same subnet
4220 Description: Nodes will repeatedly try to reconnect to an interface which it
4221 cannot reach and report an error to the log.
4222 Details : Extra peer list entries will be created by lconf with some peers
4223 unreachable. lconf now validates the peer before adding it.
4226 Frequency : Only if a default stripe is set on the filesystem root.
4228 Description: Setting a default stripe on the filesystem root prevented the
4229 filesystem from being remounted.
4230 Details : The client was sending extra request flags in the root getattr
4231 request and did not allocate a reply buffer for the dir EA.
4234 Frequency : occasional, higher if lots of files are accessed by one client
4235 Bugzilla : 6159, 6097
4236 Description: Client trips assertion regarding lsm mismatch/magic
4237 Details : While revalidating inodes the VFS looks up inodes with ifind()
4238 and in rare cases can find an inode that is being freed.
4239 The ll_test_inode() code will free the lsm during ifind()
4240 when it finds an existing inode and then the VFS later attaches
4241 this free lsm to a new inode.
4245 Bugzilla : 6422, 7030
4246 Description: MDS deadlock between mkdir and client eviction
4247 Details : Creating a new file via mkdir or mknod (starting a transaction
4248 and getting the ns lock) can deadlock with client eviction
4249 (gets ns lock and trying to finish a synchronous transaction).
4252 Frequency : occasional
4253 Description: While starting a server, the fsfilt_ext3 module could not be
4255 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4256 kernels. Previously, lconf would still use the ext3 name
4257 when trying to load modules. Now, it will correctly use
4258 ext3 on 2.4 and ldiskfs on 2.6.
4260 Severity : enhancement
4261 Description: The default stripe count has been changed to 1
4262 Details : The interpretation of the default stripe count (0, to lfs
4263 or lmc) has been changed to mean striping across a single
4264 OST, rather than all available. For general usage we have
4265 found a stripe count of 1 or 2 works best.
4267 Severity : enhancement
4268 Description: Add support for compiling against Cray portals.
4269 Details : Conditional compiling for some areas that are different
4273 Frequency : occasional
4274 Bugzilla : 6409, 6834
4275 Description: Creating files with an explicit stripe count may lead to
4276 a failed assertion on the MDS
4277 Details : If some OSTs are full or unavailable, creating files may
4278 trigger a failed assertion on the MDS. Now, Lustre will
4279 try to use other servers or return an error to the
4283 Frequency : occasional
4285 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4286 caused lock timeout during readahead (since 1.4.2).
4287 Details : Processes doing readahead might match a lock that hasn't been
4288 granted yet if there are overlapping and conflicting lock
4289 requests. The readahead process waits on ungranted lock
4290 (original lock is CBPENDING), while OST waits for that process
4291 to cancel CBPENDING read lock and eventually evicts client.
4293 Severity : enhancement
4295 Description: Initial enabling of flock support for clients
4296 Details : Implements fcntl advisory locking and file status functions.
4297 This feature is provided as an optional mount flag (default
4298 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4299 locking are implemented yet, and those that are are not guaranteed
4300 to be completely correct in production environments.
4301 mount -t lustre -o [flock|noflock] ...
4304 Frequency : occasional
4306 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4307 assertion in the ext3 JBD (journaling) layer.
4308 Details : The b_committed_data struct is protected by the big kernel lock
4309 in 2.4 kernels, serializing journal_commit_transaction() and
4310 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4311 access to this struct. In 2.6 kernels there is finer grained
4312 locking to improve SMP performance of the JBD layer.
4316 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4321 Description: merge and cleanup kernel patches.
4324 -----------------------------------------------------------------------------
4326 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4328 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4329 versions of Lustre. Please update all servers and clients to
4330 version 1.4.2 or later at the same time. You must also run
4331 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4332 to update the configuration logs.
4334 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4335 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4336 - swab llog records of type '0' so we get proper header size/idx (5861)
4337 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4338 - fix rename of one directory over another leaking an inode (5953)
4339 - avoid SetPageDirty on 2.6 (5981)
4340 - don't re-add just-being-destroyed locks to the waiting list (5653)
4341 - when creating new directories, inherit the parent's custom
4342 striping settings if present parent (3048)
4343 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4344 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4345 - don't hold client locks on temporary worklist from l_lru (5666)
4346 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4347 - automatically evict dead clients (5921)
4348 - Update file size properly in create+truncate+fstat case (6196)
4349 - Do not unhash mountpoint dentries, do not allow removal of
4351 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4352 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4353 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4354 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4355 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4356 - Correctly handle reads of files with no objects (6243)
4357 - lctl recover will also mark a device active if deactivate used (5933)
4359 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4360 - allow --write-conf on an MDS with different nettype than client (5619)
4361 - don't write config llogs to MDS for mounts not from that MDS (5617)
4362 - lconf should create multiple TCP connections from a client (5201)
4363 - init scripts are now turned off by default; run chkconfig --on
4364 lustre and chkconfig --on lustrefs to use them
4365 - upcalls are no longer needed for clients to recover to failover
4367 - add --abort-recovery option to lconf to abort recovery on device
4369 - add support for an arbitrary number of OSTs (3026)
4370 - Quota support protocol changes.
4371 - forward compatibility changes to wire structs (6007)
4372 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4373 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4374 - improved "lustre status" script
4375 - initialize blocksize for non-regular files (6062)
4376 - added --disable-server and --disable-client configure options (5782)
4377 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4378 - Vanilla 2.4.29 support
4379 - increase maximum number of obd devices to 520 (6242)
4380 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4381 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4382 SLES 9 2.6.5 kernels
4383 - limit stripes per file to 160 (the maximum EA size) (6093)
4385 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4388 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4389 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4390 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4391 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4392 - small ext3 extents cleanups and fixes (5733)
4393 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4394 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4395 - avoid needless client->OST connect, fix handle mismatch (5317)
4396 - fix DLM error path that led to out-of-sync client, long delays (5779)
4397 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4398 - fix several locking issues related to i_size (5492,5624,5654,5672)
4399 - don't move pending lock onto export if it is already evicted (5683)
4400 - fix kernel oops when creating .foo in unlinked directory (5548)
4401 - fix deadlock in obdfilter statistics vs. object create (5811)
4402 - use time_{before,after} to avoid timer jiffies wrap (5882)
4403 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4404 - Do not leak request if server was not able to process it (5154)
4405 - If mds_open unable to find parent dir, make that negative lookup(5154)
4406 - don't create new directories with extent-mapping (5909, 5936)
4408 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4409 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4410 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4411 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4412 - support for RHEL4 kernel on i686 (5773)
4413 - provide error messages when incompatible logs are encountered (5898)
4415 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4416 * version 1.4.0.10 (1.4.1 release candidate 1)
4418 - don't keep a lock reference when lock is not granted (4238)
4419 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4420 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4421 - drop import inflight refcount on signal_completed_replay error (5255)
4422 - unlock page after async write error during send (3677)
4423 - handle missing objects in filter_preprw_read properly (5265)
4424 - no transno return for symlink open, don't save no-trasno open (3440)
4425 - don't try to complete elan receive that already failed (4012)
4426 - free RPC server reply state on error (5406)
4427 - clean up thread from ptlrpc_start_thread() on error (5160)
4428 - readahead could read extra page into cache that wasn't ejected (5388)
4429 - prevent races in class_attach/setup/cleanup/detach (5260)
4430 - don't dereference de->d_inode after l_dput of de (5458)
4431 - use "int" for stripe value returned from lock_to_stripe (5544)
4432 - mballoc allocation and error-checking fixes in 2.6 (5504)
4433 - block device patches to fix I/O request sizes in 2.6 (5482)
4434 - look up hostnames for IB nals (5602)
4435 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4436 - don't start multiple acceptors for the same port (5277)
4437 - fix incorrect LASSERT in mds_getattr_name (5635)
4438 - export a proc file for general "ping" checking (5628)
4439 - fix "lfs check" to not block when the MDS is down (5628)
4441 - service request history (4965)
4442 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4443 - create an "evict_client" /proc entry on OSTs, like the MDS has
4444 - fix mount usage message, return errors per mount(8) (5168)
4445 - change grep [] to grep "[]" in tests so they work in more UMLs
4446 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4447 - remove ancient LOV_MAGIC_V0 EA support (5047)
4448 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4449 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4450 - allow readahead allocations to fail when low on memory (5383)
4451 - mmap locking landed again, after considerable improvement (2828)
4452 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4454 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4457 - send OST transaction number in read/write reply to free req (4966)
4458 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4459 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4460 - the watchdog thread now runs as interruptible (5246)
4461 - flock/lockf fixes (but it's still disabled, pending 5135)
4462 - don't use EXT3 constants in llite code (5094)
4463 - memory shortage at startup could cause assertion (5176)
4465 - reorganization of lov code
4466 - single portals codebase
4468 - add extents/mballoc support (5025)
4469 - direct I/O reads in the obdfilter (4048)
4470 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4472 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4475 - send OST transaction number in read/write reply to free req (4966)
4476 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4477 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4478 - the watchdog thread now runs as interruptible (5246)
4479 - handle missing objects in filter_preprw_read properly (5265)
4480 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4481 - drop import inflight refcount on signal_completed_replay error (5255)
4482 - unlock page after async write error during send (3677)
4483 - return original error code on reconstructed replies (3761)
4484 - no transno return for symlink open, don't save no-trasno open (3440)
4486 - add pid to ldlm debugging output (4922)
4487 - bump the watchdog timeouts -- we can't handle 30sec yet
4488 - extra debugging for orphan dentry/inode bug (5259)
4490 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4493 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4494 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4495 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4496 - fix NULL dereference in /proc/sys/portals/routes (4827)
4497 - allow failed mdc_close() operations to be interrupted (4561)
4498 - stop precreate on OST before MDS would time out on it (4778)
4499 - don't send partial-page writes before EOF from client (4410)
4500 - discard client grant for sub-page writes on large-page clients (4520)
4501 - don't free dentries not owned by NFS code, check generation (4806)
4502 - fix lsm leak if mds_create_objects() fails (4801)
4503 - limit debug_daemon file size, always print CERROR messages (4789)
4504 - use transno after validating reply (3892)
4505 - process timed out requests if import state changes (3754)
4506 - update mtime on OST during writes, return in glimpse (4829)
4507 - add mkfsoptions to LDAP (4679)
4508 - use ->max_readahead method instead of zapping global ra (5039)
4509 - don't interrupt __l_wait_event() during strace
4511 - add software watchdogs to catch hung threads quickly (4941)
4512 - make lustrefs init script start after nfs is mounted
4513 - fix CWARN/ERROR duplication (4930)
4514 - return async write errors to application if possible (2248)
4515 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4516 - print NAL number in %x format (4645)
4517 - update barely-supported suse-2.4.21-171 series (4842)
4518 - support for sles 9 %post scripts
4519 - support for building 2.6 kernel-source packages
4520 - support for sles km_* packages
4522 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4525 - ignore -ENOENT errors in osc_destroy (3639)
4526 - notify osc create thread that OSC is being cleaned up (4600)
4527 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4528 - reconstruct ost_handle() like mds_handle() (4657)
4529 - create a new thread to do import eviction to avoid deadlock (3969)
4530 - let lconf resolve symlinked-to devices (4629)
4531 - don't unlink "objects" from directory with default EA (4554)
4532 - hold socknal file ref over connect in case target is down (4394)
4533 - allow more than 32000 subdirectories in a single directory (3244)
4534 - fix blocks count for O_DIRECT writes (3751)
4535 - OST returns ENOSPC from object create when no space left (4539)
4536 - don't send truncate RPC if file size isn't changing (4410)
4537 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4538 - bind to privileged port in socknal and tcpnal (3689)
4540 - rate limit CERROR/CWARN console message to avoid overload (4519)
4541 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4542 - basic mmap support (3918)
4543 - kernel patch series update from b1_4 (4711)
4545 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4548 - avoid crash during MDS cleanup with OST shut down (2775)
4549 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4550 - don't use bad inodes on the MDS (3744)
4551 - dynamic object preallocation to improve recovery speed (4236)
4552 - don't hold spinlock over lock dumping or change debug flags (4401)
4553 - don't zero obd_dev when it is force cleaned (3651)
4554 - print grants to console if they go negative (4431)
4555 - "lctl deactivate" will stop automatic recovery attempts (3406)
4556 - look for existing locks in ldlm_handle_enqueue() (3764)
4557 - don't resolve lock handle twice in recovery avoiding race (4401)
4558 - revalidate should check working dir is a directory (4134)
4560 - don't always mark "slow" obdfilter messages as errors (4418)
4562 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4565 - don't close LustreDB during write_conf until it is done (3860)
4566 - fix typo in lconf for_each_profile (3821)
4567 - allow dumping logs from multiple threads at one time (3820)
4568 - don't allow multiple threads in OSC recovery (3812)
4569 - fix debug_size parameters (3864)
4570 - fix mds_postrecov to initialize import for llog ctxt (3121)
4571 - replace config semaphore with spinlock (3306)
4572 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4573 - don't allow enqueue to complete on a destroyed export (3822)
4574 - down write_lock before checking llog header bitmap (3825)
4575 - recover from lock replay timeout (3764)
4576 - up llog sem before sending rpc (3652)
4577 - reduce ns lock hold times when setting kms (3267)
4578 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4579 - fix NULL deref and obd_dev leak on setup error (3312)
4580 - replace some LBUG about llog ops with error handling (3841)
4581 - don't match INVALID dentries from d_lookup and spin (3784)
4582 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4583 - fix invalid assertion in ptlrpc_set_wait (3880)
4585 - add libwrap support for the TCP acceptor (3996)
4586 - add /proc/sys/portals/routes for non-root route listing (3994)
4587 - allow setting MDS UUID in .xml (2580)
4588 - print the stack of a process that LBUGs (4228)
4590 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4593 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4594 - make sure to unset replay flag from failed open requests (3440)
4595 - if default stripe count is 0, use OST count for inode size (3636)
4596 - update parent mtime/ctime on client for create/unlink (2611)
4597 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4598 - free recovery state on server during a forced cleanup (3571)
4599 - unregister_reply for resent reqs (3063)
4600 - loop back devices mounting and status check on 2.6 (3563)
4601 - fix resource-creation race that can provoke i_size == 0 (3513)
4602 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4603 - more debugging for page-accounting assertion (3746)
4604 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4605 - avoid deadlock after precreation failure (3758)
4606 - fix race and lock order deadlock in orphan handling (3450, 3750)
4607 - add validity checks when grabbing inodes from l_ast_data (3599)
4609 - add /proc/.../recovery_status to obdfilter (3428)
4610 - lightweight CDEBUG infrastructure, debug daemon (3668)
4611 - change default OSC RPC parameters to be better on small clusters
4612 - turn off OST read cache for files smaller than 32MB
4613 - install man pages and include them in rpms (3100)
4614 - add new init script for (un)mounting lustre filesystems (2593)
4615 - run chkconfig in %post for init scripts (3701)
4616 - drop scimac NAL (unmaintained)
4618 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4621 - clean kiobufs before and after use (3485)
4622 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4623 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4624 - change init script to fail more clearly if not run as root (1528)
4625 - allow clients to reconnect during replay (1742)
4626 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4627 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4628 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4629 - fix cond_resched (3554)
4630 - only evict unfinished clients after recovery (3515)
4631 - allow bulk resend, prevent data loss (3570)
4632 - dynamic ptlrpc request buffer allocation (2102)
4633 - don't allow unlinking open directory if it isn't empty (2904)
4634 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4635 - remove extraneous obd dereference causing LASSERT failure (3334)
4636 - don't use get_cycles() when creating temp. files on the mds (3156)
4637 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4638 - handle EEXIST for set-stripe, set proper directory name (3336)
4640 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4641 - fix ksocknal_fmb_callback() error messages (2918)
4643 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4646 - don't copy lvb into (possibly NULL) reply on error (2983)
4647 - don't deref dentry after dput, don't free lvb on error (2922)
4648 - use the kms to determine writeback rpc length (2947)
4649 - increment oti_logcookies when osc is inactive (2948)
4650 - update client's i_blocks count via lvb messages (2543)
4651 - handle intent open/close of special files properly (1557)
4652 - mount MDS with errors=remount-ro, like obdfilter (2009)
4653 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4654 - don't use cancelling-locks' kms values (2947)
4655 - use highest lock extent for kms, not last one (2925)
4656 - don't dereference ERR_PTR() dentry in error handling path (3107)
4657 - fix thread race in portals_debug_dumplog() (3122)
4658 - create lprocfs device entries at setup instead of at attach (1519)
4659 - common AST error handler, don't evict client on completion race (3145)
4660 - zero nameidata in detach_mnt in 2.6 (3118)
4661 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4662 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4663 - initialize RPC timeout timer earlier for 2.6 (3219)
4664 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4665 - print nal/nid for unknown nid (3258)
4666 - additional checks for oscc recovery before doing precreate (3284)
4667 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4668 - don't crash in mdc_close for bad permissions on open (3285)
4669 - zero i_rdev for non-device files (3147)
4670 - clear page->private before handing to FS, better assertion (3119)
4671 - tune the read pipeline (3236)
4672 - fix incorrect decref of invalidated dentry (2350)
4673 - provide read-ahead stats and refine rpc in flight stats (3328)
4674 - don't hold journal transaction open across create RPC (3313)
4675 - update atime on MDS at close time (3265)
4676 - close LDAP connection when recovering to avoid server load (3315)
4677 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4678 - don't leak open file on MDS after open resend (3325)
4679 - serialize filter_precreate and filter_destroy_precreated (3329)
4680 - loop device shouldn't call sync_dev() for nul device (3092)
4681 - clear page cache after eviction (2766)
4682 - resynchronize MDS->OST in background (2824)
4683 - refuse to mount the same filesystem twice on same mountpoint (3394)
4684 - allow llmount to create routes for mounting behind routers (3320)
4685 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4686 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4687 - fix rare problem with rename on htree directories (3417)
4689 - allow default OST striping configuration per directory (1414)
4690 - fix compilation for qswnal for 2.6 kernels (3125)
4691 - increase maximum number of MDS request buffers for large systems
4692 - change liblustreapi to be useful for external progs like lfsck (3098)
4693 - increase local configuration timeout for slow disks (3353)
4694 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4696 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4699 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4700 - don't overwrite extent policy data in reply if lock was blocked (2901)
4701 - drop filter export grants atomically with removal from device (2663)
4702 - del obd_self_export from work_list in class_disconnect_exports (2908)
4703 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4704 - swab reply message in mdc_close, other PPC fixes (2464)
4705 - fix destroying of named logs (2325)
4706 - overwrite old logs when running lconf --write_conf (2264)
4707 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4708 - fix race in target_handle_connect (2898)
4709 - mds_reint_create() should take same inode create lock (2926)
4710 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4711 - don't close files for self_export to avoid uninitialized obd (2936)
4712 - allow MDS with the same name as client node (2939)
4713 - hold dentry reference for closed log files for unlink (2325)
4714 - reserve space for all logs during transactions (2059)
4715 - don't evict page beyond end of stripe extent (2925)
4716 - don't oops on a deleted current working directory (2399)
4717 - handle hard links to targets without a parent properly (2517)
4718 - don't dereference NULL lock when racing during eviction (2867)
4719 - don't grow lock extents when lots of conflicting locks (2919)
4721 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4724 - account for cache space usage on clients to avoid data loss (974)
4725 - lfsck support in lustre kernel code (2349)
4726 - reduce journal credits needed for BRW writes (2370)
4727 - orphan handling to avoid losing space on client/server crashes
4728 - ptlrpcd can be blocked, stopping ALL progress (2477)
4729 - use lock value blocks to assist in proper KMS, faster stat (1021)
4730 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4731 - recovery for initial connections (2355)
4732 - fixes for mds_cleanup_orphans (1934)
4733 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4734 - block all file creations until orphan recovery completes (1901)
4735 - client remove rq_connection from request struct (2423)
4736 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4737 - recovery timer race (2670)
4738 - mdc_close recovey bug (2532)
4739 - ptlrpc cleanup bug (2710)
4740 - mds timeout on local locks (2588)
4741 - namespace lock held during RPCs (2431)
4742 - handle interrupted sync write properly (2503)
4743 - don't try to handle a message that hasn't been replied to (2699)
4744 - client assert failure during cleanup after abort recovery (2701)
4745 - leak mdc device after failed mount (2712)
4746 - ptlrpc_check_set allows timedout requests to complete (2714)
4747 - wait for inflight reqs when ptlrpcd finishes (2710)
4748 - make sure unregistered services are removed from the srv_list
4749 - reset bulk XID's when resending them (caught by 1138 test)
4750 - unregister_bulk after timeout
4751 - fix lconf error (2694)
4752 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4753 - readahead locks pages, leaves pending causing memory pressure (2673)
4754 - increase OST request buffers to 4096 on large machines (2729)
4755 - fix up permission of existing directories in simple_mkdir (2661)
4756 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4757 - don't assign transno to errored transactions (2742)
4758 - don't delete objects on OST if given a bogus objid from MDS (2751)
4759 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4760 - if rq_no_resend, then timeout request after recovery (2432)
4761 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4762 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4763 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4764 - MDS recovery shouldn't delete objects that it has given out (2730)
4765 - if enqueue arrives after completion, don't clobber LVB (2819)
4766 - don't unlock pages twice when trigger_group_io returns error (2814)
4767 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4768 - don't write pages to disk if there was an error (1450)
4769 - don't ping imports that have recovery disabled (2676)
4770 - take buffered bytes into account when balancing socknal conn (2817)
4771 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4772 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4773 - remove little-endian swabbing of llog records (1987)
4774 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4775 - retry reposting request buffers if they fail (1191)
4776 - grow extent at grant time to avoid granting a revoked lock (2809)
4777 - lock revoke doesn't evict page if covered by a second lock (2765)
4778 - disable VM readahead to avoid reading outside lock extents (2805)
4780 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4781 - updated kernel patches for hp-2.4.20 kernel (2681)
4783 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4786 - fix truncated write corruption (2366)
4787 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4789 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4790 - ensure len(uuid) < 37 in lmc (1171)
4791 - fix ia64 OOPS in llog_test (2255)
4792 - zero end of page at obdfilter for partial page writes (2648)
4793 - don't leave stale dentries around after renames (bug 2428)
4794 - fix timeouts when evicting a client with a single lock held (2642)
4795 - set deadline for the initial HELLO message to drain (2634)
4796 - print out dotted-quad IP addresses in the socknal (2302)
4798 - additional debugging for MDS client eviction problem (2443)
4799 - fix mkfsoptions support for osts (2603, 2604)
4801 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4804 - add series for the vanilla 2.6.0 kernel
4805 - add series for the vanilla 2.4.24 kernel
4806 - add series for a cray x86/64 UL kernel drop
4807 - fix xattr patches for the vanilla 2.4.19 series
4809 - generate true UUIDs in lmc (1171)
4810 - have portals stack dumping break in UML (2466)
4811 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4812 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4813 - wake callback waiting threads on client eviction (2460)
4814 - Add --ptldebug and --subsystem to lmc (1719)
4815 - update assertion to allow safe interrupt allocation
4816 - set rq_no_resend for cancel requests (2432)
4817 - recalculate ptlrpcd timeout after resend (2494)
4818 - call vfs_rmdir when removing pending directories (2368)
4819 - fix renaming a file to itself (2429)
4820 - lmc creates a default one-stripe lov (2454)
4821 - expand procfs space to handle large clusters (2326)
4822 - increase UML stack to avoid overflow
4823 - update lconf's list of debug and subsystem masks
4824 - fix lfs find --obd (2510)
4825 - /proc tunable for disabling filter read caching (2591)
4826 - stop rpm packages from altering slapd.conf (2301)
4827 - disable nagle in the socknal under 0conf (2578)
4828 - choose mds inode size based on stripe count (2572)
4829 - fix kernel-source rpm problems (2516)
4831 - add --disable-doc to avoid pdf generation (2421)
4832 - update documentation, tests, type-os, comments
4833 - avoid format warnings on ia64
4834 - remove the TOE NAL
4835 - tiny code cleanups by removing unused fields
4837 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4840 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4841 - remove the most blatant lies from BUILDING file (2371)
4842 - change default debug level to reasonable production setting
4843 - reduce client side cache size to reduce cache flush time
4844 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4845 - make TCP zerocopy and pinger support enabled by default (2476)
4846 - sync writes completed after process exits caused crashes (2319)
4847 - maintain correct mount count on the MDS (2356)
4848 - backout 1557, because 2316 wasn't really fixed
4849 - better file I/O statistics gathering in /proc
4850 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4851 - another kernel patch to fix zero-copy TCP function export
4852 - don't take duplicate lock when processing re-sent getattr (2420)
4853 - lctl uses obd_self_export instead of creating new conn (2353)
4854 - MDS/OST recovery case which requires object creation asserted (2425)
4855 - move lfs from /usr/sbin to /usr/bin in packages
4856 - fix race between mds_client_add and mds_client_free (2417)
4857 - use kmalloc instead of slabs in portals (2430)
4858 - don't create duplicate records when a failover MDS is present (2442)
4859 - remove unnecessary mount age check (2332)
4860 - don't remove directory inodes from locks prematurely (2451)
4861 - don't break if MDS service name is the same as hostname (2103)
4862 - fix races in client write RPC generation when cache full (2482)
4864 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4867 - remove now-unused request->rq_obd (278)
4868 - if an allocation fails, print out how much memory we've used (1933)
4869 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4870 - add the "configurable stack size" patch to most series files (1256)
4871 - ability to write large log records, for 100+ OST configs (2306)
4872 - fix NULL deref when filter_prep fails (2314)
4873 - fix operator precedence error in filter_sync
4874 - dynamic allocation of socknal TX descriptors (2315)
4875 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4876 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4877 - ensure that CWARN messages in Portals always get to the syslog
4878 - __init/__exit are not for prototype decls (ldlm_init/exit)
4879 - x86-64 compile warning fixes
4880 - fix gateway LMC keyword conflict (2318)
4881 - fix MDS lock inversions in getattr/reint paths (1844)
4882 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4883 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4884 - DLM race condition prevented some lock evictions (2328)
4885 - ENOMEM detection and retry on socknal sends (2230)
4886 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4887 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4888 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4889 - better error messages when a client is rejected during recovery (1505)
4890 - avoid cancelling locks which were never granted, after failure (2330)
4891 - fix i_sem/journal inversion in mds_client_add (2333)
4892 - fix truncate/getattr lock cycle deadlock (2334)
4893 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4894 - fix two rare exit paths which could leak an l_lock() ref (2321)
4895 - fencepost error in MDS/OST orphan recovery (2226)
4896 - make log record alignment 8 bytes (1988)
4897 - lstripe now fails when requested offset > ost_count (2237)
4898 - ensure that all kernel series have a complete list.h (1607)
4899 - fix crashes in special-file operations (2316)
4900 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4902 - allow configurable automake binary, for testing new versions
4903 - small update to the lfs documentation
4905 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4907 * fix negative export reference count in fsfilt_sync (2312)
4909 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4910 * release candidate 0.9.1
4912 - orphans are moved into the PENDING directory for possible recovery
4913 - replayed opens now open by fid for orphan/rename safety (1042)
4914 - last close of an orphan inode generates a transno (683)
4915 - chdir() and mount() now pin the directory entry (1020)
4916 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4917 - discard very old requests without processing them (1502)
4918 - remove some common, well-understood CERRORs (1505)
4919 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4920 - clear "grant" flags in OST replies until OST grant code lands (1644)
4921 - fix read performance by not clobbering i_blksize on client (1598)
4922 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4923 - make LRU size a /proc tunable, clears locks when reduced (707)
4924 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4925 - server locks take references on exports now (1558)
4926 - build fixes for 2.4.20-rh trees (1663)
4927 - return an error from lov_create if all OSCs are inactive (1751)
4928 - fix import levels when a reconnect happens without a timeout (1597)
4929 - exit early from mds_open if we get a lookup error (1749)
4930 - partial page read at EOF wouldn't wait for disk before sending (1642)
4931 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4932 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4933 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4934 - write/truncate lock inversion (1639)
4935 - Don't auto-load obdclass, portals modules during cleanup (1495)
4936 - fix timestamps from jumping to "now" (1763)
4937 - extra journal assertions (1648)
4938 - add an extra multiunlink test (1771)
4939 - fix read_record/write_record API (1776)
4940 - fix leak of offset_extent, possible incorrect i_size later (1772)
4941 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4942 - Debugging for the kqswnal_get_idle_tx problems (1820)
4943 - Allow recovery to be attempted multiple times (1536)
4944 - Write out MDS last_rcvd file after it is first created (1600)
4945 - Fix tx_descriptor leak in failed transmit situations (1827)
4946 - ext3 journaling fixes for assertion failure after IO error (1871)
4947 - class_export_put() on freed export after completion AST error (1896)
4948 - Fix revalidate looping in VFS (1322)
4949 - Don't access a freed export during MDS_REINT timeout (1521)
4950 - Add open-unlink recovery support on the MDS (1673,1764)
4951 - Return an error if no MDS data was read from last_rcvd (1946)
4952 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4953 - Don't setup a disk device that doesn't match exported UUID (317)
4954 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4955 - avoid committing NULL handle in force close
4956 - local.sh is now a one-stripe LOV configuration
4957 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4958 - fix ext3 htree duplicate directory entry corruption (1516)
4959 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4960 - update to new LOV EA format (2097)
4961 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4963 2003-06-15 Phil Schwan <phil@clusterfs.com>
4966 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4967 - per-import/export recovery handling (958, 931, 959)
4968 - multiple last-rcvd slots, for serving multiple FSes (949)
4969 - connections are again shared between multiple imp/exports (963, 964)
4970 - "umount -f" would hang if any requests needed to be sent (393, 978)
4971 - avoid pinning large req buffer by copying for queued messages (989)
4972 - add "uuid" to "lctl device" command to help upcalls (991)
4973 - "open" RPCs with transnos would confuse recovery counters (1037)
4974 - do proper endian conversion of all wire messages (288, 340, 891)
4975 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4976 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4977 - fix ptlrpc_connection leak in target_handle_connect (1174)
4978 - fix import refcounting bug in OST and MDS cleanup (1134)
4979 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4980 - fix very unlikely obd_types race condition (501)
4981 - remove osc_open hack for echo_client (1187)
4982 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4983 - a failure in read_inode2 leads to deadlock (1139)
4984 - cancel ack-locks as soon as transaction is committed (1072)
4985 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4986 - make sure to commitrw after any preprw to avoid deadlock (1162)
4987 - failing to execute a file in a lustre FS would lock inode (1203)
4988 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4989 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4990 - obd_uuid2tgt would walk off the end of the list (1255)
4991 - on IA64 the osc would give portals incorrect bulk size (1258)
4992 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4993 - fix lock inversion caused by new llite matching code (1282)
4994 - limit the number of dirty pages on a client to 10MB (1286)
4995 - timed out locks were not being corrected cancelled (1289)
4996 - fix O_DIRECT above 4GB on IA-32 (1292)
4997 * major user-visible changes
4998 - fail out/fail over policy now controlled by the upcall (993)
5000 - add OBD_PING to check server availability and failure (954)
5001 - lustre messages are now sent in sending host order (288, 340, 891)
5002 - add eadatalen to MDS getattr reply (340)
5003 - OST read replies may contain second buffer, with per-page status (593)
5005 2003-03-11 Phil Schwan <phil@clusterfs.com>
5008 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5009 - failed open()s could cause deadlock; fixed (867, 869)
5010 - stop cancelling OST locks when files are closed (481)
5011 - overlapping XID spaces caused network corruption (851, 853)
5012 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5013 - setattr_raw added, to do single-RPC, server-side setattrs
5014 - lmc/lconf syntax change for OST UUIDs
5015 - fix crashy race condition between ptlrpc_free_req and osc_close
5016 - don't use request in mdc_enqueue if we hit a timeout (889)
5017 - don't set the inode i_size for regular files from the MDS (896)
5018 - handle out of order completion AST (842)
5019 - don't LBUG if a lock request times out after receiving AST (913)
5020 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5021 - if a bad lock AST arrives, send an error instead of dropping entirely
5022 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5023 - fix leak in bulk IO when only partially completed (899, 900, 926)
5024 - fix O_DIRECT for ia64 (55)
5025 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5026 - C-z after timeout could hang a process forever; fixed (977)
5028 - client-side I/O cache (678, 924, 929, 941, 970)
5030 - READPAGE and SETATTRs which don't take server-side locks get
5033 2003-02-11 Phil Schwan <phil@clusterfs.com>
5036 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5037 - Fix fsfilt_extN_readpage() to read a full page of directory
5038 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5039 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5040 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5041 - Most intent operations were removed, in favour of a new RPC mode
5042 that does a single RPC to the server and bypasses most of the VFS
5043 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5044 - Aggressively cancel local locks on DLM servers
5045 - mds_reint_unlink sends EA to the client if it's the last nlink.
5046 client uses that EA to unlink OST objects.
5047 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5048 - recursive symlinks were fixed (439)
5049 - fixed NULL deref in DEBUG_REQ
5050 - filter_update_lastobjid no longer calls sync, which annoyed extN
5051 - fixed multi-client small-writes to a single file problem (445)
5052 - fixed mtime updates during file writes (607)
5053 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5054 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5055 - fixed Linux OST size reporting problem (444, 656)
5056 - OST now updates object mtime with writes or setattr (607, 619)
5057 - client verifies file size before zeroing page past EOF (445)
5058 - OST now writes last allocated objid to disk with allocation (108)
5059 - LOV on echo now works (409)
5061 - mds_reint_unlink sends a new buffer, with the EA included. this
5062 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5063 set if a regular file was being unlinked, and it was the last link
5064 - use PtlGet from the target for bulk writes (315)
5065 - OST now updates object mtime with writes or setattr (607, 619)
5066 - LDLM now has a grant-time callback to revalidate locked items, if
5068 - Many MDS operations were reorganized to combat race conditions
5070 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5071 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5073 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5076 - Fully reactivate OST imports after reconnection (512, others)
5077 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5078 - More graceful error handling for truncating on dead OST (515)
5079 - Don't error out unless we're actually accessing dead stripes (474)
5080 - Fix garbage sizes when stripes are missing (410)
5081 - LRU counters were broken, causing constant lock purge (433, 432)
5082 - garbage on read from stripes with failed OSTs (441)
5083 - mark OSCs as active before reconnecting during recovery (438)
5084 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5085 - lfind did not preserve OST order in output (443)
5086 - symlinks cause hung clients, incorrect data (439)
5087 - stop dereferencing request after dropping refcount (457)
5088 - don't LASSERT(spin_is_locked) on non-SMP (455)
5089 - fixes for many rename() bugs
5090 - fstat didn't correctly synchronize attributes (399)
5091 - server must handle lock cancellation during blocking AST prep (487)
5092 - bulk descriptors were free()d too soon (511)
5093 - fix paths in lconf, which would load incorrect modules (451, 507)
5094 - fix confusing lconf 'host not found' error message (386)
5095 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5096 - fix race condition in mdc_blocking_ast() for inode access (526)
5097 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5098 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5099 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5100 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5101 - fix lstripe in several ways (kernel side) (527)
5102 - fix request leak in ldlm_cli_enqueue (262)
5103 - incorrect OSC was marked inactive after OST failure
5104 - call mds_fs_cleanup before unmounting filesystem (524)
5105 - fix races between taking ns_lock and ldlm_lock_change_resource
5106 - fix races updating LOV export open file list
5107 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5108 - fix recovery NULL deref in ldlm_cli_cancel_unused
5109 - fix some DLM races by using new hash table for lock handles (419)
5110 - permit the client to specify desired inodes, at replay
5111 - duplicate requests when we queue them for replay reintegration
5112 - fix last_rcvd offset calculation
5113 - sync after each recovered transaction, so we always make progress
5114 - never, not always, ERESTART requests without transnos
5115 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5117 - skip replay if the MDS says that the client is already connected
5118 - don't check for a recovery-enabled export to match lctl's UUID
5119 - don't INC_USE_COUNT for phantom exports
5120 - don't crash when cleaning up phantom exports (567)
5121 - don't double-finish or set replay data for errored mdc_open requests
5122 - abort requests when they time out, so we don't get old replies
5123 - send/receive replies for AST messages again
5124 - if the client says that it doesn't have the lock, cancel it on the
5126 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5127 mark it as destroyed, it will all work out when decref is called
5128 - fix module use counts (22, 581)
5130 - ASTs now expect a reply (server cancels lock on error reply)
5132 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5135 - fix many simultaneous client startup (392)
5136 - fix dentry->d_it clobbering
5137 - credentials weren't being shipped for readdir/getattr operations
5138 - remove invalid assertions triggered during some concurrent MD
5140 - proper Lustre versions added (336, 389)
5141 - fix memory leak for create error case (398)
5142 - fix LOV locking bug that would get cli/srv out of sync
5143 - fix echo client over LOV (409)
5144 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5145 - fix double-O_EXCL intent crash (424)
5146 - avoid sending multiple lock CANCELs (352)
5148 - MDS can do multi-client recovery (modulo bugs in new code)
5150 - many updates, edits, cleanups
5152 2002-11-18 Phil Schwan <phil@clusterfs.com>
5155 - fix null d_it dereference (346)
5156 - fix full OST/dbench hang (333)
5157 - fix permission problem with file removal (286)
5158 - fix removal of OSCs from LOV when they fail
5159 - fix NULL deref during bulk timeout (214)
5160 - fix problems related to multiple filesystems on one MDS (241)
5161 - fixed serious subtle metadata locking bugs
5162 - free locks on clients when inodes are removed due to memory
5164 - fix inode pointer in lock data (285)
5165 - partial support for multiple MDS on a single host (241)
5166 - data locks weren't cancelled at clear_inode time (290, 311)
5167 - intent locks could lead to unbounded lock growth (205)
5168 - added a maximum lock count, an LRU list, and a flusher
5169 - fix multiple rename (365)
5170 - properly abstracted the echo client
5171 - OSC locked 1 byte too many; fixed
5172 - rewrote brw callback code:
5173 - fixed recovery bugs related to LOVs (306)
5174 - fixed too-many-pages-in-one-write crash (191)
5175 - fixed (again) crash in sync_io_timeout (214)
5176 - probably fixed callback-related race (385)
5178 - Add capability to MDS protocol
5179 - LDLM cancellations and callbacks on different portals
5181 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5184 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5185 - defer open object destruction to close time (601981, 138)
5186 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5187 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5188 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5189 - don't call obd_setattr in ll_file_release for destroyed objects
5190 * protocol change to lustre_msg: move |version| and add |flags|
5191 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5192 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5193 * added replay of create, unlink, link and rename operations during
5194 MDS failover; recovery should be much more robust now
5195 * remove failed OSCs from LOVs (only lov_create uses this so far)
5196 * the lustre-HOWTO was brought (more) up to date (582544)
5198 2002-10-23 Phil Schwan <phil@clusterfs.com>
5201 - in-use dentries weren't being reused properly (617851)
5202 - prevent multiple LDLM setup (599178)
5203 - fix LOV size calculations for truncate (617853)
5204 - fix client handling of MDS intent errors (POSIX)
5205 - fix permission bug in lovstripe.c test (624321)
5206 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5207 - truncate past end of file could corrupt data
5208 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5209 - a race in recovery could return ETIMEDOUT to apps (623947)
5210 - building outside the source directory was fixed
5211 * the lustre-HOWTO was brought (more) up to date (582544)
5212 * major progress was made on recovery functionality
5214 2002-10-10 Phil Schwan <phil@clusterfs.com>
5217 - recovery deadlock fix
5218 - rm -rf causes LBUG fix (617817)
5219 - file open by multiple tasks fix (618962)
5220 - directory permissions bugs (602707 and 620007)
5221 - journal_stop fixed with locking (611313)
5222 - O_APPEND failures resolved (618273, perhaps 614459)
5223 - lconf PATH fix (619770)
5224 - IA64 build fix (621450)
5225 - RPC buffer sizes scale with amount of memory
5227 2002-10-01 Phil Schwan <phil@clusterfs.com>
5230 - locks would be cancelled without throwing away data pages,
5231 resulting in inconsistent data (605627)
5232 - inode attributes were not always being refreshed (605627, 612449)
5233 - lconf now continues to cleanup after lctl reports an error
5234 - MDS now enforces user permissions (602707)
5235 - lprocfs cleanup fixed, but not yet enabled (614157)
5236 - fixed infinite server hang, should a client not respond to an AST
5237 - avoid going into recovery if user calls readlink() with a buffer
5238 that's too small (613941)
5239 - AST RPCs no longer require replies (614867) -- this may be changed
5240 - don't crash server if client sends an IOV that's too big (611336)
5241 - fixed lock conversion deadlock (611892)
5242 - fixed the following of symlinks (614622)
5243 * recovery: the server can remove locks from a client that dies, other
5244 clients can make progress
5245 * more extN patch fixes
5246 * compile-time configurable ptlrpc buffer allocations
5248 - collaborative read cache document
5249 - Lustre Lite Performance CDR document-in-progress
5251 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5254 - fix typo in patch-2.4.18
5256 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5259 - clear ptlrpc request each time in handle_incoming_request()
5260 - unlink of files now destroys the object on the OST
5262 2002-09-19 Peter Braam <braam@clusterfs.com>
5264 * add hard link support
5265 * change obdfile creation method
5266 * kernel patch changed
5268 2002-09-19 Peter Braam <braam@clusterfs.com>
5271 - stack overflow bug in extN fixed
5273 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5275 * documentation updates
5276 - add man pages for config tools
5277 - update tests/README to describe testing with new config tools
5278 - finish metadata API descriptions
5279 * bug fixes and cleanups
5280 - statfs workaround for 16TB limit
5281 - LOV stripe allocation improved, can stripe on subset of OSTs
5282 - LOV file size/IO offset was wrong for files > 4GB in size
5283 - object EA data was being dropped, caused files to be unreadable
5284 - memory overflow with non-LOV OST caused memory corruption
5285 - fixed regression tests to work with new config tools, obdfilter
5286 - fixed bug when directory size became larger than 1 block
5287 - fixed bug (for single client case) when PWD was deleted
5288 - invalidate local directory pages when doing intent-based ops
5289 - avoid LDLM oops when lock callback contained bad data
5291 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5293 * documentation updates
5294 * bug fixes and cleanups
5295 - configuration tools
5298 - 64-bit compile warnings
5299 - 64-bit internal statfs data
5301 * test_brw on persistent OST devices
5303 * lprocfs (disabled)
5305 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5307 * documentation updates
5308 * bug fixes and cleanups
5309 * configuration tools
5311 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5314 * many small fixes to 0_5_4
5315 * io/network handling
5316 * thinkos in MDS operations
5318 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5321 * crucial basic fixes to 0.5.3
5322 * IOR, Iozone work over Elan
5325 2002-08-07 Phil Schwan <phil@clusterfs.com>
5326 * version 0_5_3, our first alpha
5327 * we use the new Portals iovs
5328 * documentation updates
5329 * bug fixes and cleanups
5330 * small changes in the DLM wire protocol
5332 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5333 * version 0_5_1 with some initial stability,
5334 * locking on MD and file I/O.
5335 * documentation updates
5336 * several bug fixes since 0.5.0
5337 * small changes in wire protocol
5339 2002-07-18 Phil Schwan <phil@clusterfs.com>
5341 * delivered as Lustre Light Alpha
5342 * fixed a crash after handling invalid MDS requests
5343 * fixed directory pages for architectures with non-4k pages sizes
5345 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5346 * release version v0_4_4
5347 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5349 * Fixes a number of interruption problems with OST operations.
5350 * Update documentation for portals header changes
5351 * Move all wire protocol structs/defines to lustre_idl.h
5352 * Fixes symlink length bug.
5353 * Add tcpdump to repository.
5355 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5356 * release version v0_4_3
5357 * Fixes statfs for inodes on extN.
5358 * Fixes bug in runtests which would delete /etc/hosts.
5359 * Use 64-bit object IDs wherever possible (not into VFS though)
5360 Remove ost_get_info, which is unused by lustre, and out of date.
5362 2002-07-03 Peter Braam <braam@clusterfs.com>
5363 * release version v0_4_2 Fixes a lookup error (type not passed)
5364 * move forward to head of Portals
5365 * move forward to latest Lustre kernel
5367 2002-06-25 Peter Braam <braam@clusterfs.com>
5368 * release version v0_4_1. Hopefully stable on single node use.