1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.1.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
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: 1.40.4-cfs1
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).
16 Frequency : rare on shutdown ost
18 Description: If llog cancel was not send before clean_exports phase, this can
19 produce deadlock in llog code.
20 Details : If llog thread has last reference to obd and call class_import_put
21 this produce deadlock because llog_cleanup_commit_master wait when
22 last llog_commit_thread exited, but this never success because was
23 called from llog_commit_thread.
27 Description: allow userland application know is lost one of stripes.
28 Details : fill lvb_blocks with error code on ost and return it to
29 application if error flag found.
33 Description: NULL lov_tgts causing MDS oops
34 Details : more safe checks for NULL lov_tgts for avoid oops.
36 Severity : enhancement
38 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
40 Severity : enhancement
42 Description: Add rhel5 support to HEAD.
46 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
47 Details : If an OST is down the MDS will hang indefinitely in
48 obd_statfs_rqset() waiting for the statfs data. While for
49 MDS QOS usage of statfs, it should not stuck in waiting.
51 Severity : enhancement
53 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
55 Severity : enhancement
57 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
61 Description: make mgs_setparam() handle fsname containing dash
62 Details : fsname containing a dash does not work with lctl conf_param
64 Severity : enhancement
66 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
68 Severity : enhancement
70 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
74 Frequency : rare, at shutdown
75 Description: access already free / zero obd_namespace.
76 Details : if client_disconnect_export was called without force flag set,
77 and exist connect request in flight, this can produce access to
78 NULL pointer (or already free pointer) when connect_interpret
79 store ocd flags in obd_namespace.
83 Frequency : only at startup
84 Description: not alloc memory with spinlock held.
85 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
88 Severity : enhancement
90 Description: make lustre randomly fail allocating memory
91 Details : Make lustre randomly failed allocating memory for testing purpose.
93 Severity : enhancement
95 Description: lost problems with lov objid file
96 Details : Fixes some scability and access to not inited memory problems
97 in work with lov objdid file.
102 Description: lfs find does not continue on file error
103 Details : Continue other files processing when a file/dir is absent.
107 Description: Inconsistent usage of lustre_pack_reply()
108 Details : Standardize the usage of lustre_pack_reply() such that it
109 always generate a CERROR on failure.
112 Frequency : occasional
114 Description: MDS hang or stay in waiting lock
115 Details : If client receive lock with CBPENDING flag ldlm need send lock
116 cancel as separate rpc, to avoid situation when cancel request
117 can't processed due all i/o threads stay in wait lock.
121 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
125 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
129 Description: Update to SLES9 kernel-2.6.5-7.287.3.
133 Description: Update to RHEL4 latest kernel.
135 Severity : enhancement
137 Description: Build SLES10 patchless client fails
138 Details : The configure was broken by run ./configure with
139 --with-linux-obj=.... argument for patchless client. When the
140 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
141 can't search header adequately. Use absolute path such as
142 -I($LINUX)/include instead.
146 Description: interrupt oig_wait produce painc on resend.
147 Details : brw_redo_request can be used for resend requests from ptlrpcd and
148 private set, and this produce situation when rq_ptlrpcd_data not
149 copyed to new allocated request and triggered LBUG on assert
150 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
151 ptlrpcd set if request is changed and can be safety changed to use
156 Description: LASSERT_{REQ,REP}SWAB macros are buggy
157 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
158 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
159 these macros and replace them with their difinitions instead.
163 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
164 Details : Update which_patch & target file for SLES10 latest kernel.
167 Description: Remove client patches from SLES 10 kernel.
168 Details : This causes SLES 10 clients to behave as patchless clients
169 even on a Lustre-patched (server) kernel.
171 Severity : enhancement
173 Description: self-adjustable client's lru lists
174 Details : use adaptive algorithm for managing client cached locks lru
175 lists according to current server load, other client's work
176 pattern, memory activities, etc. Both, server and client
177 side namespaces provide number of proc tunables for controlling
180 Severity : enhancement
182 Description: light-weight GSS support
183 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
184 and reduce performance overhead.
186 Severity : enhancement
188 Description: Linux keyring support
189 Details : Support using service of Linux keyring for Lustre GSS internal
190 context refresh/cache mechanism.
194 Description: Fix errors in lfs documentation
195 Details : Fixes man pages
199 Description: don't use GFP_* in generic Lustre code.
200 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
204 Description: obdclass is limited by single OBD_ALLOC(idarray)
205 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
209 Description: warnings with build patchless client with vanila 2.6.19 and up
210 Details : change old ctl_table style and replace ctl_table/ctl_table_header
211 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
215 Frequency : Only for SLES
216 Description: sanity_quota fail test_1
217 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
218 file, which makes TSTID[2] inunique.
222 Frequency : Only for SLES9
223 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
224 Details : Update target/ChangeLog/which_patch .
228 Frequency : Only for test_10 in sanity-quota.sh
229 Description: a bug in quota test code
230 Details : checking a test flag in a wrong place
234 Frequency : Only for test_13 in sanity-quota.sh
235 Description: a bug in quota test script
236 Details : shouldn't check overfull allocated quota
240 Description: jbd statistics
241 Details : Port older jbd statistics patch for sles10
243 Severity : enhancement
245 Description: Add printing inode info into message about error in writepage.
250 Description: reply_lock_interpret crash due to race with it and lock cancel.
251 Details : Do not replay locks that are being cancelled. Do not reference
252 locks by their address during replay, just by their handle.
256 Frequency : When flocks are used.
257 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
258 Details : Flock locks might destroy just granted lock if it could be merged
259 with another existing flock, this is done in completion handler,
260 so teach ldlm_cli_enquque_fini that this is a valid case for
266 Description: Oops in read and write path when failing to allocate lock.
267 Details : Check if lock allocation failed and return error back.
271 Description: lstripe command fails for valid OST index
272 Details : The stripe offset is compared to lov->desc.ld_tgt_count
273 instead of lov->desc.ld_active_tgt_count.
277 Description: sanity.sh failed test 103
278 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
283 Description: df doesn't work properly if diskfs blocksize != 4K
284 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
288 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
289 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
293 Description: obd_health_check_timeout too short
294 Details : set obd_health_check_timeout as 1.5x of obd_timeout
298 Description: llapi_file_create() does not allow some changes
299 Details : add llapi_file_open() that allows specifying the mode and
300 open flags, and also returns an open file handle.
304 Description: merge and cleanup kernel patches.
305 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
309 Description: Add journal checksum support.(Kernel part)
310 Details : The journal checksum feature adds two new flags i.e
311 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
312 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
313 indicates that the commit block contains the checksum for
314 the blocks described by the descriptor blocks. Now commit
315 record can be sent to disk without waiting for descriptor
316 blocks to be written to disk. This behavior is controlled
317 using JBD2_FEATURE_ASYNC_COMMIT flag.
321 Description: OSS needs mutliple precreate threads
322 Details : Add ability to start more than one create thread per OSS.
326 Description: Sanity.sh test_65k failed: setstripe should have succeeded
327 Details : execute lfs setstripe on client
331 Description: mds_obd_create error creating tmp object
332 Details : When the user sets quota on root, llog will be affected and can't
333 create files and write files.
336 Frequency : Always on ia64 patchless client, and possibly others.
338 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
339 Details : This allows the patchless client to be loaded on architectures
344 Description: RedHat Update kernel for RHEL5
345 Details : Add the kernel config file for RHEL5.
349 Description: RedHat Update kernel for RHEL5
350 Details : Modify the kernel config file more closer RHEL5.
354 Description: Build failure against Centos5 (RHEL5)
355 Details : Define PAGE_SIZE when it isn't present.
359 Description: test_7 on sanity-quota.sh can't be used on separate machines
360 Details : add facet to handle it
364 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
365 Details : with NFS, the anon dentry's parent was set to itself in
366 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
367 corresponding dentry other than use rec->ur_name.
369 Severity : enhancement
371 Description: lfs setstripe enhancement
372 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
376 Description: enable data checksumming by default
377 Details : enable checksum by default, allow --disable-checksum
378 configure option and "-o nochecksum" mount option. Checksums
379 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
383 Description: lustre support for RHEL5
384 Details : Add support for RHEL5.
388 Description: Client eviction due to failover config
389 Details : after a connection loss, the lustre client should attempt to
390 reconnect to the last active server first before trying the
391 other potential connections.
395 Description: when mds and osts use different quota unit(32bit and 64bit),
396 quota will be released repeatly.
397 Details : void sending multiple quota reqs to mds, which will keep the
398 status between the reqs.
402 Description: osts not allocated evenly to files
403 Details : change the condition to increase offset_idx
406 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
408 Description: Data corruption for OSTs that are formatted larger than 2TB
410 Details : When generating the bio request for lustre file writes the
411 sector number would overflow a temporary variable before being
412 used for the IO. The data reads correctly from Lustre (which
413 will overflow in a similar manner) but other file data or
414 filesystem metadata may be corrupted in some cases.
418 Description: Tune the kernel for good SCSI performance.
419 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
420 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
425 Description: rewrite ext2-derived code in obdclass/uuid.c
426 Details : rewrite inherited code (uuid parsing code from ext2 utils)
427 from scratch preserving functionality.
431 Description: "lfs find -obd UUID" prints directories
432 Details : "lfs find -obd UUID" will return all directory names instead
433 of just file names. It is incorrect because the directories
434 do not reside on the OSTs.
438 Description: use i_size_read and i_size_write in 2.6 port
439 Details : replace inode->i_size access with i_size_read/write()
442 Frequency : when removing large files
444 Description: scheduling issue during removal of large Lustre files
445 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
446 It causes scheduling issues when removing large files (17TB in the
450 Frequency : only with liblustre clients on XT3
452 Description: evictions taking too long
453 Details : allow llrd to evict clients directly on OSTs
456 Frequency : only on ppc
458 Description: /proc/fs/lustre/devices broken on ppc
459 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
460 We should make sure the type of 'index' is loff_t and then cast
461 explicitly as needed below. Do not assign an explicitly cast
465 Frequency : only for rhel5
467 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
468 Details : Modify the target file & which_kernel.
470 Severity : enhancement
472 Description: omit set fsid for export NFS
473 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
474 as FS_REQUIRES_DEV for avoid problems with generate fsid.
478 Frequency : Always, for kernels after 2.6.16
479 Description: Fix warning idr_remove called for id=.. which is not allocated.
480 Details : Last kernels save old s_dev before kill super and not allow
481 to restore from callback - restore it before call kill_anon_super.
485 Description: buffer overruns could theoretically occur
486 Details : llapi_semantic_traverse() modifies the "path" argument by
487 appending values to the end of the origin string, and a
488 overrun may occur. Adding buffer overrun check in liblustreapi.
492 Description: Fix error on 'ls .' at the top of the Lustre mount.
493 Details : Don't revalidate dentry if it is a root dentry.
497 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
498 Details : Modify vm-tunables-rhel4.patch.
502 Description: Kernel config for 2.6.18-vanilla.
503 Details : Modify targets/2.6-vanilla.target.in.
504 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
505 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
506 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
507 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
511 Description: improve handling recoverable errors
512 Details : if request processig with error which can be recoverable on server
513 request should be resend, otherwise page released from cache and
518 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
519 Details : Modify target file & which_patch.
520 A flaw was found in the IA32 system call emulation provided
521 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
522 value could be stored in the %RAX register, which could trigger an
523 out-of-bounds system call table access. An untrusted local user
524 could exploit this flaw to run code in the kernel
525 (ie a root privilege escalation). (CVE-2007-4573).
529 Description: O_DIRECT bypasses client statistics.
530 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
531 read_ahead_stats were updated. copied the stats section from
532 osc_send_oap_rpc() into async_internal().
536 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
540 Description: change order of libsysio includes
541 Details : '#include sysio.h' should always come before '#include xtio.h'
545 Description: handle "serious error: objid * already exists" more gracefully
546 Details : If LAST_ID value on disk is smaller than the objects existing in
547 the O/0/d* directories, it indicates disk corruption and causes an
548 LBUG(). If the object is 0-length, then we should use the existing
549 object. This will help to avoid a full fsck in most cases.
551 Severity : enhancement
553 Description: adapt the lustre_config script to support the upgrade case
554 Details : Add "-u" option for lustre_config script to support upgrading 1.4
555 server targets to 1.6 in parallel.
559 Description: To avoid grant space > avaible space when the disk is almost
560 full. Without this patch you might see the error "grant XXXX >
561 available" or some LBUG about grant, when the disk is almost
563 Details : In filter_check_grant, for non_grant cache write, we should
564 check the left space by if (*left > ungranted + bytes), instead
565 of (*left > ungranted), because only we are sure the left space
566 is enough for another "bytes", then the ungrant space should be
567 increase. In client, we should update cl_avail_grant only there
568 is OBD_MD_FLGRANT in the reply.
572 Description: Update RHEL 4 kernel to fix local root privilege escalation.
573 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
574 described in CVE-2007-4573. This problem could allow untrusted
575 local users to gain root access.
578 Frequency : when using O_DIRECT and quotas
580 Description: Incorrect file ownership on O_DIRECT output files
581 Details : block usage reported by 'lfs quota' does not take into account
582 files that have been written with O_DIRECT.
587 Description: touch file failed when fs is not full
588 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
589 otherwise we can get ENOSP while fs is not full.
593 Description: parallel lock callbacks
594 Details : Instead of sending blocking and completion callbacks as separated
595 requests, adding them to a set and sending in parallel.
598 Frequency : only for Cray XT3
599 Bugzilla : 12829/13455
600 Description: Changing primary group doesn't change the group lustre assigns to
602 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
603 overridden with the primary group provided by the group upcall,
604 whereas the supplied fsgid can be trusted if it is in the list of
605 supplementary groups returned by the group upcall.
607 Severity : enhancement
609 Description: Allow masking D_WARNING, D_ERROR messages from console
610 Details : Console messages can now be disabled via lnet.printk.
614 Description: User code with malformed file open parameter crashes client node
615 Details : Before packing join_file req, all the related reference should be
616 checked carefully in case some malformed flags cause fake
617 join_file req on client.
621 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
622 Details : ldlm_completion_ast() assumes that a lock is granted when the req
623 mode is equal to the granted mode. However, it should also check
624 that LDLM_FL_CP_REQD is not set.
628 Description: Heavy nfs access might result in deadlocks
629 Details : After ELC code landed, it is now improper to enqueue any mds
630 locks under och_sem, because enqueue might want to decide to
631 cancel open locks for same inode we are holding och_sem for.
633 --------------------------------------------------------------------------------
635 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
637 * Support for kernels:
638 2.6.5-7.283 (SLES 9),
639 2.6.9-55.EL (RHEL 4),
640 2.6.16.46-0.14 (SLES 10),
641 2.6.18.8 vanilla (kernel.org)
642 * Client support for unpatched kernels:
643 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
644 2.6.16 - 2.6.22 vanilla (kernel.org)
645 * Due to recently discovered recovery problems, we do not recommend
646 using patchless RHEL 4 clients with this or any earlier release.
647 * Recommended e2fsprogs version: 1.39.cfs8
648 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
649 * Starting with this release, the ldiskfs backing filesystem required
650 by Lustre is now in its own package, lustre-ldiskfs. This package
651 should be installed. It is versioned separately from Lustre and
652 may be released separately in future.
656 Description: block reactivating mgc import until all deactivates complete
657 Details : Fix race when failing back MDT/MGS to itself (testing)
660 Frequency : at statup only
662 Description: mds_lov_synchronize race leads to various problems
663 Details : simultaneous MDT->OST connections at startup can cause the
664 sync to abort, leaving the OSC in a bad state.
666 Severity : enhancement
668 Description: add optional extra BUILD_VERSION info
669 Details : add a new environment variable (namely LUSTRE_VERS) which allows
670 to override the lustre version.
673 Frequency : 2.6.18 servers only
675 Description: ll_kern_mount() doesn't release the module reference
676 Details : The ldiskfs module reference count never drops down to 0
677 because ll_kern_mount() doesn't release the module reference.
682 Description: server LBUG when using old ost_num_threads parameter
683 Details : Accept the old ost_num_threads parameter but warn that it
684 is deprecated, and fix an off-by-one error that caused an LBUG.
689 Description: Transient SCSI error results in persistent IO issue
690 Details : iobuf->dr_error is not reinitialized to 0 between two
694 Frequency : sometimes when underlying device returns I/O errors
696 Description: OSTs not going read-only during write failures
697 Details : OSTs are not remounted read-only when the journal commit threads
698 get I/O errors because fsfilt_ext3 calls journal_start/stop()
699 instead of the ext3 wrappers.
703 Description: poor connect scaling with increasing client count
704 Details : Don't run filter_grant_sanity_check for more than 100 exports
705 to improve scaling for large numbers of clients.
708 Frequency : SLES10 only
710 Description: sanity-quota.sh quotacheck failed: rc = -22
711 Details : Quotas cannot be enabled on SLES10.
714 Frequency : liblustre clients only
716 Description: getdirentries does not give error when run on compute nodes
717 Details : getdirentries does not fail when the size specified as an argument
718 is too small to contain at least one entry
720 Severity : enhancement
722 Description: Add LNET router traceability for debug purposes
723 Details : If a checksum failure occurs with a router as part of the
724 IO path, the NID of the last router that forwarded the bulk data
725 is printed so it can be identified.
730 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
731 Details : Due to a race condition, liblustre clients were occasionally
734 Severity : enhancement
736 Description: lfs setstripe use optional parameters instead of postional
739 Severity : enhancement
741 Description: Nanosecond timestamp support for ldiskfs
742 Details : The on-disk ldiskfs filesystem has added support for nanosecond
743 resolution timestamps. There is not yet support for this at
744 the Lustre filesystem level.
747 Frequency : during server recovery
749 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
750 Details : request with rq_no_resend flag not awake l_wait_event if they get a
754 Frequency : nfs export on patchless client
756 Description: connectathon hang when test nfs export over patchless client
757 Details : Disconnected dentry cannot be found with lookup, so we do not need
758 to unhash it or make it invalid
761 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
762 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
763 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
768 Description: open req refcounting wrong on reconnect
769 Details : If reconnect happened between getting open reply from server and
770 call to mdc_set_replay_data in ll_file_open, we will schedule
771 replay for unreferenced request that we are about to free.
772 Subsequent close will crash in variety of ways.
773 Check that request is still eligible for replay in
774 mdc_set_replay_data().
779 Description: disable writes to filesystem when reading health_check file
780 Details : the default for reading the health_check proc file has changed
781 to NOT do a journal transaction and write to disk, because this
782 can cause reads of the /proc file to hang and block HA state
783 checking on a healthy but otherwise heavily loaded system. It
784 is possible to return to the previous behaviour during configure
785 with --enable-health-write.
787 Severity : enhancement
789 Description: 64-bit inode version
790 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
791 to the inode. This will be required for version-based recovery.
796 Description: MDS fails to start if a duplicate client export is detected
797 Details : in some rare cases it was possible for a client to connect to
798 an MDS multiple times. Upon recovery the MDS would detect this
799 and fail during startup. Handle this more gracefully.
801 Severity : enhancement
803 Description: Add -o localflock option to simulate old noflock
805 Details : This will achieve local-only flock/fcntl locks
811 Description: log_commit_thread vs filter_destroy race leads to crash
812 Details : Take import reference before releasing llog record semaphore
817 Description: Wrong request locking in request set processing
818 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
819 imp_delayed_list, in this place should be used imp_lock.
822 Frequency : when reconnection
824 Description: Grant Leak when osc reconnect to OST
825 Details : When osc reconnect ost, OST(filter) should check whether it
826 should grant more space to client by comparing fed_grant and
827 cl_avail_grant, and return the granted space to client instead
828 of "new granted" space, because client will call osc_init_grant
829 to update the client grant space info.
832 Frequency : when client reconnect to OST
834 Description: Grant Leak when osc do resend and replay bulk write
835 Details : When osc reconnect to OST, OST(filter)should clear grant info of
836 bulk write request, because the grant info while be sync between
837 OSC and OST when reconnect, and we should ignore the grant info
838 these of resend/replay write req.
843 Description: Grant space more than avaiable left space sometimes.
844 Details : When then OST is about to be full, if two bulk writing from
845 different clients came to OST. Accord the avaliable space of the
846 OST, the first req should be permitted, and the second one
847 should be denied by ENOSPC. But if the seconde arrived before
848 the first one is commited. The OST might wrongly permit second
849 writing, which will cause grant space > avaiable space.
852 Frequency : when client is evicted
854 Description: Grant might be wrongly erased when osc is evicted by OST
855 Details : when the import is evicted by server, it will fork another
856 thread ptlrpc_invalidate_import_thread to invalidate the
857 import, where the grant will be set to 0. While the original
858 thread will update the grant it got when connecting. So if
859 the former happened latter, the grant will be wrongly errased
860 because of this race.
865 Description: Checking Stale with correct fid
866 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
867 to check whether it is stale, because sometimes, we want the
868 enqueue happened anyway, and op_data.fid2 will not be initialized.
870 Severity : enhancement
872 Description: update patchless client
873 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
876 Frequency : only with 2.4 kernel
878 Description: random memory corruption
879 Details : size of struct ll_inode_info is to big for union inode.u and this
880 can be cause of random memory corruption.
885 Description: Memory leak in recovery
886 Details : Lov_mds_md was not free in an error handler in mds_create_object.
887 It should also check obd_fail before fsfilt_start, otherwise if
888 fsfilt_start return -EROFS,(failover mds during mds recovery).
889 then the req will return with repmsg->transno = 0 and rc = EROFS.
890 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
891 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
892 be freed no matter whether fsfilt_commit success or not.
895 Frequency : only with huge count clients
897 Description: Prevents from taking the superblock lock in llap_from_page for
899 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
900 for llap_from_page call in ll_removepage prevents from taking the
901 superblock lock for a soon died page.
906 Description: Not check open intent error before release open handle
907 Details : in some rare cases, the open intent error is not checked before
908 release open handle, which may cause
909 ASSERTION(open_req->rq_transno != 0), because it tries to release
910 the failed open handle.
915 Description: Set cat log bitmap only after create log success.
916 Details : in some rare cases, the cat log bitmap is set too early. and it
917 should be set only after create log success.
921 Description: Accessing a block bevice can re-enable I/O when Lustre is
922 tearing down a device.
923 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
928 Description: service threads may hog cpus when there are a lot of requests
929 Details : Insert cond_resched to give other threads a chance to use some CPU
934 Description: the cat log was not initialized in recovery
935 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
936 unlink log on mds will not be initialized until mds post
937 recovery. And also in mds post recovery, the unlink log will
938 initialization will be done asynchronausly, so there will be race
939 between add unlink log and unlink log initialization.
943 Description: brw_stats were being printed incorrectly
944 Details : brw_stats were being printed as log2 but all of them were not
945 recorded as log2. Also remove some code duplication arising from
946 filter_tally_{read,write}.
950 Frequency : rare, only in recovery.
951 Description: ASSERTION(req->rq_type != LI_POISON) failed
952 Details : imp_lock should be held while iterating over imp_sending_list for
953 prevent destroy request after get timeout in ptlrpc_queue_wait.
957 Description: replay-single.sh test 52 fails
958 Details : A lock's skiplist need to be cleanup when it being unlinked
959 from its resource list.
963 Description: Short directio read returns full requested size rather than
965 Details : Direct I/O operations should return actual amount of bytes
966 transferred rather than requested size.
968 Severity : enhancement
970 Description: metadata RPC reduction (e.g. for rm performance)
971 Details : decrease the amount of synchronous RPC between clients and servers
972 by canceling conflicing lock before the operation on the client
973 and packing thier handles into the main operation RPC to server.
975 Severity : enhancement
977 Description: Async OSC create to avoid the blocking unnecessarily.
978 Details : If a OST has no remain object, system will block on the creating
979 when need to create a new object on this OST. Now, ways use
980 pre-created objects when available, instead of blocking on an
981 empty osc while others are not empty. If we must block, we block
982 for the shortest possible period of time.
984 Severity : enhancement
986 Description: refine locking for avoid write wrong info into lov_objid file
987 Details : fix possible races with add new target and write/update data in
990 --------------------------------------------------------------------------------
992 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
997 Frequency : on some architectures
999 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1000 Details : Uninitialized flags sometimes cause configuration commands to
1004 Frequency : patchless clients only
1006 Description: missing __iget() symbol export
1007 Details : The __iget() symbol export is missing. To avoid the need for
1008 this on patchless clients the deathrow inode reaper is turned
1009 off, and we depend on the VM to clean up old inodes. This
1010 dependency was during via the fix for bug 12181.
1012 --------------------------------------------------------------------------------
1014 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1016 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1017 INTEROPERATE with older versions automatically. In many cases a
1018 special upgrade step is needed. Please read the
1019 user documentation before upgrading any part of a live system.
1020 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1021 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1022 * WARNING: Lustre configuration and startup changes are required with
1023 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1029 Severity : enhancement
1031 Description: MountConf
1032 Details : Lustre configuration is now managed via mkfs and mount
1033 commands instead of lmc and lconf. New obd types (MGS, MGC)
1034 are added for dynamic configuration management. See
1035 https://mail.clusterfs.com/wikis/lustre/MountConf for
1038 Severity : enhancement
1040 Description: dynamic OST addition
1041 Details : OSTs can now be added to a live filesystem
1043 Severity : enhancement
1045 Description: startup order invariance
1046 Details : MDTs and OSTs can be started in any order. Clients only
1047 require the MDT to complete startup.
1049 Severity : enhancement
1051 Description: parallel, asynchronous orphan cleanup
1052 Details : orphan cleanup is now performed in separate threads for each
1053 OST, allowing parallel non-blocking operation.
1055 Severity : enhancement
1057 Description: optimized stripe assignment
1058 Details : stripe assignments are now made based on ost space available,
1059 ost previous usage, and OSS previous usage, in order to try
1060 to optimize storage space and networking resources.
1062 Severity : enhancement
1064 Description: Permanently set tunables
1065 Details : All writable /proc/fs/lustre tunables can now be permanently
1066 set on a per-server basis, at mkfs time or on a live system.
1068 Severity : enhancement
1070 Description: Lustre message v2
1071 Details : Add lustre message format v2.
1073 Severity : enhancement
1075 Description: client OST exclusion list
1076 Details : Clients can be started with a list of OSTs that should be
1077 declared "inactive" for known non-responsive OSTs.
1081 Description: SPEC SFS validation failure on NFS v2 over lustre.
1082 Details : Changes the blocksize for regular files to be 2x RPC size,
1083 and not depend on stripe size.
1085 Severity : enhancement
1087 Description: Multiple MD RPCs in flight.
1088 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1089 To avoid overly-overloading MDS, introduce a limit on number of
1090 MDS RPCs in flight for a single client and add /proc controls
1091 to adjust this limit.
1093 Severity : enhancement
1095 Description: client read/write statistics
1096 Details : Add client read/write call usage stats for performance
1097 analysis of user processes.
1098 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1099 file access. extents_stats shows chunk size distribution.
1100 extents_stats_per_process show chunk size distribution per
1103 Severity : enhancement
1105 Description: mds statistics
1106 Details : Add detailed mds operations statistics in
1107 /proc/fs/lustre/mds/*/stats.
1111 Description: Failure of copying files with lustre special EAs.
1112 Details : Client side always return success for setxattr call for lustre
1113 special xattr (currently only "trusted.lov").
1118 Description: Refcount LNET uuids
1119 Details : The global LNET uuid list grew linearly with every startup;
1120 refcount repeated list entries instead of always adding to
1123 Severity : enhancement
1125 Description: Dynamic service threads
1126 Details : Within a small range, start extra service threads
1127 automatically when the request queue builds up.
1130 Frequency : mixed-endian client/server environments
1132 Description: mixed-endian crashes
1133 Details : The new msg_v2 system had some failures in mixed-endian
1136 Severity : enhancement
1138 Description: Easy OST removal
1139 Details : OSTs can be permanently deactivated with e.g. 'lctl
1140 conf_param lustre-OST0001.osc.active=0'
1142 Severity : enhancement
1144 Description: MGS proc entries
1145 Details : Added basic proc entries for the MGS showing what filesystems
1148 Severity : enhancement
1150 Description: provide MGS failover
1151 Details : Added config lock reacquisition after MGS server failover.
1153 Severity : enhancement
1155 Description: add Linux 2.4 support
1156 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1160 Description: a large application tries to do I/O to the same resource and dies
1161 in the middle of it.
1162 Details : Check the req->rq_arrival time after the call to
1163 ost_brw_lock_get(), but before we do anything about
1164 processing it & sending the BULK transfer request. This
1165 should help move old stale pending locks off the queue as
1166 quickly as obd_timeout.
1169 Frequency : when an incorrect nid is specified during startup
1171 Description: ptlrpc connect to non-existant node causes kernel crash
1172 Details : LNET can't be re-entered from an event callback, which
1173 happened when we expire a message after the export has been
1174 cleaned up. Instead, hand the zombie cleanup off to another
1177 Severity : enhancement
1179 Description: plain/inodebits lock performance improvement
1180 Details : Grouping plain/inodebits in granted list by their request modes
1181 and bits policy, thus improving the performance of search through
1185 Frequency : only if OST filesystem is corrupted
1187 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1188 Details : for a short time RPCs with bulk IO are in the replay list,
1189 but replay of bulk IOs is unimplemented. If the OST filesystem
1190 is corrupted due to disk cache incoherency and then replay is
1191 started it is possible to trip an assertion. Avoid putting
1192 committed RPCs into the replay list at all to avoid this issue.
1195 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1197 Description: System hang on startup
1198 Details : This bug allowed the liblustre (e.g. catamount) client to
1199 return to the app before handling all startup RPCs. This
1200 could leave the node unresponsive to lustre network traffic
1201 and manifested as a server ptllnd timeout.
1203 Severity : enhancement
1205 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1206 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1207 boolean to control whether to print peer debug info when a
1208 client's RPC times out.
1211 Frequency : only for kernels with patches from Lustre below 1.4.3
1213 Description: Remove old rdonly API
1214 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1217 Frequency : only for devices with external journals
1219 Description: Set external device read-only also
1220 Details : During a commanded failover stop, we set the disk device
1221 read-only while the server shuts down. We now also set any
1222 external journal device read-only at the same time.
1225 Frequency : when upgrading from 1.4 while trying to change parameters
1227 Description: The wrong (new) MDC name was used when setting parameters for
1228 upgraded MDT's. Also allows changing of OSC (and MDC)
1229 parameters if --writeconf is specified at tunefs upgrade time.
1232 Frequency : when setting specific ost indicies
1234 Description: QOS code breaks on skipped indicies
1235 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1236 created with --index need not be sequential.
1238 Severity : enhancement
1240 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1241 Details : The uninit_groups feature works in conjunction with the kernel
1242 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1243 up the pass1 processing of e2fsck. This is a read-only feature
1244 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1245 mount filesystems that have had this feature enabled.
1247 Severity : enhancement
1249 Description: Improve multi-block allocation algorithm to avoid fragmentation
1250 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1251 allocation locality and avoid filesystem fragmentation.
1253 ------------------------------------------------------------------------------
1255 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1257 * Support for kernels:
1258 2.4.21-47.0.1.EL (RHEL 3)
1259 2.6.5-7.283 (SLES 9)
1260 2.6.9-42.0.10.EL (RHEL 4)
1261 2.6.12.6 vanilla (kernel.org)
1262 2.6.16.27-0.9 (SLES 10)
1263 * Recommended e2fsprogs version: 1.39.cfs5
1265 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1269 Frequency : occasional, depends on client load and configuration
1270 Bugzilla : 12181, 12203
1271 Description: data loss for recently-modified files
1273 Details : In some cases it is possible that recently written or created
1274 files may not be written to disk in a timely manner (this should
1275 normally be within 30s unless client IO load is very high).
1276 The problem appears as zero-length files or files that are a
1277 multiple of 1MB in size after a client crash or client eviction
1278 that are missing data at the end of the file.
1280 This problem is more likely to be hit on clients where files are
1281 repeatedly created and unlinked in the same directory, clients
1282 have a large amount of RAM, have many CPUs, the filesystem has
1283 many OSTs, the clients are rebooted frequently, and/or the files
1284 are not accessed by other nodes after being written.
1286 The presence of the problem can be detected by looking at
1287 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1288 smaller than the second (nr_unused) then dirty files will not
1289 be flushed automatically to disk. "sync; sleep 10" should be
1290 run several times on the node before unmounting it to update
1291 Lustre (this is also safe to run on nodes without this problem).
1293 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1294 that can cause this same problem, so customers using that kernel
1295 also need to update the kernel in addition to Lustre. In order
1296 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1298 It is normal that files written just before a client crash (less
1299 than 30s) may not yet have been flushed to disk, even for local
1303 Frequency : frequent on thin XT3 nodes
1305 Description: UUID collision on thin XT3 Linux nodes
1306 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1307 randomly, since we relied on an insufficiently-seeded PRNG.
1312 Description: OSS hangs after "All ost request buffers busy"
1313 Details : A deadlock between quota and journal operations caused OSS
1314 hangs after printing "All ost request buffers busy."
1317 Frequency : always on liblustre builds
1319 Description: Cleanup compiler warnings on liblustre
1322 Frequency : always on liblustre builds on XT3
1324 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1328 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1329 > 32000 subdirectories
1330 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1331 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1332 subdirectory count crosses 32000. This will aid e2fsck to
1333 correctly handle more than 32000 subdirectories.
1336 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1338 Description: System hang on startup
1339 Details : This bug allowed the liblustre (e.g. catamount) client to
1340 return to the app before handling all startup RPCs. This
1341 could leave the node unresponsive to lustre network traffic
1342 and manifested as a server ptllnd timeout.
1344 Severity : enhancement
1346 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1347 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1348 boolean to control whether to print peer debug info when a
1349 client's RPC times out.
1354 Description: make O_SYNC working on 2.6 kernels
1355 Details : 2.6 kernels use different method for mark pages for write,
1356 so need add a code to lustre for O_SYNC work.
1361 Description: Failure to close file and release space on NFS
1362 Details : Put inode details into lock acquired in ll_intent_file_open.
1363 Use mdc_intent_lock in ll_intent_open to properly
1364 detect all kind of errors unhandled by mdc_enqueue.
1369 Description: proc file read during shutdown sometimes raced obd removal,
1371 Details : Add lock to prevent obd access after proc file removal.
1374 Frequency : Only for files larger than 4GB on 32-bit clients.
1376 Description: improperly doing page alignment of locks
1377 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1378 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1383 Description: RPCs being resent when they shouldn't be
1384 Details : Some RPCs that should not be resent are being resent. This
1385 can cause inconsistencies in the RPC state machine. Do not
1386 resend such requests.
1389 Frequency : rare, only with NFS export
1391 Description: Crash on NFS re-export node
1392 Details : under very unusual load conditions an assertion is hit in
1393 ll_intent_file_open()
1396 Frequency : only if OST filesystem is corrupted
1398 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1399 Details : for a short time RPCs with bulk IO are in the replay list,
1400 but replay of bulk IOs is unimplemented. If the OST filesystem
1401 is corrupted due to disk cache incoherency and then replay is
1402 started it is possible to trip an assertion. Avoid putting
1403 committed RPCs into the replay list at all to avoid this issue.
1408 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1409 Details : Large single O_DIRECT read and write calls can fail to allocate
1410 a sufficiently large buffer to process the request. In case of
1411 allocation failure the allocation is retried with a smaller
1412 buffer and broken into smaller requests.
1414 Severity : enhancement
1416 Description: Add -o localflock option to simulate old noflock behaviour.
1417 Details : This will achieve local-only flock/fcntl locks coherentness.
1422 Description: versioning check is incomplete
1423 Details : Checking the version difference of client vs. server, report
1424 error if the gap is too big.
1429 Description: add support PG_writeback bit
1430 Details : add support for PG_writeback bit for Lustre, for more carefull
1431 work with page cache in 2.6 kernel. This also fix some deadlocks
1432 and remove hack for work O_SYNC with 2.6 kernel.
1434 Severity : enhancement
1436 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1437 Details : The uninit_groups feature works in conjunction with the kernel
1438 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1439 up the pass1 processing of e2fsck. This is a read-only feature
1440 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1441 mount filesystems that have had this feature enabled.
1443 Severity : enhancement
1445 Description: Improve multi-block allocation algorithm to avoid fragmentation
1446 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1447 allocation locality and avoid filesystem fragmentation.
1449 ------------------------------------------------------------------------------
1451 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1453 * Support for kernels:
1454 2.6.9-42.0.3.EL (RHEL 4)
1455 2.6.5-7.276 (SLES 9)
1456 2.4.21-47.0.1.EL (RHEL 3)
1457 2.6.12.6 vanilla (kernel.org)
1458 2.6.16.21-0.8 (SLES10)
1459 * Recommended e2fsprogs version: 1.39.cfs2-0
1461 * The backwards-compatible /proc/sys/portals symlink has been removed
1462 in this release. Before upgrading, please ensure that you change
1463 any configuration scripts or /etc/sysctl.conf files that access
1464 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1465 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1466 in advance of the upgrade on any system running Lustre 1.4.6 or
1467 newer, since /proc/sys/lnet was added in that version.
1468 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1472 Frequency : only when quota is used
1474 Description: avoid scanning export list for quota master
1475 Details : Change the algorithms to avoid scanning export list in order
1476 to improve the efficiency.
1479 Frequency : MDS failover only, very rarely
1481 Description: "went back in time" messages on mds failover
1482 Details : The greatest transno may be lost when the current operation
1483 finishes with an error (transno==0) and the client's last_rcvd
1484 record is over-written. Save the greatest transno in the
1485 mds_last_transno for this case.
1488 Frequency : always for specific kernels and striping counts
1490 Description: client may get "Matching packet too big" without ACL support
1491 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1492 when trying to access files in certain configurations. The
1493 clients should in fact be denied when mounting because they do
1494 not understand ACLs.
1497 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1499 Description: many clients connecting with IO in progress causes connect timeouts
1500 Details : Avoid synchronous journal commits to avoid delays caused by many
1501 clients connecting/disconnecting when bulk IO is in progress.
1502 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1503 OST_IO_PORTAL to avoid delays behind potentially many pending
1507 Frequency : occasionally with multiple writers to a single file
1509 Description: shared writes to file may result in wrong size reported by stat()
1510 Details : Allow growing of kms when extent lock is cancelled
1513 Frequency : always with random mmap IO to multi-striped file
1515 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1516 Details : Only if the hole is at the end of OST object so that kms is too
1517 small. Fix is to increase kms accordingly in ll_nopage.
1520 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1522 Description: writes to a missing object would leak memory on the OST
1523 Details : If there is an inconsistency between the MDS and OST filesystems,
1524 such that the MDS references an object that doesn't exist, writes
1525 to that object will leak memory due to incorrect cleanup in the
1526 error handling path, eventually running out of memory on the OST.
1531 Description: Creating too long symlink causes lustre errors
1532 Details : Check symlink and name lengths before sending requests to MDS.
1535 Frequency : only if flock is enabled (not on by default)
1537 Description: posix locks not released on fd closure on 2.6.9+
1538 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1539 kernels, this caused such locks not to be released on fd close and
1540 then assertions on fs unmount about still used locks.
1543 Frequency : MDS failover only, very rarely
1545 Description: clients may get ASSERTION(granted_lock != NULL)
1546 Details : When request was taking a long time, and a client was resending
1547 a getattr by name lock request. The were multiple lock requests
1548 with the same client lock handle and
1549 mds_getattr_name->fixup_handle_for_resent_request found one of the
1550 lock handles but later failed with ASSERTION(granted_lock != NULL).
1555 Description: handle->h_buffer_credits > 0, assertion failure
1556 Details : h_buffer_credits is zero after truncate, causing assertion
1557 failure. This patch extends the transaction or creates a new
1561 Frequency : NFS re-export or patchless client
1562 Bugzilla : 11179, 10796
1563 Description: Crash on NFS re-export node (__d_move)
1564 Details : We do not want to hash the dentry if we don't have a lock.
1565 But if this dentry is later used in d_move, we'd hit uninitialised
1566 list head d_hash, so we just do this to init d_hash field but
1567 leave dentry unhashed.
1570 Frequency : NFS re-export or patchless client
1572 Description: NFS exports has problem with symbolic link
1573 Details : lustre client didn't properly install dentry when re-exported
1574 to NFS or running patchless client.
1577 Frequency : NFS re-export or patchless client
1579 Description: Various nfs/patchless fixes.
1580 Details : fixes reuse disconected alias for lookup process - this fixes
1581 warning "find_exported_dentry: npd != pd",
1582 fix permission error with open files at nfs.
1583 fix apply umask when do revalidate.
1586 Frequency : occasional
1588 Description: Crash on NFS re-export node
1589 Details : calling clear_page() on the wrong pointer triggered oops in
1590 generic_mapping_read().
1593 Frequency : rarely, using O_DIRECT IO
1595 Description: unaligned directio crashes client with LASSERT
1596 Details : check for unaligned buffers before trying any requests.
1599 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1601 Description: stale data returned from RAID cache
1602 Details : If only a small amount of IO is done to the RAID device before
1603 reading it again it is possible to get stale data from the RAID
1604 cache instead of reading it from disk.
1607 Frequency : always for sles10 kernel
1609 Description: sles10 support
1610 Details : ll_follow_link: compile fixes and using of nd_set_link
1611 under newer kernels.
1614 Frequency : depends on arch, kernel and compiler version, always on sles10
1617 Description: recursive or deep enough symlinks cause stack overflow
1618 Details : getting rid of large stack-allocated variable in
1622 Frequency : depends on hardware
1624 Description: lustre write performance loss in the SLES10 kernel
1625 Details : the performance loss is caused by using of write barriers in the
1626 ext3 code. The SLES10 kernel turns barrier support on by
1627 default. The fix is to undo that change for ldiskfs.
1629 ------------------------------------------------------------------------------
1631 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1633 * Support for kernels:
1634 2.6.9-42.0.3EL (RHEL 4)
1635 2.6.5-7.276 (SLES 9)
1636 2.4.21-47.0.1.EL (RHEL 3)
1637 2.6.12.6 vanilla (kernel.org)
1641 Frequency : quota enabled and large files being deleted
1643 Description: releasing more than 4GB of quota at once hangs OST
1644 Details : If a user deletes more than 4GB of files on a single OST it
1645 will cause the OST to spin in an infinite loop. Release
1646 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1651 Description: statfs data retrieved from /proc may be stale or zero
1652 Details : When reading per-device statfs data from /proc, in the
1653 {kbytes,files}_{total,free,avail} files, it may appear
1654 as zero or be out of date.
1657 Frequency : systems with MD RAID1 external journal devices
1659 Description: lconf's call to blkid is confused by RAID1 journal devices
1660 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1661 of returning all block devices that match the journal UUID.
1664 Frequency : always, for aggregate stripe size over 4GB
1666 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1667 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1668 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1671 Frequency : always if "lfs find" used on a local file/directory
1673 Description: "lfs find" segfaults if used on a local file/directory
1674 Details : The case where a directory component was not specified wasn't
1675 handled correctly. Handle this properly.
1678 Frequency : always on ppc64
1680 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1681 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1682 perform endianness conversion.
1685 Frequency : rarely (truncate to non-zero file size after write under load)
1686 Bugzilla : 10730, 10687
1687 Description: Files padded with zeros to next 4K multiple
1688 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1689 it is possible that files that are truncated to a non-zero size
1690 immediately after being written are filled with zero bytes beyond
1691 the truncated size. No file data is lost.
1693 Severity : enhancement
1695 Description: Allow recovery/failover for liblustre clients.
1696 Details : liblustre clients were unaware of failover configurations until
1699 Severity : enhancement
1701 Description: user file locks should fail when not mounting with flock option
1702 Details : Set up an error-returning stub in ll_file_operations.lock field
1703 to prevent incorrect behaviour when client is mounted without
1704 flock option. Also, set up properly f_op->flock field for
1708 Frequency : always on ia64
1710 Description: "lfs df" loops on printing out MDS statfs information
1711 Details : The obd_ioctl_data was not initialized and in some systems
1712 this caused a failure during the ioctl that did not return
1713 an error. Initialize the struct and return an error on failure.
1716 Frequency : SLES 9 only
1718 Description: Error of copying files with lustre special EAs as root
1719 Details : Client side always return success for setxattr call for lustre
1720 special xattr (currently only "trusted.lov").
1723 Frequency : rarely on clusters with both ia64+i386 clients
1725 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1726 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1727 clients are doing concurrent writes to the same file it is
1728 possible that the ia64 clients may LASSERT because the OST
1729 extent locks are not PAGE_SIZE aligned. Ensure that grown
1730 locks are always aligned on the request boundary.
1733 Frequency : specific use, occasional
1735 Description: Overwriting in use executable truncates on-disk binary image
1736 Details : If one node attempts to overwrite an executable in use by
1737 another node, we now correctly return ETXTBSY instead of
1738 truncating the file.
1740 Severity : enhancement
1742 Description: Async OSC create to avoid the blocking unnecessarily.
1743 Details : If a OST has no remain object, system will block on the creating
1744 when need to create a new object on this OST. Now, ways use
1745 pre-created objects when available, instead of blocking on an
1746 empty osc while others are not empty. If we must block, we block
1747 for the shortest possible period of time.
1752 Description: chmod on Lustre root is propagated to other clients
1753 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1754 invalid by the follow_mount time.
1757 Frequency : liblustre clients only
1759 Description: Race in 'instant cancel' lock handling could lead to such locks
1760 never to be granted in case of SMP MDS
1761 Details : Do not destroy not yet granted but cbpending locks in
1765 Frequency : replay/resend of open
1767 Description: non null lock assetion failure in mds_intent_policy
1768 Details : Trying to replay/resend lockless open requests resulted in
1769 mds_open() returning 0 with no lock. Now it sets a flag if
1770 a lock is going to be returned.
1772 Severity : enhancement
1774 Description: Checksum enhancements
1775 Details : New checksum enhancements allow for resending RPCs that failed
1778 Severity : enhancement
1780 Description: Tunables on number of dirty pages in cacche
1781 Details : Allow to set limit on number of dirty pages cached.
1786 Description: client crash on unmount - lock still has references
1787 Details : In some error handling cases it was possible to leak a lock
1788 reference on a client while accessing a file. This was not
1789 harmful to the client during operation, but would cause the
1790 client to crash when the filesystem is unmounted.
1793 Frequency : specific case, rare
1795 Description: ETXTBSY on mds though file not in use by client
1796 Details : ETXTBSY is no longer incorrectly returned when attempting to
1797 chmod or chown a directory that the user previously tried to
1798 execute or a currently-executing binary.
1801 Frequency : extremely rare except on liblustre-based clients
1803 Description: Lustre space not freed when files are deleted
1804 Details : Clean up open-unlinked files after client eviction. Previously
1805 the unlink was skipped and the files remained as orphans.
1810 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1811 Details : In some cases it was possible to send a blocking callback to
1812 a client doing a glimpse, even though that client didn't get
1813 a lock granted. When the glimpse lock is cancelled on the OST
1814 the freed lock is left on the waiting list and corrupted the list.
1817 Frequency : all core dumps
1819 Description: Broke core dumps to lustre
1820 Details : Negative dentry may be unhashed if parent does not have UPDATE
1821 lock, but some callers, e.g. do_coredump, expect dentry to be
1822 hashed after successful create, hash it in ll_create_it.
1824 ------------------------------------------------------------------------------
1826 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1828 * Support for kernels:
1829 2.6.9-42.0.2.EL (RHEL 4)
1830 2.6.5-7.276 (SLES 9)
1831 2.4.21-40.EL (RHEL 3)
1832 2.6.12.6 vanilla (kernel.org)
1836 Frequency : always on RHEL 3
1838 Description: Number of open files grows over time
1839 Details : The number of open files grows over time, whether or not
1840 Lustre is started. This was due to a filp leak introduced
1841 by one of our kernel patches.
1843 ------------------------------------------------------------------------------
1845 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1847 * Support for kernels:
1848 2.6.9-42.EL (RHEL 4)
1849 2.6.5-7.276 (SLES 9)
1850 2.4.21-40.EL (RHEL 3)
1851 2.6.12.6 vanilla (kernel.org)
1856 Bugzilla : 5719, 9635, 9792, 9684
1857 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1858 Details : If a server is under heavy load and cannot reply to new
1859 connection requests before the client resends the (re)connect,
1860 the connection handling code can behave badly if two service
1861 threads are concurrently handing separate (re)connections from
1862 the same client. Add better locking to the connection handling
1863 code, and ensure that only a single connection will be processed
1864 for a given client UUID, even if the lock is dropped.
1866 Severity : enhancement
1868 Description: add TCP zero-copy support to kernel
1869 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1870 sends if the hardware supports scatter-gather and checksumming.
1871 This allows socklnd to do client-write and server-read more
1872 efficiently and reduce CPU utilization from skbuf copying.
1875 Frequency : only if NFS exporting from client
1877 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1878 Details : If setting attributes on a file created under NFS that had
1879 never been opened it would be possible to oops the client
1880 if the file had no objects.
1883 Frequency : always for liblustre
1885 Description: liblustre client does MDS+OSTs setattr RPC for each write
1886 Details : When doing a write from a liblustre client, the client
1887 incorrectly issued an RPC to the MDS and each OST the file was
1888 striped over in order to update the timestamps. When writing
1889 with small chunks and many clients this could overwhelm the MDS
1890 with RPCs. In all cases it would slow down the write because
1891 these RPCs are unnecessary.
1893 Severity : enhancement
1895 Description: allow number of MDS service threads to be changed at module load
1896 Details : It is now possible to change the number of MDS service threads
1897 running. Adding "options mds mds_num_threads={N}" to the MDS's
1898 /etc/modprobe.conf will set the number of threads for the next
1899 time Lustre is restarted (assuming the "mds" module is also
1900 reloaded at that time). The default number of threads will
1901 stay the same, 32 for most systems.
1906 Description: OST crash if filesystem is unformatted or corrupt
1907 Details : If an OST is started on a device that has never been formatted
1908 or if the filesystem is corrupt and cannot even mount then the
1909 error handling cleanup routines would dereference a NULL pointer.
1914 Description: NULL pointer deref in llap_from_page.
1915 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1916 incorrect) mapping if the page was truncated/reclaimed while it was
1917 searched for. Check for this condition and skip such pages when
1918 doing readahead. Introduce extra check to llap_from_page() to
1919 verify page->mapping->host is non-NULL (so page is not anonymous).
1922 Frequency : Sometimes when using sys_sendfile
1924 Description: "page not covered by a lock" warnings from ll_readpage
1925 Details : sendfile called ll_readpage without right page locks present.
1926 Now we introduced ll_file_sendfile that does necessary locking
1927 around call to generic_file_sendfile() much like we do in
1931 Frequency : with certain MDS communication failures at client mount time
1933 Description: NULL pointer deref after failed client mount
1934 Details : a client connection request may delayed by the network layer
1935 and not be sent until after the PTLRPC layer has timed out the
1936 request. If the client fails the mount immediately it will try
1937 to clean up before the network times out the request. Add a
1938 reference from the request import to the obd device and delay
1939 the cleanup until the network drops the request.
1942 Frequency : occasionally during client (re)connect
1944 Description: assertion failure during client (re)connect
1945 Details : processing a client connection request may be delayed by the
1946 client or server longer than the client connect timeout. This
1947 causes the client to resend the connection request. If the
1948 original connection request is replied in this interval, the
1949 client may trip an assertion failure in ptlrpc_connect_interpret()
1950 which thought it would be the only running connect process.
1953 Frequency : only with obd_echo servers and clients that are rebooted
1955 Description: kernel BUG accessing uninitialized data structure
1956 Details : When running an obd_echo server it did not start the ping_evictor
1957 thread, and when a client was evicted an uninitialized data
1958 structure was accessed. Start the ping_evictor in the RPC
1959 service startup instead of the OBD startup.
1961 Severity : enhancement
1962 Bugzilla : 10193 (patchless)
1963 Description: Remove dependency on various unexported kernel interfaces.
1964 Details : No longer need reparent_to_init, exit_mm, exit_files,
1965 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1968 Frequency : rare (only users of deprecated and unsupported LDAP config)
1970 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1971 Details : LDAP apparently contains 'lustreName' attributes instead of
1972 'name'. A simple remapping of the name is sufficient.
1975 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1977 Description: waiting_locks_callback trips kernel BUG if client is evicted
1978 Details : Running with the dump_on_timeout debug flag turned on makes
1979 it possible that the waiting_locks_callback() can try to dump
1980 the Lustre kernel debug logs from an interrupt handler. Defer
1981 this log dumping to the expired_lock_main() thread.
1983 Severity : enhancement
1985 Description: Support NFS exporting on 2.6 kernels.
1986 Details : Implement non-rawops metadata methods for NFS server to use without
1987 changing NFS server code.
1990 Frequency : very rare (synthetic metadata workload only)
1992 Description: two racing renames might cause an MDS thread to deadlock
1993 Details : Running the "racer" program may cause one MDS thread to rename
1994 a file from being the source of a rename to being the target of
1995 a rename at exactly the same time that another thread is doing
1996 so, and the second thread has already enqueued these locks after
1997 doing a lookup of the target and is trying to relock them in
1998 order. Ensure that we don't try to re-lock the same resource.
2001 Frequency : only very large systems with liblustre clients
2003 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2004 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2005 all OSTs in parallel. This allows the eviction of stale liblustre
2006 clients to proceed much faster than if they were done in series,
2007 and also offers similar improvements for other set_info RPCs.
2012 Description: excessive CPU usage during initial read phase on client
2013 Details : During the initial read phase on a client, it would agressively
2014 retry readahead on the file, consuming too much CPU and impacting
2015 performance (since 1.4.5.8). Improve the readahead algorithm
2016 to avoid this, and also improve some other common cases (read
2017 of small files in particular, where "small" is files smaller than
2018 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2023 Description: MDS crash when receiving packet with unknown intent.
2024 Details : Do not LBUG in unknown intent case, just return -EFAULT
2026 Severity : enhancement
2027 Bugzilla : 9293, 9385
2028 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2029 Details : Do not serialize getattr (non-intent version) and statfs.
2032 Frequency : occasional, when OST network is overloaded/intermittent
2034 Description: client evicted by OST after bulk IO timeout
2035 Details : If a client sends a bulk IO request (read or write) the OST
2036 may evict the client if it is unresposive to its data GET/PUT
2037 request. This is incorrect if the network is overloaded (takes
2038 too long to transfer the RPC data) or dropped the OST GET/PUT
2039 request. There is no need to evict the client at all, since
2040 the pinger and/or lock callbacks will handle this, and the
2041 client can restart the bulk request.
2044 Frequency : Always when mmapping file with no objects
2046 Description: client crashes when mmapping file with no objects
2047 Details : Check that we actually have objects in a file before doing any
2048 operations on objects in ll_vm_open, ll_vm_close and
2054 Description: Request leak when working with deleted CWD
2055 Details : Introduce advanced request refcount tracking for requests
2056 referenced from lustre intent.
2058 Severity : Enhancement
2060 Description: Cache open file handles on client.
2061 Details : MDS now will return special lock along with openhandle, if
2062 requested and client is allowed to hold openhandle, even if unused,
2063 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2064 closing files for every read/write openration.
2066 Severity : Enhancement
2068 Description: Cache open negative dentries on client when possible.
2069 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2070 negative dentries on lock revocation.
2075 Description: Remounting a client read-only wasn't possible with a zconf mount
2076 Details : It wasn't possible to remount a client read-only with llmount.
2078 Severity : enhancement
2079 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2080 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2081 MPICH is included to add Lustre-specific ADIO interfaces.
2082 This is based closely on the UFS ADIO layer and only differs
2083 in file creation, in order to allow the OST striping to be set.
2084 This is user-contributed code and not supported by CFS.
2089 Description: extended inode attributes (immutable, append-only) work improperly
2090 when 2.4 and 2.6 kernels are used on client/server or vice versa
2091 Details : Introduce kernel-independent values for these flags.
2093 Severity : enhancement
2096 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2097 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2098 granularity. Now, Fractional MB granularity is be supported,
2099 this is very useful for low memory system.
2101 Severity : enhancement
2103 Description: Getattr by fid
2104 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2105 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2106 have locks and avoids getattr rpc requests for it on every stat.
2109 Frequency : Always, for filesystems larger than 2TB
2111 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2112 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2113 filesystems larger than 2TB to be mounted with mballoc enabled.
2116 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2118 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2119 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2120 filesystems larger than 2TB is mounted then the kernel will
2121 silently corrupt the start of the filesystem. CONFIG_LBD is
2122 enabled for all CFS-supported kernels, but the possibility of
2123 this happening with a modified kernel config exists.
2125 Severity : enhancement
2127 Description: add client O_DIRECT support for 2.6 kernels
2128 Details : It is now possible to do O_DIRECT reads and writes to files
2129 in the Lustre client mountpoint on 2.6 kernel clients.
2131 Severity : enhancement
2133 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2134 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2135 parallel, not waiting for response from every OST before sending
2136 a rpc to the next OST.
2141 Description: setattr vs write race when updating file timestamps
2142 Details : Client processes that update a file timestamp into the past
2143 right after writing to the file (e.g. tar) it is possible that
2144 the updated file modification time can be reset to the current
2145 time due to a race between processing the setattr and write RPC.
2147 Severity : enhancement
2149 Description: Bring 'lfs find' closer in line with regular Linux find.
2150 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2151 -print0 options and obtains all the needed info through the lustre
2154 Severity : enhancement
2156 Description: support up to 1024 configured devices on one node
2157 Details : change obd_dev array from statically allocated to dynamically
2158 allocated structs as they are first used to reduce memory usage
2163 Description: Flush dirty partially truncated pages during truncate
2164 Details : Immediatelly flush partially truncated pages in filter_setattr,
2165 this way we completely avoid having any pages in page cache on OST
2166 and can retire ugly workarounds during writes to flush such pages.
2171 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2172 Details : protect inode from truncation within vfs_unlink() context
2173 just take a reference before calling vfs_unlink() and release it
2174 when parent's i_sem is free.
2179 Description: last_id value checked outside lock on OST caused LASSERT failure
2180 Details : If there were multiple MDS->OST object precreate requests in
2181 flight, it was possible that the OST's last object id was checked
2182 outside a lock and incorrectly tripped an assertion. Move checks
2183 inside locks, and discard old precreate requests.
2186 Frequency : always, if extents are used on OSTs
2188 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2189 Details : OSTs using the extents format would not zero the high 16 bits of
2190 the index physical block number. This is not a problem for any
2191 OST filesystems smaller than 16TB, and no kernels support ext3
2192 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2193 new/modified files) and can be fixed for existing filesystems
2194 with e2fsprogs-1.39-cfs1.
2199 Description: import connection selection may be incorrect if timer wraps
2200 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2201 import connections to be ignored if the 32-bit jiffies counter
2202 wraps. Use a 64-bit jiffies counter.
2205 Frequency : very large clusters immediately after boot
2207 Description: LNET request buffers exhausted under heavy short-term load
2208 Details : If a large number of client requests are generated on a service
2209 that has previously never seen so many requests it is possible
2210 that the request buffer growth cannot keep up with the spike in
2211 demand. Instead of dropping incoming requests, they are held in
2212 the LND until the RPC service can accept more requests.
2215 Frequency : Sometimes during replay
2217 Description: Assertion failure in ll_local_open after replay.
2218 Details : If replay happened on an open request reply before we were able
2219 to set replay handler, reply will become not swabbed tripping the
2220 assertion in ll_local_open. Now we set the handler right after
2221 recognising of open request
2224 Frequency : very rare
2226 Description: kernel reports "badness in vsnprintf"
2227 Details : Reading from the "recovery_status" /proc file in small chunks
2228 may cause a negative length in lprocfs_obd_rd_recovery_status()
2229 call to vsnprintf() (which is otherwise harmless). Exit early
2230 if there is no more space in the output buffer.
2232 Severity : enhancement
2234 Description: clear OBD RPC statistics by writing to them
2235 Details : It is now possible to clear the OBD RPC statistics by writing
2236 to the "stats" file.
2241 Description: Client mtime is not the same on different clients after utimes
2242 Details : In some cases, the client was using the utimes() syscall on
2243 a file cached on another node. The clients now validate the
2244 ctime from the MDS + OSTs to determine which one is right.
2249 Description: Inability to activate failout mode
2250 Details : lconf script incorrectly assumed that in pythong string's numeric
2251 value is used in comparisons.
2254 Frequency : always with multiple stripes per file
2256 Description: Inefficient object allocation for mutli-stripe files
2257 Details : When selecting which OSTs to stripe files over, for files with
2258 a stripe count that divides evenly into the number of OSTs,
2259 the MDS is always picking the same starting OST for each file.
2260 Return the OST selection heuristic to the original design.
2265 Description: mount failures may take full timeout to return an error
2266 Details : Under some heavy load conditions it is possible that a
2267 failed mount can wait for the full obd_timeout interval,
2268 possibly several minutes, before reporting an error.
2269 Instead return an error as soon as the status is known.
2271 Frequency : quota enabled and large files being deleted
2273 Description: releasing more than 4GB of quota at once hangs OST
2274 Details : If a user deletes more than 4GB of files on a single OST it
2275 will cause the OST to spin in an infinite loop. Release
2276 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2281 Description: statfs data retrieved from /proc may be stale or zero
2282 Details : When reading per-device statfs data from /proc, in the
2283 {kbytes,files}_{total,free,avail} files, it may appear
2284 as zero or be out of date.
2287 Frequency : systems with MD RAID1 external journal devices
2289 Description: lconf's call to blkid is confused by RAID1 journal devices
2290 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2291 of returning all block devices that match the journal UUID.
2294 Frequency : always, for aggregate stripe size over 4GB
2296 Description: assertion fails when trying to use 4GB stripe size
2297 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2298 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2301 Frequency : always on ppc64
2303 Description: the first write on an ext3 filesystem with mballoc got stuck
2304 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2305 perform endianness conversion.
2307 ------------------------------------------------------------------------------
2309 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2311 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2312 INTEROPERATE with older versions automatically. Please read the
2313 user documentation before upgrading any part of a live system.
2314 * WARNING: Lustre networking configuration changes are required with
2315 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2318 * Support for kernels:
2319 2.6.9-22.0.2.EL (RHEL 4)
2320 2.6.5-7.244 (SLES 9)
2321 2.6.12.6 vanilla (kernel.org)
2324 Severity : enhancement
2325 Bugzilla : 7981/8208
2326 Description: Introduced Lustre Networking (LNET)
2327 Details : LNET is new networking infrastructure for Lustre, it includes
2328 a reorganized network configuration mode (see the user
2329 documentation for full details) as well as support for routing
2330 between different network fabrics. Lustre Networking Devices
2331 (LNDs) for the supported network fabrics have also been
2332 created for this new infrastructure.
2334 Severity : enhancement
2335 Description: Introduced Access control lists
2336 Details : clients can set ACLs on files and directories in order to have
2337 more fine-grained permissions than the standard Unix UGO+RWX.
2338 The MDS must be started with the "-o acl" mount option.
2340 Severity : enhancement
2341 Description: Introduced filesystem quotas
2342 Details : Administrators may now establish per-user quotas on the
2345 Severity : enhancement
2347 Description: Configuration change for the XT3
2348 The PTLLND is now used to run Lustre over Portals on the XT3
2349 The configure option(s) --with-cray-portals are no longer used.
2350 Rather --with-portals=<path-to-portals-includes> is used to
2351 enable building on the XT3. In addition to enable XT3 specific
2352 features the option --enable-cray-xt3 must be used.
2357 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2358 Details : A race between allocating a new llcd and re-getting the llcd_lock
2359 allowed another thread to grab newly-allocated llcd.
2361 Severity : enhancement
2363 Description: 2.6 OST async journal commit and locking fix to improve performance
2364 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2365 2.6 kernels are now async as they already were in 2.4 kernels so
2366 that they can commit concurrently with the network bulk transfer.
2367 For block-allocated files the filter allocation semaphore is held
2368 to avoid filesystem fragmentation during allocation. BKL lock
2369 removed for 2.6 xattr operations where it is no longer needed.
2374 Description: lconf incorrectly determined whether two IP networks could talk
2375 Details : In some more complicated routing and multiple-network
2376 configurations, lconf will avoid trying to make a network
2377 connection to a disjoint part of the IP space. It was doing the
2378 math incorrectly for one set of cases.
2383 Description: Fix for potential infinite loop processing records in an llog.
2384 Details : If an llog record is corrupted/zeroed, it is possible to loop
2385 forever in llog_process(). Validate the llog record length
2386 and skip the remainder of the block on error.
2389 Frequency : occasional (liblustre only)
2391 Description: liblustre could not open files whose last component is a symlink
2392 Details : sysio_path_walk() would incorrectly pass the open intent to
2393 intermediate path components.
2396 Frequency : rare (liblustre only with non-standard tuning)
2397 Bugzilla : 7201 (7350)
2398 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2399 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2400 after it has been released, only do this for referenced locks
2402 Severity : enhancement
2404 Description: specifying an (invalid) directory default stripe_size of -1
2405 would reset the directory default striping
2406 Details : stripe_size -1 was used internally to signal directory stripe
2407 removal, now use "all default" to signal dir stripe removal
2408 as a directory striping of "all default" is not useful
2411 Frequency : common for large clusters running liblustre clients
2413 Description: doing an ls when liblustre clients are running is slow
2414 Details : sending a glimpse AST to a liblustre client waits for every AST
2415 to time out, as liblustre clients will not respond. Since they
2416 cannot cache data we refresh the OST lock LVB from disk instead.
2418 Severity : enhancement
2420 Description: doing an ls at the same time as file IO can be slow
2421 Details : enqueue and other "small" requests can be blocked behind many
2422 large IO requests. Create a new OST IO portal for non-IO
2423 requests so they can be processed faster.
2426 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2428 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2429 created file with mode 000
2430 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2431 allow an ftruncate() as a non-root user to a file with mode 000.
2432 Lustre now respects this flag to disable mode checks when
2433 truncating a file owned by the user
2436 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2439 Description: Revoking locks from clients that went dead or catatonic might take
2441 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2442 cancellation of such locks instant on servers without waiting for
2443 any reply from clients. Clients drops these locks when cancel
2444 notification from server is received without replying.
2447 Frequency : liblustre-only, when liblustre client dies or becomes busy
2449 Description: Doing ls on Linux clients can take a long time with active
2451 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2452 granting such locks to it in the first place if possible. Locks
2453 are taken by proxy on the OST during the read or write and
2454 dropped immediately afterward. Add connect flags handling, do
2455 not grant locks to liblustre clients for glimpse ASTs.
2457 Severity : enhancement
2459 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2460 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2461 sequential reads and behaves badly with applications doing random
2462 reads. Improve it by reading ahead at least read region, and
2463 avoiding excessive large RPC for small reads.
2465 Severity : enhancement
2467 Description: Creating more than 1000 files for a single job may cause a load
2468 imbalance on the OSTs if there are also a large number of OSTs.
2469 Details : qos_prep_create() uses an OST index reseed value that is an
2470 even multiple of the number of available OSTs so that if the
2471 reseed happens in the middle of the object allocation it will
2472 still utilize the OSTs as uniformly as possible.
2477 Description: OST or MDS may oops in ping_evictor_main()
2478 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2479 but doesn't restart at beginning of obd_exports_timed list
2482 Severity : enhancement
2484 Description: improve by-nid export eviction on the MDS and OST
2485 Details : allow multiple exports with the same NID to be evicted at one
2486 time without re-searching the exports list.
2489 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2491 Description: MDS may oops in groups_free()
2492 Details : in rare race conditions a newly allocated group_info struct is
2493 freed again, and this can be NULL. The 2.4 compatibility code
2494 for groups_free() checked for a NULL pointer, but 2.6 did not.
2497 Frequency : common for liblustre clients doing little filesystem IO
2498 Bugzilla : 9352, 7313
2499 Description: server may evict liblustre clients accessing contended locks
2500 Details : if a client is granted a lock or receives a completion AST
2501 with a blocking AST already set it would not reply to the AST
2502 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2505 Frequency : lfs setstripe, only systems with more than 160 OSTs
2507 Description: unable to set striping with a starting offset beyond OST 160
2508 Details : llapi_create_file() incorrectly limited the starting stripe
2509 index to the maximum single-file stripe count.
2512 Frequency : LDAP users only
2514 Description: lconf did not handle in-kernel recovery with LDAP properly
2515 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2517 Severity : enhancement
2519 Description: bind OST threads to NUMA nodes to improve performance
2520 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2521 node and do their allocations there to localize memory access
2523 Severity : enhancement
2525 Description: llmount can determine client NID directly from Myrinet (GM)
2526 Details : the client NID code from gmnalnid was moved directly into
2527 llmount, removing the need to use this or specifying the
2528 client NID explicitly when mounting GM clients with zeroconf
2531 Frequency : if client is started with down MDS
2533 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2534 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2535 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2536 ptlrpc_import_delayed_req() because we didn't check if the
2537 request was interrupted, and we also didn't break out of the
2538 event loop if there was no timeout
2543 Description: data loss during non-page-aligned writes to a single file from
2544 both multiple nodes and multiple threads on one node at same time
2545 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2546 inconsistency led to false short-reads, that were cached and later
2547 used by ->prepare_write() to fill in partially written page,
2548 leading to data loss.
2551 Frequency : always, if lconf --abort_recovery used
2553 Description: lconf --abort_recovery fails with 'Operation not supported'
2554 Details : lconf was attempting to abort recovery on the MDT device and not
2557 Severity : enhancement
2559 Description: remove cleanup logs
2560 Details : replace lconf-generated cleanup logs with lustre internal
2561 cleanup routines. Eliminates the need for client-cleanup and
2564 Severity : enhancement
2566 Description: add support for EAs (user and system) on lustre filesystems
2567 Details : it is now possible to store extended attributes in the Lustre
2568 client filesystem, and with the user_xattr mount option it
2569 is possible to allow users to store EAs on their files also
2571 Severity : enhancement
2573 Description: Add possibility (config option) to show minimal available OST free
2575 Details : When compiled with --enable-mindf configure option, statfs(2)
2576 (and so, df) will return least minimal free space available from
2577 all OSTs as amount of free space on FS, instead of summary of
2578 free spaces of all OSTs.
2580 Severity : enhancement
2582 Description: do not expand extent locks acquired on OST-side
2583 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2584 by server: they are not cached anyway.
2587 Frequency : when mmap is used/binaries executed from Lustre
2589 Description: Unmmap pages before throwing them away from read cache.
2590 Details : llap_shrink cache now attempts to unmap pages before discarding
2591 them (if unmapping failed - do not discard). SLES9 kernel has
2592 extra checks that trigger if this unmapping is not done first.
2597 Description: lconf didn't resolve symlinks before checking to see whether a
2598 given mountpoint was already in use
2601 Frequency : when migrating failover services
2602 Bugzilla : 6395, 9514
2603 Description: When migrating a subset of services from a node (e.g. failback
2604 from a failover service node) the remaining services would
2605 time out and evict clients.
2606 Details : lconf --force (implied by --failover) sets the global obd_timeout
2607 to 5 seconds in order to quickly disconnect, but this caused
2608 other RPCs to time out too quickly. Do not change the global
2609 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2611 Severity : enhancement
2612 Frequency : if MDS is started with down OST
2613 Bugzilla : 9439,5706
2614 Description: Allow startup/shutdown of an MDS without depending on the
2615 availability of the OSTs.
2616 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2617 Add appropriate locking and lov-osc refcounts for safe
2618 cleaning. Add osc abort_inflight calls in case the
2619 synchronize never started.
2622 Frequency : occasional (Cray XT3 only)
2624 Description: root not authorized to access files in CRAY_PORTALS environment
2625 Details : The client process capabilities were not honoured on the MDS in
2626 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2627 been accessed by an authorized user then root was able to access
2628 the file on the local client also. The root user capabilities
2629 are now allowed on the MDS, as this environment has secure UID.
2632 Frequency : occasional
2634 Description: ldiskfs "too long searching" message happens too often
2635 Details : A debugging message (otherwise harmless) prints too often on
2636 the OST console. This has been reduced to only happen when
2637 there are fragmentation problems on the filesystem.
2642 Description: Division by zero in statfs when all OSCs are inactive
2643 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2644 lov_statfs divides by value returned by lov_get_stripecnt().
2648 Bugzilla : 9489, 3273
2649 Description: First write from each client to each OST was only 4kB in size,
2650 to initialize client writeback cache, which caused sub-optimal
2651 RPCs and poor layout on disk for the first writen file.
2652 Details : Clients now request an initial cache grant at (re)connect time
2653 and so that they can start streaming writes to the cache right
2654 away and always do full-sized RPCs if there is enough data.
2655 If the OST is rebooted the client also re-establishes its grant
2656 so that client cached writes will be honoured under the grant.
2661 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2662 Details : Now I/O RPCs go to different portal number and (presumably) fast
2663 lock requests (and glimses) and other RPCs get their own service
2664 threads pool that should be able to service those RPCs
2667 Severity : enhancement
2669 Description: Ability to exchange lustre version between client and servers and
2670 issue warnings at client side if client is too old. Also for
2671 liblustre clients there is ability to refuse connection of too old
2673 Details : New 'version' field is added to connect data structure that is
2674 filled with version info. That info is later checked by server and
2678 Frequency : rare, liblustre only.
2679 Bugzilla : 9296, 9581
2680 Description: Two simultaneous writes from liblustre at offset within same page
2681 might proceed at the same time overwriting eachother with stale
2683 Details : I/O lock withing llu_file_prwv was released too early, before data
2684 actually was hitting the wire. Extended lock-holding time until
2685 server acknowledges receiving data.
2688 Frequency : extremely rare. Never observed in practice.
2690 Description: avoid generating lustre_handle cookie of 0.
2691 Details : class_handle_hash() generates handle cookies by incrementing
2692 global counter, and can hit 0 occasionaly (this is unlikely, but
2693 not impossible, because initial value of cookie counter is
2694 selected randonly). Value of 0 is used as a sentinel meaning
2695 "unassigned handle" --- avoid it. Also coalesce two critical
2696 sections in this function into one.
2698 Severity : enhancement
2700 Description: allow liblustre clients to delegate truncate locking to OST
2701 Details : To avoid overhead of locking, liblustre client instructs OST to
2702 take extent lock in ost_punch() on client's behalf. New connection
2703 flag is added to handle backward compatibility.
2705 Severity : enhancement
2706 Bugzilla : 4928, 7341, 9758
2707 Description: allow number of OST service threads to be specified
2708 Details : a module parameter allows the number of OST service threads
2709 to be specified via "options ost ost_num_threads={N}" in the
2710 OSS's /etc/modules.conf or /etc/modprobe.conf.
2714 Bugzilla : 6146, 9635, 9895
2715 Description: servers crash with bad pointer in target_handle_connect()
2716 Details : In rare cases when a client is reconnecting it was possible that
2717 the connection request was the last reference for that export.
2718 We would temporarily drop the export reference and get a new
2719 one, but this may have been the last reference and the export
2720 was just destroyed. Get new reference before dropping old one.
2722 Severity : enhancement
2723 Frequency : if client is started with failover MDS
2725 Description: Allow multiple MDS hostnames in the mount command
2726 Details : Try to read the configuration from all specified MDS
2727 hostnames during a client mount in case the "primary"
2730 Severity : enhancement
2732 Description: Stop sending data to evicted clients as soon as possible.
2733 Details : Check if the client we are about to send or are sending data to
2734 was evicted already. (Check is done every second of waiting,
2735 for which l_wait_event interface was extended to allow checking
2736 of exit condition at specified intervals).
2739 Frequency : rare, normally only when NFS exporting is done from client
2741 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2743 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2746 Frequency : always, if srand() is called before liblustre initialization
2748 Description: Liblustre uses system PRNG disturbing its usage by user application
2749 Details : Introduce internal to lustre fast and high-quality PRNG for
2750 lustre usage and make liblustre and some other places in generic
2751 lustre code to use it.
2753 Severity : enhancement
2754 Bugzilla : 9477, 9557, 9870
2755 Description: Verify that the MDS configuration logs are updated when xml is
2756 Details : Check if the .xml configuration logs are newer than the config
2757 logs stored on the MDS and report an error if this is the case.
2758 Request --write-conf, or allow starting with --old_conf.
2760 Severity : enhancement
2762 Description: Handle symlinks in the path when checking if Lustre is mounted.
2763 Details : Resolve intermediate symlinks when checking if a client has
2764 mounted a filesystem to avoid duplicate client mounts.
2769 Description: lconf can hit an error exception but still return success.
2770 Details : The lconf command catches the Command error exception at the top
2771 level script context and will exit with the associated exit
2772 status, but doesn't ensure that this exit status is non-zero.
2777 Description: failure of ptlrpc thread startup can cause oops
2778 Details : Starting a ptlrpc service thread can fail if there are a large
2779 number of threads or the server memory is very fragmented.
2780 Handle this without oopsing.
2783 Frequency : always, only if liblustre and non-default acceptor port was used
2785 Description: liblustre cannot connect to servers with non-default acceptor port
2786 Details : tcpnal_set_default_params() was not called and was therefore
2787 ignoring the environment varaible TCPNAL_PORT, as well as other
2788 TCPNAL_ environment variables
2793 Description: two objects could be created on the same OST for a single file
2794 Details : If an OST is down, in some cases it was possible to create two
2795 objects on a single OST for a single file. No problems other
2796 than potential performance impact and spurious error messages.
2800 Bugzilla : 5681, 9562
2801 Description: Client may oops in ll_unhash_aliases
2802 Details : Client dcache may become inconsistent in race condition.
2803 In some cases "getcwd" can fail if the current directory is
2809 Description: Inode refcounting problems in NFS export code
2810 Details : link_raw functions used to call d_instantiate without obtaining
2811 extra inode reference first.
2815 Bugzilla : 9942, 9903
2816 Description: Referencing freed requests leading to crash, memleaks with NFS.
2817 Details : We used to require that call to ll_revalidate_it was always
2818 followed by ll_lookup_it. Also with revalidate_special() it is
2819 possible to call ll_revalidate_it() twice for the same dentry
2820 even if first occurence returned success. This fix changes semantic
2821 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2822 reference on a request referred from the intent.
2823 ll_intent_release() then releases such a request.
2826 Frequency : rare, normally benchmark loads only
2828 Description: unlinked inodes were kept in memory on the client
2829 Details : If a client is repeatedly creating and unlinking files it
2830 can accumulate a lot of stale inodes in the inode slab cache.
2831 If there is no other client load running this can cause the
2832 client node to run out of memory. Instead flush old inodes
2833 from client cache that have the same inode number as a new inode.
2836 Frequency : SLES9 2.6.5 kernel and long filenames only
2837 Bugzilla : 9969, 10379
2838 Description: utime reports stale NFS file handle
2839 Details : SLES9 uses out-of-dentry names in some cases, which confused
2840 the lustre dentry revalidation. Change it to always use the
2844 Frequency : rare, unless heavy write-truncate concurrency is continuous
2845 Bugzilla : 4180, 6984, 7171, 9963, 9331
2846 Description: OST becomes very slow and/or deadlocked during object unlink
2847 Details : filter_destroy() was holding onto the parent directory lock
2848 while truncating+unlinking objects. For very large objects this
2849 may block other threads for a long time and slow overall OST
2850 responsiveness. It may also be possible to get a lock ordering
2851 deadlock in this case, or run out of journal credits because of
2852 the combined truncate+unlink. Solution is to do object truncate
2853 first in one transaction without parent lock, and then do the
2854 final unlink in a new transaction with the parent lock. This
2855 reduces the lock hold time dramatically.
2858 Frequency : rare, 2.4 kernels only
2860 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2861 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2862 before unmounting their filesystem in order to avoid deadlock.
2863 The kernel_locked() function in 2.4 kernels only checks whether
2864 the kernel lock is held, not whether it is this process that is
2865 holding it as 2.6 kernels do.
2870 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2871 Details : The client ptlrpc code may be trying to reconnect to a down
2872 server before a previous connection attempt has timed out.
2873 Increase the reconnect interval to be longer than the connection
2874 timeout interval to avoid sending duplicate connections to
2878 Frequency : echo_client brw_test command
2880 Description: fix echo_client to work with OST preallocated code
2881 Details : OST preallocation code (5137) didn't take echo_client IO path
2882 into account: echo_client calls filter methods outside of any
2883 OST thread and, hence, there is no per-thread preallocated
2884 pages and buffers to use. Solution: hijack pga pages for IO. As
2885 a byproduct, this avoids unnecessary data copying.
2889 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2890 Description: Client can oops in mdc_commit_close() after open replay
2891 Details : It was possible for the MDS to return an open request with no
2892 transaction number in mds_finish_transno() if the client was
2893 evicted, but without actually returning an error. Clients
2894 would later try to replay that open and may trip an assertion
2895 Simplify the client close codepath, and always return an error
2896 from the MDS in case the open is not successful.
2899 Frequency : rare, 2.6 OSTs only
2901 Description: OST may deadlock under high load on fragmented files
2902 Details : If there was a heavy load and highly-fragmented OST filesystems
2903 it was possible to have all the OST threads deadlock waiting on
2904 allocation of biovecs, because the biovecs were not released
2905 until the entire RPC IO was completed. Instead, release biovecs
2906 as soon as they are complete to ensure forward IO progress.
2908 Severity : enhancement
2910 Description: Support for specifying external journal device at mount
2911 Details : If an OST or MDS device is formatted with an external journal
2912 device, this device major/minor is stored in the ext3 superblock
2913 and may not be valid for failover. Allow detecting and
2914 specifying the external journal at mount time.
2919 Description: Mounting an MDS with pending unlinked files may cause oops
2920 Details : target_finish_recovery() calls mds_postrecov() which returned
2921 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2922 considers this an error and immediately begins cleaning up the
2923 lov, just after starting the mds_lov process
2925 Severity : enhancement
2927 Description: Implement 'lfs df' to report actual free space on per-OST basis
2928 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2929 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2930 usage of objects; '-h' to report in human readable format.
2932 ------------------------------------------------------------------------------
2934 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2941 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2942 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2943 at startup during orphan unlinking. Instead, skip buddy update
2944 before setup, as it will be regenerated after recovery is complete.
2949 Description: If an OST is inactive, its locks might reference stale inodes.
2950 Details : lov_change_cbdata() must iterate over all namespaces, even if
2951 they are inactive to clear inode references from the lock.
2953 Severity : enhancement
2954 Frequency : occasional, if non-standard max_dirty_mb used
2956 Description: Client will block write RPCs if not enough grant
2957 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2958 then the client will block writes while waiting for another RPC
2959 to complete instead of consuming its dirty limit. With change
2960 we get improved performance when max_dirty_mb is small.
2962 Severity : enhancement
2963 Bugzilla : 3389, 6253
2964 Description: Add support for supplementary groups on the MDS.
2965 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2966 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2967 lookups for user supplementary groups into a cache.
2971 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2972 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2973 for liblustre O_EXCL is enforced by MDS.
2977 Description: readdir never returns NULL in liblustre.
2978 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2979 dentry in struct dirent.
2983 Frequency : liblustre only, depends on application IO pattern
2984 Description: liblustre clients evicted if not contacting servers
2985 Details : Don't put liblustre clients into the ping_evictor list, so
2986 they will not be evicted by the pinger ever.
2988 Severity : enhancement
2990 Description: Add ability to evict clients by NID from MDS.
2991 Details : By echoing "nid:$NID" string into
2992 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2993 $NID would be instantly evicted from this MDS and from all active
2994 OSTs connected to it.
2998 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2999 Details : lookup_it_finish() used to query file size from OSTs that was not
3004 Description: service threads change working directory to that of init
3005 Details : Starting lustre service threads may pin the working directory
3006 of the parent thread, making that filesystem busy. Threads
3007 now change to the working directory of init to avoid this.
3011 Frequency : during shutdown only
3012 Description: shutdown with a failed MDS or OST can cause unmount to hang
3013 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3018 Frequency : 2.6 only
3019 Description: chmod/chown may include an extra supplementary group
3020 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3021 supplementary group and if none is specified this is used.
3024 Bugzilla : 5479 (6816)
3026 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3027 Details : If lookup is in progress on a file that is unlinked we might try
3028 to revalidate the inode and fail in revalidate after lookup is
3029 complete and ll_file_open() enqueues the open again but
3030 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3033 Frequency : always, if lconf --abort_recovery used
3035 Description: lconf --abort_recovery fails with 'Operation not supported'
3036 Details : lconf was attempting to abort recovery on the MDT device and not
3039 ------------------------------------------------------------------------------
3041 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3046 Frequency : rare (only unsupported configurations with a node running as an
3048 Bugzilla : 6514, 5137
3049 Description: Mounting a Lustre file system on a node running as an OST could
3051 Details : OSTs now preallocates memory needed to write out data at
3052 startup, instead of when needed, to avoid having to
3053 allocate memory in possibly low memory situations.
3054 Specifically, if the file system is mounted on on OST,
3055 memory pressure could force it to try to write out data,
3056 which it needed to allocate memory to do. Due to the low
3057 memory, it would be unable to do so and the node would
3058 become unresponsive.
3060 Severity : enhancement
3062 Description: Addition of lconf --service command line option
3063 Details : lconf now accepts a '--service <arg>' option, which is
3064 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3066 Severity : enhancement
3068 Description: Failover mode is now the default for OSTs.
3069 Details : By default, OSTs will now run in failover mode. To return to
3070 the old behaviour, add '--failout' to the lmc line for OSTs.
3072 Severity : enhancement
3074 Description: Health checks are now provided for MDS and OSTs
3075 Details : Additional detailed health check information on MSD and OSTs
3076 is now provided through the procfs health_check value.
3079 Frequency : occasional, depends on IO load
3081 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3082 after numerous create/delete cycles
3083 Details : The ext3 inode allocation policy would not allocate new inodes
3084 very well on the OSTs because there are no new directories
3085 being created. Instead we look for groups with free space if
3086 the parent directories are nearly full.
3091 Description: Network or server problems during mount may cause partially
3092 mounted clients instead of returning an error.
3093 Details : The config llog parsing code may overwrite the error return
3094 code during mount error handling, returning success instead
3100 Description: MDS can fail to allocate large reply buffers
3101 Details : After long uptimes the MDS can fail to allocate large reply
3102 buffers (e.g. zconf client mount config records) due to memory
3103 fragmentation or consumption by the buffer cache. Preallocate
3104 some large reply buffers so that these replies can be sent even
3105 under memory pressure.
3109 Frequency : rare (liblustre)
3110 Description: fsx running with liblustre complained that using truncate() to
3111 extend the file doesn't work. This patch corrects that issue.
3112 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3113 ATTR_SIZE and lsm use in llu_setattr_raw.
3117 Frequency : rare, only 2.6 kernels
3118 Description: Unusual file access patterns on the MDS may result in inode
3119 data being lost in very rare circumstances.
3120 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3121 code in the 2.6 kernel caused the inode and/or EA data not to
3122 be read from disk, causing single-file corruption.
3126 Frequency : rare, only 2.6 filesystems using extents
3127 Description: Heavy concurrent write and delete load may cause data corruption.
3128 Details : It was possible under high-load situations to have an extent
3129 metadata block in the block device cache from a just-unlinked
3130 file overwrite a newly-allocated data block. We now unmap any
3131 metadata buffers that alias just-allocated data blocks.
3135 Frequency : filesystems with default stripe_count larger than 77
3136 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3137 Details : lconf specifies an inode size of 4096 bytes when the default
3138 stripe_count is larger than 77. This conflicts with the default
3139 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3140 this case to avoid pinning too much memory for large EAs.
3142 ------------------------------------------------------------------------------
3144 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3149 Frequency : rare (extremely heavy IO load with hundreds of clients)
3151 Description: Client is evicted, gets IO error writing to file
3152 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3153 caused clients to be evicted for AST timeouts. The fixes in
3154 bug 5192 mean we no longer need to have such short AST timeouts
3155 so ldlm_timeout has been increased.
3158 Frequency : occasional during --force or --failover shutdown under load
3159 Bugzilla : 5949, 4834
3160 Description: Server oops/LBUG if stopped with --force or --failover under load
3161 Details : a collection of import/export refcount and cleanup ordering
3162 issues fixed for safer force cleanup
3165 Frequency : only filesystems larger than 120 OSTs
3166 Bugzilla : 5990, 6223
3167 Description: lfs getstripe would oops on a very large filesystem
3168 Details : lov_getconfig used kfree on vmalloc'd memory
3171 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3172 Bugzilla : 6242, 6243
3173 Description: reading from files that had been truncated to a non-zero size
3174 but never opened returned no data
3175 Details : ll_file_read() reads zeros from no-object files to EOF
3180 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3181 Details : There's little harm in aborting MDS/OSS recovery and letting it
3182 try again, so I removed the LASSERT and return an error instead.
3184 Severity : enhancement
3186 Description: New debugging infrastructure for tracking down data corruption
3187 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3188 (b) cover more of the client-side code path, and (c) try to narrow
3189 down where problems occurred
3193 Bugzilla : 3819, 4364, 4397, 6313
3194 Description: Racing close and eviction MDS could cause assertion in mds_close
3195 Details : It was possible to get multiple mfd references during close and
3196 client eviction, leading to one thread referencing a freed mfd.
3198 Severity: : enhancement
3199 Bugzilla : 3262, 6359
3200 Description: Attempts to reconnect to servers are now more aggressive.
3201 Details : This builds on the enhanced upcall-less recovery that was added
3202 in 1.4.2. When trying to reconnect to servers, clients will
3203 now try each server in the failover group every 10 seconds. By
3204 default, clients would previously try one server every 25 seconds.
3209 Description: After recovery, certain operations trigger a failed
3210 assertion on a client.
3211 Details : Failing over an mds, using lconf -d --failover, while a
3212 client was doing a readdir() call would cause the client to
3213 LBUG after recovery completed and the readdir() was resent.
3215 Severity : enhancement
3217 Description: Default groups are now added by lconf
3218 Details : You can now run lconf --group <servicename> without having to
3219 manually add groups with lmc.
3222 Frequency : occasional
3224 Description: Nodes with an elan id of 0 trigger a failed assertion
3227 Frequency : always when accessing e.g. tty/console device nodes
3229 Description: tty and some other devices nodes cannot be used on lustre
3230 Details : file's private_data field is used by device data and lustre
3231 values in there got lost. New field was added to struct file to
3232 store fs-specific private data.
3235 Frequency : when exporting Lustre via NFS
3237 Description: NFSD failed occasionally when looking up a path component
3238 Details : NFSD is looking up ".." which was broken in ext3 directories
3239 that had grown large enough to become hashed.
3242 Frequency : Clusters with multiple interfaces not on the same subnet
3244 Description: Nodes will repeatedly try to reconnect to an interface which it
3245 cannot reach and report an error to the log.
3246 Details : Extra peer list entries will be created by lconf with some peers
3247 unreachable. lconf now validates the peer before adding it.
3250 Frequency : Only if a default stripe is set on the filesystem root.
3252 Description: Setting a default stripe on the filesystem root prevented the
3253 filesystem from being remounted.
3254 Details : The client was sending extra request flags in the root getattr
3255 request and did not allocate a reply buffer for the dir EA.
3258 Frequency : occasional, higher if lots of files are accessed by one client
3259 Bugzilla : 6159, 6097
3260 Description: Client trips assertion regarding lsm mismatch/magic
3261 Details : While revalidating inodes the VFS looks up inodes with ifind()
3262 and in rare cases can find an inode that is being freed.
3263 The ll_test_inode() code will free the lsm during ifind()
3264 when it finds an existing inode and then the VFS later attaches
3265 this free lsm to a new inode.
3269 Bugzilla : 6422, 7030
3270 Description: MDS deadlock between mkdir and client eviction
3271 Details : Creating a new file via mkdir or mknod (starting a transaction
3272 and getting the ns lock) can deadlock with client eviction
3273 (gets ns lock and trying to finish a synchronous transaction).
3276 Frequency : occasional
3277 Description: While starting a server, the fsfilt_ext3 module could not be
3279 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3280 kernels. Previously, lconf would still use the ext3 name
3281 when trying to load modules. Now, it will correctly use
3282 ext3 on 2.4 and ldiskfs on 2.6.
3284 Severity : enhancement
3285 Description: The default stripe count has been changed to 1
3286 Details : The interpretation of the default stripe count (0, to lfs
3287 or lmc) has been changed to mean striping across a single
3288 OST, rather than all available. For general usage we have
3289 found a stripe count of 1 or 2 works best.
3291 Severity : enhancement
3292 Description: Add support for compiling against Cray portals.
3293 Details : Conditional compiling for some areas that are different
3297 Frequency : occasional
3298 Bugzilla : 6409, 6834
3299 Description: Creating files with an explicit stripe count may lead to
3300 a failed assertion on the MDS
3301 Details : If some OSTs are full or unavailable, creating files may
3302 trigger a failed assertion on the MDS. Now, Lustre will
3303 try to use other servers or return an error to the
3307 Frequency : occasional
3309 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3310 caused lock timeout during readahead (since 1.4.2).
3311 Details : Processes doing readahead might match a lock that hasn't been
3312 granted yet if there are overlapping and conflicting lock
3313 requests. The readahead process waits on ungranted lock
3314 (original lock is CBPENDING), while OST waits for that process
3315 to cancel CBPENDING read lock and eventually evicts client.
3317 Severity : enhancement
3319 Description: Initial enabling of flock support for clients
3320 Details : Implements fcntl advisory locking and file status functions.
3321 This feature is provided as an optional mount flag (default
3322 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3323 locking are implemented yet, and those that are are not guaranteed
3324 to be completely correct in production environments.
3325 mount -t lustre -o [flock|noflock] ...
3328 Frequency : occasional
3330 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3331 assertion in the ext3 JBD (journaling) layer.
3332 Details : The b_committed_data struct is protected by the big kernel lock
3333 in 2.4 kernels, serializing journal_commit_transaction() and
3334 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3335 access to this struct. In 2.6 kernels there is finer grained
3336 locking to improve SMP performance of the JBD layer.
3340 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3345 Description: merge and cleanup kernel patches.
3348 -----------------------------------------------------------------------------
3350 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3352 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3353 versions of Lustre. Please update all servers and clients to
3354 version 1.4.2 or later at the same time. You must also run
3355 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3356 to update the configuration logs.
3358 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3359 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3360 - swab llog records of type '0' so we get proper header size/idx (5861)
3361 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3362 - fix rename of one directory over another leaking an inode (5953)
3363 - avoid SetPageDirty on 2.6 (5981)
3364 - don't re-add just-being-destroyed locks to the waiting list (5653)
3365 - when creating new directories, inherit the parent's custom
3366 striping settings if present parent (3048)
3367 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3368 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3369 - don't hold client locks on temporary worklist from l_lru (5666)
3370 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3371 - automatically evict dead clients (5921)
3372 - Update file size properly in create+truncate+fstat case (6196)
3373 - Do not unhash mountpoint dentries, do not allow removal of
3375 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3376 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3377 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3378 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3379 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3380 - Correctly handle reads of files with no objects (6243)
3381 - lctl recover will also mark a device active if deactivate used (5933)
3383 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3384 - allow --write-conf on an MDS with different nettype than client (5619)
3385 - don't write config llogs to MDS for mounts not from that MDS (5617)
3386 - lconf should create multiple TCP connections from a client (5201)
3387 - init scripts are now turned off by default; run chkconfig --on
3388 lustre and chkconfig --on lustrefs to use them
3389 - upcalls are no longer needed for clients to recover to failover
3391 - add --abort-recovery option to lconf to abort recovery on device
3393 - add support for an arbitrary number of OSTs (3026)
3394 - Quota support protocol changes.
3395 - forward compatibility changes to wire structs (6007)
3396 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3397 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3398 - improved "lustre status" script
3399 - initialize blocksize for non-regular files (6062)
3400 - added --disable-server and --disable-client configure options (5782)
3401 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3402 - Vanilla 2.4.29 support
3403 - increase maximum number of obd devices to 520 (6242)
3404 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3405 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3406 SLES 9 2.6.5 kernels
3407 - limit stripes per file to 160 (the maximum EA size) (6093)
3409 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3412 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3413 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3414 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3415 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3416 - small ext3 extents cleanups and fixes (5733)
3417 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3418 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3419 - avoid needless client->OST connect, fix handle mismatch (5317)
3420 - fix DLM error path that led to out-of-sync client, long delays (5779)
3421 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3422 - fix several locking issues related to i_size (5492,5624,5654,5672)
3423 - don't move pending lock onto export if it is already evicted (5683)
3424 - fix kernel oops when creating .foo in unlinked directory (5548)
3425 - fix deadlock in obdfilter statistics vs. object create (5811)
3426 - use time_{before,after} to avoid timer jiffies wrap (5882)
3427 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3428 - Do not leak request if server was not able to process it (5154)
3429 - If mds_open unable to find parent dir, make that negative lookup(5154)
3430 - don't create new directories with extent-mapping (5909, 5936)
3432 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3433 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3434 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3435 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3436 - support for RHEL4 kernel on i686 (5773)
3437 - provide error messages when incompatible logs are encountered (5898)
3439 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3440 * version 1.4.0.10 (1.4.1 release candidate 1)
3442 - don't keep a lock reference when lock is not granted (4238)
3443 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3444 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3445 - drop import inflight refcount on signal_completed_replay error (5255)
3446 - unlock page after async write error during send (3677)
3447 - handle missing objects in filter_preprw_read properly (5265)
3448 - no transno return for symlink open, don't save no-trasno open (3440)
3449 - don't try to complete elan receive that already failed (4012)
3450 - free RPC server reply state on error (5406)
3451 - clean up thread from ptlrpc_start_thread() on error (5160)
3452 - readahead could read extra page into cache that wasn't ejected (5388)
3453 - prevent races in class_attach/setup/cleanup/detach (5260)
3454 - don't dereference de->d_inode after l_dput of de (5458)
3455 - use "int" for stripe value returned from lock_to_stripe (5544)
3456 - mballoc allocation and error-checking fixes in 2.6 (5504)
3457 - block device patches to fix I/O request sizes in 2.6 (5482)
3458 - look up hostnames for IB nals (5602)
3459 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3460 - don't start multiple acceptors for the same port (5277)
3461 - fix incorrect LASSERT in mds_getattr_name (5635)
3462 - export a proc file for general "ping" checking (5628)
3463 - fix "lfs check" to not block when the MDS is down (5628)
3465 - service request history (4965)
3466 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3467 - create an "evict_client" /proc entry on OSTs, like the MDS has
3468 - fix mount usage message, return errors per mount(8) (5168)
3469 - change grep [] to grep "[]" in tests so they work in more UMLs
3470 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3471 - remove ancient LOV_MAGIC_V0 EA support (5047)
3472 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3473 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3474 - allow readahead allocations to fail when low on memory (5383)
3475 - mmap locking landed again, after considerable improvement (2828)
3476 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3478 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3481 - send OST transaction number in read/write reply to free req (4966)
3482 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3483 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3484 - the watchdog thread now runs as interruptible (5246)
3485 - flock/lockf fixes (but it's still disabled, pending 5135)
3486 - don't use EXT3 constants in llite code (5094)
3487 - memory shortage at startup could cause assertion (5176)
3489 - reorganization of lov code
3490 - single portals codebase
3492 - add extents/mballoc support (5025)
3493 - direct I/O reads in the obdfilter (4048)
3494 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3496 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3499 - send OST transaction number in read/write reply to free req (4966)
3500 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3501 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3502 - the watchdog thread now runs as interruptible (5246)
3503 - handle missing objects in filter_preprw_read properly (5265)
3504 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3505 - drop import inflight refcount on signal_completed_replay error (5255)
3506 - unlock page after async write error during send (3677)
3507 - return original error code on reconstructed replies (3761)
3508 - no transno return for symlink open, don't save no-trasno open (3440)
3510 - add pid to ldlm debugging output (4922)
3511 - bump the watchdog timeouts -- we can't handle 30sec yet
3512 - extra debugging for orphan dentry/inode bug (5259)
3514 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3517 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3518 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3519 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3520 - fix NULL dereference in /proc/sys/portals/routes (4827)
3521 - allow failed mdc_close() operations to be interrupted (4561)
3522 - stop precreate on OST before MDS would time out on it (4778)
3523 - don't send partial-page writes before EOF from client (4410)
3524 - discard client grant for sub-page writes on large-page clients (4520)
3525 - don't free dentries not owned by NFS code, check generation (4806)
3526 - fix lsm leak if mds_create_objects() fails (4801)
3527 - limit debug_daemon file size, always print CERROR messages (4789)
3528 - use transno after validating reply (3892)
3529 - process timed out requests if import state changes (3754)
3530 - update mtime on OST during writes, return in glimpse (4829)
3531 - add mkfsoptions to LDAP (4679)
3532 - use ->max_readahead method instead of zapping global ra (5039)
3533 - don't interrupt __l_wait_event() during strace
3535 - add software watchdogs to catch hung threads quickly (4941)
3536 - make lustrefs init script start after nfs is mounted
3537 - fix CWARN/ERROR duplication (4930)
3538 - return async write errors to application if possible (2248)
3539 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3540 - print NAL number in %x format (4645)
3541 - update barely-supported suse-2.4.21-171 series (4842)
3542 - support for sles 9 %post scripts
3543 - support for building 2.6 kernel-source packages
3544 - support for sles km_* packages
3546 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3549 - ignore -ENOENT errors in osc_destroy (3639)
3550 - notify osc create thread that OSC is being cleaned up (4600)
3551 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3552 - reconstruct ost_handle() like mds_handle() (4657)
3553 - create a new thread to do import eviction to avoid deadlock (3969)
3554 - let lconf resolve symlinked-to devices (4629)
3555 - don't unlink "objects" from directory with default EA (4554)
3556 - hold socknal file ref over connect in case target is down (4394)
3557 - allow more than 32000 subdirectories in a single directory (3244)
3558 - fix blocks count for O_DIRECT writes (3751)
3559 - OST returns ENOSPC from object create when no space left (4539)
3560 - don't send truncate RPC if file size isn't changing (4410)
3561 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3562 - bind to privileged port in socknal and tcpnal (3689)
3564 - rate limit CERROR/CWARN console message to avoid overload (4519)
3565 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3566 - basic mmap support (3918)
3567 - kernel patch series update from b1_4 (4711)
3569 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3572 - avoid crash during MDS cleanup with OST shut down (2775)
3573 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3574 - don't use bad inodes on the MDS (3744)
3575 - dynamic object preallocation to improve recovery speed (4236)
3576 - don't hold spinlock over lock dumping or change debug flags (4401)
3577 - don't zero obd_dev when it is force cleaned (3651)
3578 - print grants to console if they go negative (4431)
3579 - "lctl deactivate" will stop automatic recovery attempts (3406)
3580 - look for existing locks in ldlm_handle_enqueue() (3764)
3581 - don't resolve lock handle twice in recovery avoiding race (4401)
3582 - revalidate should check working dir is a directory (4134)
3584 - don't always mark "slow" obdfilter messages as errors (4418)
3586 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3589 - don't close LustreDB during write_conf until it is done (3860)
3590 - fix typo in lconf for_each_profile (3821)
3591 - allow dumping logs from multiple threads at one time (3820)
3592 - don't allow multiple threads in OSC recovery (3812)
3593 - fix debug_size parameters (3864)
3594 - fix mds_postrecov to initialize import for llog ctxt (3121)
3595 - replace config semaphore with spinlock (3306)
3596 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3597 - don't allow enqueue to complete on a destroyed export (3822)
3598 - down write_lock before checking llog header bitmap (3825)
3599 - recover from lock replay timeout (3764)
3600 - up llog sem before sending rpc (3652)
3601 - reduce ns lock hold times when setting kms (3267)
3602 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3603 - fix NULL deref and obd_dev leak on setup error (3312)
3604 - replace some LBUG about llog ops with error handling (3841)
3605 - don't match INVALID dentries from d_lookup and spin (3784)
3606 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3607 - fix invalid assertion in ptlrpc_set_wait (3880)
3609 - add libwrap support for the TCP acceptor (3996)
3610 - add /proc/sys/portals/routes for non-root route listing (3994)
3611 - allow setting MDS UUID in .xml (2580)
3612 - print the stack of a process that LBUGs (4228)
3614 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3617 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3618 - make sure to unset replay flag from failed open requests (3440)
3619 - if default stripe count is 0, use OST count for inode size (3636)
3620 - update parent mtime/ctime on client for create/unlink (2611)
3621 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3622 - free recovery state on server during a forced cleanup (3571)
3623 - unregister_reply for resent reqs (3063)
3624 - loop back devices mounting and status check on 2.6 (3563)
3625 - fix resource-creation race that can provoke i_size == 0 (3513)
3626 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3627 - more debugging for page-accounting assertion (3746)
3628 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3629 - avoid deadlock after precreation failure (3758)
3630 - fix race and lock order deadlock in orphan handling (3450, 3750)
3631 - add validity checks when grabbing inodes from l_ast_data (3599)
3633 - add /proc/.../recovery_status to obdfilter (3428)
3634 - lightweight CDEBUG infrastructure, debug daemon (3668)
3635 - change default OSC RPC parameters to be better on small clusters
3636 - turn off OST read cache for files smaller than 32MB
3637 - install man pages and include them in rpms (3100)
3638 - add new init script for (un)mounting lustre filesystems (2593)
3639 - run chkconfig in %post for init scripts (3701)
3640 - drop scimac NAL (unmaintained)
3642 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3645 - clean kiobufs before and after use (3485)
3646 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3647 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3648 - change init script to fail more clearly if not run as root (1528)
3649 - allow clients to reconnect during replay (1742)
3650 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3651 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3652 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3653 - fix cond_resched (3554)
3654 - only evict unfinished clients after recovery (3515)
3655 - allow bulk resend, prevent data loss (3570)
3656 - dynamic ptlrpc request buffer allocation (2102)
3657 - don't allow unlinking open directory if it isn't empty (2904)
3658 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3659 - remove extraneous obd dereference causing LASSERT failure (3334)
3660 - don't use get_cycles() when creating temp. files on the mds (3156)
3661 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3662 - handle EEXIST for set-stripe, set proper directory name (3336)
3664 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3665 - fix ksocknal_fmb_callback() error messages (2918)
3667 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3670 - don't copy lvb into (possibly NULL) reply on error (2983)
3671 - don't deref dentry after dput, don't free lvb on error (2922)
3672 - use the kms to determine writeback rpc length (2947)
3673 - increment oti_logcookies when osc is inactive (2948)
3674 - update client's i_blocks count via lvb messages (2543)
3675 - handle intent open/close of special files properly (1557)
3676 - mount MDS with errors=remount-ro, like obdfilter (2009)
3677 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3678 - don't use cancelling-locks' kms values (2947)
3679 - use highest lock extent for kms, not last one (2925)
3680 - don't dereference ERR_PTR() dentry in error handling path (3107)
3681 - fix thread race in portals_debug_dumplog() (3122)
3682 - create lprocfs device entries at setup instead of at attach (1519)
3683 - common AST error handler, don't evict client on completion race (3145)
3684 - zero nameidata in detach_mnt in 2.6 (3118)
3685 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3686 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3687 - initialize RPC timeout timer earlier for 2.6 (3219)
3688 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3689 - print nal/nid for unknown nid (3258)
3690 - additional checks for oscc recovery before doing precreate (3284)
3691 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3692 - don't crash in mdc_close for bad permissions on open (3285)
3693 - zero i_rdev for non-device files (3147)
3694 - clear page->private before handing to FS, better assertion (3119)
3695 - tune the read pipeline (3236)
3696 - fix incorrect decref of invalidated dentry (2350)
3697 - provide read-ahead stats and refine rpc in flight stats (3328)
3698 - don't hold journal transaction open across create RPC (3313)
3699 - update atime on MDS at close time (3265)
3700 - close LDAP connection when recovering to avoid server load (3315)
3701 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3702 - don't leak open file on MDS after open resend (3325)
3703 - serialize filter_precreate and filter_destroy_precreated (3329)
3704 - loop device shouldn't call sync_dev() for nul device (3092)
3705 - clear page cache after eviction (2766)
3706 - resynchronize MDS->OST in background (2824)
3707 - refuse to mount the same filesystem twice on same mountpoint (3394)
3708 - allow llmount to create routes for mounting behind routers (3320)
3709 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3710 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3711 - fix rare problem with rename on htree directories (3417)
3713 - allow default OST striping configuration per directory (1414)
3714 - fix compilation for qswnal for 2.6 kernels (3125)
3715 - increase maximum number of MDS request buffers for large systems
3716 - change liblustreapi to be useful for external progs like lfsck (3098)
3717 - increase local configuration timeout for slow disks (3353)
3718 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3720 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3723 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3724 - don't overwrite extent policy data in reply if lock was blocked (2901)
3725 - drop filter export grants atomically with removal from device (2663)
3726 - del obd_self_export from work_list in class_disconnect_exports (2908)
3727 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3728 - swab reply message in mdc_close, other PPC fixes (2464)
3729 - fix destroying of named logs (2325)
3730 - overwrite old logs when running lconf --write_conf (2264)
3731 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3732 - fix race in target_handle_connect (2898)
3733 - mds_reint_create() should take same inode create lock (2926)
3734 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3735 - don't close files for self_export to avoid uninitialized obd (2936)
3736 - allow MDS with the same name as client node (2939)
3737 - hold dentry reference for closed log files for unlink (2325)
3738 - reserve space for all logs during transactions (2059)
3739 - don't evict page beyond end of stripe extent (2925)
3740 - don't oops on a deleted current working directory (2399)
3741 - handle hard links to targets without a parent properly (2517)
3742 - don't dereference NULL lock when racing during eviction (2867)
3743 - don't grow lock extents when lots of conflicting locks (2919)
3745 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3748 - account for cache space usage on clients to avoid data loss (974)
3749 - lfsck support in lustre kernel code (2349)
3750 - reduce journal credits needed for BRW writes (2370)
3751 - orphan handling to avoid losing space on client/server crashes
3752 - ptlrpcd can be blocked, stopping ALL progress (2477)
3753 - use lock value blocks to assist in proper KMS, faster stat (1021)
3754 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3755 - recovery for initial connections (2355)
3756 - fixes for mds_cleanup_orphans (1934)
3757 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3758 - block all file creations until orphan recovery completes (1901)
3759 - client remove rq_connection from request struct (2423)
3760 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3761 - recovery timer race (2670)
3762 - mdc_close recovey bug (2532)
3763 - ptlrpc cleanup bug (2710)
3764 - mds timeout on local locks (2588)
3765 - namespace lock held during RPCs (2431)
3766 - handle interrupted sync write properly (2503)
3767 - don't try to handle a message that hasn't been replied to (2699)
3768 - client assert failure during cleanup after abort recovery (2701)
3769 - leak mdc device after failed mount (2712)
3770 - ptlrpc_check_set allows timedout requests to complete (2714)
3771 - wait for inflight reqs when ptlrpcd finishes (2710)
3772 - make sure unregistered services are removed from the srv_list
3773 - reset bulk XID's when resending them (caught by 1138 test)
3774 - unregister_bulk after timeout
3775 - fix lconf error (2694)
3776 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3777 - readahead locks pages, leaves pending causing memory pressure (2673)
3778 - increase OST request buffers to 4096 on large machines (2729)
3779 - fix up permission of existing directories in simple_mkdir (2661)
3780 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3781 - don't assign transno to errored transactions (2742)
3782 - don't delete objects on OST if given a bogus objid from MDS (2751)
3783 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3784 - if rq_no_resend, then timeout request after recovery (2432)
3785 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3786 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3787 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3788 - MDS recovery shouldn't delete objects that it has given out (2730)
3789 - if enqueue arrives after completion, don't clobber LVB (2819)
3790 - don't unlock pages twice when trigger_group_io returns error (2814)
3791 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3792 - don't write pages to disk if there was an error (1450)
3793 - don't ping imports that have recovery disabled (2676)
3794 - take buffered bytes into account when balancing socknal conn (2817)
3795 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3796 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3797 - remove little-endian swabbing of llog records (1987)
3798 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3799 - retry reposting request buffers if they fail (1191)
3800 - grow extent at grant time to avoid granting a revoked lock (2809)
3801 - lock revoke doesn't evict page if covered by a second lock (2765)
3802 - disable VM readahead to avoid reading outside lock extents (2805)
3804 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3805 - updated kernel patches for hp-2.4.20 kernel (2681)
3807 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3810 - fix truncated write corruption (2366)
3811 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3813 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3814 - ensure len(uuid) < 37 in lmc (1171)
3815 - fix ia64 OOPS in llog_test (2255)
3816 - zero end of page at obdfilter for partial page writes (2648)
3817 - don't leave stale dentries around after renames (bug 2428)
3818 - fix timeouts when evicting a client with a single lock held (2642)
3819 - set deadline for the initial HELLO message to drain (2634)
3820 - print out dotted-quad IP addresses in the socknal (2302)
3822 - additional debugging for MDS client eviction problem (2443)
3823 - fix mkfsoptions support for osts (2603, 2604)
3825 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3828 - add series for the vanilla 2.6.0 kernel
3829 - add series for the vanilla 2.4.24 kernel
3830 - add series for a cray x86/64 UL kernel drop
3831 - fix xattr patches for the vanilla 2.4.19 series
3833 - generate true UUIDs in lmc (1171)
3834 - have portals stack dumping break in UML (2466)
3835 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3836 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3837 - wake callback waiting threads on client eviction (2460)
3838 - Add --ptldebug and --subsystem to lmc (1719)
3839 - update assertion to allow safe interrupt allocation
3840 - set rq_no_resend for cancel requests (2432)
3841 - recalculate ptlrpcd timeout after resend (2494)
3842 - call vfs_rmdir when removing pending directories (2368)
3843 - fix renaming a file to itself (2429)
3844 - lmc creates a default one-stripe lov (2454)
3845 - expand procfs space to handle large clusters (2326)
3846 - increase UML stack to avoid overflow
3847 - update lconf's list of debug and subsystem masks
3848 - fix lfs find --obd (2510)
3849 - /proc tunable for disabling filter read caching (2591)
3850 - stop rpm packages from altering slapd.conf (2301)
3851 - disable nagle in the socknal under 0conf (2578)
3852 - choose mds inode size based on stripe count (2572)
3853 - fix kernel-source rpm problems (2516)
3855 - add --disable-doc to avoid pdf generation (2421)
3856 - update documentation, tests, type-os, comments
3857 - avoid format warnings on ia64
3858 - remove the TOE NAL
3859 - tiny code cleanups by removing unused fields
3861 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3864 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3865 - remove the most blatant lies from BUILDING file (2371)
3866 - change default debug level to reasonable production setting
3867 - reduce client side cache size to reduce cache flush time
3868 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3869 - make TCP zerocopy and pinger support enabled by default (2476)
3870 - sync writes completed after process exits caused crashes (2319)
3871 - maintain correct mount count on the MDS (2356)
3872 - backout 1557, because 2316 wasn't really fixed
3873 - better file I/O statistics gathering in /proc
3874 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3875 - another kernel patch to fix zero-copy TCP function export
3876 - don't take duplicate lock when processing re-sent getattr (2420)
3877 - lctl uses obd_self_export instead of creating new conn (2353)
3878 - MDS/OST recovery case which requires object creation asserted (2425)
3879 - move lfs from /usr/sbin to /usr/bin in packages
3880 - fix race between mds_client_add and mds_client_free (2417)
3881 - use kmalloc instead of slabs in portals (2430)
3882 - don't create duplicate records when a failover MDS is present (2442)
3883 - remove unnecessary mount age check (2332)
3884 - don't remove directory inodes from locks prematurely (2451)
3885 - don't break if MDS service name is the same as hostname (2103)
3886 - fix races in client write RPC generation when cache full (2482)
3888 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3891 - remove now-unused request->rq_obd (278)
3892 - if an allocation fails, print out how much memory we've used (1933)
3893 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3894 - add the "configurable stack size" patch to most series files (1256)
3895 - ability to write large log records, for 100+ OST configs (2306)
3896 - fix NULL deref when filter_prep fails (2314)
3897 - fix operator precedence error in filter_sync
3898 - dynamic allocation of socknal TX descriptors (2315)
3899 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3900 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3901 - ensure that CWARN messages in Portals always get to the syslog
3902 - __init/__exit are not for prototype decls (ldlm_init/exit)
3903 - x86-64 compile warning fixes
3904 - fix gateway LMC keyword conflict (2318)
3905 - fix MDS lock inversions in getattr/reint paths (1844)
3906 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3907 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3908 - DLM race condition prevented some lock evictions (2328)
3909 - ENOMEM detection and retry on socknal sends (2230)
3910 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3911 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3912 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3913 - better error messages when a client is rejected during recovery (1505)
3914 - avoid cancelling locks which were never granted, after failure (2330)
3915 - fix i_sem/journal inversion in mds_client_add (2333)
3916 - fix truncate/getattr lock cycle deadlock (2334)
3917 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3918 - fix two rare exit paths which could leak an l_lock() ref (2321)
3919 - fencepost error in MDS/OST orphan recovery (2226)
3920 - make log record alignment 8 bytes (1988)
3921 - lstripe now fails when requested offset > ost_count (2237)
3922 - ensure that all kernel series have a complete list.h (1607)
3923 - fix crashes in special-file operations (2316)
3924 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3926 - allow configurable automake binary, for testing new versions
3927 - small update to the lfs documentation
3929 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3931 * fix negative export reference count in fsfilt_sync (2312)
3933 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3934 * release candidate 0.9.1
3936 - orphans are moved into the PENDING directory for possible recovery
3937 - replayed opens now open by fid for orphan/rename safety (1042)
3938 - last close of an orphan inode generates a transno (683)
3939 - chdir() and mount() now pin the directory entry (1020)
3940 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3941 - discard very old requests without processing them (1502)
3942 - remove some common, well-understood CERRORs (1505)
3943 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3944 - clear "grant" flags in OST replies until OST grant code lands (1644)
3945 - fix read performance by not clobbering i_blksize on client (1598)
3946 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3947 - make LRU size a /proc tunable, clears locks when reduced (707)
3948 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3949 - server locks take references on exports now (1558)
3950 - build fixes for 2.4.20-rh trees (1663)
3951 - return an error from lov_create if all OSCs are inactive (1751)
3952 - fix import levels when a reconnect happens without a timeout (1597)
3953 - exit early from mds_open if we get a lookup error (1749)
3954 - partial page read at EOF wouldn't wait for disk before sending (1642)
3955 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3956 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3957 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3958 - write/truncate lock inversion (1639)
3959 - Don't auto-load obdclass, portals modules during cleanup (1495)
3960 - fix timestamps from jumping to "now" (1763)
3961 - extra journal assertions (1648)
3962 - add an extra multiunlink test (1771)
3963 - fix read_record/write_record API (1776)
3964 - fix leak of offset_extent, possible incorrect i_size later (1772)
3965 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3966 - Debugging for the kqswnal_get_idle_tx problems (1820)
3967 - Allow recovery to be attempted multiple times (1536)
3968 - Write out MDS last_rcvd file after it is first created (1600)
3969 - Fix tx_descriptor leak in failed transmit situations (1827)
3970 - ext3 journaling fixes for assertion failure after IO error (1871)
3971 - class_export_put() on freed export after completion AST error (1896)
3972 - Fix revalidate looping in VFS (1322)
3973 - Don't access a freed export during MDS_REINT timeout (1521)
3974 - Add open-unlink recovery support on the MDS (1673,1764)
3975 - Return an error if no MDS data was read from last_rcvd (1946)
3976 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3977 - Don't setup a disk device that doesn't match exported UUID (317)
3978 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3979 - avoid committing NULL handle in force close
3980 - local.sh is now a one-stripe LOV configuration
3981 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3982 - fix ext3 htree duplicate directory entry corruption (1516)
3983 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3984 - update to new LOV EA format (2097)
3985 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3987 2003-06-15 Phil Schwan <phil@clusterfs.com>
3990 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3991 - per-import/export recovery handling (958, 931, 959)
3992 - multiple last-rcvd slots, for serving multiple FSes (949)
3993 - connections are again shared between multiple imp/exports (963, 964)
3994 - "umount -f" would hang if any requests needed to be sent (393, 978)
3995 - avoid pinning large req buffer by copying for queued messages (989)
3996 - add "uuid" to "lctl device" command to help upcalls (991)
3997 - "open" RPCs with transnos would confuse recovery counters (1037)
3998 - do proper endian conversion of all wire messages (288, 340, 891)
3999 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4000 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4001 - fix ptlrpc_connection leak in target_handle_connect (1174)
4002 - fix import refcounting bug in OST and MDS cleanup (1134)
4003 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4004 - fix very unlikely obd_types race condition (501)
4005 - remove osc_open hack for echo_client (1187)
4006 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4007 - a failure in read_inode2 leads to deadlock (1139)
4008 - cancel ack-locks as soon as transaction is committed (1072)
4009 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4010 - make sure to commitrw after any preprw to avoid deadlock (1162)
4011 - failing to execute a file in a lustre FS would lock inode (1203)
4012 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4013 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4014 - obd_uuid2tgt would walk off the end of the list (1255)
4015 - on IA64 the osc would give portals incorrect bulk size (1258)
4016 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4017 - fix lock inversion caused by new llite matching code (1282)
4018 - limit the number of dirty pages on a client to 10MB (1286)
4019 - timed out locks were not being corrected cancelled (1289)
4020 - fix O_DIRECT above 4GB on IA-32 (1292)
4021 * major user-visible changes
4022 - fail out/fail over policy now controlled by the upcall (993)
4024 - add OBD_PING to check server availability and failure (954)
4025 - lustre messages are now sent in sending host order (288, 340, 891)
4026 - add eadatalen to MDS getattr reply (340)
4027 - OST read replies may contain second buffer, with per-page status (593)
4029 2003-03-11 Phil Schwan <phil@clusterfs.com>
4032 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4033 - failed open()s could cause deadlock; fixed (867, 869)
4034 - stop cancelling OST locks when files are closed (481)
4035 - overlapping XID spaces caused network corruption (851, 853)
4036 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4037 - setattr_raw added, to do single-RPC, server-side setattrs
4038 - lmc/lconf syntax change for OST UUIDs
4039 - fix crashy race condition between ptlrpc_free_req and osc_close
4040 - don't use request in mdc_enqueue if we hit a timeout (889)
4041 - don't set the inode i_size for regular files from the MDS (896)
4042 - handle out of order completion AST (842)
4043 - don't LBUG if a lock request times out after receiving AST (913)
4044 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4045 - if a bad lock AST arrives, send an error instead of dropping entirely
4046 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4047 - fix leak in bulk IO when only partially completed (899, 900, 926)
4048 - fix O_DIRECT for ia64 (55)
4049 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4050 - C-z after timeout could hang a process forever; fixed (977)
4052 - client-side I/O cache (678, 924, 929, 941, 970)
4054 - READPAGE and SETATTRs which don't take server-side locks get
4057 2003-02-11 Phil Schwan <phil@clusterfs.com>
4060 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4061 - Fix fsfilt_extN_readpage() to read a full page of directory
4062 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4063 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4064 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4065 - Most intent operations were removed, in favour of a new RPC mode
4066 that does a single RPC to the server and bypasses most of the VFS
4067 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4068 - Aggressively cancel local locks on DLM servers
4069 - mds_reint_unlink sends EA to the client if it's the last nlink.
4070 client uses that EA to unlink OST objects.
4071 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4072 - recursive symlinks were fixed (439)
4073 - fixed NULL deref in DEBUG_REQ
4074 - filter_update_lastobjid no longer calls sync, which annoyed extN
4075 - fixed multi-client small-writes to a single file problem (445)
4076 - fixed mtime updates during file writes (607)
4077 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4078 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4079 - fixed Linux OST size reporting problem (444, 656)
4080 - OST now updates object mtime with writes or setattr (607, 619)
4081 - client verifies file size before zeroing page past EOF (445)
4082 - OST now writes last allocated objid to disk with allocation (108)
4083 - LOV on echo now works (409)
4085 - mds_reint_unlink sends a new buffer, with the EA included. this
4086 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4087 set if a regular file was being unlinked, and it was the last link
4088 - use PtlGet from the target for bulk writes (315)
4089 - OST now updates object mtime with writes or setattr (607, 619)
4090 - LDLM now has a grant-time callback to revalidate locked items, if
4092 - Many MDS operations were reorganized to combat race conditions
4094 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4095 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4097 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4100 - Fully reactivate OST imports after reconnection (512, others)
4101 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4102 - More graceful error handling for truncating on dead OST (515)
4103 - Don't error out unless we're actually accessing dead stripes (474)
4104 - Fix garbage sizes when stripes are missing (410)
4105 - LRU counters were broken, causing constant lock purge (433, 432)
4106 - garbage on read from stripes with failed OSTs (441)
4107 - mark OSCs as active before reconnecting during recovery (438)
4108 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4109 - lfind did not preserve OST order in output (443)
4110 - symlinks cause hung clients, incorrect data (439)
4111 - stop dereferencing request after dropping refcount (457)
4112 - don't LASSERT(spin_is_locked) on non-SMP (455)
4113 - fixes for many rename() bugs
4114 - fstat didn't correctly synchronize attributes (399)
4115 - server must handle lock cancellation during blocking AST prep (487)
4116 - bulk descriptors were free()d too soon (511)
4117 - fix paths in lconf, which would load incorrect modules (451, 507)
4118 - fix confusing lconf 'host not found' error message (386)
4119 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4120 - fix race condition in mdc_blocking_ast() for inode access (526)
4121 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4122 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4123 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4124 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4125 - fix lstripe in several ways (kernel side) (527)
4126 - fix request leak in ldlm_cli_enqueue (262)
4127 - incorrect OSC was marked inactive after OST failure
4128 - call mds_fs_cleanup before unmounting filesystem (524)
4129 - fix races between taking ns_lock and ldlm_lock_change_resource
4130 - fix races updating LOV export open file list
4131 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4132 - fix recovery NULL deref in ldlm_cli_cancel_unused
4133 - fix some DLM races by using new hash table for lock handles (419)
4134 - permit the client to specify desired inodes, at replay
4135 - duplicate requests when we queue them for replay reintegration
4136 - fix last_rcvd offset calculation
4137 - sync after each recovered transaction, so we always make progress
4138 - never, not always, ERESTART requests without transnos
4139 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4141 - skip replay if the MDS says that the client is already connected
4142 - don't check for a recovery-enabled export to match lctl's UUID
4143 - don't INC_USE_COUNT for phantom exports
4144 - don't crash when cleaning up phantom exports (567)
4145 - don't double-finish or set replay data for errored mdc_open requests
4146 - abort requests when they time out, so we don't get old replies
4147 - send/receive replies for AST messages again
4148 - if the client says that it doesn't have the lock, cancel it on the
4150 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4151 mark it as destroyed, it will all work out when decref is called
4152 - fix module use counts (22, 581)
4154 - ASTs now expect a reply (server cancels lock on error reply)
4156 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4159 - fix many simultaneous client startup (392)
4160 - fix dentry->d_it clobbering
4161 - credentials weren't being shipped for readdir/getattr operations
4162 - remove invalid assertions triggered during some concurrent MD
4164 - proper Lustre versions added (336, 389)
4165 - fix memory leak for create error case (398)
4166 - fix LOV locking bug that would get cli/srv out of sync
4167 - fix echo client over LOV (409)
4168 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4169 - fix double-O_EXCL intent crash (424)
4170 - avoid sending multiple lock CANCELs (352)
4172 - MDS can do multi-client recovery (modulo bugs in new code)
4174 - many updates, edits, cleanups
4176 2002-11-18 Phil Schwan <phil@clusterfs.com>
4179 - fix null d_it dereference (346)
4180 - fix full OST/dbench hang (333)
4181 - fix permission problem with file removal (286)
4182 - fix removal of OSCs from LOV when they fail
4183 - fix NULL deref during bulk timeout (214)
4184 - fix problems related to multiple filesystems on one MDS (241)
4185 - fixed serious subtle metadata locking bugs
4186 - free locks on clients when inodes are removed due to memory
4188 - fix inode pointer in lock data (285)
4189 - partial support for multiple MDS on a single host (241)
4190 - data locks weren't cancelled at clear_inode time (290, 311)
4191 - intent locks could lead to unbounded lock growth (205)
4192 - added a maximum lock count, an LRU list, and a flusher
4193 - fix multiple rename (365)
4194 - properly abstracted the echo client
4195 - OSC locked 1 byte too many; fixed
4196 - rewrote brw callback code:
4197 - fixed recovery bugs related to LOVs (306)
4198 - fixed too-many-pages-in-one-write crash (191)
4199 - fixed (again) crash in sync_io_timeout (214)
4200 - probably fixed callback-related race (385)
4202 - Add capability to MDS protocol
4203 - LDLM cancellations and callbacks on different portals
4205 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4208 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4209 - defer open object destruction to close time (601981, 138)
4210 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4211 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4212 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4213 - don't call obd_setattr in ll_file_release for destroyed objects
4214 * protocol change to lustre_msg: move |version| and add |flags|
4215 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4216 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4217 * added replay of create, unlink, link and rename operations during
4218 MDS failover; recovery should be much more robust now
4219 * remove failed OSCs from LOVs (only lov_create uses this so far)
4220 * the lustre-HOWTO was brought (more) up to date (582544)
4222 2002-10-23 Phil Schwan <phil@clusterfs.com>
4225 - in-use dentries weren't being reused properly (617851)
4226 - prevent multiple LDLM setup (599178)
4227 - fix LOV size calculations for truncate (617853)
4228 - fix client handling of MDS intent errors (POSIX)
4229 - fix permission bug in lovstripe.c test (624321)
4230 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4231 - truncate past end of file could corrupt data
4232 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4233 - a race in recovery could return ETIMEDOUT to apps (623947)
4234 - building outside the source directory was fixed
4235 * the lustre-HOWTO was brought (more) up to date (582544)
4236 * major progress was made on recovery functionality
4238 2002-10-10 Phil Schwan <phil@clusterfs.com>
4241 - recovery deadlock fix
4242 - rm -rf causes LBUG fix (617817)
4243 - file open by multiple tasks fix (618962)
4244 - directory permissions bugs (602707 and 620007)
4245 - journal_stop fixed with locking (611313)
4246 - O_APPEND failures resolved (618273, perhaps 614459)
4247 - lconf PATH fix (619770)
4248 - IA64 build fix (621450)
4249 - RPC buffer sizes scale with amount of memory
4251 2002-10-01 Phil Schwan <phil@clusterfs.com>
4254 - locks would be cancelled without throwing away data pages,
4255 resulting in inconsistent data (605627)
4256 - inode attributes were not always being refreshed (605627, 612449)
4257 - lconf now continues to cleanup after lctl reports an error
4258 - MDS now enforces user permissions (602707)
4259 - lprocfs cleanup fixed, but not yet enabled (614157)
4260 - fixed infinite server hang, should a client not respond to an AST
4261 - avoid going into recovery if user calls readlink() with a buffer
4262 that's too small (613941)
4263 - AST RPCs no longer require replies (614867) -- this may be changed
4264 - don't crash server if client sends an IOV that's too big (611336)
4265 - fixed lock conversion deadlock (611892)
4266 - fixed the following of symlinks (614622)
4267 * recovery: the server can remove locks from a client that dies, other
4268 clients can make progress
4269 * more extN patch fixes
4270 * compile-time configurable ptlrpc buffer allocations
4272 - collaborative read cache document
4273 - Lustre Lite Performance CDR document-in-progress
4275 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4278 - fix typo in patch-2.4.18
4280 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4283 - clear ptlrpc request each time in handle_incoming_request()
4284 - unlink of files now destroys the object on the OST
4286 2002-09-19 Peter Braam <braam@clusterfs.com>
4288 * add hard link support
4289 * change obdfile creation method
4290 * kernel patch changed
4292 2002-09-19 Peter Braam <braam@clusterfs.com>
4295 - stack overflow bug in extN fixed
4297 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4299 * documentation updates
4300 - add man pages for config tools
4301 - update tests/README to describe testing with new config tools
4302 - finish metadata API descriptions
4303 * bug fixes and cleanups
4304 - statfs workaround for 16TB limit
4305 - LOV stripe allocation improved, can stripe on subset of OSTs
4306 - LOV file size/IO offset was wrong for files > 4GB in size
4307 - object EA data was being dropped, caused files to be unreadable
4308 - memory overflow with non-LOV OST caused memory corruption
4309 - fixed regression tests to work with new config tools, obdfilter
4310 - fixed bug when directory size became larger than 1 block
4311 - fixed bug (for single client case) when PWD was deleted
4312 - invalidate local directory pages when doing intent-based ops
4313 - avoid LDLM oops when lock callback contained bad data
4315 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4317 * documentation updates
4318 * bug fixes and cleanups
4319 - configuration tools
4322 - 64-bit compile warnings
4323 - 64-bit internal statfs data
4325 * test_brw on persistent OST devices
4327 * lprocfs (disabled)
4329 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4331 * documentation updates
4332 * bug fixes and cleanups
4333 * configuration tools
4335 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4338 * many small fixes to 0_5_4
4339 * io/network handling
4340 * thinkos in MDS operations
4342 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4345 * crucial basic fixes to 0.5.3
4346 * IOR, Iozone work over Elan
4349 2002-08-07 Phil Schwan <phil@clusterfs.com>
4350 * version 0_5_3, our first alpha
4351 * we use the new Portals iovs
4352 * documentation updates
4353 * bug fixes and cleanups
4354 * small changes in the DLM wire protocol
4356 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4357 * version 0_5_1 with some initial stability,
4358 * locking on MD and file I/O.
4359 * documentation updates
4360 * several bug fixes since 0.5.0
4361 * small changes in wire protocol
4363 2002-07-18 Phil Schwan <phil@clusterfs.com>
4365 * delivered as Lustre Light Alpha
4366 * fixed a crash after handling invalid MDS requests
4367 * fixed directory pages for architectures with non-4k pages sizes
4369 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4370 * release version v0_4_4
4371 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4373 * Fixes a number of interruption problems with OST operations.
4374 * Update documentation for portals header changes
4375 * Move all wire protocol structs/defines to lustre_idl.h
4376 * Fixes symlink length bug.
4377 * Add tcpdump to repository.
4379 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4380 * release version v0_4_3
4381 * Fixes statfs for inodes on extN.
4382 * Fixes bug in runtests which would delete /etc/hosts.
4383 * Use 64-bit object IDs wherever possible (not into VFS though)
4384 Remove ost_get_info, which is unused by lustre, and out of date.
4386 2002-07-03 Peter Braam <braam@clusterfs.com>
4387 * release version v0_4_2 Fixes a lookup error (type not passed)
4388 * move forward to head of Portals
4389 * move forward to latest Lustre kernel
4391 2002-06-25 Peter Braam <braam@clusterfs.com>
4392 * release version v0_4_1. Hopefully stable on single node use.