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).
17 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
18 Details : access to bitfield in structure is always rounded to long
19 and this produce problem with not atomic change any bit.
23 Description: Lustre make rpms failed.
24 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
25 builds Lustre rpms from tarball.
28 Frequency : rare on shutdown ost
30 Description: If llog cancel was not send before clean_exports phase, this can
31 produce deadlock in llog code.
32 Details : If llog thread has last reference to obd and call class_import_put
33 this produce deadlock because llog_cleanup_commit_master wait when
34 last llog_commit_thread exited, but this never success because was
35 called from llog_commit_thread.
39 Description: allow userland application know is lost one of stripes.
40 Details : fill lvb_blocks with error code on ost and return it to
41 application if error flag found.
45 Description: NULL lov_tgts causing MDS oops
46 Details : more safe checks for NULL lov_tgts for avoid oops.
48 Severity : enhancement
50 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
52 Severity : enhancement
54 Description: Add rhel5 support to HEAD.
58 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
59 Details : If an OST is down the MDS will hang indefinitely in
60 obd_statfs_rqset() waiting for the statfs data. While for
61 MDS QOS usage of statfs, it should not stuck in waiting.
63 Severity : enhancement
65 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
67 Severity : enhancement
69 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
73 Description: make mgs_setparam() handle fsname containing dash
74 Details : fsname containing a dash does not work with lctl conf_param
76 Severity : enhancement
78 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
80 Severity : enhancement
82 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
86 Frequency : rare, at shutdown
87 Description: access already free / zero obd_namespace.
88 Details : if client_disconnect_export was called without force flag set,
89 and exist connect request in flight, this can produce access to
90 NULL pointer (or already free pointer) when connect_interpret
91 store ocd flags in obd_namespace.
95 Frequency : only at startup
96 Description: not alloc memory with spinlock held.
97 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
100 Severity : enhancement
102 Description: make lustre randomly fail allocating memory
103 Details : Make lustre randomly failed allocating memory for testing purpose.
105 Severity : enhancement
107 Description: lost problems with lov objid file
108 Details : Fixes some scability and access to not inited memory problems
109 in work with lov objdid file.
114 Description: lfs find does not continue on file error
115 Details : Continue other files processing when a file/dir is absent.
119 Description: Inconsistent usage of lustre_pack_reply()
120 Details : Standardize the usage of lustre_pack_reply() such that it
121 always generate a CERROR on failure.
124 Frequency : occasional
126 Description: MDS hang or stay in waiting lock
127 Details : If client receive lock with CBPENDING flag ldlm need send lock
128 cancel as separate rpc, to avoid situation when cancel request
129 can't processed due all i/o threads stay in wait lock.
133 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
137 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
141 Description: Update to SLES9 kernel-2.6.5-7.287.3.
145 Description: Update to RHEL4 latest kernel.
147 Severity : enhancement
149 Description: Build SLES10 patchless client fails
150 Details : The configure was broken by run ./configure with
151 --with-linux-obj=.... argument for patchless client. When the
152 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
153 can't search header adequately. Use absolute path such as
154 -I($LINUX)/include instead.
158 Description: interrupt oig_wait produce painc on resend.
159 Details : brw_redo_request can be used for resend requests from ptlrpcd and
160 private set, and this produce situation when rq_ptlrpcd_data not
161 copyed to new allocated request and triggered LBUG on assert
162 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
163 ptlrpcd set if request is changed and can be safety changed to use
168 Description: LASSERT_{REQ,REP}SWAB macros are buggy
169 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
170 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
171 these macros and replace them with their difinitions instead.
175 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
176 Details : Update which_patch & target file for SLES10 latest kernel.
179 Description: Remove client patches from SLES 10 kernel.
180 Details : This causes SLES 10 clients to behave as patchless clients
181 even on a Lustre-patched (server) kernel.
183 Severity : enhancement
185 Description: self-adjustable client's lru lists
186 Details : use adaptive algorithm for managing client cached locks lru
187 lists according to current server load, other client's work
188 pattern, memory activities, etc. Both, server and client
189 side namespaces provide number of proc tunables for controlling
192 Severity : enhancement
194 Description: light-weight GSS support
195 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
196 and reduce performance overhead.
198 Severity : enhancement
200 Description: Linux keyring support
201 Details : Support using service of Linux keyring for Lustre GSS internal
202 context refresh/cache mechanism.
206 Description: Fix errors in lfs documentation
207 Details : Fixes man pages
211 Description: don't use GFP_* in generic Lustre code.
212 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
216 Description: obdclass is limited by single OBD_ALLOC(idarray)
217 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
221 Description: warnings with build patchless client with vanila 2.6.19 and up
222 Details : change old ctl_table style and replace ctl_table/ctl_table_header
223 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
227 Frequency : Only for SLES
228 Description: sanity_quota fail test_1
229 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
230 file, which makes TSTID[2] inunique.
234 Frequency : Only for SLES9
235 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
236 Details : Update target/ChangeLog/which_patch .
240 Frequency : Only for test_10 in sanity-quota.sh
241 Description: a bug in quota test code
242 Details : checking a test flag in a wrong place
246 Frequency : Only for test_13 in sanity-quota.sh
247 Description: a bug in quota test script
248 Details : shouldn't check overfull allocated quota
252 Description: jbd statistics
253 Details : Port older jbd statistics patch for sles10
255 Severity : enhancement
257 Description: Add printing inode info into message about error in writepage.
262 Description: reply_lock_interpret crash due to race with it and lock cancel.
263 Details : Do not replay locks that are being cancelled. Do not reference
264 locks by their address during replay, just by their handle.
268 Frequency : When flocks are used.
269 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
270 Details : Flock locks might destroy just granted lock if it could be merged
271 with another existing flock, this is done in completion handler,
272 so teach ldlm_cli_enquque_fini that this is a valid case for
278 Description: Oops in read and write path when failing to allocate lock.
279 Details : Check if lock allocation failed and return error back.
283 Description: lstripe command fails for valid OST index
284 Details : The stripe offset is compared to lov->desc.ld_tgt_count
285 instead of lov->desc.ld_active_tgt_count.
289 Description: sanity.sh failed test 103
290 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
295 Description: df doesn't work properly if diskfs blocksize != 4K
296 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
300 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
301 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
305 Description: obd_health_check_timeout too short
306 Details : set obd_health_check_timeout as 1.5x of obd_timeout
310 Description: llapi_file_create() does not allow some changes
311 Details : add llapi_file_open() that allows specifying the mode and
312 open flags, and also returns an open file handle.
316 Description: merge and cleanup kernel patches.
317 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
321 Description: Add journal checksum support.(Kernel part)
322 Details : The journal checksum feature adds two new flags i.e
323 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
324 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
325 indicates that the commit block contains the checksum for
326 the blocks described by the descriptor blocks. Now commit
327 record can be sent to disk without waiting for descriptor
328 blocks to be written to disk. This behavior is controlled
329 using JBD2_FEATURE_ASYNC_COMMIT flag.
333 Description: OSS needs mutliple precreate threads
334 Details : Add ability to start more than one create thread per OSS.
338 Description: Sanity.sh test_65k failed: setstripe should have succeeded
339 Details : execute lfs setstripe on client
343 Description: mds_obd_create error creating tmp object
344 Details : When the user sets quota on root, llog will be affected and can't
345 create files and write files.
348 Frequency : Always on ia64 patchless client, and possibly others.
350 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
351 Details : This allows the patchless client to be loaded on architectures
356 Description: RedHat Update kernel for RHEL5
357 Details : Add the kernel config file for RHEL5.
361 Description: RedHat Update kernel for RHEL5
362 Details : Modify the kernel config file more closer RHEL5.
366 Description: Build failure against Centos5 (RHEL5)
367 Details : Define PAGE_SIZE when it isn't present.
371 Description: test_7 on sanity-quota.sh can't be used on separate machines
372 Details : add facet to handle it
376 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
377 Details : with NFS, the anon dentry's parent was set to itself in
378 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
379 corresponding dentry other than use rec->ur_name.
381 Severity : enhancement
383 Description: lfs setstripe enhancement
384 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
388 Description: enable data checksumming by default
389 Details : enable checksum by default, allow --disable-checksum
390 configure option and "-o nochecksum" mount option. Checksums
391 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
395 Description: lustre support for RHEL5
396 Details : Add support for RHEL5.
400 Description: Client eviction due to failover config
401 Details : after a connection loss, the lustre client should attempt to
402 reconnect to the last active server first before trying the
403 other potential connections.
407 Description: when mds and osts use different quota unit(32bit and 64bit),
408 quota will be released repeatly.
409 Details : void sending multiple quota reqs to mds, which will keep the
410 status between the reqs.
414 Description: osts not allocated evenly to files
415 Details : change the condition to increase offset_idx
418 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
420 Description: Data corruption for OSTs that are formatted larger than 2TB
422 Details : When generating the bio request for lustre file writes the
423 sector number would overflow a temporary variable before being
424 used for the IO. The data reads correctly from Lustre (which
425 will overflow in a similar manner) but other file data or
426 filesystem metadata may be corrupted in some cases.
430 Description: Tune the kernel for good SCSI performance.
431 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
432 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
437 Description: rewrite ext2-derived code in obdclass/uuid.c
438 Details : rewrite inherited code (uuid parsing code from ext2 utils)
439 from scratch preserving functionality.
443 Description: "lfs find -obd UUID" prints directories
444 Details : "lfs find -obd UUID" will return all directory names instead
445 of just file names. It is incorrect because the directories
446 do not reside on the OSTs.
450 Description: use i_size_read and i_size_write in 2.6 port
451 Details : replace inode->i_size access with i_size_read/write()
454 Frequency : when removing large files
456 Description: scheduling issue during removal of large Lustre files
457 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
458 It causes scheduling issues when removing large files (17TB in the
462 Frequency : only with liblustre clients on XT3
464 Description: evictions taking too long
465 Details : allow llrd to evict clients directly on OSTs
468 Frequency : only on ppc
470 Description: /proc/fs/lustre/devices broken on ppc
471 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
472 We should make sure the type of 'index' is loff_t and then cast
473 explicitly as needed below. Do not assign an explicitly cast
477 Frequency : only for rhel5
479 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
480 Details : Modify the target file & which_kernel.
482 Severity : enhancement
484 Description: omit set fsid for export NFS
485 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
486 as FS_REQUIRES_DEV for avoid problems with generate fsid.
490 Frequency : Always, for kernels after 2.6.16
491 Description: Fix warning idr_remove called for id=.. which is not allocated.
492 Details : Last kernels save old s_dev before kill super and not allow
493 to restore from callback - restore it before call kill_anon_super.
497 Description: buffer overruns could theoretically occur
498 Details : llapi_semantic_traverse() modifies the "path" argument by
499 appending values to the end of the origin string, and a
500 overrun may occur. Adding buffer overrun check in liblustreapi.
504 Description: Fix error on 'ls .' at the top of the Lustre mount.
505 Details : Don't revalidate dentry if it is a root dentry.
509 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
510 Details : Modify vm-tunables-rhel4.patch.
514 Description: Kernel config for 2.6.18-vanilla.
515 Details : Modify targets/2.6-vanilla.target.in.
516 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
517 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
518 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
519 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
523 Description: improve handling recoverable errors
524 Details : if request processig with error which can be recoverable on server
525 request should be resend, otherwise page released from cache and
530 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
531 Details : Modify target file & which_patch.
532 A flaw was found in the IA32 system call emulation provided
533 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
534 value could be stored in the %RAX register, which could trigger an
535 out-of-bounds system call table access. An untrusted local user
536 could exploit this flaw to run code in the kernel
537 (ie a root privilege escalation). (CVE-2007-4573).
541 Description: O_DIRECT bypasses client statistics.
542 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
543 read_ahead_stats were updated. copied the stats section from
544 osc_send_oap_rpc() into async_internal().
548 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
552 Description: change order of libsysio includes
553 Details : '#include sysio.h' should always come before '#include xtio.h'
557 Description: handle "serious error: objid * already exists" more gracefully
558 Details : If LAST_ID value on disk is smaller than the objects existing in
559 the O/0/d* directories, it indicates disk corruption and causes an
560 LBUG(). If the object is 0-length, then we should use the existing
561 object. This will help to avoid a full fsck in most cases.
563 Severity : enhancement
565 Description: adapt the lustre_config script to support the upgrade case
566 Details : Add "-u" option for lustre_config script to support upgrading 1.4
567 server targets to 1.6 in parallel.
571 Description: To avoid grant space > avaible space when the disk is almost
572 full. Without this patch you might see the error "grant XXXX >
573 available" or some LBUG about grant, when the disk is almost
575 Details : In filter_check_grant, for non_grant cache write, we should
576 check the left space by if (*left > ungranted + bytes), instead
577 of (*left > ungranted), because only we are sure the left space
578 is enough for another "bytes", then the ungrant space should be
579 increase. In client, we should update cl_avail_grant only there
580 is OBD_MD_FLGRANT in the reply.
584 Description: Update RHEL 4 kernel to fix local root privilege escalation.
585 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
586 described in CVE-2007-4573. This problem could allow untrusted
587 local users to gain root access.
590 Frequency : when using O_DIRECT and quotas
592 Description: Incorrect file ownership on O_DIRECT output files
593 Details : block usage reported by 'lfs quota' does not take into account
594 files that have been written with O_DIRECT.
599 Description: touch file failed when fs is not full
600 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
601 otherwise we can get ENOSP while fs is not full.
605 Description: parallel lock callbacks
606 Details : Instead of sending blocking and completion callbacks as separated
607 requests, adding them to a set and sending in parallel.
610 Frequency : only for Cray XT3
611 Bugzilla : 12829/13455
612 Description: Changing primary group doesn't change the group lustre assigns to
614 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
615 overridden with the primary group provided by the group upcall,
616 whereas the supplied fsgid can be trusted if it is in the list of
617 supplementary groups returned by the group upcall.
619 Severity : enhancement
621 Description: Allow masking D_WARNING, D_ERROR messages from console
622 Details : Console messages can now be disabled via lnet.printk.
626 Description: User code with malformed file open parameter crashes client node
627 Details : Before packing join_file req, all the related reference should be
628 checked carefully in case some malformed flags cause fake
629 join_file req on client.
633 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
634 Details : ldlm_completion_ast() assumes that a lock is granted when the req
635 mode is equal to the granted mode. However, it should also check
636 that LDLM_FL_CP_REQD is not set.
640 Description: Heavy nfs access might result in deadlocks
641 Details : After ELC code landed, it is now improper to enqueue any mds
642 locks under och_sem, because enqueue might want to decide to
643 cancel open locks for same inode we are holding och_sem for.
647 Description: Client eviction while running blogbench
648 Details : A lot of unlink operations with concurrent I/O can lead to a
649 deadlock causing evictions. To address the problem, the number of
650 oustanding OST_DESTROY requests is now throttled to
651 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
652 callbacks are processed in priority.
654 --------------------------------------------------------------------------------
656 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
658 * Support for kernels:
659 2.6.5-7.283 (SLES 9),
660 2.6.9-55.EL (RHEL 4),
661 2.6.16.46-0.14 (SLES 10),
662 2.6.18.8 vanilla (kernel.org)
663 * Client support for unpatched kernels:
664 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
665 2.6.16 - 2.6.22 vanilla (kernel.org)
666 * Due to recently discovered recovery problems, we do not recommend
667 using patchless RHEL 4 clients with this or any earlier release.
668 * Recommended e2fsprogs version: 1.39.cfs8
669 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
670 * Starting with this release, the ldiskfs backing filesystem required
671 by Lustre is now in its own package, lustre-ldiskfs. This package
672 should be installed. It is versioned separately from Lustre and
673 may be released separately in future.
677 Description: block reactivating mgc import until all deactivates complete
678 Details : Fix race when failing back MDT/MGS to itself (testing)
681 Frequency : at statup only
683 Description: mds_lov_synchronize race leads to various problems
684 Details : simultaneous MDT->OST connections at startup can cause the
685 sync to abort, leaving the OSC in a bad state.
687 Severity : enhancement
689 Description: add optional extra BUILD_VERSION info
690 Details : add a new environment variable (namely LUSTRE_VERS) which allows
691 to override the lustre version.
694 Frequency : 2.6.18 servers only
696 Description: ll_kern_mount() doesn't release the module reference
697 Details : The ldiskfs module reference count never drops down to 0
698 because ll_kern_mount() doesn't release the module reference.
703 Description: server LBUG when using old ost_num_threads parameter
704 Details : Accept the old ost_num_threads parameter but warn that it
705 is deprecated, and fix an off-by-one error that caused an LBUG.
710 Description: Transient SCSI error results in persistent IO issue
711 Details : iobuf->dr_error is not reinitialized to 0 between two
715 Frequency : sometimes when underlying device returns I/O errors
717 Description: OSTs not going read-only during write failures
718 Details : OSTs are not remounted read-only when the journal commit threads
719 get I/O errors because fsfilt_ext3 calls journal_start/stop()
720 instead of the ext3 wrappers.
724 Description: poor connect scaling with increasing client count
725 Details : Don't run filter_grant_sanity_check for more than 100 exports
726 to improve scaling for large numbers of clients.
729 Frequency : SLES10 only
731 Description: sanity-quota.sh quotacheck failed: rc = -22
732 Details : Quotas cannot be enabled on SLES10.
735 Frequency : liblustre clients only
737 Description: getdirentries does not give error when run on compute nodes
738 Details : getdirentries does not fail when the size specified as an argument
739 is too small to contain at least one entry
741 Severity : enhancement
743 Description: Add LNET router traceability for debug purposes
744 Details : If a checksum failure occurs with a router as part of the
745 IO path, the NID of the last router that forwarded the bulk data
746 is printed so it can be identified.
751 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
752 Details : Due to a race condition, liblustre clients were occasionally
755 Severity : enhancement
757 Description: lfs setstripe use optional parameters instead of postional
760 Severity : enhancement
762 Description: Nanosecond timestamp support for ldiskfs
763 Details : The on-disk ldiskfs filesystem has added support for nanosecond
764 resolution timestamps. There is not yet support for this at
765 the Lustre filesystem level.
768 Frequency : during server recovery
770 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
771 Details : request with rq_no_resend flag not awake l_wait_event if they get a
775 Frequency : nfs export on patchless client
777 Description: connectathon hang when test nfs export over patchless client
778 Details : Disconnected dentry cannot be found with lookup, so we do not need
779 to unhash it or make it invalid
782 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
783 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
784 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
789 Description: open req refcounting wrong on reconnect
790 Details : If reconnect happened between getting open reply from server and
791 call to mdc_set_replay_data in ll_file_open, we will schedule
792 replay for unreferenced request that we are about to free.
793 Subsequent close will crash in variety of ways.
794 Check that request is still eligible for replay in
795 mdc_set_replay_data().
800 Description: disable writes to filesystem when reading health_check file
801 Details : the default for reading the health_check proc file has changed
802 to NOT do a journal transaction and write to disk, because this
803 can cause reads of the /proc file to hang and block HA state
804 checking on a healthy but otherwise heavily loaded system. It
805 is possible to return to the previous behaviour during configure
806 with --enable-health-write.
808 Severity : enhancement
810 Description: 64-bit inode version
811 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
812 to the inode. This will be required for version-based recovery.
817 Description: MDS fails to start if a duplicate client export is detected
818 Details : in some rare cases it was possible for a client to connect to
819 an MDS multiple times. Upon recovery the MDS would detect this
820 and fail during startup. Handle this more gracefully.
822 Severity : enhancement
824 Description: Add -o localflock option to simulate old noflock
826 Details : This will achieve local-only flock/fcntl locks
832 Description: log_commit_thread vs filter_destroy race leads to crash
833 Details : Take import reference before releasing llog record semaphore
838 Description: Wrong request locking in request set processing
839 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
840 imp_delayed_list, in this place should be used imp_lock.
843 Frequency : when reconnection
845 Description: Grant Leak when osc reconnect to OST
846 Details : When osc reconnect ost, OST(filter) should check whether it
847 should grant more space to client by comparing fed_grant and
848 cl_avail_grant, and return the granted space to client instead
849 of "new granted" space, because client will call osc_init_grant
850 to update the client grant space info.
853 Frequency : when client reconnect to OST
855 Description: Grant Leak when osc do resend and replay bulk write
856 Details : When osc reconnect to OST, OST(filter)should clear grant info of
857 bulk write request, because the grant info while be sync between
858 OSC and OST when reconnect, and we should ignore the grant info
859 these of resend/replay write req.
864 Description: Grant space more than avaiable left space sometimes.
865 Details : When then OST is about to be full, if two bulk writing from
866 different clients came to OST. Accord the avaliable space of the
867 OST, the first req should be permitted, and the second one
868 should be denied by ENOSPC. But if the seconde arrived before
869 the first one is commited. The OST might wrongly permit second
870 writing, which will cause grant space > avaiable space.
873 Frequency : when client is evicted
875 Description: Grant might be wrongly erased when osc is evicted by OST
876 Details : when the import is evicted by server, it will fork another
877 thread ptlrpc_invalidate_import_thread to invalidate the
878 import, where the grant will be set to 0. While the original
879 thread will update the grant it got when connecting. So if
880 the former happened latter, the grant will be wrongly errased
881 because of this race.
886 Description: Checking Stale with correct fid
887 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
888 to check whether it is stale, because sometimes, we want the
889 enqueue happened anyway, and op_data.fid2 will not be initialized.
891 Severity : enhancement
893 Description: update patchless client
894 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
897 Frequency : only with 2.4 kernel
899 Description: random memory corruption
900 Details : size of struct ll_inode_info is to big for union inode.u and this
901 can be cause of random memory corruption.
906 Description: Memory leak in recovery
907 Details : Lov_mds_md was not free in an error handler in mds_create_object.
908 It should also check obd_fail before fsfilt_start, otherwise if
909 fsfilt_start return -EROFS,(failover mds during mds recovery).
910 then the req will return with repmsg->transno = 0 and rc = EROFS.
911 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
912 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
913 be freed no matter whether fsfilt_commit success or not.
916 Frequency : only with huge count clients
918 Description: Prevents from taking the superblock lock in llap_from_page for
920 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
921 for llap_from_page call in ll_removepage prevents from taking the
922 superblock lock for a soon died page.
927 Description: Not check open intent error before release open handle
928 Details : in some rare cases, the open intent error is not checked before
929 release open handle, which may cause
930 ASSERTION(open_req->rq_transno != 0), because it tries to release
931 the failed open handle.
936 Description: Set cat log bitmap only after create log success.
937 Details : in some rare cases, the cat log bitmap is set too early. and it
938 should be set only after create log success.
942 Description: Accessing a block bevice can re-enable I/O when Lustre is
943 tearing down a device.
944 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
949 Description: service threads may hog cpus when there are a lot of requests
950 Details : Insert cond_resched to give other threads a chance to use some CPU
955 Description: the cat log was not initialized in recovery
956 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
957 unlink log on mds will not be initialized until mds post
958 recovery. And also in mds post recovery, the unlink log will
959 initialization will be done asynchronausly, so there will be race
960 between add unlink log and unlink log initialization.
964 Description: brw_stats were being printed incorrectly
965 Details : brw_stats were being printed as log2 but all of them were not
966 recorded as log2. Also remove some code duplication arising from
967 filter_tally_{read,write}.
971 Frequency : rare, only in recovery.
972 Description: ASSERTION(req->rq_type != LI_POISON) failed
973 Details : imp_lock should be held while iterating over imp_sending_list for
974 prevent destroy request after get timeout in ptlrpc_queue_wait.
978 Description: replay-single.sh test 52 fails
979 Details : A lock's skiplist need to be cleanup when it being unlinked
980 from its resource list.
984 Description: Short directio read returns full requested size rather than
986 Details : Direct I/O operations should return actual amount of bytes
987 transferred rather than requested size.
989 Severity : enhancement
991 Description: metadata RPC reduction (e.g. for rm performance)
992 Details : decrease the amount of synchronous RPC between clients and servers
993 by canceling conflicing lock before the operation on the client
994 and packing thier handles into the main operation RPC to server.
996 Severity : enhancement
998 Description: Async OSC create to avoid the blocking unnecessarily.
999 Details : If a OST has no remain object, system will block on the creating
1000 when need to create a new object on this OST. Now, ways use
1001 pre-created objects when available, instead of blocking on an
1002 empty osc while others are not empty. If we must block, we block
1003 for the shortest possible period of time.
1005 Severity : enhancement
1007 Description: refine locking for avoid write wrong info into lov_objid file
1008 Details : fix possible races with add new target and write/update data in
1011 --------------------------------------------------------------------------------
1013 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1018 Frequency : on some architectures
1020 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1021 Details : Uninitialized flags sometimes cause configuration commands to
1025 Frequency : patchless clients only
1027 Description: missing __iget() symbol export
1028 Details : The __iget() symbol export is missing. To avoid the need for
1029 this on patchless clients the deathrow inode reaper is turned
1030 off, and we depend on the VM to clean up old inodes. This
1031 dependency was during via the fix for bug 12181.
1033 --------------------------------------------------------------------------------
1035 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1037 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1038 INTEROPERATE with older versions automatically. In many cases a
1039 special upgrade step is needed. Please read the
1040 user documentation before upgrading any part of a live system.
1041 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1042 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1043 * WARNING: Lustre configuration and startup changes are required with
1044 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1050 Severity : enhancement
1052 Description: MountConf
1053 Details : Lustre configuration is now managed via mkfs and mount
1054 commands instead of lmc and lconf. New obd types (MGS, MGC)
1055 are added for dynamic configuration management. See
1056 https://mail.clusterfs.com/wikis/lustre/MountConf for
1059 Severity : enhancement
1061 Description: dynamic OST addition
1062 Details : OSTs can now be added to a live filesystem
1064 Severity : enhancement
1066 Description: startup order invariance
1067 Details : MDTs and OSTs can be started in any order. Clients only
1068 require the MDT to complete startup.
1070 Severity : enhancement
1072 Description: parallel, asynchronous orphan cleanup
1073 Details : orphan cleanup is now performed in separate threads for each
1074 OST, allowing parallel non-blocking operation.
1076 Severity : enhancement
1078 Description: optimized stripe assignment
1079 Details : stripe assignments are now made based on ost space available,
1080 ost previous usage, and OSS previous usage, in order to try
1081 to optimize storage space and networking resources.
1083 Severity : enhancement
1085 Description: Permanently set tunables
1086 Details : All writable /proc/fs/lustre tunables can now be permanently
1087 set on a per-server basis, at mkfs time or on a live system.
1089 Severity : enhancement
1091 Description: Lustre message v2
1092 Details : Add lustre message format v2.
1094 Severity : enhancement
1096 Description: client OST exclusion list
1097 Details : Clients can be started with a list of OSTs that should be
1098 declared "inactive" for known non-responsive OSTs.
1102 Description: SPEC SFS validation failure on NFS v2 over lustre.
1103 Details : Changes the blocksize for regular files to be 2x RPC size,
1104 and not depend on stripe size.
1106 Severity : enhancement
1108 Description: Multiple MD RPCs in flight.
1109 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1110 To avoid overly-overloading MDS, introduce a limit on number of
1111 MDS RPCs in flight for a single client and add /proc controls
1112 to adjust this limit.
1114 Severity : enhancement
1116 Description: client read/write statistics
1117 Details : Add client read/write call usage stats for performance
1118 analysis of user processes.
1119 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1120 file access. extents_stats shows chunk size distribution.
1121 extents_stats_per_process show chunk size distribution per
1124 Severity : enhancement
1126 Description: mds statistics
1127 Details : Add detailed mds operations statistics in
1128 /proc/fs/lustre/mds/*/stats.
1132 Description: Failure of copying files with lustre special EAs.
1133 Details : Client side always return success for setxattr call for lustre
1134 special xattr (currently only "trusted.lov").
1139 Description: Refcount LNET uuids
1140 Details : The global LNET uuid list grew linearly with every startup;
1141 refcount repeated list entries instead of always adding to
1144 Severity : enhancement
1146 Description: Dynamic service threads
1147 Details : Within a small range, start extra service threads
1148 automatically when the request queue builds up.
1151 Frequency : mixed-endian client/server environments
1153 Description: mixed-endian crashes
1154 Details : The new msg_v2 system had some failures in mixed-endian
1157 Severity : enhancement
1159 Description: Easy OST removal
1160 Details : OSTs can be permanently deactivated with e.g. 'lctl
1161 conf_param lustre-OST0001.osc.active=0'
1163 Severity : enhancement
1165 Description: MGS proc entries
1166 Details : Added basic proc entries for the MGS showing what filesystems
1169 Severity : enhancement
1171 Description: provide MGS failover
1172 Details : Added config lock reacquisition after MGS server failover.
1174 Severity : enhancement
1176 Description: add Linux 2.4 support
1177 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1181 Description: a large application tries to do I/O to the same resource and dies
1182 in the middle of it.
1183 Details : Check the req->rq_arrival time after the call to
1184 ost_brw_lock_get(), but before we do anything about
1185 processing it & sending the BULK transfer request. This
1186 should help move old stale pending locks off the queue as
1187 quickly as obd_timeout.
1190 Frequency : when an incorrect nid is specified during startup
1192 Description: ptlrpc connect to non-existant node causes kernel crash
1193 Details : LNET can't be re-entered from an event callback, which
1194 happened when we expire a message after the export has been
1195 cleaned up. Instead, hand the zombie cleanup off to another
1198 Severity : enhancement
1200 Description: plain/inodebits lock performance improvement
1201 Details : Grouping plain/inodebits in granted list by their request modes
1202 and bits policy, thus improving the performance of search through
1206 Frequency : only if OST filesystem is corrupted
1208 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1209 Details : for a short time RPCs with bulk IO are in the replay list,
1210 but replay of bulk IOs is unimplemented. If the OST filesystem
1211 is corrupted due to disk cache incoherency and then replay is
1212 started it is possible to trip an assertion. Avoid putting
1213 committed RPCs into the replay list at all to avoid this issue.
1216 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1218 Description: System hang on startup
1219 Details : This bug allowed the liblustre (e.g. catamount) client to
1220 return to the app before handling all startup RPCs. This
1221 could leave the node unresponsive to lustre network traffic
1222 and manifested as a server ptllnd timeout.
1224 Severity : enhancement
1226 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1227 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1228 boolean to control whether to print peer debug info when a
1229 client's RPC times out.
1232 Frequency : only for kernels with patches from Lustre below 1.4.3
1234 Description: Remove old rdonly API
1235 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1238 Frequency : only for devices with external journals
1240 Description: Set external device read-only also
1241 Details : During a commanded failover stop, we set the disk device
1242 read-only while the server shuts down. We now also set any
1243 external journal device read-only at the same time.
1246 Frequency : when upgrading from 1.4 while trying to change parameters
1248 Description: The wrong (new) MDC name was used when setting parameters for
1249 upgraded MDT's. Also allows changing of OSC (and MDC)
1250 parameters if --writeconf is specified at tunefs upgrade time.
1253 Frequency : when setting specific ost indicies
1255 Description: QOS code breaks on skipped indicies
1256 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1257 created with --index need not be sequential.
1259 Severity : enhancement
1261 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1262 Details : The uninit_groups feature works in conjunction with the kernel
1263 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1264 up the pass1 processing of e2fsck. This is a read-only feature
1265 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1266 mount filesystems that have had this feature enabled.
1268 Severity : enhancement
1270 Description: Improve multi-block allocation algorithm to avoid fragmentation
1271 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1272 allocation locality and avoid filesystem fragmentation.
1274 ------------------------------------------------------------------------------
1276 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1278 * Support for kernels:
1279 2.4.21-47.0.1.EL (RHEL 3)
1280 2.6.5-7.283 (SLES 9)
1281 2.6.9-42.0.10.EL (RHEL 4)
1282 2.6.12.6 vanilla (kernel.org)
1283 2.6.16.27-0.9 (SLES 10)
1284 * Recommended e2fsprogs version: 1.39.cfs5
1286 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1290 Frequency : occasional, depends on client load and configuration
1291 Bugzilla : 12181, 12203
1292 Description: data loss for recently-modified files
1294 Details : In some cases it is possible that recently written or created
1295 files may not be written to disk in a timely manner (this should
1296 normally be within 30s unless client IO load is very high).
1297 The problem appears as zero-length files or files that are a
1298 multiple of 1MB in size after a client crash or client eviction
1299 that are missing data at the end of the file.
1301 This problem is more likely to be hit on clients where files are
1302 repeatedly created and unlinked in the same directory, clients
1303 have a large amount of RAM, have many CPUs, the filesystem has
1304 many OSTs, the clients are rebooted frequently, and/or the files
1305 are not accessed by other nodes after being written.
1307 The presence of the problem can be detected by looking at
1308 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1309 smaller than the second (nr_unused) then dirty files will not
1310 be flushed automatically to disk. "sync; sleep 10" should be
1311 run several times on the node before unmounting it to update
1312 Lustre (this is also safe to run on nodes without this problem).
1314 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1315 that can cause this same problem, so customers using that kernel
1316 also need to update the kernel in addition to Lustre. In order
1317 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1319 It is normal that files written just before a client crash (less
1320 than 30s) may not yet have been flushed to disk, even for local
1324 Frequency : frequent on thin XT3 nodes
1326 Description: UUID collision on thin XT3 Linux nodes
1327 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1328 randomly, since we relied on an insufficiently-seeded PRNG.
1333 Description: OSS hangs after "All ost request buffers busy"
1334 Details : A deadlock between quota and journal operations caused OSS
1335 hangs after printing "All ost request buffers busy."
1338 Frequency : always on liblustre builds
1340 Description: Cleanup compiler warnings on liblustre
1343 Frequency : always on liblustre builds on XT3
1345 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1349 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1350 > 32000 subdirectories
1351 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1352 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1353 subdirectory count crosses 32000. This will aid e2fsck to
1354 correctly handle more than 32000 subdirectories.
1357 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1359 Description: System hang on startup
1360 Details : This bug allowed the liblustre (e.g. catamount) client to
1361 return to the app before handling all startup RPCs. This
1362 could leave the node unresponsive to lustre network traffic
1363 and manifested as a server ptllnd timeout.
1365 Severity : enhancement
1367 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1368 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1369 boolean to control whether to print peer debug info when a
1370 client's RPC times out.
1375 Description: make O_SYNC working on 2.6 kernels
1376 Details : 2.6 kernels use different method for mark pages for write,
1377 so need add a code to lustre for O_SYNC work.
1382 Description: Failure to close file and release space on NFS
1383 Details : Put inode details into lock acquired in ll_intent_file_open.
1384 Use mdc_intent_lock in ll_intent_open to properly
1385 detect all kind of errors unhandled by mdc_enqueue.
1390 Description: proc file read during shutdown sometimes raced obd removal,
1392 Details : Add lock to prevent obd access after proc file removal.
1395 Frequency : Only for files larger than 4GB on 32-bit clients.
1397 Description: improperly doing page alignment of locks
1398 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1399 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1404 Description: RPCs being resent when they shouldn't be
1405 Details : Some RPCs that should not be resent are being resent. This
1406 can cause inconsistencies in the RPC state machine. Do not
1407 resend such requests.
1410 Frequency : rare, only with NFS export
1412 Description: Crash on NFS re-export node
1413 Details : under very unusual load conditions an assertion is hit in
1414 ll_intent_file_open()
1417 Frequency : only if OST filesystem is corrupted
1419 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1420 Details : for a short time RPCs with bulk IO are in the replay list,
1421 but replay of bulk IOs is unimplemented. If the OST filesystem
1422 is corrupted due to disk cache incoherency and then replay is
1423 started it is possible to trip an assertion. Avoid putting
1424 committed RPCs into the replay list at all to avoid this issue.
1429 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1430 Details : Large single O_DIRECT read and write calls can fail to allocate
1431 a sufficiently large buffer to process the request. In case of
1432 allocation failure the allocation is retried with a smaller
1433 buffer and broken into smaller requests.
1435 Severity : enhancement
1437 Description: Add -o localflock option to simulate old noflock behaviour.
1438 Details : This will achieve local-only flock/fcntl locks coherentness.
1443 Description: versioning check is incomplete
1444 Details : Checking the version difference of client vs. server, report
1445 error if the gap is too big.
1450 Description: add support PG_writeback bit
1451 Details : add support for PG_writeback bit for Lustre, for more carefull
1452 work with page cache in 2.6 kernel. This also fix some deadlocks
1453 and remove hack for work O_SYNC with 2.6 kernel.
1455 Severity : enhancement
1457 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1458 Details : The uninit_groups feature works in conjunction with the kernel
1459 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1460 up the pass1 processing of e2fsck. This is a read-only feature
1461 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1462 mount filesystems that have had this feature enabled.
1464 Severity : enhancement
1466 Description: Improve multi-block allocation algorithm to avoid fragmentation
1467 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1468 allocation locality and avoid filesystem fragmentation.
1470 ------------------------------------------------------------------------------
1472 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1474 * Support for kernels:
1475 2.6.9-42.0.3.EL (RHEL 4)
1476 2.6.5-7.276 (SLES 9)
1477 2.4.21-47.0.1.EL (RHEL 3)
1478 2.6.12.6 vanilla (kernel.org)
1479 2.6.16.21-0.8 (SLES10)
1480 * Recommended e2fsprogs version: 1.39.cfs2-0
1482 * The backwards-compatible /proc/sys/portals symlink has been removed
1483 in this release. Before upgrading, please ensure that you change
1484 any configuration scripts or /etc/sysctl.conf files that access
1485 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1486 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1487 in advance of the upgrade on any system running Lustre 1.4.6 or
1488 newer, since /proc/sys/lnet was added in that version.
1489 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1493 Frequency : only when quota is used
1495 Description: avoid scanning export list for quota master
1496 Details : Change the algorithms to avoid scanning export list in order
1497 to improve the efficiency.
1500 Frequency : MDS failover only, very rarely
1502 Description: "went back in time" messages on mds failover
1503 Details : The greatest transno may be lost when the current operation
1504 finishes with an error (transno==0) and the client's last_rcvd
1505 record is over-written. Save the greatest transno in the
1506 mds_last_transno for this case.
1509 Frequency : always for specific kernels and striping counts
1511 Description: client may get "Matching packet too big" without ACL support
1512 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1513 when trying to access files in certain configurations. The
1514 clients should in fact be denied when mounting because they do
1515 not understand ACLs.
1518 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1520 Description: many clients connecting with IO in progress causes connect timeouts
1521 Details : Avoid synchronous journal commits to avoid delays caused by many
1522 clients connecting/disconnecting when bulk IO is in progress.
1523 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1524 OST_IO_PORTAL to avoid delays behind potentially many pending
1528 Frequency : occasionally with multiple writers to a single file
1530 Description: shared writes to file may result in wrong size reported by stat()
1531 Details : Allow growing of kms when extent lock is cancelled
1534 Frequency : always with random mmap IO to multi-striped file
1536 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1537 Details : Only if the hole is at the end of OST object so that kms is too
1538 small. Fix is to increase kms accordingly in ll_nopage.
1541 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1543 Description: writes to a missing object would leak memory on the OST
1544 Details : If there is an inconsistency between the MDS and OST filesystems,
1545 such that the MDS references an object that doesn't exist, writes
1546 to that object will leak memory due to incorrect cleanup in the
1547 error handling path, eventually running out of memory on the OST.
1552 Description: Creating too long symlink causes lustre errors
1553 Details : Check symlink and name lengths before sending requests to MDS.
1556 Frequency : only if flock is enabled (not on by default)
1558 Description: posix locks not released on fd closure on 2.6.9+
1559 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1560 kernels, this caused such locks not to be released on fd close and
1561 then assertions on fs unmount about still used locks.
1564 Frequency : MDS failover only, very rarely
1566 Description: clients may get ASSERTION(granted_lock != NULL)
1567 Details : When request was taking a long time, and a client was resending
1568 a getattr by name lock request. The were multiple lock requests
1569 with the same client lock handle and
1570 mds_getattr_name->fixup_handle_for_resent_request found one of the
1571 lock handles but later failed with ASSERTION(granted_lock != NULL).
1576 Description: handle->h_buffer_credits > 0, assertion failure
1577 Details : h_buffer_credits is zero after truncate, causing assertion
1578 failure. This patch extends the transaction or creates a new
1582 Frequency : NFS re-export or patchless client
1583 Bugzilla : 11179, 10796
1584 Description: Crash on NFS re-export node (__d_move)
1585 Details : We do not want to hash the dentry if we don't have a lock.
1586 But if this dentry is later used in d_move, we'd hit uninitialised
1587 list head d_hash, so we just do this to init d_hash field but
1588 leave dentry unhashed.
1591 Frequency : NFS re-export or patchless client
1593 Description: NFS exports has problem with symbolic link
1594 Details : lustre client didn't properly install dentry when re-exported
1595 to NFS or running patchless client.
1598 Frequency : NFS re-export or patchless client
1600 Description: Various nfs/patchless fixes.
1601 Details : fixes reuse disconected alias for lookup process - this fixes
1602 warning "find_exported_dentry: npd != pd",
1603 fix permission error with open files at nfs.
1604 fix apply umask when do revalidate.
1607 Frequency : occasional
1609 Description: Crash on NFS re-export node
1610 Details : calling clear_page() on the wrong pointer triggered oops in
1611 generic_mapping_read().
1614 Frequency : rarely, using O_DIRECT IO
1616 Description: unaligned directio crashes client with LASSERT
1617 Details : check for unaligned buffers before trying any requests.
1620 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1622 Description: stale data returned from RAID cache
1623 Details : If only a small amount of IO is done to the RAID device before
1624 reading it again it is possible to get stale data from the RAID
1625 cache instead of reading it from disk.
1628 Frequency : always for sles10 kernel
1630 Description: sles10 support
1631 Details : ll_follow_link: compile fixes and using of nd_set_link
1632 under newer kernels.
1635 Frequency : depends on arch, kernel and compiler version, always on sles10
1638 Description: recursive or deep enough symlinks cause stack overflow
1639 Details : getting rid of large stack-allocated variable in
1643 Frequency : depends on hardware
1645 Description: lustre write performance loss in the SLES10 kernel
1646 Details : the performance loss is caused by using of write barriers in the
1647 ext3 code. The SLES10 kernel turns barrier support on by
1648 default. The fix is to undo that change for ldiskfs.
1650 ------------------------------------------------------------------------------
1652 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1654 * Support for kernels:
1655 2.6.9-42.0.3EL (RHEL 4)
1656 2.6.5-7.276 (SLES 9)
1657 2.4.21-47.0.1.EL (RHEL 3)
1658 2.6.12.6 vanilla (kernel.org)
1662 Frequency : quota enabled and large files being deleted
1664 Description: releasing more than 4GB of quota at once hangs OST
1665 Details : If a user deletes more than 4GB of files on a single OST it
1666 will cause the OST to spin in an infinite loop. Release
1667 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1672 Description: statfs data retrieved from /proc may be stale or zero
1673 Details : When reading per-device statfs data from /proc, in the
1674 {kbytes,files}_{total,free,avail} files, it may appear
1675 as zero or be out of date.
1678 Frequency : systems with MD RAID1 external journal devices
1680 Description: lconf's call to blkid is confused by RAID1 journal devices
1681 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1682 of returning all block devices that match the journal UUID.
1685 Frequency : always, for aggregate stripe size over 4GB
1687 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1688 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1689 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1692 Frequency : always if "lfs find" used on a local file/directory
1694 Description: "lfs find" segfaults if used on a local file/directory
1695 Details : The case where a directory component was not specified wasn't
1696 handled correctly. Handle this properly.
1699 Frequency : always on ppc64
1701 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1702 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1703 perform endianness conversion.
1706 Frequency : rarely (truncate to non-zero file size after write under load)
1707 Bugzilla : 10730, 10687
1708 Description: Files padded with zeros to next 4K multiple
1709 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1710 it is possible that files that are truncated to a non-zero size
1711 immediately after being written are filled with zero bytes beyond
1712 the truncated size. No file data is lost.
1714 Severity : enhancement
1716 Description: Allow recovery/failover for liblustre clients.
1717 Details : liblustre clients were unaware of failover configurations until
1720 Severity : enhancement
1722 Description: user file locks should fail when not mounting with flock option
1723 Details : Set up an error-returning stub in ll_file_operations.lock field
1724 to prevent incorrect behaviour when client is mounted without
1725 flock option. Also, set up properly f_op->flock field for
1729 Frequency : always on ia64
1731 Description: "lfs df" loops on printing out MDS statfs information
1732 Details : The obd_ioctl_data was not initialized and in some systems
1733 this caused a failure during the ioctl that did not return
1734 an error. Initialize the struct and return an error on failure.
1737 Frequency : SLES 9 only
1739 Description: Error of copying files with lustre special EAs as root
1740 Details : Client side always return success for setxattr call for lustre
1741 special xattr (currently only "trusted.lov").
1744 Frequency : rarely on clusters with both ia64+i386 clients
1746 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1747 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1748 clients are doing concurrent writes to the same file it is
1749 possible that the ia64 clients may LASSERT because the OST
1750 extent locks are not PAGE_SIZE aligned. Ensure that grown
1751 locks are always aligned on the request boundary.
1754 Frequency : specific use, occasional
1756 Description: Overwriting in use executable truncates on-disk binary image
1757 Details : If one node attempts to overwrite an executable in use by
1758 another node, we now correctly return ETXTBSY instead of
1759 truncating the file.
1761 Severity : enhancement
1763 Description: Async OSC create to avoid the blocking unnecessarily.
1764 Details : If a OST has no remain object, system will block on the creating
1765 when need to create a new object on this OST. Now, ways use
1766 pre-created objects when available, instead of blocking on an
1767 empty osc while others are not empty. If we must block, we block
1768 for the shortest possible period of time.
1773 Description: chmod on Lustre root is propagated to other clients
1774 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1775 invalid by the follow_mount time.
1778 Frequency : liblustre clients only
1780 Description: Race in 'instant cancel' lock handling could lead to such locks
1781 never to be granted in case of SMP MDS
1782 Details : Do not destroy not yet granted but cbpending locks in
1786 Frequency : replay/resend of open
1788 Description: non null lock assetion failure in mds_intent_policy
1789 Details : Trying to replay/resend lockless open requests resulted in
1790 mds_open() returning 0 with no lock. Now it sets a flag if
1791 a lock is going to be returned.
1793 Severity : enhancement
1795 Description: Checksum enhancements
1796 Details : New checksum enhancements allow for resending RPCs that failed
1799 Severity : enhancement
1801 Description: Tunables on number of dirty pages in cacche
1802 Details : Allow to set limit on number of dirty pages cached.
1807 Description: client crash on unmount - lock still has references
1808 Details : In some error handling cases it was possible to leak a lock
1809 reference on a client while accessing a file. This was not
1810 harmful to the client during operation, but would cause the
1811 client to crash when the filesystem is unmounted.
1814 Frequency : specific case, rare
1816 Description: ETXTBSY on mds though file not in use by client
1817 Details : ETXTBSY is no longer incorrectly returned when attempting to
1818 chmod or chown a directory that the user previously tried to
1819 execute or a currently-executing binary.
1822 Frequency : extremely rare except on liblustre-based clients
1824 Description: Lustre space not freed when files are deleted
1825 Details : Clean up open-unlinked files after client eviction. Previously
1826 the unlink was skipped and the files remained as orphans.
1831 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1832 Details : In some cases it was possible to send a blocking callback to
1833 a client doing a glimpse, even though that client didn't get
1834 a lock granted. When the glimpse lock is cancelled on the OST
1835 the freed lock is left on the waiting list and corrupted the list.
1838 Frequency : all core dumps
1840 Description: Broke core dumps to lustre
1841 Details : Negative dentry may be unhashed if parent does not have UPDATE
1842 lock, but some callers, e.g. do_coredump, expect dentry to be
1843 hashed after successful create, hash it in ll_create_it.
1845 ------------------------------------------------------------------------------
1847 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1849 * Support for kernels:
1850 2.6.9-42.0.2.EL (RHEL 4)
1851 2.6.5-7.276 (SLES 9)
1852 2.4.21-40.EL (RHEL 3)
1853 2.6.12.6 vanilla (kernel.org)
1857 Frequency : always on RHEL 3
1859 Description: Number of open files grows over time
1860 Details : The number of open files grows over time, whether or not
1861 Lustre is started. This was due to a filp leak introduced
1862 by one of our kernel patches.
1864 ------------------------------------------------------------------------------
1866 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1868 * Support for kernels:
1869 2.6.9-42.EL (RHEL 4)
1870 2.6.5-7.276 (SLES 9)
1871 2.4.21-40.EL (RHEL 3)
1872 2.6.12.6 vanilla (kernel.org)
1877 Bugzilla : 5719, 9635, 9792, 9684
1878 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1879 Details : If a server is under heavy load and cannot reply to new
1880 connection requests before the client resends the (re)connect,
1881 the connection handling code can behave badly if two service
1882 threads are concurrently handing separate (re)connections from
1883 the same client. Add better locking to the connection handling
1884 code, and ensure that only a single connection will be processed
1885 for a given client UUID, even if the lock is dropped.
1887 Severity : enhancement
1889 Description: add TCP zero-copy support to kernel
1890 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1891 sends if the hardware supports scatter-gather and checksumming.
1892 This allows socklnd to do client-write and server-read more
1893 efficiently and reduce CPU utilization from skbuf copying.
1896 Frequency : only if NFS exporting from client
1898 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1899 Details : If setting attributes on a file created under NFS that had
1900 never been opened it would be possible to oops the client
1901 if the file had no objects.
1904 Frequency : always for liblustre
1906 Description: liblustre client does MDS+OSTs setattr RPC for each write
1907 Details : When doing a write from a liblustre client, the client
1908 incorrectly issued an RPC to the MDS and each OST the file was
1909 striped over in order to update the timestamps. When writing
1910 with small chunks and many clients this could overwhelm the MDS
1911 with RPCs. In all cases it would slow down the write because
1912 these RPCs are unnecessary.
1914 Severity : enhancement
1916 Description: allow number of MDS service threads to be changed at module load
1917 Details : It is now possible to change the number of MDS service threads
1918 running. Adding "options mds mds_num_threads={N}" to the MDS's
1919 /etc/modprobe.conf will set the number of threads for the next
1920 time Lustre is restarted (assuming the "mds" module is also
1921 reloaded at that time). The default number of threads will
1922 stay the same, 32 for most systems.
1927 Description: OST crash if filesystem is unformatted or corrupt
1928 Details : If an OST is started on a device that has never been formatted
1929 or if the filesystem is corrupt and cannot even mount then the
1930 error handling cleanup routines would dereference a NULL pointer.
1935 Description: NULL pointer deref in llap_from_page.
1936 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1937 incorrect) mapping if the page was truncated/reclaimed while it was
1938 searched for. Check for this condition and skip such pages when
1939 doing readahead. Introduce extra check to llap_from_page() to
1940 verify page->mapping->host is non-NULL (so page is not anonymous).
1943 Frequency : Sometimes when using sys_sendfile
1945 Description: "page not covered by a lock" warnings from ll_readpage
1946 Details : sendfile called ll_readpage without right page locks present.
1947 Now we introduced ll_file_sendfile that does necessary locking
1948 around call to generic_file_sendfile() much like we do in
1952 Frequency : with certain MDS communication failures at client mount time
1954 Description: NULL pointer deref after failed client mount
1955 Details : a client connection request may delayed by the network layer
1956 and not be sent until after the PTLRPC layer has timed out the
1957 request. If the client fails the mount immediately it will try
1958 to clean up before the network times out the request. Add a
1959 reference from the request import to the obd device and delay
1960 the cleanup until the network drops the request.
1963 Frequency : occasionally during client (re)connect
1965 Description: assertion failure during client (re)connect
1966 Details : processing a client connection request may be delayed by the
1967 client or server longer than the client connect timeout. This
1968 causes the client to resend the connection request. If the
1969 original connection request is replied in this interval, the
1970 client may trip an assertion failure in ptlrpc_connect_interpret()
1971 which thought it would be the only running connect process.
1974 Frequency : only with obd_echo servers and clients that are rebooted
1976 Description: kernel BUG accessing uninitialized data structure
1977 Details : When running an obd_echo server it did not start the ping_evictor
1978 thread, and when a client was evicted an uninitialized data
1979 structure was accessed. Start the ping_evictor in the RPC
1980 service startup instead of the OBD startup.
1982 Severity : enhancement
1983 Bugzilla : 10193 (patchless)
1984 Description: Remove dependency on various unexported kernel interfaces.
1985 Details : No longer need reparent_to_init, exit_mm, exit_files,
1986 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1989 Frequency : rare (only users of deprecated and unsupported LDAP config)
1991 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1992 Details : LDAP apparently contains 'lustreName' attributes instead of
1993 'name'. A simple remapping of the name is sufficient.
1996 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1998 Description: waiting_locks_callback trips kernel BUG if client is evicted
1999 Details : Running with the dump_on_timeout debug flag turned on makes
2000 it possible that the waiting_locks_callback() can try to dump
2001 the Lustre kernel debug logs from an interrupt handler. Defer
2002 this log dumping to the expired_lock_main() thread.
2004 Severity : enhancement
2006 Description: Support NFS exporting on 2.6 kernels.
2007 Details : Implement non-rawops metadata methods for NFS server to use without
2008 changing NFS server code.
2011 Frequency : very rare (synthetic metadata workload only)
2013 Description: two racing renames might cause an MDS thread to deadlock
2014 Details : Running the "racer" program may cause one MDS thread to rename
2015 a file from being the source of a rename to being the target of
2016 a rename at exactly the same time that another thread is doing
2017 so, and the second thread has already enqueued these locks after
2018 doing a lookup of the target and is trying to relock them in
2019 order. Ensure that we don't try to re-lock the same resource.
2022 Frequency : only very large systems with liblustre clients
2024 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2025 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2026 all OSTs in parallel. This allows the eviction of stale liblustre
2027 clients to proceed much faster than if they were done in series,
2028 and also offers similar improvements for other set_info RPCs.
2033 Description: excessive CPU usage during initial read phase on client
2034 Details : During the initial read phase on a client, it would agressively
2035 retry readahead on the file, consuming too much CPU and impacting
2036 performance (since 1.4.5.8). Improve the readahead algorithm
2037 to avoid this, and also improve some other common cases (read
2038 of small files in particular, where "small" is files smaller than
2039 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2044 Description: MDS crash when receiving packet with unknown intent.
2045 Details : Do not LBUG in unknown intent case, just return -EFAULT
2047 Severity : enhancement
2048 Bugzilla : 9293, 9385
2049 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2050 Details : Do not serialize getattr (non-intent version) and statfs.
2053 Frequency : occasional, when OST network is overloaded/intermittent
2055 Description: client evicted by OST after bulk IO timeout
2056 Details : If a client sends a bulk IO request (read or write) the OST
2057 may evict the client if it is unresposive to its data GET/PUT
2058 request. This is incorrect if the network is overloaded (takes
2059 too long to transfer the RPC data) or dropped the OST GET/PUT
2060 request. There is no need to evict the client at all, since
2061 the pinger and/or lock callbacks will handle this, and the
2062 client can restart the bulk request.
2065 Frequency : Always when mmapping file with no objects
2067 Description: client crashes when mmapping file with no objects
2068 Details : Check that we actually have objects in a file before doing any
2069 operations on objects in ll_vm_open, ll_vm_close and
2075 Description: Request leak when working with deleted CWD
2076 Details : Introduce advanced request refcount tracking for requests
2077 referenced from lustre intent.
2079 Severity : Enhancement
2081 Description: Cache open file handles on client.
2082 Details : MDS now will return special lock along with openhandle, if
2083 requested and client is allowed to hold openhandle, even if unused,
2084 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2085 closing files for every read/write openration.
2087 Severity : Enhancement
2089 Description: Cache open negative dentries on client when possible.
2090 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2091 negative dentries on lock revocation.
2096 Description: Remounting a client read-only wasn't possible with a zconf mount
2097 Details : It wasn't possible to remount a client read-only with llmount.
2099 Severity : enhancement
2100 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2101 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2102 MPICH is included to add Lustre-specific ADIO interfaces.
2103 This is based closely on the UFS ADIO layer and only differs
2104 in file creation, in order to allow the OST striping to be set.
2105 This is user-contributed code and not supported by CFS.
2110 Description: extended inode attributes (immutable, append-only) work improperly
2111 when 2.4 and 2.6 kernels are used on client/server or vice versa
2112 Details : Introduce kernel-independent values for these flags.
2114 Severity : enhancement
2117 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2118 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2119 granularity. Now, Fractional MB granularity is be supported,
2120 this is very useful for low memory system.
2122 Severity : enhancement
2124 Description: Getattr by fid
2125 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2126 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2127 have locks and avoids getattr rpc requests for it on every stat.
2130 Frequency : Always, for filesystems larger than 2TB
2132 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2133 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2134 filesystems larger than 2TB to be mounted with mballoc enabled.
2137 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2139 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2140 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2141 filesystems larger than 2TB is mounted then the kernel will
2142 silently corrupt the start of the filesystem. CONFIG_LBD is
2143 enabled for all CFS-supported kernels, but the possibility of
2144 this happening with a modified kernel config exists.
2146 Severity : enhancement
2148 Description: add client O_DIRECT support for 2.6 kernels
2149 Details : It is now possible to do O_DIRECT reads and writes to files
2150 in the Lustre client mountpoint on 2.6 kernel clients.
2152 Severity : enhancement
2154 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2155 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2156 parallel, not waiting for response from every OST before sending
2157 a rpc to the next OST.
2162 Description: setattr vs write race when updating file timestamps
2163 Details : Client processes that update a file timestamp into the past
2164 right after writing to the file (e.g. tar) it is possible that
2165 the updated file modification time can be reset to the current
2166 time due to a race between processing the setattr and write RPC.
2168 Severity : enhancement
2170 Description: Bring 'lfs find' closer in line with regular Linux find.
2171 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2172 -print0 options and obtains all the needed info through the lustre
2175 Severity : enhancement
2177 Description: support up to 1024 configured devices on one node
2178 Details : change obd_dev array from statically allocated to dynamically
2179 allocated structs as they are first used to reduce memory usage
2184 Description: Flush dirty partially truncated pages during truncate
2185 Details : Immediatelly flush partially truncated pages in filter_setattr,
2186 this way we completely avoid having any pages in page cache on OST
2187 and can retire ugly workarounds during writes to flush such pages.
2192 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2193 Details : protect inode from truncation within vfs_unlink() context
2194 just take a reference before calling vfs_unlink() and release it
2195 when parent's i_sem is free.
2200 Description: last_id value checked outside lock on OST caused LASSERT failure
2201 Details : If there were multiple MDS->OST object precreate requests in
2202 flight, it was possible that the OST's last object id was checked
2203 outside a lock and incorrectly tripped an assertion. Move checks
2204 inside locks, and discard old precreate requests.
2207 Frequency : always, if extents are used on OSTs
2209 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2210 Details : OSTs using the extents format would not zero the high 16 bits of
2211 the index physical block number. This is not a problem for any
2212 OST filesystems smaller than 16TB, and no kernels support ext3
2213 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2214 new/modified files) and can be fixed for existing filesystems
2215 with e2fsprogs-1.39-cfs1.
2220 Description: import connection selection may be incorrect if timer wraps
2221 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2222 import connections to be ignored if the 32-bit jiffies counter
2223 wraps. Use a 64-bit jiffies counter.
2226 Frequency : very large clusters immediately after boot
2228 Description: LNET request buffers exhausted under heavy short-term load
2229 Details : If a large number of client requests are generated on a service
2230 that has previously never seen so many requests it is possible
2231 that the request buffer growth cannot keep up with the spike in
2232 demand. Instead of dropping incoming requests, they are held in
2233 the LND until the RPC service can accept more requests.
2236 Frequency : Sometimes during replay
2238 Description: Assertion failure in ll_local_open after replay.
2239 Details : If replay happened on an open request reply before we were able
2240 to set replay handler, reply will become not swabbed tripping the
2241 assertion in ll_local_open. Now we set the handler right after
2242 recognising of open request
2245 Frequency : very rare
2247 Description: kernel reports "badness in vsnprintf"
2248 Details : Reading from the "recovery_status" /proc file in small chunks
2249 may cause a negative length in lprocfs_obd_rd_recovery_status()
2250 call to vsnprintf() (which is otherwise harmless). Exit early
2251 if there is no more space in the output buffer.
2253 Severity : enhancement
2255 Description: clear OBD RPC statistics by writing to them
2256 Details : It is now possible to clear the OBD RPC statistics by writing
2257 to the "stats" file.
2262 Description: Client mtime is not the same on different clients after utimes
2263 Details : In some cases, the client was using the utimes() syscall on
2264 a file cached on another node. The clients now validate the
2265 ctime from the MDS + OSTs to determine which one is right.
2270 Description: Inability to activate failout mode
2271 Details : lconf script incorrectly assumed that in pythong string's numeric
2272 value is used in comparisons.
2275 Frequency : always with multiple stripes per file
2277 Description: Inefficient object allocation for mutli-stripe files
2278 Details : When selecting which OSTs to stripe files over, for files with
2279 a stripe count that divides evenly into the number of OSTs,
2280 the MDS is always picking the same starting OST for each file.
2281 Return the OST selection heuristic to the original design.
2286 Description: mount failures may take full timeout to return an error
2287 Details : Under some heavy load conditions it is possible that a
2288 failed mount can wait for the full obd_timeout interval,
2289 possibly several minutes, before reporting an error.
2290 Instead return an error as soon as the status is known.
2292 Frequency : quota enabled and large files being deleted
2294 Description: releasing more than 4GB of quota at once hangs OST
2295 Details : If a user deletes more than 4GB of files on a single OST it
2296 will cause the OST to spin in an infinite loop. Release
2297 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2302 Description: statfs data retrieved from /proc may be stale or zero
2303 Details : When reading per-device statfs data from /proc, in the
2304 {kbytes,files}_{total,free,avail} files, it may appear
2305 as zero or be out of date.
2308 Frequency : systems with MD RAID1 external journal devices
2310 Description: lconf's call to blkid is confused by RAID1 journal devices
2311 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2312 of returning all block devices that match the journal UUID.
2315 Frequency : always, for aggregate stripe size over 4GB
2317 Description: assertion fails when trying to use 4GB stripe size
2318 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2319 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2322 Frequency : always on ppc64
2324 Description: the first write on an ext3 filesystem with mballoc got stuck
2325 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2326 perform endianness conversion.
2328 ------------------------------------------------------------------------------
2330 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2332 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2333 INTEROPERATE with older versions automatically. Please read the
2334 user documentation before upgrading any part of a live system.
2335 * WARNING: Lustre networking configuration changes are required with
2336 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2339 * Support for kernels:
2340 2.6.9-22.0.2.EL (RHEL 4)
2341 2.6.5-7.244 (SLES 9)
2342 2.6.12.6 vanilla (kernel.org)
2345 Severity : enhancement
2346 Bugzilla : 7981/8208
2347 Description: Introduced Lustre Networking (LNET)
2348 Details : LNET is new networking infrastructure for Lustre, it includes
2349 a reorganized network configuration mode (see the user
2350 documentation for full details) as well as support for routing
2351 between different network fabrics. Lustre Networking Devices
2352 (LNDs) for the supported network fabrics have also been
2353 created for this new infrastructure.
2355 Severity : enhancement
2356 Description: Introduced Access control lists
2357 Details : clients can set ACLs on files and directories in order to have
2358 more fine-grained permissions than the standard Unix UGO+RWX.
2359 The MDS must be started with the "-o acl" mount option.
2361 Severity : enhancement
2362 Description: Introduced filesystem quotas
2363 Details : Administrators may now establish per-user quotas on the
2366 Severity : enhancement
2368 Description: Configuration change for the XT3
2369 The PTLLND is now used to run Lustre over Portals on the XT3
2370 The configure option(s) --with-cray-portals are no longer used.
2371 Rather --with-portals=<path-to-portals-includes> is used to
2372 enable building on the XT3. In addition to enable XT3 specific
2373 features the option --enable-cray-xt3 must be used.
2378 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2379 Details : A race between allocating a new llcd and re-getting the llcd_lock
2380 allowed another thread to grab newly-allocated llcd.
2382 Severity : enhancement
2384 Description: 2.6 OST async journal commit and locking fix to improve performance
2385 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2386 2.6 kernels are now async as they already were in 2.4 kernels so
2387 that they can commit concurrently with the network bulk transfer.
2388 For block-allocated files the filter allocation semaphore is held
2389 to avoid filesystem fragmentation during allocation. BKL lock
2390 removed for 2.6 xattr operations where it is no longer needed.
2395 Description: lconf incorrectly determined whether two IP networks could talk
2396 Details : In some more complicated routing and multiple-network
2397 configurations, lconf will avoid trying to make a network
2398 connection to a disjoint part of the IP space. It was doing the
2399 math incorrectly for one set of cases.
2404 Description: Fix for potential infinite loop processing records in an llog.
2405 Details : If an llog record is corrupted/zeroed, it is possible to loop
2406 forever in llog_process(). Validate the llog record length
2407 and skip the remainder of the block on error.
2410 Frequency : occasional (liblustre only)
2412 Description: liblustre could not open files whose last component is a symlink
2413 Details : sysio_path_walk() would incorrectly pass the open intent to
2414 intermediate path components.
2417 Frequency : rare (liblustre only with non-standard tuning)
2418 Bugzilla : 7201 (7350)
2419 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2420 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2421 after it has been released, only do this for referenced locks
2423 Severity : enhancement
2425 Description: specifying an (invalid) directory default stripe_size of -1
2426 would reset the directory default striping
2427 Details : stripe_size -1 was used internally to signal directory stripe
2428 removal, now use "all default" to signal dir stripe removal
2429 as a directory striping of "all default" is not useful
2432 Frequency : common for large clusters running liblustre clients
2434 Description: doing an ls when liblustre clients are running is slow
2435 Details : sending a glimpse AST to a liblustre client waits for every AST
2436 to time out, as liblustre clients will not respond. Since they
2437 cannot cache data we refresh the OST lock LVB from disk instead.
2439 Severity : enhancement
2441 Description: doing an ls at the same time as file IO can be slow
2442 Details : enqueue and other "small" requests can be blocked behind many
2443 large IO requests. Create a new OST IO portal for non-IO
2444 requests so they can be processed faster.
2447 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2449 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2450 created file with mode 000
2451 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2452 allow an ftruncate() as a non-root user to a file with mode 000.
2453 Lustre now respects this flag to disable mode checks when
2454 truncating a file owned by the user
2457 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2460 Description: Revoking locks from clients that went dead or catatonic might take
2462 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2463 cancellation of such locks instant on servers without waiting for
2464 any reply from clients. Clients drops these locks when cancel
2465 notification from server is received without replying.
2468 Frequency : liblustre-only, when liblustre client dies or becomes busy
2470 Description: Doing ls on Linux clients can take a long time with active
2472 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2473 granting such locks to it in the first place if possible. Locks
2474 are taken by proxy on the OST during the read or write and
2475 dropped immediately afterward. Add connect flags handling, do
2476 not grant locks to liblustre clients for glimpse ASTs.
2478 Severity : enhancement
2480 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2481 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2482 sequential reads and behaves badly with applications doing random
2483 reads. Improve it by reading ahead at least read region, and
2484 avoiding excessive large RPC for small reads.
2486 Severity : enhancement
2488 Description: Creating more than 1000 files for a single job may cause a load
2489 imbalance on the OSTs if there are also a large number of OSTs.
2490 Details : qos_prep_create() uses an OST index reseed value that is an
2491 even multiple of the number of available OSTs so that if the
2492 reseed happens in the middle of the object allocation it will
2493 still utilize the OSTs as uniformly as possible.
2498 Description: OST or MDS may oops in ping_evictor_main()
2499 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2500 but doesn't restart at beginning of obd_exports_timed list
2503 Severity : enhancement
2505 Description: improve by-nid export eviction on the MDS and OST
2506 Details : allow multiple exports with the same NID to be evicted at one
2507 time without re-searching the exports list.
2510 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2512 Description: MDS may oops in groups_free()
2513 Details : in rare race conditions a newly allocated group_info struct is
2514 freed again, and this can be NULL. The 2.4 compatibility code
2515 for groups_free() checked for a NULL pointer, but 2.6 did not.
2518 Frequency : common for liblustre clients doing little filesystem IO
2519 Bugzilla : 9352, 7313
2520 Description: server may evict liblustre clients accessing contended locks
2521 Details : if a client is granted a lock or receives a completion AST
2522 with a blocking AST already set it would not reply to the AST
2523 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2526 Frequency : lfs setstripe, only systems with more than 160 OSTs
2528 Description: unable to set striping with a starting offset beyond OST 160
2529 Details : llapi_create_file() incorrectly limited the starting stripe
2530 index to the maximum single-file stripe count.
2533 Frequency : LDAP users only
2535 Description: lconf did not handle in-kernel recovery with LDAP properly
2536 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2538 Severity : enhancement
2540 Description: bind OST threads to NUMA nodes to improve performance
2541 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2542 node and do their allocations there to localize memory access
2544 Severity : enhancement
2546 Description: llmount can determine client NID directly from Myrinet (GM)
2547 Details : the client NID code from gmnalnid was moved directly into
2548 llmount, removing the need to use this or specifying the
2549 client NID explicitly when mounting GM clients with zeroconf
2552 Frequency : if client is started with down MDS
2554 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2555 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2556 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2557 ptlrpc_import_delayed_req() because we didn't check if the
2558 request was interrupted, and we also didn't break out of the
2559 event loop if there was no timeout
2564 Description: data loss during non-page-aligned writes to a single file from
2565 both multiple nodes and multiple threads on one node at same time
2566 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2567 inconsistency led to false short-reads, that were cached and later
2568 used by ->prepare_write() to fill in partially written page,
2569 leading to data loss.
2572 Frequency : always, if lconf --abort_recovery used
2574 Description: lconf --abort_recovery fails with 'Operation not supported'
2575 Details : lconf was attempting to abort recovery on the MDT device and not
2578 Severity : enhancement
2580 Description: remove cleanup logs
2581 Details : replace lconf-generated cleanup logs with lustre internal
2582 cleanup routines. Eliminates the need for client-cleanup and
2585 Severity : enhancement
2587 Description: add support for EAs (user and system) on lustre filesystems
2588 Details : it is now possible to store extended attributes in the Lustre
2589 client filesystem, and with the user_xattr mount option it
2590 is possible to allow users to store EAs on their files also
2592 Severity : enhancement
2594 Description: Add possibility (config option) to show minimal available OST free
2596 Details : When compiled with --enable-mindf configure option, statfs(2)
2597 (and so, df) will return least minimal free space available from
2598 all OSTs as amount of free space on FS, instead of summary of
2599 free spaces of all OSTs.
2601 Severity : enhancement
2603 Description: do not expand extent locks acquired on OST-side
2604 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2605 by server: they are not cached anyway.
2608 Frequency : when mmap is used/binaries executed from Lustre
2610 Description: Unmmap pages before throwing them away from read cache.
2611 Details : llap_shrink cache now attempts to unmap pages before discarding
2612 them (if unmapping failed - do not discard). SLES9 kernel has
2613 extra checks that trigger if this unmapping is not done first.
2618 Description: lconf didn't resolve symlinks before checking to see whether a
2619 given mountpoint was already in use
2622 Frequency : when migrating failover services
2623 Bugzilla : 6395, 9514
2624 Description: When migrating a subset of services from a node (e.g. failback
2625 from a failover service node) the remaining services would
2626 time out and evict clients.
2627 Details : lconf --force (implied by --failover) sets the global obd_timeout
2628 to 5 seconds in order to quickly disconnect, but this caused
2629 other RPCs to time out too quickly. Do not change the global
2630 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2632 Severity : enhancement
2633 Frequency : if MDS is started with down OST
2634 Bugzilla : 9439,5706
2635 Description: Allow startup/shutdown of an MDS without depending on the
2636 availability of the OSTs.
2637 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2638 Add appropriate locking and lov-osc refcounts for safe
2639 cleaning. Add osc abort_inflight calls in case the
2640 synchronize never started.
2643 Frequency : occasional (Cray XT3 only)
2645 Description: root not authorized to access files in CRAY_PORTALS environment
2646 Details : The client process capabilities were not honoured on the MDS in
2647 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2648 been accessed by an authorized user then root was able to access
2649 the file on the local client also. The root user capabilities
2650 are now allowed on the MDS, as this environment has secure UID.
2653 Frequency : occasional
2655 Description: ldiskfs "too long searching" message happens too often
2656 Details : A debugging message (otherwise harmless) prints too often on
2657 the OST console. This has been reduced to only happen when
2658 there are fragmentation problems on the filesystem.
2663 Description: Division by zero in statfs when all OSCs are inactive
2664 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2665 lov_statfs divides by value returned by lov_get_stripecnt().
2669 Bugzilla : 9489, 3273
2670 Description: First write from each client to each OST was only 4kB in size,
2671 to initialize client writeback cache, which caused sub-optimal
2672 RPCs and poor layout on disk for the first writen file.
2673 Details : Clients now request an initial cache grant at (re)connect time
2674 and so that they can start streaming writes to the cache right
2675 away and always do full-sized RPCs if there is enough data.
2676 If the OST is rebooted the client also re-establishes its grant
2677 so that client cached writes will be honoured under the grant.
2682 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2683 Details : Now I/O RPCs go to different portal number and (presumably) fast
2684 lock requests (and glimses) and other RPCs get their own service
2685 threads pool that should be able to service those RPCs
2688 Severity : enhancement
2690 Description: Ability to exchange lustre version between client and servers and
2691 issue warnings at client side if client is too old. Also for
2692 liblustre clients there is ability to refuse connection of too old
2694 Details : New 'version' field is added to connect data structure that is
2695 filled with version info. That info is later checked by server and
2699 Frequency : rare, liblustre only.
2700 Bugzilla : 9296, 9581
2701 Description: Two simultaneous writes from liblustre at offset within same page
2702 might proceed at the same time overwriting eachother with stale
2704 Details : I/O lock withing llu_file_prwv was released too early, before data
2705 actually was hitting the wire. Extended lock-holding time until
2706 server acknowledges receiving data.
2709 Frequency : extremely rare. Never observed in practice.
2711 Description: avoid generating lustre_handle cookie of 0.
2712 Details : class_handle_hash() generates handle cookies by incrementing
2713 global counter, and can hit 0 occasionaly (this is unlikely, but
2714 not impossible, because initial value of cookie counter is
2715 selected randonly). Value of 0 is used as a sentinel meaning
2716 "unassigned handle" --- avoid it. Also coalesce two critical
2717 sections in this function into one.
2719 Severity : enhancement
2721 Description: allow liblustre clients to delegate truncate locking to OST
2722 Details : To avoid overhead of locking, liblustre client instructs OST to
2723 take extent lock in ost_punch() on client's behalf. New connection
2724 flag is added to handle backward compatibility.
2726 Severity : enhancement
2727 Bugzilla : 4928, 7341, 9758
2728 Description: allow number of OST service threads to be specified
2729 Details : a module parameter allows the number of OST service threads
2730 to be specified via "options ost ost_num_threads={N}" in the
2731 OSS's /etc/modules.conf or /etc/modprobe.conf.
2735 Bugzilla : 6146, 9635, 9895
2736 Description: servers crash with bad pointer in target_handle_connect()
2737 Details : In rare cases when a client is reconnecting it was possible that
2738 the connection request was the last reference for that export.
2739 We would temporarily drop the export reference and get a new
2740 one, but this may have been the last reference and the export
2741 was just destroyed. Get new reference before dropping old one.
2743 Severity : enhancement
2744 Frequency : if client is started with failover MDS
2746 Description: Allow multiple MDS hostnames in the mount command
2747 Details : Try to read the configuration from all specified MDS
2748 hostnames during a client mount in case the "primary"
2751 Severity : enhancement
2753 Description: Stop sending data to evicted clients as soon as possible.
2754 Details : Check if the client we are about to send or are sending data to
2755 was evicted already. (Check is done every second of waiting,
2756 for which l_wait_event interface was extended to allow checking
2757 of exit condition at specified intervals).
2760 Frequency : rare, normally only when NFS exporting is done from client
2762 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2764 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2767 Frequency : always, if srand() is called before liblustre initialization
2769 Description: Liblustre uses system PRNG disturbing its usage by user application
2770 Details : Introduce internal to lustre fast and high-quality PRNG for
2771 lustre usage and make liblustre and some other places in generic
2772 lustre code to use it.
2774 Severity : enhancement
2775 Bugzilla : 9477, 9557, 9870
2776 Description: Verify that the MDS configuration logs are updated when xml is
2777 Details : Check if the .xml configuration logs are newer than the config
2778 logs stored on the MDS and report an error if this is the case.
2779 Request --write-conf, or allow starting with --old_conf.
2781 Severity : enhancement
2783 Description: Handle symlinks in the path when checking if Lustre is mounted.
2784 Details : Resolve intermediate symlinks when checking if a client has
2785 mounted a filesystem to avoid duplicate client mounts.
2790 Description: lconf can hit an error exception but still return success.
2791 Details : The lconf command catches the Command error exception at the top
2792 level script context and will exit with the associated exit
2793 status, but doesn't ensure that this exit status is non-zero.
2798 Description: failure of ptlrpc thread startup can cause oops
2799 Details : Starting a ptlrpc service thread can fail if there are a large
2800 number of threads or the server memory is very fragmented.
2801 Handle this without oopsing.
2804 Frequency : always, only if liblustre and non-default acceptor port was used
2806 Description: liblustre cannot connect to servers with non-default acceptor port
2807 Details : tcpnal_set_default_params() was not called and was therefore
2808 ignoring the environment varaible TCPNAL_PORT, as well as other
2809 TCPNAL_ environment variables
2814 Description: two objects could be created on the same OST for a single file
2815 Details : If an OST is down, in some cases it was possible to create two
2816 objects on a single OST for a single file. No problems other
2817 than potential performance impact and spurious error messages.
2821 Bugzilla : 5681, 9562
2822 Description: Client may oops in ll_unhash_aliases
2823 Details : Client dcache may become inconsistent in race condition.
2824 In some cases "getcwd" can fail if the current directory is
2830 Description: Inode refcounting problems in NFS export code
2831 Details : link_raw functions used to call d_instantiate without obtaining
2832 extra inode reference first.
2836 Bugzilla : 9942, 9903
2837 Description: Referencing freed requests leading to crash, memleaks with NFS.
2838 Details : We used to require that call to ll_revalidate_it was always
2839 followed by ll_lookup_it. Also with revalidate_special() it is
2840 possible to call ll_revalidate_it() twice for the same dentry
2841 even if first occurence returned success. This fix changes semantic
2842 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2843 reference on a request referred from the intent.
2844 ll_intent_release() then releases such a request.
2847 Frequency : rare, normally benchmark loads only
2849 Description: unlinked inodes were kept in memory on the client
2850 Details : If a client is repeatedly creating and unlinking files it
2851 can accumulate a lot of stale inodes in the inode slab cache.
2852 If there is no other client load running this can cause the
2853 client node to run out of memory. Instead flush old inodes
2854 from client cache that have the same inode number as a new inode.
2857 Frequency : SLES9 2.6.5 kernel and long filenames only
2858 Bugzilla : 9969, 10379
2859 Description: utime reports stale NFS file handle
2860 Details : SLES9 uses out-of-dentry names in some cases, which confused
2861 the lustre dentry revalidation. Change it to always use the
2865 Frequency : rare, unless heavy write-truncate concurrency is continuous
2866 Bugzilla : 4180, 6984, 7171, 9963, 9331
2867 Description: OST becomes very slow and/or deadlocked during object unlink
2868 Details : filter_destroy() was holding onto the parent directory lock
2869 while truncating+unlinking objects. For very large objects this
2870 may block other threads for a long time and slow overall OST
2871 responsiveness. It may also be possible to get a lock ordering
2872 deadlock in this case, or run out of journal credits because of
2873 the combined truncate+unlink. Solution is to do object truncate
2874 first in one transaction without parent lock, and then do the
2875 final unlink in a new transaction with the parent lock. This
2876 reduces the lock hold time dramatically.
2879 Frequency : rare, 2.4 kernels only
2881 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2882 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2883 before unmounting their filesystem in order to avoid deadlock.
2884 The kernel_locked() function in 2.4 kernels only checks whether
2885 the kernel lock is held, not whether it is this process that is
2886 holding it as 2.6 kernels do.
2891 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2892 Details : The client ptlrpc code may be trying to reconnect to a down
2893 server before a previous connection attempt has timed out.
2894 Increase the reconnect interval to be longer than the connection
2895 timeout interval to avoid sending duplicate connections to
2899 Frequency : echo_client brw_test command
2901 Description: fix echo_client to work with OST preallocated code
2902 Details : OST preallocation code (5137) didn't take echo_client IO path
2903 into account: echo_client calls filter methods outside of any
2904 OST thread and, hence, there is no per-thread preallocated
2905 pages and buffers to use. Solution: hijack pga pages for IO. As
2906 a byproduct, this avoids unnecessary data copying.
2910 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2911 Description: Client can oops in mdc_commit_close() after open replay
2912 Details : It was possible for the MDS to return an open request with no
2913 transaction number in mds_finish_transno() if the client was
2914 evicted, but without actually returning an error. Clients
2915 would later try to replay that open and may trip an assertion
2916 Simplify the client close codepath, and always return an error
2917 from the MDS in case the open is not successful.
2920 Frequency : rare, 2.6 OSTs only
2922 Description: OST may deadlock under high load on fragmented files
2923 Details : If there was a heavy load and highly-fragmented OST filesystems
2924 it was possible to have all the OST threads deadlock waiting on
2925 allocation of biovecs, because the biovecs were not released
2926 until the entire RPC IO was completed. Instead, release biovecs
2927 as soon as they are complete to ensure forward IO progress.
2929 Severity : enhancement
2931 Description: Support for specifying external journal device at mount
2932 Details : If an OST or MDS device is formatted with an external journal
2933 device, this device major/minor is stored in the ext3 superblock
2934 and may not be valid for failover. Allow detecting and
2935 specifying the external journal at mount time.
2940 Description: Mounting an MDS with pending unlinked files may cause oops
2941 Details : target_finish_recovery() calls mds_postrecov() which returned
2942 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2943 considers this an error and immediately begins cleaning up the
2944 lov, just after starting the mds_lov process
2946 Severity : enhancement
2948 Description: Implement 'lfs df' to report actual free space on per-OST basis
2949 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2950 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2951 usage of objects; '-h' to report in human readable format.
2953 ------------------------------------------------------------------------------
2955 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2962 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2963 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2964 at startup during orphan unlinking. Instead, skip buddy update
2965 before setup, as it will be regenerated after recovery is complete.
2970 Description: If an OST is inactive, its locks might reference stale inodes.
2971 Details : lov_change_cbdata() must iterate over all namespaces, even if
2972 they are inactive to clear inode references from the lock.
2974 Severity : enhancement
2975 Frequency : occasional, if non-standard max_dirty_mb used
2977 Description: Client will block write RPCs if not enough grant
2978 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2979 then the client will block writes while waiting for another RPC
2980 to complete instead of consuming its dirty limit. With change
2981 we get improved performance when max_dirty_mb is small.
2983 Severity : enhancement
2984 Bugzilla : 3389, 6253
2985 Description: Add support for supplementary groups on the MDS.
2986 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2987 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2988 lookups for user supplementary groups into a cache.
2992 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2993 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2994 for liblustre O_EXCL is enforced by MDS.
2998 Description: readdir never returns NULL in liblustre.
2999 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3000 dentry in struct dirent.
3004 Frequency : liblustre only, depends on application IO pattern
3005 Description: liblustre clients evicted if not contacting servers
3006 Details : Don't put liblustre clients into the ping_evictor list, so
3007 they will not be evicted by the pinger ever.
3009 Severity : enhancement
3011 Description: Add ability to evict clients by NID from MDS.
3012 Details : By echoing "nid:$NID" string into
3013 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3014 $NID would be instantly evicted from this MDS and from all active
3015 OSTs connected to it.
3019 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3020 Details : lookup_it_finish() used to query file size from OSTs that was not
3025 Description: service threads change working directory to that of init
3026 Details : Starting lustre service threads may pin the working directory
3027 of the parent thread, making that filesystem busy. Threads
3028 now change to the working directory of init to avoid this.
3032 Frequency : during shutdown only
3033 Description: shutdown with a failed MDS or OST can cause unmount to hang
3034 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3039 Frequency : 2.6 only
3040 Description: chmod/chown may include an extra supplementary group
3041 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3042 supplementary group and if none is specified this is used.
3045 Bugzilla : 5479 (6816)
3047 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3048 Details : If lookup is in progress on a file that is unlinked we might try
3049 to revalidate the inode and fail in revalidate after lookup is
3050 complete and ll_file_open() enqueues the open again but
3051 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3054 Frequency : always, if lconf --abort_recovery used
3056 Description: lconf --abort_recovery fails with 'Operation not supported'
3057 Details : lconf was attempting to abort recovery on the MDT device and not
3060 ------------------------------------------------------------------------------
3062 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3067 Frequency : rare (only unsupported configurations with a node running as an
3069 Bugzilla : 6514, 5137
3070 Description: Mounting a Lustre file system on a node running as an OST could
3072 Details : OSTs now preallocates memory needed to write out data at
3073 startup, instead of when needed, to avoid having to
3074 allocate memory in possibly low memory situations.
3075 Specifically, if the file system is mounted on on OST,
3076 memory pressure could force it to try to write out data,
3077 which it needed to allocate memory to do. Due to the low
3078 memory, it would be unable to do so and the node would
3079 become unresponsive.
3081 Severity : enhancement
3083 Description: Addition of lconf --service command line option
3084 Details : lconf now accepts a '--service <arg>' option, which is
3085 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3087 Severity : enhancement
3089 Description: Failover mode is now the default for OSTs.
3090 Details : By default, OSTs will now run in failover mode. To return to
3091 the old behaviour, add '--failout' to the lmc line for OSTs.
3093 Severity : enhancement
3095 Description: Health checks are now provided for MDS and OSTs
3096 Details : Additional detailed health check information on MSD and OSTs
3097 is now provided through the procfs health_check value.
3100 Frequency : occasional, depends on IO load
3102 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3103 after numerous create/delete cycles
3104 Details : The ext3 inode allocation policy would not allocate new inodes
3105 very well on the OSTs because there are no new directories
3106 being created. Instead we look for groups with free space if
3107 the parent directories are nearly full.
3112 Description: Network or server problems during mount may cause partially
3113 mounted clients instead of returning an error.
3114 Details : The config llog parsing code may overwrite the error return
3115 code during mount error handling, returning success instead
3121 Description: MDS can fail to allocate large reply buffers
3122 Details : After long uptimes the MDS can fail to allocate large reply
3123 buffers (e.g. zconf client mount config records) due to memory
3124 fragmentation or consumption by the buffer cache. Preallocate
3125 some large reply buffers so that these replies can be sent even
3126 under memory pressure.
3130 Frequency : rare (liblustre)
3131 Description: fsx running with liblustre complained that using truncate() to
3132 extend the file doesn't work. This patch corrects that issue.
3133 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3134 ATTR_SIZE and lsm use in llu_setattr_raw.
3138 Frequency : rare, only 2.6 kernels
3139 Description: Unusual file access patterns on the MDS may result in inode
3140 data being lost in very rare circumstances.
3141 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3142 code in the 2.6 kernel caused the inode and/or EA data not to
3143 be read from disk, causing single-file corruption.
3147 Frequency : rare, only 2.6 filesystems using extents
3148 Description: Heavy concurrent write and delete load may cause data corruption.
3149 Details : It was possible under high-load situations to have an extent
3150 metadata block in the block device cache from a just-unlinked
3151 file overwrite a newly-allocated data block. We now unmap any
3152 metadata buffers that alias just-allocated data blocks.
3156 Frequency : filesystems with default stripe_count larger than 77
3157 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3158 Details : lconf specifies an inode size of 4096 bytes when the default
3159 stripe_count is larger than 77. This conflicts with the default
3160 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3161 this case to avoid pinning too much memory for large EAs.
3163 ------------------------------------------------------------------------------
3165 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3170 Frequency : rare (extremely heavy IO load with hundreds of clients)
3172 Description: Client is evicted, gets IO error writing to file
3173 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3174 caused clients to be evicted for AST timeouts. The fixes in
3175 bug 5192 mean we no longer need to have such short AST timeouts
3176 so ldlm_timeout has been increased.
3179 Frequency : occasional during --force or --failover shutdown under load
3180 Bugzilla : 5949, 4834
3181 Description: Server oops/LBUG if stopped with --force or --failover under load
3182 Details : a collection of import/export refcount and cleanup ordering
3183 issues fixed for safer force cleanup
3186 Frequency : only filesystems larger than 120 OSTs
3187 Bugzilla : 5990, 6223
3188 Description: lfs getstripe would oops on a very large filesystem
3189 Details : lov_getconfig used kfree on vmalloc'd memory
3192 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3193 Bugzilla : 6242, 6243
3194 Description: reading from files that had been truncated to a non-zero size
3195 but never opened returned no data
3196 Details : ll_file_read() reads zeros from no-object files to EOF
3201 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3202 Details : There's little harm in aborting MDS/OSS recovery and letting it
3203 try again, so I removed the LASSERT and return an error instead.
3205 Severity : enhancement
3207 Description: New debugging infrastructure for tracking down data corruption
3208 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3209 (b) cover more of the client-side code path, and (c) try to narrow
3210 down where problems occurred
3214 Bugzilla : 3819, 4364, 4397, 6313
3215 Description: Racing close and eviction MDS could cause assertion in mds_close
3216 Details : It was possible to get multiple mfd references during close and
3217 client eviction, leading to one thread referencing a freed mfd.
3219 Severity: : enhancement
3220 Bugzilla : 3262, 6359
3221 Description: Attempts to reconnect to servers are now more aggressive.
3222 Details : This builds on the enhanced upcall-less recovery that was added
3223 in 1.4.2. When trying to reconnect to servers, clients will
3224 now try each server in the failover group every 10 seconds. By
3225 default, clients would previously try one server every 25 seconds.
3230 Description: After recovery, certain operations trigger a failed
3231 assertion on a client.
3232 Details : Failing over an mds, using lconf -d --failover, while a
3233 client was doing a readdir() call would cause the client to
3234 LBUG after recovery completed and the readdir() was resent.
3236 Severity : enhancement
3238 Description: Default groups are now added by lconf
3239 Details : You can now run lconf --group <servicename> without having to
3240 manually add groups with lmc.
3243 Frequency : occasional
3245 Description: Nodes with an elan id of 0 trigger a failed assertion
3248 Frequency : always when accessing e.g. tty/console device nodes
3250 Description: tty and some other devices nodes cannot be used on lustre
3251 Details : file's private_data field is used by device data and lustre
3252 values in there got lost. New field was added to struct file to
3253 store fs-specific private data.
3256 Frequency : when exporting Lustre via NFS
3258 Description: NFSD failed occasionally when looking up a path component
3259 Details : NFSD is looking up ".." which was broken in ext3 directories
3260 that had grown large enough to become hashed.
3263 Frequency : Clusters with multiple interfaces not on the same subnet
3265 Description: Nodes will repeatedly try to reconnect to an interface which it
3266 cannot reach and report an error to the log.
3267 Details : Extra peer list entries will be created by lconf with some peers
3268 unreachable. lconf now validates the peer before adding it.
3271 Frequency : Only if a default stripe is set on the filesystem root.
3273 Description: Setting a default stripe on the filesystem root prevented the
3274 filesystem from being remounted.
3275 Details : The client was sending extra request flags in the root getattr
3276 request and did not allocate a reply buffer for the dir EA.
3279 Frequency : occasional, higher if lots of files are accessed by one client
3280 Bugzilla : 6159, 6097
3281 Description: Client trips assertion regarding lsm mismatch/magic
3282 Details : While revalidating inodes the VFS looks up inodes with ifind()
3283 and in rare cases can find an inode that is being freed.
3284 The ll_test_inode() code will free the lsm during ifind()
3285 when it finds an existing inode and then the VFS later attaches
3286 this free lsm to a new inode.
3290 Bugzilla : 6422, 7030
3291 Description: MDS deadlock between mkdir and client eviction
3292 Details : Creating a new file via mkdir or mknod (starting a transaction
3293 and getting the ns lock) can deadlock with client eviction
3294 (gets ns lock and trying to finish a synchronous transaction).
3297 Frequency : occasional
3298 Description: While starting a server, the fsfilt_ext3 module could not be
3300 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3301 kernels. Previously, lconf would still use the ext3 name
3302 when trying to load modules. Now, it will correctly use
3303 ext3 on 2.4 and ldiskfs on 2.6.
3305 Severity : enhancement
3306 Description: The default stripe count has been changed to 1
3307 Details : The interpretation of the default stripe count (0, to lfs
3308 or lmc) has been changed to mean striping across a single
3309 OST, rather than all available. For general usage we have
3310 found a stripe count of 1 or 2 works best.
3312 Severity : enhancement
3313 Description: Add support for compiling against Cray portals.
3314 Details : Conditional compiling for some areas that are different
3318 Frequency : occasional
3319 Bugzilla : 6409, 6834
3320 Description: Creating files with an explicit stripe count may lead to
3321 a failed assertion on the MDS
3322 Details : If some OSTs are full or unavailable, creating files may
3323 trigger a failed assertion on the MDS. Now, Lustre will
3324 try to use other servers or return an error to the
3328 Frequency : occasional
3330 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3331 caused lock timeout during readahead (since 1.4.2).
3332 Details : Processes doing readahead might match a lock that hasn't been
3333 granted yet if there are overlapping and conflicting lock
3334 requests. The readahead process waits on ungranted lock
3335 (original lock is CBPENDING), while OST waits for that process
3336 to cancel CBPENDING read lock and eventually evicts client.
3338 Severity : enhancement
3340 Description: Initial enabling of flock support for clients
3341 Details : Implements fcntl advisory locking and file status functions.
3342 This feature is provided as an optional mount flag (default
3343 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3344 locking are implemented yet, and those that are are not guaranteed
3345 to be completely correct in production environments.
3346 mount -t lustre -o [flock|noflock] ...
3349 Frequency : occasional
3351 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3352 assertion in the ext3 JBD (journaling) layer.
3353 Details : The b_committed_data struct is protected by the big kernel lock
3354 in 2.4 kernels, serializing journal_commit_transaction() and
3355 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3356 access to this struct. In 2.6 kernels there is finer grained
3357 locking to improve SMP performance of the JBD layer.
3361 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3366 Description: merge and cleanup kernel patches.
3369 -----------------------------------------------------------------------------
3371 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3373 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3374 versions of Lustre. Please update all servers and clients to
3375 version 1.4.2 or later at the same time. You must also run
3376 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3377 to update the configuration logs.
3379 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3380 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3381 - swab llog records of type '0' so we get proper header size/idx (5861)
3382 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3383 - fix rename of one directory over another leaking an inode (5953)
3384 - avoid SetPageDirty on 2.6 (5981)
3385 - don't re-add just-being-destroyed locks to the waiting list (5653)
3386 - when creating new directories, inherit the parent's custom
3387 striping settings if present parent (3048)
3388 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3389 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3390 - don't hold client locks on temporary worklist from l_lru (5666)
3391 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3392 - automatically evict dead clients (5921)
3393 - Update file size properly in create+truncate+fstat case (6196)
3394 - Do not unhash mountpoint dentries, do not allow removal of
3396 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3397 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3398 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3399 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3400 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3401 - Correctly handle reads of files with no objects (6243)
3402 - lctl recover will also mark a device active if deactivate used (5933)
3404 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3405 - allow --write-conf on an MDS with different nettype than client (5619)
3406 - don't write config llogs to MDS for mounts not from that MDS (5617)
3407 - lconf should create multiple TCP connections from a client (5201)
3408 - init scripts are now turned off by default; run chkconfig --on
3409 lustre and chkconfig --on lustrefs to use them
3410 - upcalls are no longer needed for clients to recover to failover
3412 - add --abort-recovery option to lconf to abort recovery on device
3414 - add support for an arbitrary number of OSTs (3026)
3415 - Quota support protocol changes.
3416 - forward compatibility changes to wire structs (6007)
3417 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3418 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3419 - improved "lustre status" script
3420 - initialize blocksize for non-regular files (6062)
3421 - added --disable-server and --disable-client configure options (5782)
3422 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3423 - Vanilla 2.4.29 support
3424 - increase maximum number of obd devices to 520 (6242)
3425 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3426 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3427 SLES 9 2.6.5 kernels
3428 - limit stripes per file to 160 (the maximum EA size) (6093)
3430 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3433 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3434 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3435 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3436 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3437 - small ext3 extents cleanups and fixes (5733)
3438 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3439 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3440 - avoid needless client->OST connect, fix handle mismatch (5317)
3441 - fix DLM error path that led to out-of-sync client, long delays (5779)
3442 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3443 - fix several locking issues related to i_size (5492,5624,5654,5672)
3444 - don't move pending lock onto export if it is already evicted (5683)
3445 - fix kernel oops when creating .foo in unlinked directory (5548)
3446 - fix deadlock in obdfilter statistics vs. object create (5811)
3447 - use time_{before,after} to avoid timer jiffies wrap (5882)
3448 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3449 - Do not leak request if server was not able to process it (5154)
3450 - If mds_open unable to find parent dir, make that negative lookup(5154)
3451 - don't create new directories with extent-mapping (5909, 5936)
3453 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3454 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3455 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3456 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3457 - support for RHEL4 kernel on i686 (5773)
3458 - provide error messages when incompatible logs are encountered (5898)
3460 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3461 * version 1.4.0.10 (1.4.1 release candidate 1)
3463 - don't keep a lock reference when lock is not granted (4238)
3464 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3465 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3466 - drop import inflight refcount on signal_completed_replay error (5255)
3467 - unlock page after async write error during send (3677)
3468 - handle missing objects in filter_preprw_read properly (5265)
3469 - no transno return for symlink open, don't save no-trasno open (3440)
3470 - don't try to complete elan receive that already failed (4012)
3471 - free RPC server reply state on error (5406)
3472 - clean up thread from ptlrpc_start_thread() on error (5160)
3473 - readahead could read extra page into cache that wasn't ejected (5388)
3474 - prevent races in class_attach/setup/cleanup/detach (5260)
3475 - don't dereference de->d_inode after l_dput of de (5458)
3476 - use "int" for stripe value returned from lock_to_stripe (5544)
3477 - mballoc allocation and error-checking fixes in 2.6 (5504)
3478 - block device patches to fix I/O request sizes in 2.6 (5482)
3479 - look up hostnames for IB nals (5602)
3480 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3481 - don't start multiple acceptors for the same port (5277)
3482 - fix incorrect LASSERT in mds_getattr_name (5635)
3483 - export a proc file for general "ping" checking (5628)
3484 - fix "lfs check" to not block when the MDS is down (5628)
3486 - service request history (4965)
3487 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3488 - create an "evict_client" /proc entry on OSTs, like the MDS has
3489 - fix mount usage message, return errors per mount(8) (5168)
3490 - change grep [] to grep "[]" in tests so they work in more UMLs
3491 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3492 - remove ancient LOV_MAGIC_V0 EA support (5047)
3493 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3494 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3495 - allow readahead allocations to fail when low on memory (5383)
3496 - mmap locking landed again, after considerable improvement (2828)
3497 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3499 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3502 - send OST transaction number in read/write reply to free req (4966)
3503 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3504 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3505 - the watchdog thread now runs as interruptible (5246)
3506 - flock/lockf fixes (but it's still disabled, pending 5135)
3507 - don't use EXT3 constants in llite code (5094)
3508 - memory shortage at startup could cause assertion (5176)
3510 - reorganization of lov code
3511 - single portals codebase
3513 - add extents/mballoc support (5025)
3514 - direct I/O reads in the obdfilter (4048)
3515 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3517 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3520 - send OST transaction number in read/write reply to free req (4966)
3521 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3522 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3523 - the watchdog thread now runs as interruptible (5246)
3524 - handle missing objects in filter_preprw_read properly (5265)
3525 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3526 - drop import inflight refcount on signal_completed_replay error (5255)
3527 - unlock page after async write error during send (3677)
3528 - return original error code on reconstructed replies (3761)
3529 - no transno return for symlink open, don't save no-trasno open (3440)
3531 - add pid to ldlm debugging output (4922)
3532 - bump the watchdog timeouts -- we can't handle 30sec yet
3533 - extra debugging for orphan dentry/inode bug (5259)
3535 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3538 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3539 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3540 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3541 - fix NULL dereference in /proc/sys/portals/routes (4827)
3542 - allow failed mdc_close() operations to be interrupted (4561)
3543 - stop precreate on OST before MDS would time out on it (4778)
3544 - don't send partial-page writes before EOF from client (4410)
3545 - discard client grant for sub-page writes on large-page clients (4520)
3546 - don't free dentries not owned by NFS code, check generation (4806)
3547 - fix lsm leak if mds_create_objects() fails (4801)
3548 - limit debug_daemon file size, always print CERROR messages (4789)
3549 - use transno after validating reply (3892)
3550 - process timed out requests if import state changes (3754)
3551 - update mtime on OST during writes, return in glimpse (4829)
3552 - add mkfsoptions to LDAP (4679)
3553 - use ->max_readahead method instead of zapping global ra (5039)
3554 - don't interrupt __l_wait_event() during strace
3556 - add software watchdogs to catch hung threads quickly (4941)
3557 - make lustrefs init script start after nfs is mounted
3558 - fix CWARN/ERROR duplication (4930)
3559 - return async write errors to application if possible (2248)
3560 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3561 - print NAL number in %x format (4645)
3562 - update barely-supported suse-2.4.21-171 series (4842)
3563 - support for sles 9 %post scripts
3564 - support for building 2.6 kernel-source packages
3565 - support for sles km_* packages
3567 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3570 - ignore -ENOENT errors in osc_destroy (3639)
3571 - notify osc create thread that OSC is being cleaned up (4600)
3572 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3573 - reconstruct ost_handle() like mds_handle() (4657)
3574 - create a new thread to do import eviction to avoid deadlock (3969)
3575 - let lconf resolve symlinked-to devices (4629)
3576 - don't unlink "objects" from directory with default EA (4554)
3577 - hold socknal file ref over connect in case target is down (4394)
3578 - allow more than 32000 subdirectories in a single directory (3244)
3579 - fix blocks count for O_DIRECT writes (3751)
3580 - OST returns ENOSPC from object create when no space left (4539)
3581 - don't send truncate RPC if file size isn't changing (4410)
3582 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3583 - bind to privileged port in socknal and tcpnal (3689)
3585 - rate limit CERROR/CWARN console message to avoid overload (4519)
3586 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3587 - basic mmap support (3918)
3588 - kernel patch series update from b1_4 (4711)
3590 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3593 - avoid crash during MDS cleanup with OST shut down (2775)
3594 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3595 - don't use bad inodes on the MDS (3744)
3596 - dynamic object preallocation to improve recovery speed (4236)
3597 - don't hold spinlock over lock dumping or change debug flags (4401)
3598 - don't zero obd_dev when it is force cleaned (3651)
3599 - print grants to console if they go negative (4431)
3600 - "lctl deactivate" will stop automatic recovery attempts (3406)
3601 - look for existing locks in ldlm_handle_enqueue() (3764)
3602 - don't resolve lock handle twice in recovery avoiding race (4401)
3603 - revalidate should check working dir is a directory (4134)
3605 - don't always mark "slow" obdfilter messages as errors (4418)
3607 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3610 - don't close LustreDB during write_conf until it is done (3860)
3611 - fix typo in lconf for_each_profile (3821)
3612 - allow dumping logs from multiple threads at one time (3820)
3613 - don't allow multiple threads in OSC recovery (3812)
3614 - fix debug_size parameters (3864)
3615 - fix mds_postrecov to initialize import for llog ctxt (3121)
3616 - replace config semaphore with spinlock (3306)
3617 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3618 - don't allow enqueue to complete on a destroyed export (3822)
3619 - down write_lock before checking llog header bitmap (3825)
3620 - recover from lock replay timeout (3764)
3621 - up llog sem before sending rpc (3652)
3622 - reduce ns lock hold times when setting kms (3267)
3623 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3624 - fix NULL deref and obd_dev leak on setup error (3312)
3625 - replace some LBUG about llog ops with error handling (3841)
3626 - don't match INVALID dentries from d_lookup and spin (3784)
3627 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3628 - fix invalid assertion in ptlrpc_set_wait (3880)
3630 - add libwrap support for the TCP acceptor (3996)
3631 - add /proc/sys/portals/routes for non-root route listing (3994)
3632 - allow setting MDS UUID in .xml (2580)
3633 - print the stack of a process that LBUGs (4228)
3635 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3638 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3639 - make sure to unset replay flag from failed open requests (3440)
3640 - if default stripe count is 0, use OST count for inode size (3636)
3641 - update parent mtime/ctime on client for create/unlink (2611)
3642 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3643 - free recovery state on server during a forced cleanup (3571)
3644 - unregister_reply for resent reqs (3063)
3645 - loop back devices mounting and status check on 2.6 (3563)
3646 - fix resource-creation race that can provoke i_size == 0 (3513)
3647 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3648 - more debugging for page-accounting assertion (3746)
3649 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3650 - avoid deadlock after precreation failure (3758)
3651 - fix race and lock order deadlock in orphan handling (3450, 3750)
3652 - add validity checks when grabbing inodes from l_ast_data (3599)
3654 - add /proc/.../recovery_status to obdfilter (3428)
3655 - lightweight CDEBUG infrastructure, debug daemon (3668)
3656 - change default OSC RPC parameters to be better on small clusters
3657 - turn off OST read cache for files smaller than 32MB
3658 - install man pages and include them in rpms (3100)
3659 - add new init script for (un)mounting lustre filesystems (2593)
3660 - run chkconfig in %post for init scripts (3701)
3661 - drop scimac NAL (unmaintained)
3663 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3666 - clean kiobufs before and after use (3485)
3667 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3668 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3669 - change init script to fail more clearly if not run as root (1528)
3670 - allow clients to reconnect during replay (1742)
3671 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3672 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3673 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3674 - fix cond_resched (3554)
3675 - only evict unfinished clients after recovery (3515)
3676 - allow bulk resend, prevent data loss (3570)
3677 - dynamic ptlrpc request buffer allocation (2102)
3678 - don't allow unlinking open directory if it isn't empty (2904)
3679 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3680 - remove extraneous obd dereference causing LASSERT failure (3334)
3681 - don't use get_cycles() when creating temp. files on the mds (3156)
3682 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3683 - handle EEXIST for set-stripe, set proper directory name (3336)
3685 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3686 - fix ksocknal_fmb_callback() error messages (2918)
3688 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3691 - don't copy lvb into (possibly NULL) reply on error (2983)
3692 - don't deref dentry after dput, don't free lvb on error (2922)
3693 - use the kms to determine writeback rpc length (2947)
3694 - increment oti_logcookies when osc is inactive (2948)
3695 - update client's i_blocks count via lvb messages (2543)
3696 - handle intent open/close of special files properly (1557)
3697 - mount MDS with errors=remount-ro, like obdfilter (2009)
3698 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3699 - don't use cancelling-locks' kms values (2947)
3700 - use highest lock extent for kms, not last one (2925)
3701 - don't dereference ERR_PTR() dentry in error handling path (3107)
3702 - fix thread race in portals_debug_dumplog() (3122)
3703 - create lprocfs device entries at setup instead of at attach (1519)
3704 - common AST error handler, don't evict client on completion race (3145)
3705 - zero nameidata in detach_mnt in 2.6 (3118)
3706 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3707 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3708 - initialize RPC timeout timer earlier for 2.6 (3219)
3709 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3710 - print nal/nid for unknown nid (3258)
3711 - additional checks for oscc recovery before doing precreate (3284)
3712 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3713 - don't crash in mdc_close for bad permissions on open (3285)
3714 - zero i_rdev for non-device files (3147)
3715 - clear page->private before handing to FS, better assertion (3119)
3716 - tune the read pipeline (3236)
3717 - fix incorrect decref of invalidated dentry (2350)
3718 - provide read-ahead stats and refine rpc in flight stats (3328)
3719 - don't hold journal transaction open across create RPC (3313)
3720 - update atime on MDS at close time (3265)
3721 - close LDAP connection when recovering to avoid server load (3315)
3722 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3723 - don't leak open file on MDS after open resend (3325)
3724 - serialize filter_precreate and filter_destroy_precreated (3329)
3725 - loop device shouldn't call sync_dev() for nul device (3092)
3726 - clear page cache after eviction (2766)
3727 - resynchronize MDS->OST in background (2824)
3728 - refuse to mount the same filesystem twice on same mountpoint (3394)
3729 - allow llmount to create routes for mounting behind routers (3320)
3730 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3731 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3732 - fix rare problem with rename on htree directories (3417)
3734 - allow default OST striping configuration per directory (1414)
3735 - fix compilation for qswnal for 2.6 kernels (3125)
3736 - increase maximum number of MDS request buffers for large systems
3737 - change liblustreapi to be useful for external progs like lfsck (3098)
3738 - increase local configuration timeout for slow disks (3353)
3739 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3741 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3744 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3745 - don't overwrite extent policy data in reply if lock was blocked (2901)
3746 - drop filter export grants atomically with removal from device (2663)
3747 - del obd_self_export from work_list in class_disconnect_exports (2908)
3748 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3749 - swab reply message in mdc_close, other PPC fixes (2464)
3750 - fix destroying of named logs (2325)
3751 - overwrite old logs when running lconf --write_conf (2264)
3752 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3753 - fix race in target_handle_connect (2898)
3754 - mds_reint_create() should take same inode create lock (2926)
3755 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3756 - don't close files for self_export to avoid uninitialized obd (2936)
3757 - allow MDS with the same name as client node (2939)
3758 - hold dentry reference for closed log files for unlink (2325)
3759 - reserve space for all logs during transactions (2059)
3760 - don't evict page beyond end of stripe extent (2925)
3761 - don't oops on a deleted current working directory (2399)
3762 - handle hard links to targets without a parent properly (2517)
3763 - don't dereference NULL lock when racing during eviction (2867)
3764 - don't grow lock extents when lots of conflicting locks (2919)
3766 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3769 - account for cache space usage on clients to avoid data loss (974)
3770 - lfsck support in lustre kernel code (2349)
3771 - reduce journal credits needed for BRW writes (2370)
3772 - orphan handling to avoid losing space on client/server crashes
3773 - ptlrpcd can be blocked, stopping ALL progress (2477)
3774 - use lock value blocks to assist in proper KMS, faster stat (1021)
3775 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3776 - recovery for initial connections (2355)
3777 - fixes for mds_cleanup_orphans (1934)
3778 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3779 - block all file creations until orphan recovery completes (1901)
3780 - client remove rq_connection from request struct (2423)
3781 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3782 - recovery timer race (2670)
3783 - mdc_close recovey bug (2532)
3784 - ptlrpc cleanup bug (2710)
3785 - mds timeout on local locks (2588)
3786 - namespace lock held during RPCs (2431)
3787 - handle interrupted sync write properly (2503)
3788 - don't try to handle a message that hasn't been replied to (2699)
3789 - client assert failure during cleanup after abort recovery (2701)
3790 - leak mdc device after failed mount (2712)
3791 - ptlrpc_check_set allows timedout requests to complete (2714)
3792 - wait for inflight reqs when ptlrpcd finishes (2710)
3793 - make sure unregistered services are removed from the srv_list
3794 - reset bulk XID's when resending them (caught by 1138 test)
3795 - unregister_bulk after timeout
3796 - fix lconf error (2694)
3797 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3798 - readahead locks pages, leaves pending causing memory pressure (2673)
3799 - increase OST request buffers to 4096 on large machines (2729)
3800 - fix up permission of existing directories in simple_mkdir (2661)
3801 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3802 - don't assign transno to errored transactions (2742)
3803 - don't delete objects on OST if given a bogus objid from MDS (2751)
3804 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3805 - if rq_no_resend, then timeout request after recovery (2432)
3806 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3807 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3808 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3809 - MDS recovery shouldn't delete objects that it has given out (2730)
3810 - if enqueue arrives after completion, don't clobber LVB (2819)
3811 - don't unlock pages twice when trigger_group_io returns error (2814)
3812 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3813 - don't write pages to disk if there was an error (1450)
3814 - don't ping imports that have recovery disabled (2676)
3815 - take buffered bytes into account when balancing socknal conn (2817)
3816 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3817 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3818 - remove little-endian swabbing of llog records (1987)
3819 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3820 - retry reposting request buffers if they fail (1191)
3821 - grow extent at grant time to avoid granting a revoked lock (2809)
3822 - lock revoke doesn't evict page if covered by a second lock (2765)
3823 - disable VM readahead to avoid reading outside lock extents (2805)
3825 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3826 - updated kernel patches for hp-2.4.20 kernel (2681)
3828 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3831 - fix truncated write corruption (2366)
3832 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3834 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3835 - ensure len(uuid) < 37 in lmc (1171)
3836 - fix ia64 OOPS in llog_test (2255)
3837 - zero end of page at obdfilter for partial page writes (2648)
3838 - don't leave stale dentries around after renames (bug 2428)
3839 - fix timeouts when evicting a client with a single lock held (2642)
3840 - set deadline for the initial HELLO message to drain (2634)
3841 - print out dotted-quad IP addresses in the socknal (2302)
3843 - additional debugging for MDS client eviction problem (2443)
3844 - fix mkfsoptions support for osts (2603, 2604)
3846 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3849 - add series for the vanilla 2.6.0 kernel
3850 - add series for the vanilla 2.4.24 kernel
3851 - add series for a cray x86/64 UL kernel drop
3852 - fix xattr patches for the vanilla 2.4.19 series
3854 - generate true UUIDs in lmc (1171)
3855 - have portals stack dumping break in UML (2466)
3856 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3857 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3858 - wake callback waiting threads on client eviction (2460)
3859 - Add --ptldebug and --subsystem to lmc (1719)
3860 - update assertion to allow safe interrupt allocation
3861 - set rq_no_resend for cancel requests (2432)
3862 - recalculate ptlrpcd timeout after resend (2494)
3863 - call vfs_rmdir when removing pending directories (2368)
3864 - fix renaming a file to itself (2429)
3865 - lmc creates a default one-stripe lov (2454)
3866 - expand procfs space to handle large clusters (2326)
3867 - increase UML stack to avoid overflow
3868 - update lconf's list of debug and subsystem masks
3869 - fix lfs find --obd (2510)
3870 - /proc tunable for disabling filter read caching (2591)
3871 - stop rpm packages from altering slapd.conf (2301)
3872 - disable nagle in the socknal under 0conf (2578)
3873 - choose mds inode size based on stripe count (2572)
3874 - fix kernel-source rpm problems (2516)
3876 - add --disable-doc to avoid pdf generation (2421)
3877 - update documentation, tests, type-os, comments
3878 - avoid format warnings on ia64
3879 - remove the TOE NAL
3880 - tiny code cleanups by removing unused fields
3882 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3885 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3886 - remove the most blatant lies from BUILDING file (2371)
3887 - change default debug level to reasonable production setting
3888 - reduce client side cache size to reduce cache flush time
3889 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3890 - make TCP zerocopy and pinger support enabled by default (2476)
3891 - sync writes completed after process exits caused crashes (2319)
3892 - maintain correct mount count on the MDS (2356)
3893 - backout 1557, because 2316 wasn't really fixed
3894 - better file I/O statistics gathering in /proc
3895 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3896 - another kernel patch to fix zero-copy TCP function export
3897 - don't take duplicate lock when processing re-sent getattr (2420)
3898 - lctl uses obd_self_export instead of creating new conn (2353)
3899 - MDS/OST recovery case which requires object creation asserted (2425)
3900 - move lfs from /usr/sbin to /usr/bin in packages
3901 - fix race between mds_client_add and mds_client_free (2417)
3902 - use kmalloc instead of slabs in portals (2430)
3903 - don't create duplicate records when a failover MDS is present (2442)
3904 - remove unnecessary mount age check (2332)
3905 - don't remove directory inodes from locks prematurely (2451)
3906 - don't break if MDS service name is the same as hostname (2103)
3907 - fix races in client write RPC generation when cache full (2482)
3909 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3912 - remove now-unused request->rq_obd (278)
3913 - if an allocation fails, print out how much memory we've used (1933)
3914 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3915 - add the "configurable stack size" patch to most series files (1256)
3916 - ability to write large log records, for 100+ OST configs (2306)
3917 - fix NULL deref when filter_prep fails (2314)
3918 - fix operator precedence error in filter_sync
3919 - dynamic allocation of socknal TX descriptors (2315)
3920 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3921 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3922 - ensure that CWARN messages in Portals always get to the syslog
3923 - __init/__exit are not for prototype decls (ldlm_init/exit)
3924 - x86-64 compile warning fixes
3925 - fix gateway LMC keyword conflict (2318)
3926 - fix MDS lock inversions in getattr/reint paths (1844)
3927 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3928 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3929 - DLM race condition prevented some lock evictions (2328)
3930 - ENOMEM detection and retry on socknal sends (2230)
3931 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3932 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3933 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3934 - better error messages when a client is rejected during recovery (1505)
3935 - avoid cancelling locks which were never granted, after failure (2330)
3936 - fix i_sem/journal inversion in mds_client_add (2333)
3937 - fix truncate/getattr lock cycle deadlock (2334)
3938 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3939 - fix two rare exit paths which could leak an l_lock() ref (2321)
3940 - fencepost error in MDS/OST orphan recovery (2226)
3941 - make log record alignment 8 bytes (1988)
3942 - lstripe now fails when requested offset > ost_count (2237)
3943 - ensure that all kernel series have a complete list.h (1607)
3944 - fix crashes in special-file operations (2316)
3945 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3947 - allow configurable automake binary, for testing new versions
3948 - small update to the lfs documentation
3950 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3952 * fix negative export reference count in fsfilt_sync (2312)
3954 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3955 * release candidate 0.9.1
3957 - orphans are moved into the PENDING directory for possible recovery
3958 - replayed opens now open by fid for orphan/rename safety (1042)
3959 - last close of an orphan inode generates a transno (683)
3960 - chdir() and mount() now pin the directory entry (1020)
3961 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3962 - discard very old requests without processing them (1502)
3963 - remove some common, well-understood CERRORs (1505)
3964 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3965 - clear "grant" flags in OST replies until OST grant code lands (1644)
3966 - fix read performance by not clobbering i_blksize on client (1598)
3967 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3968 - make LRU size a /proc tunable, clears locks when reduced (707)
3969 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3970 - server locks take references on exports now (1558)
3971 - build fixes for 2.4.20-rh trees (1663)
3972 - return an error from lov_create if all OSCs are inactive (1751)
3973 - fix import levels when a reconnect happens without a timeout (1597)
3974 - exit early from mds_open if we get a lookup error (1749)
3975 - partial page read at EOF wouldn't wait for disk before sending (1642)
3976 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3977 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3978 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3979 - write/truncate lock inversion (1639)
3980 - Don't auto-load obdclass, portals modules during cleanup (1495)
3981 - fix timestamps from jumping to "now" (1763)
3982 - extra journal assertions (1648)
3983 - add an extra multiunlink test (1771)
3984 - fix read_record/write_record API (1776)
3985 - fix leak of offset_extent, possible incorrect i_size later (1772)
3986 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3987 - Debugging for the kqswnal_get_idle_tx problems (1820)
3988 - Allow recovery to be attempted multiple times (1536)
3989 - Write out MDS last_rcvd file after it is first created (1600)
3990 - Fix tx_descriptor leak in failed transmit situations (1827)
3991 - ext3 journaling fixes for assertion failure after IO error (1871)
3992 - class_export_put() on freed export after completion AST error (1896)
3993 - Fix revalidate looping in VFS (1322)
3994 - Don't access a freed export during MDS_REINT timeout (1521)
3995 - Add open-unlink recovery support on the MDS (1673,1764)
3996 - Return an error if no MDS data was read from last_rcvd (1946)
3997 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3998 - Don't setup a disk device that doesn't match exported UUID (317)
3999 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4000 - avoid committing NULL handle in force close
4001 - local.sh is now a one-stripe LOV configuration
4002 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4003 - fix ext3 htree duplicate directory entry corruption (1516)
4004 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4005 - update to new LOV EA format (2097)
4006 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4008 2003-06-15 Phil Schwan <phil@clusterfs.com>
4011 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4012 - per-import/export recovery handling (958, 931, 959)
4013 - multiple last-rcvd slots, for serving multiple FSes (949)
4014 - connections are again shared between multiple imp/exports (963, 964)
4015 - "umount -f" would hang if any requests needed to be sent (393, 978)
4016 - avoid pinning large req buffer by copying for queued messages (989)
4017 - add "uuid" to "lctl device" command to help upcalls (991)
4018 - "open" RPCs with transnos would confuse recovery counters (1037)
4019 - do proper endian conversion of all wire messages (288, 340, 891)
4020 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4021 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4022 - fix ptlrpc_connection leak in target_handle_connect (1174)
4023 - fix import refcounting bug in OST and MDS cleanup (1134)
4024 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4025 - fix very unlikely obd_types race condition (501)
4026 - remove osc_open hack for echo_client (1187)
4027 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4028 - a failure in read_inode2 leads to deadlock (1139)
4029 - cancel ack-locks as soon as transaction is committed (1072)
4030 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4031 - make sure to commitrw after any preprw to avoid deadlock (1162)
4032 - failing to execute a file in a lustre FS would lock inode (1203)
4033 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4034 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4035 - obd_uuid2tgt would walk off the end of the list (1255)
4036 - on IA64 the osc would give portals incorrect bulk size (1258)
4037 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4038 - fix lock inversion caused by new llite matching code (1282)
4039 - limit the number of dirty pages on a client to 10MB (1286)
4040 - timed out locks were not being corrected cancelled (1289)
4041 - fix O_DIRECT above 4GB on IA-32 (1292)
4042 * major user-visible changes
4043 - fail out/fail over policy now controlled by the upcall (993)
4045 - add OBD_PING to check server availability and failure (954)
4046 - lustre messages are now sent in sending host order (288, 340, 891)
4047 - add eadatalen to MDS getattr reply (340)
4048 - OST read replies may contain second buffer, with per-page status (593)
4050 2003-03-11 Phil Schwan <phil@clusterfs.com>
4053 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4054 - failed open()s could cause deadlock; fixed (867, 869)
4055 - stop cancelling OST locks when files are closed (481)
4056 - overlapping XID spaces caused network corruption (851, 853)
4057 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4058 - setattr_raw added, to do single-RPC, server-side setattrs
4059 - lmc/lconf syntax change for OST UUIDs
4060 - fix crashy race condition between ptlrpc_free_req and osc_close
4061 - don't use request in mdc_enqueue if we hit a timeout (889)
4062 - don't set the inode i_size for regular files from the MDS (896)
4063 - handle out of order completion AST (842)
4064 - don't LBUG if a lock request times out after receiving AST (913)
4065 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4066 - if a bad lock AST arrives, send an error instead of dropping entirely
4067 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4068 - fix leak in bulk IO when only partially completed (899, 900, 926)
4069 - fix O_DIRECT for ia64 (55)
4070 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4071 - C-z after timeout could hang a process forever; fixed (977)
4073 - client-side I/O cache (678, 924, 929, 941, 970)
4075 - READPAGE and SETATTRs which don't take server-side locks get
4078 2003-02-11 Phil Schwan <phil@clusterfs.com>
4081 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4082 - Fix fsfilt_extN_readpage() to read a full page of directory
4083 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4084 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4085 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4086 - Most intent operations were removed, in favour of a new RPC mode
4087 that does a single RPC to the server and bypasses most of the VFS
4088 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4089 - Aggressively cancel local locks on DLM servers
4090 - mds_reint_unlink sends EA to the client if it's the last nlink.
4091 client uses that EA to unlink OST objects.
4092 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4093 - recursive symlinks were fixed (439)
4094 - fixed NULL deref in DEBUG_REQ
4095 - filter_update_lastobjid no longer calls sync, which annoyed extN
4096 - fixed multi-client small-writes to a single file problem (445)
4097 - fixed mtime updates during file writes (607)
4098 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4099 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4100 - fixed Linux OST size reporting problem (444, 656)
4101 - OST now updates object mtime with writes or setattr (607, 619)
4102 - client verifies file size before zeroing page past EOF (445)
4103 - OST now writes last allocated objid to disk with allocation (108)
4104 - LOV on echo now works (409)
4106 - mds_reint_unlink sends a new buffer, with the EA included. this
4107 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4108 set if a regular file was being unlinked, and it was the last link
4109 - use PtlGet from the target for bulk writes (315)
4110 - OST now updates object mtime with writes or setattr (607, 619)
4111 - LDLM now has a grant-time callback to revalidate locked items, if
4113 - Many MDS operations were reorganized to combat race conditions
4115 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4116 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4118 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4121 - Fully reactivate OST imports after reconnection (512, others)
4122 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4123 - More graceful error handling for truncating on dead OST (515)
4124 - Don't error out unless we're actually accessing dead stripes (474)
4125 - Fix garbage sizes when stripes are missing (410)
4126 - LRU counters were broken, causing constant lock purge (433, 432)
4127 - garbage on read from stripes with failed OSTs (441)
4128 - mark OSCs as active before reconnecting during recovery (438)
4129 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4130 - lfind did not preserve OST order in output (443)
4131 - symlinks cause hung clients, incorrect data (439)
4132 - stop dereferencing request after dropping refcount (457)
4133 - don't LASSERT(spin_is_locked) on non-SMP (455)
4134 - fixes for many rename() bugs
4135 - fstat didn't correctly synchronize attributes (399)
4136 - server must handle lock cancellation during blocking AST prep (487)
4137 - bulk descriptors were free()d too soon (511)
4138 - fix paths in lconf, which would load incorrect modules (451, 507)
4139 - fix confusing lconf 'host not found' error message (386)
4140 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4141 - fix race condition in mdc_blocking_ast() for inode access (526)
4142 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4143 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4144 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4145 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4146 - fix lstripe in several ways (kernel side) (527)
4147 - fix request leak in ldlm_cli_enqueue (262)
4148 - incorrect OSC was marked inactive after OST failure
4149 - call mds_fs_cleanup before unmounting filesystem (524)
4150 - fix races between taking ns_lock and ldlm_lock_change_resource
4151 - fix races updating LOV export open file list
4152 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4153 - fix recovery NULL deref in ldlm_cli_cancel_unused
4154 - fix some DLM races by using new hash table for lock handles (419)
4155 - permit the client to specify desired inodes, at replay
4156 - duplicate requests when we queue them for replay reintegration
4157 - fix last_rcvd offset calculation
4158 - sync after each recovered transaction, so we always make progress
4159 - never, not always, ERESTART requests without transnos
4160 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4162 - skip replay if the MDS says that the client is already connected
4163 - don't check for a recovery-enabled export to match lctl's UUID
4164 - don't INC_USE_COUNT for phantom exports
4165 - don't crash when cleaning up phantom exports (567)
4166 - don't double-finish or set replay data for errored mdc_open requests
4167 - abort requests when they time out, so we don't get old replies
4168 - send/receive replies for AST messages again
4169 - if the client says that it doesn't have the lock, cancel it on the
4171 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4172 mark it as destroyed, it will all work out when decref is called
4173 - fix module use counts (22, 581)
4175 - ASTs now expect a reply (server cancels lock on error reply)
4177 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4180 - fix many simultaneous client startup (392)
4181 - fix dentry->d_it clobbering
4182 - credentials weren't being shipped for readdir/getattr operations
4183 - remove invalid assertions triggered during some concurrent MD
4185 - proper Lustre versions added (336, 389)
4186 - fix memory leak for create error case (398)
4187 - fix LOV locking bug that would get cli/srv out of sync
4188 - fix echo client over LOV (409)
4189 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4190 - fix double-O_EXCL intent crash (424)
4191 - avoid sending multiple lock CANCELs (352)
4193 - MDS can do multi-client recovery (modulo bugs in new code)
4195 - many updates, edits, cleanups
4197 2002-11-18 Phil Schwan <phil@clusterfs.com>
4200 - fix null d_it dereference (346)
4201 - fix full OST/dbench hang (333)
4202 - fix permission problem with file removal (286)
4203 - fix removal of OSCs from LOV when they fail
4204 - fix NULL deref during bulk timeout (214)
4205 - fix problems related to multiple filesystems on one MDS (241)
4206 - fixed serious subtle metadata locking bugs
4207 - free locks on clients when inodes are removed due to memory
4209 - fix inode pointer in lock data (285)
4210 - partial support for multiple MDS on a single host (241)
4211 - data locks weren't cancelled at clear_inode time (290, 311)
4212 - intent locks could lead to unbounded lock growth (205)
4213 - added a maximum lock count, an LRU list, and a flusher
4214 - fix multiple rename (365)
4215 - properly abstracted the echo client
4216 - OSC locked 1 byte too many; fixed
4217 - rewrote brw callback code:
4218 - fixed recovery bugs related to LOVs (306)
4219 - fixed too-many-pages-in-one-write crash (191)
4220 - fixed (again) crash in sync_io_timeout (214)
4221 - probably fixed callback-related race (385)
4223 - Add capability to MDS protocol
4224 - LDLM cancellations and callbacks on different portals
4226 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4229 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4230 - defer open object destruction to close time (601981, 138)
4231 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4232 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4233 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4234 - don't call obd_setattr in ll_file_release for destroyed objects
4235 * protocol change to lustre_msg: move |version| and add |flags|
4236 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4237 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4238 * added replay of create, unlink, link and rename operations during
4239 MDS failover; recovery should be much more robust now
4240 * remove failed OSCs from LOVs (only lov_create uses this so far)
4241 * the lustre-HOWTO was brought (more) up to date (582544)
4243 2002-10-23 Phil Schwan <phil@clusterfs.com>
4246 - in-use dentries weren't being reused properly (617851)
4247 - prevent multiple LDLM setup (599178)
4248 - fix LOV size calculations for truncate (617853)
4249 - fix client handling of MDS intent errors (POSIX)
4250 - fix permission bug in lovstripe.c test (624321)
4251 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4252 - truncate past end of file could corrupt data
4253 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4254 - a race in recovery could return ETIMEDOUT to apps (623947)
4255 - building outside the source directory was fixed
4256 * the lustre-HOWTO was brought (more) up to date (582544)
4257 * major progress was made on recovery functionality
4259 2002-10-10 Phil Schwan <phil@clusterfs.com>
4262 - recovery deadlock fix
4263 - rm -rf causes LBUG fix (617817)
4264 - file open by multiple tasks fix (618962)
4265 - directory permissions bugs (602707 and 620007)
4266 - journal_stop fixed with locking (611313)
4267 - O_APPEND failures resolved (618273, perhaps 614459)
4268 - lconf PATH fix (619770)
4269 - IA64 build fix (621450)
4270 - RPC buffer sizes scale with amount of memory
4272 2002-10-01 Phil Schwan <phil@clusterfs.com>
4275 - locks would be cancelled without throwing away data pages,
4276 resulting in inconsistent data (605627)
4277 - inode attributes were not always being refreshed (605627, 612449)
4278 - lconf now continues to cleanup after lctl reports an error
4279 - MDS now enforces user permissions (602707)
4280 - lprocfs cleanup fixed, but not yet enabled (614157)
4281 - fixed infinite server hang, should a client not respond to an AST
4282 - avoid going into recovery if user calls readlink() with a buffer
4283 that's too small (613941)
4284 - AST RPCs no longer require replies (614867) -- this may be changed
4285 - don't crash server if client sends an IOV that's too big (611336)
4286 - fixed lock conversion deadlock (611892)
4287 - fixed the following of symlinks (614622)
4288 * recovery: the server can remove locks from a client that dies, other
4289 clients can make progress
4290 * more extN patch fixes
4291 * compile-time configurable ptlrpc buffer allocations
4293 - collaborative read cache document
4294 - Lustre Lite Performance CDR document-in-progress
4296 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4299 - fix typo in patch-2.4.18
4301 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4304 - clear ptlrpc request each time in handle_incoming_request()
4305 - unlink of files now destroys the object on the OST
4307 2002-09-19 Peter Braam <braam@clusterfs.com>
4309 * add hard link support
4310 * change obdfile creation method
4311 * kernel patch changed
4313 2002-09-19 Peter Braam <braam@clusterfs.com>
4316 - stack overflow bug in extN fixed
4318 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4320 * documentation updates
4321 - add man pages for config tools
4322 - update tests/README to describe testing with new config tools
4323 - finish metadata API descriptions
4324 * bug fixes and cleanups
4325 - statfs workaround for 16TB limit
4326 - LOV stripe allocation improved, can stripe on subset of OSTs
4327 - LOV file size/IO offset was wrong for files > 4GB in size
4328 - object EA data was being dropped, caused files to be unreadable
4329 - memory overflow with non-LOV OST caused memory corruption
4330 - fixed regression tests to work with new config tools, obdfilter
4331 - fixed bug when directory size became larger than 1 block
4332 - fixed bug (for single client case) when PWD was deleted
4333 - invalidate local directory pages when doing intent-based ops
4334 - avoid LDLM oops when lock callback contained bad data
4336 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4338 * documentation updates
4339 * bug fixes and cleanups
4340 - configuration tools
4343 - 64-bit compile warnings
4344 - 64-bit internal statfs data
4346 * test_brw on persistent OST devices
4348 * lprocfs (disabled)
4350 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4352 * documentation updates
4353 * bug fixes and cleanups
4354 * configuration tools
4356 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4359 * many small fixes to 0_5_4
4360 * io/network handling
4361 * thinkos in MDS operations
4363 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4366 * crucial basic fixes to 0.5.3
4367 * IOR, Iozone work over Elan
4370 2002-08-07 Phil Schwan <phil@clusterfs.com>
4371 * version 0_5_3, our first alpha
4372 * we use the new Portals iovs
4373 * documentation updates
4374 * bug fixes and cleanups
4375 * small changes in the DLM wire protocol
4377 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4378 * version 0_5_1 with some initial stability,
4379 * locking on MD and file I/O.
4380 * documentation updates
4381 * several bug fixes since 0.5.0
4382 * small changes in wire protocol
4384 2002-07-18 Phil Schwan <phil@clusterfs.com>
4386 * delivered as Lustre Light Alpha
4387 * fixed a crash after handling invalid MDS requests
4388 * fixed directory pages for architectures with non-4k pages sizes
4390 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4391 * release version v0_4_4
4392 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4394 * Fixes a number of interruption problems with OST operations.
4395 * Update documentation for portals header changes
4396 * Move all wire protocol structs/defines to lustre_idl.h
4397 * Fixes symlink length bug.
4398 * Add tcpdump to repository.
4400 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4401 * release version v0_4_3
4402 * Fixes statfs for inodes on extN.
4403 * Fixes bug in runtests which would delete /etc/hosts.
4404 * Use 64-bit object IDs wherever possible (not into VFS though)
4405 Remove ost_get_info, which is unused by lustre, and out of date.
4407 2002-07-03 Peter Braam <braam@clusterfs.com>
4408 * release version v0_4_2 Fixes a lookup error (type not passed)
4409 * move forward to head of Portals
4410 * move forward to latest Lustre kernel
4412 2002-06-25 Peter Braam <braam@clusterfs.com>
4413 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob