1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.6.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).
15 Severity : enhancement
17 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
19 Severity : enhancement
21 Description: Add rhel5 support to HEAD.
23 Serverity : enhancement
25 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
30 Description: Don't allow skipping OSTs if index has been specified.
31 Details : Don't allow skipping OSTs if index has been specified, make locking
32 in internal create lots better.
36 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
37 Details : access to bitfield in structure is always rounded to long
38 and this produce problem with not atomic change any bit.
42 Description: Lustre make rpms failed.
43 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
44 builds Lustre rpms from tarball.
47 Frequency : rare on shutdown ost
49 Description: If llog cancel was not send before clean_exports phase, this can
50 produce deadlock in llog code.
51 Details : If llog thread has last reference to obd and call class_import_put
52 this produce deadlock because llog_cleanup_commit_master wait when
53 last llog_commit_thread exited, but this never success because was
54 called from llog_commit_thread.
58 Description: allow userland application know is lost one of stripes.
59 Details : fill lvb_blocks with error code on ost and return it to
60 application if error flag found.
64 Description: NULL lov_tgts causing MDS oops
65 Details : more safe checks for NULL lov_tgts for avoid oops.
67 Severity : enhancement
69 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
73 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
74 Details : If an OST is down the MDS will hang indefinitely in
75 obd_statfs_rqset() waiting for the statfs data. While for
76 MDS QOS usage of statfs, it should not stuck in waiting.
78 Severity : enhancement
80 Description: remote_acl support
81 Details : Support ACL-based permission check for remote user.
82 Support setfacl/getfacl for remote user with the utils
83 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
84 the system "{s,g}etfacl" utils.
86 Severity : enhancement
88 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
90 Severity : enhancement
92 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
96 Description: make mgs_setparam() handle fsname containing dash
97 Details : fsname containing a dash does not work with lctl conf_param
99 Severity : enhancement
101 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
103 Severity : enhancement
105 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
109 Frequency : rare, at shutdown
110 Description: access already free / zero obd_namespace.
111 Details : if client_disconnect_export was called without force flag set,
112 and exist connect request in flight, this can produce access to
113 NULL pointer (or already free pointer) when connect_interpret
114 store ocd flags in obd_namespace.
118 Frequency : only at startup
119 Description: not alloc memory with spinlock held.
120 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
121 if any spinlock held.
123 Severity : enhancement
125 Description: make lustre randomly fail allocating memory
126 Details : Make lustre randomly failed allocating memory for testing purpose.
128 Severity : enhancement
130 Description: lost problems with lov objid file
131 Details : Fixes some scability and access to not inited memory problems
132 in work with lov objdid file.
137 Description: lfs find does not continue on file error
138 Details : Continue other files processing when a file/dir is absent.
142 Description: Inconsistent usage of lustre_pack_reply()
143 Details : Standardize the usage of lustre_pack_reply() such that it
144 always generate a CERROR on failure.
147 Frequency : occasional
149 Description: MDS hang or stay in waiting lock
150 Details : If client receive lock with CBPENDING flag ldlm need send lock
151 cancel as separate rpc, to avoid situation when cancel request
152 can't processed due all i/o threads stay in wait lock.
156 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
160 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
164 Description: Update to SLES9 kernel-2.6.5-7.287.3.
168 Description: Update to RHEL4 latest kernel.
170 Severity : enhancement
172 Description: Build SLES10 patchless client fails
173 Details : The configure was broken by run ./configure with
174 --with-linux-obj=.... argument for patchless client. When the
175 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
176 can't search header adequately. Use absolute path such as
177 -I($LINUX)/include instead.
181 Description: interrupt oig_wait produce painc on resend.
182 Details : brw_redo_request can be used for resend requests from ptlrpcd and
183 private set, and this produce situation when rq_ptlrpcd_data not
184 copyed to new allocated request and triggered LBUG on assert
185 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
186 ptlrpcd set if request is changed and can be safety changed to use
191 Description: LASSERT_{REQ,REP}SWAB macros are buggy
192 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
193 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
194 these macros and replace them with their difinitions instead.
198 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
199 Details : Update which_patch & target file for SLES10 latest kernel.
202 Description: Remove client patches from SLES 10 kernel.
203 Details : This causes SLES 10 clients to behave as patchless clients
204 even on a Lustre-patched (server) kernel.
206 Severity : enhancement
208 Description: self-adjustable client's lru lists
209 Details : use adaptive algorithm for managing client cached locks lru
210 lists according to current server load, other client's work
211 pattern, memory activities, etc. Both, server and client
212 side namespaces provide number of proc tunables for controlling
215 Severity : enhancement
217 Description: light-weight GSS support
218 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
219 and reduce performance overhead.
221 Severity : enhancement
223 Description: Linux keyring support
224 Details : Support using service of Linux keyring for Lustre GSS internal
225 context refresh/cache mechanism.
229 Description: Fix errors in lfs documentation
230 Details : Fixes man pages
234 Description: don't use GFP_* in generic Lustre code.
235 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
239 Description: obdclass is limited by single OBD_ALLOC(idarray)
240 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
244 Description: warnings with build patchless client with vanila 2.6.19 and up
245 Details : change old ctl_table style and replace ctl_table/ctl_table_header
246 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
250 Frequency : Only for SLES
251 Description: sanity_quota fail test_1
252 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
253 file, which makes TSTID[2] inunique.
257 Frequency : Only for SLES9
258 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
259 Details : Update target/ChangeLog/which_patch .
263 Frequency : Only for test_10 in sanity-quota.sh
264 Description: a bug in quota test code
265 Details : checking a test flag in a wrong place
269 Frequency : Only for test_13 in sanity-quota.sh
270 Description: a bug in quota test script
271 Details : shouldn't check overfull allocated quota
275 Description: jbd statistics
276 Details : Port older jbd statistics patch for sles10
278 Severity : enhancement
280 Description: Add printing inode info into message about error in writepage.
285 Description: reply_lock_interpret crash due to race with it and lock cancel.
286 Details : Do not replay locks that are being cancelled. Do not reference
287 locks by their address during replay, just by their handle.
291 Frequency : When flocks are used.
292 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
293 Details : Flock locks might destroy just granted lock if it could be merged
294 with another existing flock, this is done in completion handler,
295 so teach ldlm_cli_enquque_fini that this is a valid case for
301 Description: Oops in read and write path when failing to allocate lock.
302 Details : Check if lock allocation failed and return error back.
306 Description: lstripe command fails for valid OST index
307 Details : The stripe offset is compared to lov->desc.ld_tgt_count
308 instead of lov->desc.ld_active_tgt_count.
312 Description: sanity.sh failed test 103
313 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
318 Description: df doesn't work properly if diskfs blocksize != 4K
319 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
323 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
324 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
328 Description: obd_health_check_timeout too short
329 Details : set obd_health_check_timeout as 1.5x of obd_timeout
333 Description: llapi_file_create() does not allow some changes
334 Details : add llapi_file_open() that allows specifying the mode and
335 open flags, and also returns an open file handle.
339 Description: merge and cleanup kernel patches.
340 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
344 Description: Add journal checksum support.(Kernel part)
345 Details : The journal checksum feature adds two new flags i.e
346 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
347 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
348 indicates that the commit block contains the checksum for
349 the blocks described by the descriptor blocks. Now commit
350 record can be sent to disk without waiting for descriptor
351 blocks to be written to disk. This behavior is controlled
352 using JBD2_FEATURE_ASYNC_COMMIT flag.
356 Description: OSS needs mutliple precreate threads
357 Details : Add ability to start more than one create thread per OSS.
361 Description: Sanity.sh test_65k failed: setstripe should have succeeded
362 Details : execute lfs setstripe on client
366 Description: mds_obd_create error creating tmp object
367 Details : When the user sets quota on root, llog will be affected and can't
368 create files and write files.
371 Frequency : Always on ia64 patchless client, and possibly others.
373 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
374 Details : This allows the patchless client to be loaded on architectures
379 Description: RedHat Update kernel for RHEL5
380 Details : Add the kernel config file for RHEL5.
384 Description: RedHat Update kernel for RHEL5
385 Details : Modify the kernel config file more closer RHEL5.
389 Description: Build failure against Centos5 (RHEL5)
390 Details : Define PAGE_SIZE when it isn't present.
394 Description: test_7 on sanity-quota.sh can't be used on separate machines
395 Details : add facet to handle it
399 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
400 Details : with NFS, the anon dentry's parent was set to itself in
401 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
402 corresponding dentry other than use rec->ur_name.
404 Severity : enhancement
406 Description: lfs setstripe enhancement
407 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
411 Description: enable data checksumming by default
412 Details : enable checksum by default, allow --disable-checksum
413 configure option and "-o nochecksum" mount option. Checksums
414 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
418 Description: lustre support for RHEL5
419 Details : Add support for RHEL5.
423 Description: Client eviction due to failover config
424 Details : after a connection loss, the lustre client should attempt to
425 reconnect to the last active server first before trying the
426 other potential connections.
430 Description: when mds and osts use different quota unit(32bit and 64bit),
431 quota will be released repeatly.
432 Details : void sending multiple quota reqs to mds, which will keep the
433 status between the reqs.
437 Description: osts not allocated evenly to files
438 Details : change the condition to increase offset_idx
441 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
443 Description: Data corruption for OSTs that are formatted larger than 2TB
445 Details : When generating the bio request for lustre file writes the
446 sector number would overflow a temporary variable before being
447 used for the IO. The data reads correctly from Lustre (which
448 will overflow in a similar manner) but other file data or
449 filesystem metadata may be corrupted in some cases.
453 Description: Tune the kernel for good SCSI performance.
454 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
455 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
460 Description: rewrite ext2-derived code in obdclass/uuid.c
461 Details : rewrite inherited code (uuid parsing code from ext2 utils)
462 from scratch preserving functionality.
466 Description: "lfs find -obd UUID" prints directories
467 Details : "lfs find -obd UUID" will return all directory names instead
468 of just file names. It is incorrect because the directories
469 do not reside on the OSTs.
473 Description: use i_size_read and i_size_write in 2.6 port
474 Details : replace inode->i_size access with i_size_read/write()
477 Frequency : when removing large files
479 Description: scheduling issue during removal of large Lustre files
480 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
481 It causes scheduling issues when removing large files (17TB in the
485 Frequency : only with liblustre clients on XT3
487 Description: evictions taking too long
488 Details : allow llrd to evict clients directly on OSTs
491 Frequency : only on ppc
493 Description: /proc/fs/lustre/devices broken on ppc
494 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
495 We should make sure the type of 'index' is loff_t and then cast
496 explicitly as needed below. Do not assign an explicitly cast
500 Frequency : only for rhel5
502 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
503 Details : Modify the target file & which_kernel.
505 Severity : enhancement
507 Description: omit set fsid for export NFS
508 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
509 as FS_REQUIRES_DEV for avoid problems with generate fsid.
513 Frequency : Always, for kernels after 2.6.16
514 Description: Fix warning idr_remove called for id=.. which is not allocated.
515 Details : Last kernels save old s_dev before kill super and not allow
516 to restore from callback - restore it before call kill_anon_super.
520 Description: buffer overruns could theoretically occur
521 Details : llapi_semantic_traverse() modifies the "path" argument by
522 appending values to the end of the origin string, and a
523 overrun may occur. Adding buffer overrun check in liblustreapi.
527 Description: Fix error on 'ls .' at the top of the Lustre mount.
528 Details : Don't revalidate dentry if it is a root dentry.
532 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
533 Details : Modify vm-tunables-rhel4.patch.
537 Description: Kernel config for 2.6.18-vanilla.
538 Details : Modify targets/2.6-vanilla.target.in.
539 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
540 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
541 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
542 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
546 Description: improve handling recoverable errors
547 Details : if request processig with error which can be recoverable on server
548 request should be resend, otherwise page released from cache and
553 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
554 Details : Modify target file & which_patch.
555 A flaw was found in the IA32 system call emulation provided
556 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
557 value could be stored in the %RAX register, which could trigger an
558 out-of-bounds system call table access. An untrusted local user
559 could exploit this flaw to run code in the kernel
560 (ie a root privilege escalation). (CVE-2007-4573).
564 Description: O_DIRECT bypasses client statistics.
565 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
566 read_ahead_stats were updated. copied the stats section from
567 osc_send_oap_rpc() into async_internal().
571 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
575 Description: change order of libsysio includes
576 Details : '#include sysio.h' should always come before '#include xtio.h'
580 Description: handle "serious error: objid * already exists" more gracefully
581 Details : If LAST_ID value on disk is smaller than the objects existing in
582 the O/0/d* directories, it indicates disk corruption and causes an
583 LBUG(). If the object is 0-length, then we should use the existing
584 object. This will help to avoid a full fsck in most cases.
586 Severity : enhancement
588 Description: adapt the lustre_config script to support the upgrade case
589 Details : Add "-u" option for lustre_config script to support upgrading 1.4
590 server targets to 1.6 in parallel.
594 Description: To avoid grant space > avaible space when the disk is almost
595 full. Without this patch you might see the error "grant XXXX >
596 available" or some LBUG about grant, when the disk is almost
598 Details : In filter_check_grant, for non_grant cache write, we should
599 check the left space by if (*left > ungranted + bytes), instead
600 of (*left > ungranted), because only we are sure the left space
601 is enough for another "bytes", then the ungrant space should be
602 increase. In client, we should update cl_avail_grant only there
603 is OBD_MD_FLGRANT in the reply.
607 Description: Update RHEL 4 kernel to fix local root privilege escalation.
608 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
609 described in CVE-2007-4573. This problem could allow untrusted
610 local users to gain root access.
613 Frequency : when using O_DIRECT and quotas
615 Description: Incorrect file ownership on O_DIRECT output files
616 Details : block usage reported by 'lfs quota' does not take into account
617 files that have been written with O_DIRECT.
622 Description: touch file failed when fs is not full
623 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
624 otherwise we can get ENOSP while fs is not full.
628 Description: parallel lock callbacks
629 Details : Instead of sending blocking and completion callbacks as separated
630 requests, adding them to a set and sending in parallel.
633 Frequency : only for Cray XT3
634 Bugzilla : 12829/13455
635 Description: Changing primary group doesn't change the group lustre assigns to
637 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
638 overridden with the primary group provided by the group upcall,
639 whereas the supplied fsgid can be trusted if it is in the list of
640 supplementary groups returned by the group upcall.
642 Severity : enhancement
644 Description: Allow masking D_WARNING, D_ERROR messages from console
645 Details : Console messages can now be disabled via lnet.printk.
649 Description: User code with malformed file open parameter crashes client node
650 Details : Before packing join_file req, all the related reference should be
651 checked carefully in case some malformed flags cause fake
652 join_file req on client.
656 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
657 Details : ldlm_completion_ast() assumes that a lock is granted when the req
658 mode is equal to the granted mode. However, it should also check
659 that LDLM_FL_CP_REQD is not set.
663 Description: Heavy nfs access might result in deadlocks
664 Details : After ELC code landed, it is now improper to enqueue any mds
665 locks under och_sem, because enqueue might want to decide to
666 cancel open locks for same inode we are holding och_sem for.
670 Description: Client eviction while running blogbench
671 Details : A lot of unlink operations with concurrent I/O can lead to a
672 deadlock causing evictions. To address the problem, the number of
673 oustanding OST_DESTROY requests is now throttled to
674 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
675 callbacks are processed in priority.
677 --------------------------------------------------------------------------------
679 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
681 * Support for kernels:
682 2.6.5-7.283 (SLES 9),
683 2.6.9-55.EL (RHEL 4),
684 2.6.16.46-0.14 (SLES 10),
685 2.6.18.8 vanilla (kernel.org)
686 * Client support for unpatched kernels:
687 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
688 2.6.16 - 2.6.22 vanilla (kernel.org)
689 * Due to recently discovered recovery problems, we do not recommend
690 using patchless RHEL 4 clients with this or any earlier release.
691 * Recommended e2fsprogs version: 1.39.cfs8
692 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
693 * Starting with this release, the ldiskfs backing filesystem required
694 by Lustre is now in its own package, lustre-ldiskfs. This package
695 should be installed. It is versioned separately from Lustre and
696 may be released separately in future.
700 Description: block reactivating mgc import until all deactivates complete
701 Details : Fix race when failing back MDT/MGS to itself (testing)
704 Frequency : at statup only
706 Description: mds_lov_synchronize race leads to various problems
707 Details : simultaneous MDT->OST connections at startup can cause the
708 sync to abort, leaving the OSC in a bad state.
710 Severity : enhancement
712 Description: add optional extra BUILD_VERSION info
713 Details : add a new environment variable (namely LUSTRE_VERS) which allows
714 to override the lustre version.
717 Frequency : 2.6.18 servers only
719 Description: ll_kern_mount() doesn't release the module reference
720 Details : The ldiskfs module reference count never drops down to 0
721 because ll_kern_mount() doesn't release the module reference.
726 Description: server LBUG when using old ost_num_threads parameter
727 Details : Accept the old ost_num_threads parameter but warn that it
728 is deprecated, and fix an off-by-one error that caused an LBUG.
733 Description: Transient SCSI error results in persistent IO issue
734 Details : iobuf->dr_error is not reinitialized to 0 between two
738 Frequency : sometimes when underlying device returns I/O errors
740 Description: OSTs not going read-only during write failures
741 Details : OSTs are not remounted read-only when the journal commit threads
742 get I/O errors because fsfilt_ext3 calls journal_start/stop()
743 instead of the ext3 wrappers.
747 Description: poor connect scaling with increasing client count
748 Details : Don't run filter_grant_sanity_check for more than 100 exports
749 to improve scaling for large numbers of clients.
752 Frequency : SLES10 only
754 Description: sanity-quota.sh quotacheck failed: rc = -22
755 Details : Quotas cannot be enabled on SLES10.
758 Frequency : liblustre clients only
760 Description: getdirentries does not give error when run on compute nodes
761 Details : getdirentries does not fail when the size specified as an argument
762 is too small to contain at least one entry
764 Severity : enhancement
766 Description: Add LNET router traceability for debug purposes
767 Details : If a checksum failure occurs with a router as part of the
768 IO path, the NID of the last router that forwarded the bulk data
769 is printed so it can be identified.
774 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
775 Details : Due to a race condition, liblustre clients were occasionally
778 Severity : enhancement
780 Description: lfs setstripe use optional parameters instead of postional
783 Severity : enhancement
785 Description: Nanosecond timestamp support for ldiskfs
786 Details : The on-disk ldiskfs filesystem has added support for nanosecond
787 resolution timestamps. There is not yet support for this at
788 the Lustre filesystem level.
791 Frequency : during server recovery
793 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
794 Details : request with rq_no_resend flag not awake l_wait_event if they get a
798 Frequency : nfs export on patchless client
800 Description: connectathon hang when test nfs export over patchless client
801 Details : Disconnected dentry cannot be found with lookup, so we do not need
802 to unhash it or make it invalid
805 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
806 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
807 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
812 Description: open req refcounting wrong on reconnect
813 Details : If reconnect happened between getting open reply from server and
814 call to mdc_set_replay_data in ll_file_open, we will schedule
815 replay for unreferenced request that we are about to free.
816 Subsequent close will crash in variety of ways.
817 Check that request is still eligible for replay in
818 mdc_set_replay_data().
823 Description: disable writes to filesystem when reading health_check file
824 Details : the default for reading the health_check proc file has changed
825 to NOT do a journal transaction and write to disk, because this
826 can cause reads of the /proc file to hang and block HA state
827 checking on a healthy but otherwise heavily loaded system. It
828 is possible to return to the previous behaviour during configure
829 with --enable-health-write.
831 Severity : enhancement
833 Description: 64-bit inode version
834 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
835 to the inode. This will be required for version-based recovery.
840 Description: MDS fails to start if a duplicate client export is detected
841 Details : in some rare cases it was possible for a client to connect to
842 an MDS multiple times. Upon recovery the MDS would detect this
843 and fail during startup. Handle this more gracefully.
845 Severity : enhancement
847 Description: Add -o localflock option to simulate old noflock
849 Details : This will achieve local-only flock/fcntl locks
855 Description: log_commit_thread vs filter_destroy race leads to crash
856 Details : Take import reference before releasing llog record semaphore
861 Description: Wrong request locking in request set processing
862 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
863 imp_delayed_list, in this place should be used imp_lock.
866 Frequency : when reconnection
868 Description: Grant Leak when osc reconnect to OST
869 Details : When osc reconnect ost, OST(filter) should check whether it
870 should grant more space to client by comparing fed_grant and
871 cl_avail_grant, and return the granted space to client instead
872 of "new granted" space, because client will call osc_init_grant
873 to update the client grant space info.
876 Frequency : when client reconnect to OST
878 Description: Grant Leak when osc do resend and replay bulk write
879 Details : When osc reconnect to OST, OST(filter)should clear grant info of
880 bulk write request, because the grant info while be sync between
881 OSC and OST when reconnect, and we should ignore the grant info
882 these of resend/replay write req.
887 Description: Grant space more than avaiable left space sometimes.
888 Details : When then OST is about to be full, if two bulk writing from
889 different clients came to OST. Accord the avaliable space of the
890 OST, the first req should be permitted, and the second one
891 should be denied by ENOSPC. But if the seconde arrived before
892 the first one is commited. The OST might wrongly permit second
893 writing, which will cause grant space > avaiable space.
896 Frequency : when client is evicted
898 Description: Grant might be wrongly erased when osc is evicted by OST
899 Details : when the import is evicted by server, it will fork another
900 thread ptlrpc_invalidate_import_thread to invalidate the
901 import, where the grant will be set to 0. While the original
902 thread will update the grant it got when connecting. So if
903 the former happened latter, the grant will be wrongly errased
904 because of this race.
909 Description: Checking Stale with correct fid
910 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
911 to check whether it is stale, because sometimes, we want the
912 enqueue happened anyway, and op_data.fid2 will not be initialized.
914 Severity : enhancement
916 Description: update patchless client
917 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
920 Frequency : only with 2.4 kernel
922 Description: random memory corruption
923 Details : size of struct ll_inode_info is to big for union inode.u and this
924 can be cause of random memory corruption.
929 Description: Memory leak in recovery
930 Details : Lov_mds_md was not free in an error handler in mds_create_object.
931 It should also check obd_fail before fsfilt_start, otherwise if
932 fsfilt_start return -EROFS,(failover mds during mds recovery).
933 then the req will return with repmsg->transno = 0 and rc = EROFS.
934 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
935 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
936 be freed no matter whether fsfilt_commit success or not.
939 Frequency : only with huge count clients
941 Description: Prevents from taking the superblock lock in llap_from_page for
943 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
944 for llap_from_page call in ll_removepage prevents from taking the
945 superblock lock for a soon died page.
950 Description: Not check open intent error before release open handle
951 Details : in some rare cases, the open intent error is not checked before
952 release open handle, which may cause
953 ASSERTION(open_req->rq_transno != 0), because it tries to release
954 the failed open handle.
959 Description: Set cat log bitmap only after create log success.
960 Details : in some rare cases, the cat log bitmap is set too early. and it
961 should be set only after create log success.
965 Description: Accessing a block bevice can re-enable I/O when Lustre is
966 tearing down a device.
967 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
972 Description: service threads may hog cpus when there are a lot of requests
973 Details : Insert cond_resched to give other threads a chance to use some CPU
978 Description: the cat log was not initialized in recovery
979 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
980 unlink log on mds will not be initialized until mds post
981 recovery. And also in mds post recovery, the unlink log will
982 initialization will be done asynchronausly, so there will be race
983 between add unlink log and unlink log initialization.
987 Description: brw_stats were being printed incorrectly
988 Details : brw_stats were being printed as log2 but all of them were not
989 recorded as log2. Also remove some code duplication arising from
990 filter_tally_{read,write}.
994 Frequency : rare, only in recovery.
995 Description: ASSERTION(req->rq_type != LI_POISON) failed
996 Details : imp_lock should be held while iterating over imp_sending_list for
997 prevent destroy request after get timeout in ptlrpc_queue_wait.
1001 Description: replay-single.sh test 52 fails
1002 Details : A lock's skiplist need to be cleanup when it being unlinked
1003 from its resource list.
1007 Description: Short directio read returns full requested size rather than
1009 Details : Direct I/O operations should return actual amount of bytes
1010 transferred rather than requested size.
1012 Severity : enhancement
1014 Description: metadata RPC reduction (e.g. for rm performance)
1015 Details : decrease the amount of synchronous RPC between clients and servers
1016 by canceling conflicing lock before the operation on the client
1017 and packing thier handles into the main operation RPC to server.
1019 Severity : enhancement
1021 Description: Async OSC create to avoid the blocking unnecessarily.
1022 Details : If a OST has no remain object, system will block on the creating
1023 when need to create a new object on this OST. Now, ways use
1024 pre-created objects when available, instead of blocking on an
1025 empty osc while others are not empty. If we must block, we block
1026 for the shortest possible period of time.
1028 Severity : enhancement
1030 Description: refine locking for avoid write wrong info into lov_objid file
1031 Details : fix possible races with add new target and write/update data in
1034 --------------------------------------------------------------------------------
1036 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1041 Frequency : on some architectures
1043 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1044 Details : Uninitialized flags sometimes cause configuration commands to
1048 Frequency : patchless clients only
1050 Description: missing __iget() symbol export
1051 Details : The __iget() symbol export is missing. To avoid the need for
1052 this on patchless clients the deathrow inode reaper is turned
1053 off, and we depend on the VM to clean up old inodes. This
1054 dependency was during via the fix for bug 12181.
1056 --------------------------------------------------------------------------------
1058 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1060 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1061 INTEROPERATE with older versions automatically. In many cases a
1062 special upgrade step is needed. Please read the
1063 user documentation before upgrading any part of a live system.
1064 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1065 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1066 * WARNING: Lustre configuration and startup changes are required with
1067 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1073 Severity : enhancement
1075 Description: MountConf
1076 Details : Lustre configuration is now managed via mkfs and mount
1077 commands instead of lmc and lconf. New obd types (MGS, MGC)
1078 are added for dynamic configuration management. See
1079 https://mail.clusterfs.com/wikis/lustre/MountConf for
1082 Severity : enhancement
1084 Description: dynamic OST addition
1085 Details : OSTs can now be added to a live filesystem
1087 Severity : enhancement
1089 Description: startup order invariance
1090 Details : MDTs and OSTs can be started in any order. Clients only
1091 require the MDT to complete startup.
1093 Severity : enhancement
1095 Description: parallel, asynchronous orphan cleanup
1096 Details : orphan cleanup is now performed in separate threads for each
1097 OST, allowing parallel non-blocking operation.
1099 Severity : enhancement
1101 Description: optimized stripe assignment
1102 Details : stripe assignments are now made based on ost space available,
1103 ost previous usage, and OSS previous usage, in order to try
1104 to optimize storage space and networking resources.
1106 Severity : enhancement
1108 Description: Permanently set tunables
1109 Details : All writable /proc/fs/lustre tunables can now be permanently
1110 set on a per-server basis, at mkfs time or on a live system.
1112 Severity : enhancement
1114 Description: Lustre message v2
1115 Details : Add lustre message format v2.
1117 Severity : enhancement
1119 Description: client OST exclusion list
1120 Details : Clients can be started with a list of OSTs that should be
1121 declared "inactive" for known non-responsive OSTs.
1125 Description: SPEC SFS validation failure on NFS v2 over lustre.
1126 Details : Changes the blocksize for regular files to be 2x RPC size,
1127 and not depend on stripe size.
1129 Severity : enhancement
1131 Description: Multiple MD RPCs in flight.
1132 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1133 To avoid overly-overloading MDS, introduce a limit on number of
1134 MDS RPCs in flight for a single client and add /proc controls
1135 to adjust this limit.
1137 Severity : enhancement
1139 Description: client read/write statistics
1140 Details : Add client read/write call usage stats for performance
1141 analysis of user processes.
1142 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1143 file access. extents_stats shows chunk size distribution.
1144 extents_stats_per_process show chunk size distribution per
1147 Severity : enhancement
1149 Description: mds statistics
1150 Details : Add detailed mds operations statistics in
1151 /proc/fs/lustre/mds/*/stats.
1155 Description: Failure of copying files with lustre special EAs.
1156 Details : Client side always return success for setxattr call for lustre
1157 special xattr (currently only "trusted.lov").
1162 Description: Refcount LNET uuids
1163 Details : The global LNET uuid list grew linearly with every startup;
1164 refcount repeated list entries instead of always adding to
1167 Severity : enhancement
1169 Description: Dynamic service threads
1170 Details : Within a small range, start extra service threads
1171 automatically when the request queue builds up.
1174 Frequency : mixed-endian client/server environments
1176 Description: mixed-endian crashes
1177 Details : The new msg_v2 system had some failures in mixed-endian
1180 Severity : enhancement
1182 Description: Easy OST removal
1183 Details : OSTs can be permanently deactivated with e.g. 'lctl
1184 conf_param lustre-OST0001.osc.active=0'
1186 Severity : enhancement
1188 Description: MGS proc entries
1189 Details : Added basic proc entries for the MGS showing what filesystems
1192 Severity : enhancement
1194 Description: provide MGS failover
1195 Details : Added config lock reacquisition after MGS server failover.
1197 Severity : enhancement
1199 Description: add Linux 2.4 support
1200 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1204 Description: a large application tries to do I/O to the same resource and dies
1205 in the middle of it.
1206 Details : Check the req->rq_arrival time after the call to
1207 ost_brw_lock_get(), but before we do anything about
1208 processing it & sending the BULK transfer request. This
1209 should help move old stale pending locks off the queue as
1210 quickly as obd_timeout.
1213 Frequency : when an incorrect nid is specified during startup
1215 Description: ptlrpc connect to non-existant node causes kernel crash
1216 Details : LNET can't be re-entered from an event callback, which
1217 happened when we expire a message after the export has been
1218 cleaned up. Instead, hand the zombie cleanup off to another
1221 Severity : enhancement
1223 Description: plain/inodebits lock performance improvement
1224 Details : Grouping plain/inodebits in granted list by their request modes
1225 and bits policy, thus improving the performance of search through
1229 Frequency : only if OST filesystem is corrupted
1231 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1232 Details : for a short time RPCs with bulk IO are in the replay list,
1233 but replay of bulk IOs is unimplemented. If the OST filesystem
1234 is corrupted due to disk cache incoherency and then replay is
1235 started it is possible to trip an assertion. Avoid putting
1236 committed RPCs into the replay list at all to avoid this issue.
1239 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1241 Description: System hang on startup
1242 Details : This bug allowed the liblustre (e.g. catamount) client to
1243 return to the app before handling all startup RPCs. This
1244 could leave the node unresponsive to lustre network traffic
1245 and manifested as a server ptllnd timeout.
1247 Severity : enhancement
1249 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1250 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1251 boolean to control whether to print peer debug info when a
1252 client's RPC times out.
1255 Frequency : only for kernels with patches from Lustre below 1.4.3
1257 Description: Remove old rdonly API
1258 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1261 Frequency : only for devices with external journals
1263 Description: Set external device read-only also
1264 Details : During a commanded failover stop, we set the disk device
1265 read-only while the server shuts down. We now also set any
1266 external journal device read-only at the same time.
1269 Frequency : when upgrading from 1.4 while trying to change parameters
1271 Description: The wrong (new) MDC name was used when setting parameters for
1272 upgraded MDT's. Also allows changing of OSC (and MDC)
1273 parameters if --writeconf is specified at tunefs upgrade time.
1276 Frequency : when setting specific ost indicies
1278 Description: QOS code breaks on skipped indicies
1279 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1280 created with --index need not be sequential.
1282 Severity : enhancement
1284 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1285 Details : The uninit_groups feature works in conjunction with the kernel
1286 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1287 up the pass1 processing of e2fsck. This is a read-only feature
1288 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1289 mount filesystems that have had this feature enabled.
1291 Severity : enhancement
1293 Description: Improve multi-block allocation algorithm to avoid fragmentation
1294 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1295 allocation locality and avoid filesystem fragmentation.
1297 ------------------------------------------------------------------------------
1299 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1301 * Support for kernels:
1302 2.4.21-47.0.1.EL (RHEL 3)
1303 2.6.5-7.283 (SLES 9)
1304 2.6.9-42.0.10.EL (RHEL 4)
1305 2.6.12.6 vanilla (kernel.org)
1306 2.6.16.27-0.9 (SLES 10)
1307 * Recommended e2fsprogs version: 1.39.cfs5
1309 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1313 Frequency : occasional, depends on client load and configuration
1314 Bugzilla : 12181, 12203
1315 Description: data loss for recently-modified files
1317 Details : In some cases it is possible that recently written or created
1318 files may not be written to disk in a timely manner (this should
1319 normally be within 30s unless client IO load is very high).
1320 The problem appears as zero-length files or files that are a
1321 multiple of 1MB in size after a client crash or client eviction
1322 that are missing data at the end of the file.
1324 This problem is more likely to be hit on clients where files are
1325 repeatedly created and unlinked in the same directory, clients
1326 have a large amount of RAM, have many CPUs, the filesystem has
1327 many OSTs, the clients are rebooted frequently, and/or the files
1328 are not accessed by other nodes after being written.
1330 The presence of the problem can be detected by looking at
1331 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1332 smaller than the second (nr_unused) then dirty files will not
1333 be flushed automatically to disk. "sync; sleep 10" should be
1334 run several times on the node before unmounting it to update
1335 Lustre (this is also safe to run on nodes without this problem).
1337 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1338 that can cause this same problem, so customers using that kernel
1339 also need to update the kernel in addition to Lustre. In order
1340 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1342 It is normal that files written just before a client crash (less
1343 than 30s) may not yet have been flushed to disk, even for local
1347 Frequency : frequent on thin XT3 nodes
1349 Description: UUID collision on thin XT3 Linux nodes
1350 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1351 randomly, since we relied on an insufficiently-seeded PRNG.
1356 Description: OSS hangs after "All ost request buffers busy"
1357 Details : A deadlock between quota and journal operations caused OSS
1358 hangs after printing "All ost request buffers busy."
1361 Frequency : always on liblustre builds
1363 Description: Cleanup compiler warnings on liblustre
1366 Frequency : always on liblustre builds on XT3
1368 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1372 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1373 > 32000 subdirectories
1374 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1375 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1376 subdirectory count crosses 32000. This will aid e2fsck to
1377 correctly handle more than 32000 subdirectories.
1380 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1382 Description: System hang on startup
1383 Details : This bug allowed the liblustre (e.g. catamount) client to
1384 return to the app before handling all startup RPCs. This
1385 could leave the node unresponsive to lustre network traffic
1386 and manifested as a server ptllnd timeout.
1388 Severity : enhancement
1390 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1391 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1392 boolean to control whether to print peer debug info when a
1393 client's RPC times out.
1398 Description: make O_SYNC working on 2.6 kernels
1399 Details : 2.6 kernels use different method for mark pages for write,
1400 so need add a code to lustre for O_SYNC work.
1405 Description: Failure to close file and release space on NFS
1406 Details : Put inode details into lock acquired in ll_intent_file_open.
1407 Use mdc_intent_lock in ll_intent_open to properly
1408 detect all kind of errors unhandled by mdc_enqueue.
1413 Description: proc file read during shutdown sometimes raced obd removal,
1415 Details : Add lock to prevent obd access after proc file removal.
1418 Frequency : Only for files larger than 4GB on 32-bit clients.
1420 Description: improperly doing page alignment of locks
1421 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1422 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1427 Description: RPCs being resent when they shouldn't be
1428 Details : Some RPCs that should not be resent are being resent. This
1429 can cause inconsistencies in the RPC state machine. Do not
1430 resend such requests.
1433 Frequency : rare, only with NFS export
1435 Description: Crash on NFS re-export node
1436 Details : under very unusual load conditions an assertion is hit in
1437 ll_intent_file_open()
1440 Frequency : only if OST filesystem is corrupted
1442 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1443 Details : for a short time RPCs with bulk IO are in the replay list,
1444 but replay of bulk IOs is unimplemented. If the OST filesystem
1445 is corrupted due to disk cache incoherency and then replay is
1446 started it is possible to trip an assertion. Avoid putting
1447 committed RPCs into the replay list at all to avoid this issue.
1452 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1453 Details : Large single O_DIRECT read and write calls can fail to allocate
1454 a sufficiently large buffer to process the request. In case of
1455 allocation failure the allocation is retried with a smaller
1456 buffer and broken into smaller requests.
1458 Severity : enhancement
1460 Description: Add -o localflock option to simulate old noflock behaviour.
1461 Details : This will achieve local-only flock/fcntl locks coherentness.
1466 Description: versioning check is incomplete
1467 Details : Checking the version difference of client vs. server, report
1468 error if the gap is too big.
1473 Description: add support PG_writeback bit
1474 Details : add support for PG_writeback bit for Lustre, for more carefull
1475 work with page cache in 2.6 kernel. This also fix some deadlocks
1476 and remove hack for work O_SYNC with 2.6 kernel.
1478 Severity : enhancement
1480 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1481 Details : The uninit_groups feature works in conjunction with the kernel
1482 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1483 up the pass1 processing of e2fsck. This is a read-only feature
1484 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1485 mount filesystems that have had this feature enabled.
1487 Severity : enhancement
1489 Description: Improve multi-block allocation algorithm to avoid fragmentation
1490 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1491 allocation locality and avoid filesystem fragmentation.
1493 ------------------------------------------------------------------------------
1495 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1497 * Support for kernels:
1498 2.6.9-42.0.3.EL (RHEL 4)
1499 2.6.5-7.276 (SLES 9)
1500 2.4.21-47.0.1.EL (RHEL 3)
1501 2.6.12.6 vanilla (kernel.org)
1502 2.6.16.21-0.8 (SLES10)
1503 * Recommended e2fsprogs version: 1.39.cfs2-0
1505 * The backwards-compatible /proc/sys/portals symlink has been removed
1506 in this release. Before upgrading, please ensure that you change
1507 any configuration scripts or /etc/sysctl.conf files that access
1508 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1509 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1510 in advance of the upgrade on any system running Lustre 1.4.6 or
1511 newer, since /proc/sys/lnet was added in that version.
1512 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1516 Frequency : only when quota is used
1518 Description: avoid scanning export list for quota master
1519 Details : Change the algorithms to avoid scanning export list in order
1520 to improve the efficiency.
1523 Frequency : MDS failover only, very rarely
1525 Description: "went back in time" messages on mds failover
1526 Details : The greatest transno may be lost when the current operation
1527 finishes with an error (transno==0) and the client's last_rcvd
1528 record is over-written. Save the greatest transno in the
1529 mds_last_transno for this case.
1532 Frequency : always for specific kernels and striping counts
1534 Description: client may get "Matching packet too big" without ACL support
1535 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1536 when trying to access files in certain configurations. The
1537 clients should in fact be denied when mounting because they do
1538 not understand ACLs.
1541 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1543 Description: many clients connecting with IO in progress causes connect timeouts
1544 Details : Avoid synchronous journal commits to avoid delays caused by many
1545 clients connecting/disconnecting when bulk IO is in progress.
1546 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1547 OST_IO_PORTAL to avoid delays behind potentially many pending
1551 Frequency : occasionally with multiple writers to a single file
1553 Description: shared writes to file may result in wrong size reported by stat()
1554 Details : Allow growing of kms when extent lock is cancelled
1557 Frequency : always with random mmap IO to multi-striped file
1559 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1560 Details : Only if the hole is at the end of OST object so that kms is too
1561 small. Fix is to increase kms accordingly in ll_nopage.
1564 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1566 Description: writes to a missing object would leak memory on the OST
1567 Details : If there is an inconsistency between the MDS and OST filesystems,
1568 such that the MDS references an object that doesn't exist, writes
1569 to that object will leak memory due to incorrect cleanup in the
1570 error handling path, eventually running out of memory on the OST.
1575 Description: Creating too long symlink causes lustre errors
1576 Details : Check symlink and name lengths before sending requests to MDS.
1579 Frequency : only if flock is enabled (not on by default)
1581 Description: posix locks not released on fd closure on 2.6.9+
1582 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1583 kernels, this caused such locks not to be released on fd close and
1584 then assertions on fs unmount about still used locks.
1587 Frequency : MDS failover only, very rarely
1589 Description: clients may get ASSERTION(granted_lock != NULL)
1590 Details : When request was taking a long time, and a client was resending
1591 a getattr by name lock request. The were multiple lock requests
1592 with the same client lock handle and
1593 mds_getattr_name->fixup_handle_for_resent_request found one of the
1594 lock handles but later failed with ASSERTION(granted_lock != NULL).
1599 Description: handle->h_buffer_credits > 0, assertion failure
1600 Details : h_buffer_credits is zero after truncate, causing assertion
1601 failure. This patch extends the transaction or creates a new
1605 Frequency : NFS re-export or patchless client
1606 Bugzilla : 11179, 10796
1607 Description: Crash on NFS re-export node (__d_move)
1608 Details : We do not want to hash the dentry if we don't have a lock.
1609 But if this dentry is later used in d_move, we'd hit uninitialised
1610 list head d_hash, so we just do this to init d_hash field but
1611 leave dentry unhashed.
1614 Frequency : NFS re-export or patchless client
1616 Description: NFS exports has problem with symbolic link
1617 Details : lustre client didn't properly install dentry when re-exported
1618 to NFS or running patchless client.
1621 Frequency : NFS re-export or patchless client
1623 Description: Various nfs/patchless fixes.
1624 Details : fixes reuse disconected alias for lookup process - this fixes
1625 warning "find_exported_dentry: npd != pd",
1626 fix permission error with open files at nfs.
1627 fix apply umask when do revalidate.
1630 Frequency : occasional
1632 Description: Crash on NFS re-export node
1633 Details : calling clear_page() on the wrong pointer triggered oops in
1634 generic_mapping_read().
1637 Frequency : rarely, using O_DIRECT IO
1639 Description: unaligned directio crashes client with LASSERT
1640 Details : check for unaligned buffers before trying any requests.
1643 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1645 Description: stale data returned from RAID cache
1646 Details : If only a small amount of IO is done to the RAID device before
1647 reading it again it is possible to get stale data from the RAID
1648 cache instead of reading it from disk.
1651 Frequency : always for sles10 kernel
1653 Description: sles10 support
1654 Details : ll_follow_link: compile fixes and using of nd_set_link
1655 under newer kernels.
1658 Frequency : depends on arch, kernel and compiler version, always on sles10
1661 Description: recursive or deep enough symlinks cause stack overflow
1662 Details : getting rid of large stack-allocated variable in
1666 Frequency : depends on hardware
1668 Description: lustre write performance loss in the SLES10 kernel
1669 Details : the performance loss is caused by using of write barriers in the
1670 ext3 code. The SLES10 kernel turns barrier support on by
1671 default. The fix is to undo that change for ldiskfs.
1673 ------------------------------------------------------------------------------
1675 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1677 * Support for kernels:
1678 2.6.9-42.0.3EL (RHEL 4)
1679 2.6.5-7.276 (SLES 9)
1680 2.4.21-47.0.1.EL (RHEL 3)
1681 2.6.12.6 vanilla (kernel.org)
1685 Frequency : quota enabled and large files being deleted
1687 Description: releasing more than 4GB of quota at once hangs OST
1688 Details : If a user deletes more than 4GB of files on a single OST it
1689 will cause the OST to spin in an infinite loop. Release
1690 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1695 Description: statfs data retrieved from /proc may be stale or zero
1696 Details : When reading per-device statfs data from /proc, in the
1697 {kbytes,files}_{total,free,avail} files, it may appear
1698 as zero or be out of date.
1701 Frequency : systems with MD RAID1 external journal devices
1703 Description: lconf's call to blkid is confused by RAID1 journal devices
1704 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1705 of returning all block devices that match the journal UUID.
1708 Frequency : always, for aggregate stripe size over 4GB
1710 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1711 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1712 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1715 Frequency : always if "lfs find" used on a local file/directory
1717 Description: "lfs find" segfaults if used on a local file/directory
1718 Details : The case where a directory component was not specified wasn't
1719 handled correctly. Handle this properly.
1722 Frequency : always on ppc64
1724 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1725 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1726 perform endianness conversion.
1729 Frequency : rarely (truncate to non-zero file size after write under load)
1730 Bugzilla : 10730, 10687
1731 Description: Files padded with zeros to next 4K multiple
1732 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1733 it is possible that files that are truncated to a non-zero size
1734 immediately after being written are filled with zero bytes beyond
1735 the truncated size. No file data is lost.
1737 Severity : enhancement
1739 Description: Allow recovery/failover for liblustre clients.
1740 Details : liblustre clients were unaware of failover configurations until
1743 Severity : enhancement
1745 Description: user file locks should fail when not mounting with flock option
1746 Details : Set up an error-returning stub in ll_file_operations.lock field
1747 to prevent incorrect behaviour when client is mounted without
1748 flock option. Also, set up properly f_op->flock field for
1752 Frequency : always on ia64
1754 Description: "lfs df" loops on printing out MDS statfs information
1755 Details : The obd_ioctl_data was not initialized and in some systems
1756 this caused a failure during the ioctl that did not return
1757 an error. Initialize the struct and return an error on failure.
1760 Frequency : SLES 9 only
1762 Description: Error of copying files with lustre special EAs as root
1763 Details : Client side always return success for setxattr call for lustre
1764 special xattr (currently only "trusted.lov").
1767 Frequency : rarely on clusters with both ia64+i386 clients
1769 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1770 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1771 clients are doing concurrent writes to the same file it is
1772 possible that the ia64 clients may LASSERT because the OST
1773 extent locks are not PAGE_SIZE aligned. Ensure that grown
1774 locks are always aligned on the request boundary.
1777 Frequency : specific use, occasional
1779 Description: Overwriting in use executable truncates on-disk binary image
1780 Details : If one node attempts to overwrite an executable in use by
1781 another node, we now correctly return ETXTBSY instead of
1782 truncating the file.
1784 Severity : enhancement
1786 Description: Async OSC create to avoid the blocking unnecessarily.
1787 Details : If a OST has no remain object, system will block on the creating
1788 when need to create a new object on this OST. Now, ways use
1789 pre-created objects when available, instead of blocking on an
1790 empty osc while others are not empty. If we must block, we block
1791 for the shortest possible period of time.
1796 Description: chmod on Lustre root is propagated to other clients
1797 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1798 invalid by the follow_mount time.
1801 Frequency : liblustre clients only
1803 Description: Race in 'instant cancel' lock handling could lead to such locks
1804 never to be granted in case of SMP MDS
1805 Details : Do not destroy not yet granted but cbpending locks in
1809 Frequency : replay/resend of open
1811 Description: non null lock assetion failure in mds_intent_policy
1812 Details : Trying to replay/resend lockless open requests resulted in
1813 mds_open() returning 0 with no lock. Now it sets a flag if
1814 a lock is going to be returned.
1816 Severity : enhancement
1818 Description: Checksum enhancements
1819 Details : New checksum enhancements allow for resending RPCs that failed
1822 Severity : enhancement
1824 Description: Tunables on number of dirty pages in cacche
1825 Details : Allow to set limit on number of dirty pages cached.
1830 Description: client crash on unmount - lock still has references
1831 Details : In some error handling cases it was possible to leak a lock
1832 reference on a client while accessing a file. This was not
1833 harmful to the client during operation, but would cause the
1834 client to crash when the filesystem is unmounted.
1837 Frequency : specific case, rare
1839 Description: ETXTBSY on mds though file not in use by client
1840 Details : ETXTBSY is no longer incorrectly returned when attempting to
1841 chmod or chown a directory that the user previously tried to
1842 execute or a currently-executing binary.
1845 Frequency : extremely rare except on liblustre-based clients
1847 Description: Lustre space not freed when files are deleted
1848 Details : Clean up open-unlinked files after client eviction. Previously
1849 the unlink was skipped and the files remained as orphans.
1854 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1855 Details : In some cases it was possible to send a blocking callback to
1856 a client doing a glimpse, even though that client didn't get
1857 a lock granted. When the glimpse lock is cancelled on the OST
1858 the freed lock is left on the waiting list and corrupted the list.
1861 Frequency : all core dumps
1863 Description: Broke core dumps to lustre
1864 Details : Negative dentry may be unhashed if parent does not have UPDATE
1865 lock, but some callers, e.g. do_coredump, expect dentry to be
1866 hashed after successful create, hash it in ll_create_it.
1868 ------------------------------------------------------------------------------
1870 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1872 * Support for kernels:
1873 2.6.9-42.0.2.EL (RHEL 4)
1874 2.6.5-7.276 (SLES 9)
1875 2.4.21-40.EL (RHEL 3)
1876 2.6.12.6 vanilla (kernel.org)
1880 Frequency : always on RHEL 3
1882 Description: Number of open files grows over time
1883 Details : The number of open files grows over time, whether or not
1884 Lustre is started. This was due to a filp leak introduced
1885 by one of our kernel patches.
1887 ------------------------------------------------------------------------------
1889 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1891 * Support for kernels:
1892 2.6.9-42.EL (RHEL 4)
1893 2.6.5-7.276 (SLES 9)
1894 2.4.21-40.EL (RHEL 3)
1895 2.6.12.6 vanilla (kernel.org)
1900 Bugzilla : 5719, 9635, 9792, 9684
1901 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1902 Details : If a server is under heavy load and cannot reply to new
1903 connection requests before the client resends the (re)connect,
1904 the connection handling code can behave badly if two service
1905 threads are concurrently handing separate (re)connections from
1906 the same client. Add better locking to the connection handling
1907 code, and ensure that only a single connection will be processed
1908 for a given client UUID, even if the lock is dropped.
1910 Severity : enhancement
1912 Description: add TCP zero-copy support to kernel
1913 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1914 sends if the hardware supports scatter-gather and checksumming.
1915 This allows socklnd to do client-write and server-read more
1916 efficiently and reduce CPU utilization from skbuf copying.
1919 Frequency : only if NFS exporting from client
1921 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1922 Details : If setting attributes on a file created under NFS that had
1923 never been opened it would be possible to oops the client
1924 if the file had no objects.
1927 Frequency : always for liblustre
1929 Description: liblustre client does MDS+OSTs setattr RPC for each write
1930 Details : When doing a write from a liblustre client, the client
1931 incorrectly issued an RPC to the MDS and each OST the file was
1932 striped over in order to update the timestamps. When writing
1933 with small chunks and many clients this could overwhelm the MDS
1934 with RPCs. In all cases it would slow down the write because
1935 these RPCs are unnecessary.
1937 Severity : enhancement
1939 Description: allow number of MDS service threads to be changed at module load
1940 Details : It is now possible to change the number of MDS service threads
1941 running. Adding "options mds mds_num_threads={N}" to the MDS's
1942 /etc/modprobe.conf will set the number of threads for the next
1943 time Lustre is restarted (assuming the "mds" module is also
1944 reloaded at that time). The default number of threads will
1945 stay the same, 32 for most systems.
1950 Description: OST crash if filesystem is unformatted or corrupt
1951 Details : If an OST is started on a device that has never been formatted
1952 or if the filesystem is corrupt and cannot even mount then the
1953 error handling cleanup routines would dereference a NULL pointer.
1958 Description: NULL pointer deref in llap_from_page.
1959 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1960 incorrect) mapping if the page was truncated/reclaimed while it was
1961 searched for. Check for this condition and skip such pages when
1962 doing readahead. Introduce extra check to llap_from_page() to
1963 verify page->mapping->host is non-NULL (so page is not anonymous).
1966 Frequency : Sometimes when using sys_sendfile
1968 Description: "page not covered by a lock" warnings from ll_readpage
1969 Details : sendfile called ll_readpage without right page locks present.
1970 Now we introduced ll_file_sendfile that does necessary locking
1971 around call to generic_file_sendfile() much like we do in
1975 Frequency : with certain MDS communication failures at client mount time
1977 Description: NULL pointer deref after failed client mount
1978 Details : a client connection request may delayed by the network layer
1979 and not be sent until after the PTLRPC layer has timed out the
1980 request. If the client fails the mount immediately it will try
1981 to clean up before the network times out the request. Add a
1982 reference from the request import to the obd device and delay
1983 the cleanup until the network drops the request.
1986 Frequency : occasionally during client (re)connect
1988 Description: assertion failure during client (re)connect
1989 Details : processing a client connection request may be delayed by the
1990 client or server longer than the client connect timeout. This
1991 causes the client to resend the connection request. If the
1992 original connection request is replied in this interval, the
1993 client may trip an assertion failure in ptlrpc_connect_interpret()
1994 which thought it would be the only running connect process.
1997 Frequency : only with obd_echo servers and clients that are rebooted
1999 Description: kernel BUG accessing uninitialized data structure
2000 Details : When running an obd_echo server it did not start the ping_evictor
2001 thread, and when a client was evicted an uninitialized data
2002 structure was accessed. Start the ping_evictor in the RPC
2003 service startup instead of the OBD startup.
2005 Severity : enhancement
2006 Bugzilla : 10193 (patchless)
2007 Description: Remove dependency on various unexported kernel interfaces.
2008 Details : No longer need reparent_to_init, exit_mm, exit_files,
2009 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2012 Frequency : rare (only users of deprecated and unsupported LDAP config)
2014 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2015 Details : LDAP apparently contains 'lustreName' attributes instead of
2016 'name'. A simple remapping of the name is sufficient.
2019 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2021 Description: waiting_locks_callback trips kernel BUG if client is evicted
2022 Details : Running with the dump_on_timeout debug flag turned on makes
2023 it possible that the waiting_locks_callback() can try to dump
2024 the Lustre kernel debug logs from an interrupt handler. Defer
2025 this log dumping to the expired_lock_main() thread.
2027 Severity : enhancement
2029 Description: Support NFS exporting on 2.6 kernels.
2030 Details : Implement non-rawops metadata methods for NFS server to use without
2031 changing NFS server code.
2034 Frequency : very rare (synthetic metadata workload only)
2036 Description: two racing renames might cause an MDS thread to deadlock
2037 Details : Running the "racer" program may cause one MDS thread to rename
2038 a file from being the source of a rename to being the target of
2039 a rename at exactly the same time that another thread is doing
2040 so, and the second thread has already enqueued these locks after
2041 doing a lookup of the target and is trying to relock them in
2042 order. Ensure that we don't try to re-lock the same resource.
2045 Frequency : only very large systems with liblustre clients
2047 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2048 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2049 all OSTs in parallel. This allows the eviction of stale liblustre
2050 clients to proceed much faster than if they were done in series,
2051 and also offers similar improvements for other set_info RPCs.
2056 Description: excessive CPU usage during initial read phase on client
2057 Details : During the initial read phase on a client, it would agressively
2058 retry readahead on the file, consuming too much CPU and impacting
2059 performance (since 1.4.5.8). Improve the readahead algorithm
2060 to avoid this, and also improve some other common cases (read
2061 of small files in particular, where "small" is files smaller than
2062 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2067 Description: MDS crash when receiving packet with unknown intent.
2068 Details : Do not LBUG in unknown intent case, just return -EFAULT
2070 Severity : enhancement
2071 Bugzilla : 9293, 9385
2072 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2073 Details : Do not serialize getattr (non-intent version) and statfs.
2076 Frequency : occasional, when OST network is overloaded/intermittent
2078 Description: client evicted by OST after bulk IO timeout
2079 Details : If a client sends a bulk IO request (read or write) the OST
2080 may evict the client if it is unresposive to its data GET/PUT
2081 request. This is incorrect if the network is overloaded (takes
2082 too long to transfer the RPC data) or dropped the OST GET/PUT
2083 request. There is no need to evict the client at all, since
2084 the pinger and/or lock callbacks will handle this, and the
2085 client can restart the bulk request.
2088 Frequency : Always when mmapping file with no objects
2090 Description: client crashes when mmapping file with no objects
2091 Details : Check that we actually have objects in a file before doing any
2092 operations on objects in ll_vm_open, ll_vm_close and
2098 Description: Request leak when working with deleted CWD
2099 Details : Introduce advanced request refcount tracking for requests
2100 referenced from lustre intent.
2102 Severity : Enhancement
2104 Description: Cache open file handles on client.
2105 Details : MDS now will return special lock along with openhandle, if
2106 requested and client is allowed to hold openhandle, even if unused,
2107 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2108 closing files for every read/write openration.
2110 Severity : Enhancement
2112 Description: Cache open negative dentries on client when possible.
2113 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2114 negative dentries on lock revocation.
2119 Description: Remounting a client read-only wasn't possible with a zconf mount
2120 Details : It wasn't possible to remount a client read-only with llmount.
2122 Severity : enhancement
2123 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2124 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2125 MPICH is included to add Lustre-specific ADIO interfaces.
2126 This is based closely on the UFS ADIO layer and only differs
2127 in file creation, in order to allow the OST striping to be set.
2128 This is user-contributed code and not supported by CFS.
2133 Description: extended inode attributes (immutable, append-only) work improperly
2134 when 2.4 and 2.6 kernels are used on client/server or vice versa
2135 Details : Introduce kernel-independent values for these flags.
2137 Severity : enhancement
2140 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2141 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2142 granularity. Now, Fractional MB granularity is be supported,
2143 this is very useful for low memory system.
2145 Severity : enhancement
2147 Description: Getattr by fid
2148 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2149 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2150 have locks and avoids getattr rpc requests for it on every stat.
2153 Frequency : Always, for filesystems larger than 2TB
2155 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2156 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2157 filesystems larger than 2TB to be mounted with mballoc enabled.
2160 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2162 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2163 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2164 filesystems larger than 2TB is mounted then the kernel will
2165 silently corrupt the start of the filesystem. CONFIG_LBD is
2166 enabled for all CFS-supported kernels, but the possibility of
2167 this happening with a modified kernel config exists.
2169 Severity : enhancement
2171 Description: add client O_DIRECT support for 2.6 kernels
2172 Details : It is now possible to do O_DIRECT reads and writes to files
2173 in the Lustre client mountpoint on 2.6 kernel clients.
2175 Severity : enhancement
2177 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2178 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2179 parallel, not waiting for response from every OST before sending
2180 a rpc to the next OST.
2185 Description: setattr vs write race when updating file timestamps
2186 Details : Client processes that update a file timestamp into the past
2187 right after writing to the file (e.g. tar) it is possible that
2188 the updated file modification time can be reset to the current
2189 time due to a race between processing the setattr and write RPC.
2191 Severity : enhancement
2193 Description: Bring 'lfs find' closer in line with regular Linux find.
2194 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2195 -print0 options and obtains all the needed info through the lustre
2198 Severity : enhancement
2200 Description: support up to 1024 configured devices on one node
2201 Details : change obd_dev array from statically allocated to dynamically
2202 allocated structs as they are first used to reduce memory usage
2207 Description: Flush dirty partially truncated pages during truncate
2208 Details : Immediatelly flush partially truncated pages in filter_setattr,
2209 this way we completely avoid having any pages in page cache on OST
2210 and can retire ugly workarounds during writes to flush such pages.
2215 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2216 Details : protect inode from truncation within vfs_unlink() context
2217 just take a reference before calling vfs_unlink() and release it
2218 when parent's i_sem is free.
2223 Description: last_id value checked outside lock on OST caused LASSERT failure
2224 Details : If there were multiple MDS->OST object precreate requests in
2225 flight, it was possible that the OST's last object id was checked
2226 outside a lock and incorrectly tripped an assertion. Move checks
2227 inside locks, and discard old precreate requests.
2230 Frequency : always, if extents are used on OSTs
2232 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2233 Details : OSTs using the extents format would not zero the high 16 bits of
2234 the index physical block number. This is not a problem for any
2235 OST filesystems smaller than 16TB, and no kernels support ext3
2236 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2237 new/modified files) and can be fixed for existing filesystems
2238 with e2fsprogs-1.39-cfs1.
2243 Description: import connection selection may be incorrect if timer wraps
2244 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2245 import connections to be ignored if the 32-bit jiffies counter
2246 wraps. Use a 64-bit jiffies counter.
2249 Frequency : very large clusters immediately after boot
2251 Description: LNET request buffers exhausted under heavy short-term load
2252 Details : If a large number of client requests are generated on a service
2253 that has previously never seen so many requests it is possible
2254 that the request buffer growth cannot keep up with the spike in
2255 demand. Instead of dropping incoming requests, they are held in
2256 the LND until the RPC service can accept more requests.
2259 Frequency : Sometimes during replay
2261 Description: Assertion failure in ll_local_open after replay.
2262 Details : If replay happened on an open request reply before we were able
2263 to set replay handler, reply will become not swabbed tripping the
2264 assertion in ll_local_open. Now we set the handler right after
2265 recognising of open request
2268 Frequency : very rare
2270 Description: kernel reports "badness in vsnprintf"
2271 Details : Reading from the "recovery_status" /proc file in small chunks
2272 may cause a negative length in lprocfs_obd_rd_recovery_status()
2273 call to vsnprintf() (which is otherwise harmless). Exit early
2274 if there is no more space in the output buffer.
2276 Severity : enhancement
2278 Description: clear OBD RPC statistics by writing to them
2279 Details : It is now possible to clear the OBD RPC statistics by writing
2280 to the "stats" file.
2285 Description: Client mtime is not the same on different clients after utimes
2286 Details : In some cases, the client was using the utimes() syscall on
2287 a file cached on another node. The clients now validate the
2288 ctime from the MDS + OSTs to determine which one is right.
2293 Description: Inability to activate failout mode
2294 Details : lconf script incorrectly assumed that in pythong string's numeric
2295 value is used in comparisons.
2298 Frequency : always with multiple stripes per file
2300 Description: Inefficient object allocation for mutli-stripe files
2301 Details : When selecting which OSTs to stripe files over, for files with
2302 a stripe count that divides evenly into the number of OSTs,
2303 the MDS is always picking the same starting OST for each file.
2304 Return the OST selection heuristic to the original design.
2309 Description: mount failures may take full timeout to return an error
2310 Details : Under some heavy load conditions it is possible that a
2311 failed mount can wait for the full obd_timeout interval,
2312 possibly several minutes, before reporting an error.
2313 Instead return an error as soon as the status is known.
2315 Frequency : quota enabled and large files being deleted
2317 Description: releasing more than 4GB of quota at once hangs OST
2318 Details : If a user deletes more than 4GB of files on a single OST it
2319 will cause the OST to spin in an infinite loop. Release
2320 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2325 Description: statfs data retrieved from /proc may be stale or zero
2326 Details : When reading per-device statfs data from /proc, in the
2327 {kbytes,files}_{total,free,avail} files, it may appear
2328 as zero or be out of date.
2331 Frequency : systems with MD RAID1 external journal devices
2333 Description: lconf's call to blkid is confused by RAID1 journal devices
2334 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2335 of returning all block devices that match the journal UUID.
2338 Frequency : always, for aggregate stripe size over 4GB
2340 Description: assertion fails when trying to use 4GB stripe size
2341 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2342 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2345 Frequency : always on ppc64
2347 Description: the first write on an ext3 filesystem with mballoc got stuck
2348 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2349 perform endianness conversion.
2351 ------------------------------------------------------------------------------
2353 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2355 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2356 INTEROPERATE with older versions automatically. Please read the
2357 user documentation before upgrading any part of a live system.
2358 * WARNING: Lustre networking configuration changes are required with
2359 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2362 * Support for kernels:
2363 2.6.9-22.0.2.EL (RHEL 4)
2364 2.6.5-7.244 (SLES 9)
2365 2.6.12.6 vanilla (kernel.org)
2368 Severity : enhancement
2369 Bugzilla : 7981/8208
2370 Description: Introduced Lustre Networking (LNET)
2371 Details : LNET is new networking infrastructure for Lustre, it includes
2372 a reorganized network configuration mode (see the user
2373 documentation for full details) as well as support for routing
2374 between different network fabrics. Lustre Networking Devices
2375 (LNDs) for the supported network fabrics have also been
2376 created for this new infrastructure.
2378 Severity : enhancement
2379 Description: Introduced Access control lists
2380 Details : clients can set ACLs on files and directories in order to have
2381 more fine-grained permissions than the standard Unix UGO+RWX.
2382 The MDS must be started with the "-o acl" mount option.
2384 Severity : enhancement
2385 Description: Introduced filesystem quotas
2386 Details : Administrators may now establish per-user quotas on the
2389 Severity : enhancement
2391 Description: Configuration change for the XT3
2392 The PTLLND is now used to run Lustre over Portals on the XT3
2393 The configure option(s) --with-cray-portals are no longer used.
2394 Rather --with-portals=<path-to-portals-includes> is used to
2395 enable building on the XT3. In addition to enable XT3 specific
2396 features the option --enable-cray-xt3 must be used.
2401 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2402 Details : A race between allocating a new llcd and re-getting the llcd_lock
2403 allowed another thread to grab newly-allocated llcd.
2405 Severity : enhancement
2407 Description: 2.6 OST async journal commit and locking fix to improve performance
2408 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2409 2.6 kernels are now async as they already were in 2.4 kernels so
2410 that they can commit concurrently with the network bulk transfer.
2411 For block-allocated files the filter allocation semaphore is held
2412 to avoid filesystem fragmentation during allocation. BKL lock
2413 removed for 2.6 xattr operations where it is no longer needed.
2418 Description: lconf incorrectly determined whether two IP networks could talk
2419 Details : In some more complicated routing and multiple-network
2420 configurations, lconf will avoid trying to make a network
2421 connection to a disjoint part of the IP space. It was doing the
2422 math incorrectly for one set of cases.
2427 Description: Fix for potential infinite loop processing records in an llog.
2428 Details : If an llog record is corrupted/zeroed, it is possible to loop
2429 forever in llog_process(). Validate the llog record length
2430 and skip the remainder of the block on error.
2433 Frequency : occasional (liblustre only)
2435 Description: liblustre could not open files whose last component is a symlink
2436 Details : sysio_path_walk() would incorrectly pass the open intent to
2437 intermediate path components.
2440 Frequency : rare (liblustre only with non-standard tuning)
2441 Bugzilla : 7201 (7350)
2442 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2443 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2444 after it has been released, only do this for referenced locks
2446 Severity : enhancement
2448 Description: specifying an (invalid) directory default stripe_size of -1
2449 would reset the directory default striping
2450 Details : stripe_size -1 was used internally to signal directory stripe
2451 removal, now use "all default" to signal dir stripe removal
2452 as a directory striping of "all default" is not useful
2455 Frequency : common for large clusters running liblustre clients
2457 Description: doing an ls when liblustre clients are running is slow
2458 Details : sending a glimpse AST to a liblustre client waits for every AST
2459 to time out, as liblustre clients will not respond. Since they
2460 cannot cache data we refresh the OST lock LVB from disk instead.
2462 Severity : enhancement
2464 Description: doing an ls at the same time as file IO can be slow
2465 Details : enqueue and other "small" requests can be blocked behind many
2466 large IO requests. Create a new OST IO portal for non-IO
2467 requests so they can be processed faster.
2470 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2472 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2473 created file with mode 000
2474 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2475 allow an ftruncate() as a non-root user to a file with mode 000.
2476 Lustre now respects this flag to disable mode checks when
2477 truncating a file owned by the user
2480 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2483 Description: Revoking locks from clients that went dead or catatonic might take
2485 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2486 cancellation of such locks instant on servers without waiting for
2487 any reply from clients. Clients drops these locks when cancel
2488 notification from server is received without replying.
2491 Frequency : liblustre-only, when liblustre client dies or becomes busy
2493 Description: Doing ls on Linux clients can take a long time with active
2495 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2496 granting such locks to it in the first place if possible. Locks
2497 are taken by proxy on the OST during the read or write and
2498 dropped immediately afterward. Add connect flags handling, do
2499 not grant locks to liblustre clients for glimpse ASTs.
2501 Severity : enhancement
2503 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2504 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2505 sequential reads and behaves badly with applications doing random
2506 reads. Improve it by reading ahead at least read region, and
2507 avoiding excessive large RPC for small reads.
2509 Severity : enhancement
2511 Description: Creating more than 1000 files for a single job may cause a load
2512 imbalance on the OSTs if there are also a large number of OSTs.
2513 Details : qos_prep_create() uses an OST index reseed value that is an
2514 even multiple of the number of available OSTs so that if the
2515 reseed happens in the middle of the object allocation it will
2516 still utilize the OSTs as uniformly as possible.
2521 Description: OST or MDS may oops in ping_evictor_main()
2522 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2523 but doesn't restart at beginning of obd_exports_timed list
2526 Severity : enhancement
2528 Description: improve by-nid export eviction on the MDS and OST
2529 Details : allow multiple exports with the same NID to be evicted at one
2530 time without re-searching the exports list.
2533 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2535 Description: MDS may oops in groups_free()
2536 Details : in rare race conditions a newly allocated group_info struct is
2537 freed again, and this can be NULL. The 2.4 compatibility code
2538 for groups_free() checked for a NULL pointer, but 2.6 did not.
2541 Frequency : common for liblustre clients doing little filesystem IO
2542 Bugzilla : 9352, 7313
2543 Description: server may evict liblustre clients accessing contended locks
2544 Details : if a client is granted a lock or receives a completion AST
2545 with a blocking AST already set it would not reply to the AST
2546 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2549 Frequency : lfs setstripe, only systems with more than 160 OSTs
2551 Description: unable to set striping with a starting offset beyond OST 160
2552 Details : llapi_create_file() incorrectly limited the starting stripe
2553 index to the maximum single-file stripe count.
2556 Frequency : LDAP users only
2558 Description: lconf did not handle in-kernel recovery with LDAP properly
2559 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2561 Severity : enhancement
2563 Description: bind OST threads to NUMA nodes to improve performance
2564 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2565 node and do their allocations there to localize memory access
2567 Severity : enhancement
2569 Description: llmount can determine client NID directly from Myrinet (GM)
2570 Details : the client NID code from gmnalnid was moved directly into
2571 llmount, removing the need to use this or specifying the
2572 client NID explicitly when mounting GM clients with zeroconf
2575 Frequency : if client is started with down MDS
2577 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2578 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2579 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2580 ptlrpc_import_delayed_req() because we didn't check if the
2581 request was interrupted, and we also didn't break out of the
2582 event loop if there was no timeout
2587 Description: data loss during non-page-aligned writes to a single file from
2588 both multiple nodes and multiple threads on one node at same time
2589 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2590 inconsistency led to false short-reads, that were cached and later
2591 used by ->prepare_write() to fill in partially written page,
2592 leading to data loss.
2595 Frequency : always, if lconf --abort_recovery used
2597 Description: lconf --abort_recovery fails with 'Operation not supported'
2598 Details : lconf was attempting to abort recovery on the MDT device and not
2601 Severity : enhancement
2603 Description: remove cleanup logs
2604 Details : replace lconf-generated cleanup logs with lustre internal
2605 cleanup routines. Eliminates the need for client-cleanup and
2608 Severity : enhancement
2610 Description: add support for EAs (user and system) on lustre filesystems
2611 Details : it is now possible to store extended attributes in the Lustre
2612 client filesystem, and with the user_xattr mount option it
2613 is possible to allow users to store EAs on their files also
2615 Severity : enhancement
2617 Description: Add possibility (config option) to show minimal available OST free
2619 Details : When compiled with --enable-mindf configure option, statfs(2)
2620 (and so, df) will return least minimal free space available from
2621 all OSTs as amount of free space on FS, instead of summary of
2622 free spaces of all OSTs.
2624 Severity : enhancement
2626 Description: do not expand extent locks acquired on OST-side
2627 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2628 by server: they are not cached anyway.
2631 Frequency : when mmap is used/binaries executed from Lustre
2633 Description: Unmmap pages before throwing them away from read cache.
2634 Details : llap_shrink cache now attempts to unmap pages before discarding
2635 them (if unmapping failed - do not discard). SLES9 kernel has
2636 extra checks that trigger if this unmapping is not done first.
2641 Description: lconf didn't resolve symlinks before checking to see whether a
2642 given mountpoint was already in use
2645 Frequency : when migrating failover services
2646 Bugzilla : 6395, 9514
2647 Description: When migrating a subset of services from a node (e.g. failback
2648 from a failover service node) the remaining services would
2649 time out and evict clients.
2650 Details : lconf --force (implied by --failover) sets the global obd_timeout
2651 to 5 seconds in order to quickly disconnect, but this caused
2652 other RPCs to time out too quickly. Do not change the global
2653 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2655 Severity : enhancement
2656 Frequency : if MDS is started with down OST
2657 Bugzilla : 9439,5706
2658 Description: Allow startup/shutdown of an MDS without depending on the
2659 availability of the OSTs.
2660 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2661 Add appropriate locking and lov-osc refcounts for safe
2662 cleaning. Add osc abort_inflight calls in case the
2663 synchronize never started.
2666 Frequency : occasional (Cray XT3 only)
2668 Description: root not authorized to access files in CRAY_PORTALS environment
2669 Details : The client process capabilities were not honoured on the MDS in
2670 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2671 been accessed by an authorized user then root was able to access
2672 the file on the local client also. The root user capabilities
2673 are now allowed on the MDS, as this environment has secure UID.
2676 Frequency : occasional
2678 Description: ldiskfs "too long searching" message happens too often
2679 Details : A debugging message (otherwise harmless) prints too often on
2680 the OST console. This has been reduced to only happen when
2681 there are fragmentation problems on the filesystem.
2686 Description: Division by zero in statfs when all OSCs are inactive
2687 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2688 lov_statfs divides by value returned by lov_get_stripecnt().
2692 Bugzilla : 9489, 3273
2693 Description: First write from each client to each OST was only 4kB in size,
2694 to initialize client writeback cache, which caused sub-optimal
2695 RPCs and poor layout on disk for the first writen file.
2696 Details : Clients now request an initial cache grant at (re)connect time
2697 and so that they can start streaming writes to the cache right
2698 away and always do full-sized RPCs if there is enough data.
2699 If the OST is rebooted the client also re-establishes its grant
2700 so that client cached writes will be honoured under the grant.
2705 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2706 Details : Now I/O RPCs go to different portal number and (presumably) fast
2707 lock requests (and glimses) and other RPCs get their own service
2708 threads pool that should be able to service those RPCs
2711 Severity : enhancement
2713 Description: Ability to exchange lustre version between client and servers and
2714 issue warnings at client side if client is too old. Also for
2715 liblustre clients there is ability to refuse connection of too old
2717 Details : New 'version' field is added to connect data structure that is
2718 filled with version info. That info is later checked by server and
2722 Frequency : rare, liblustre only.
2723 Bugzilla : 9296, 9581
2724 Description: Two simultaneous writes from liblustre at offset within same page
2725 might proceed at the same time overwriting eachother with stale
2727 Details : I/O lock withing llu_file_prwv was released too early, before data
2728 actually was hitting the wire. Extended lock-holding time until
2729 server acknowledges receiving data.
2732 Frequency : extremely rare. Never observed in practice.
2734 Description: avoid generating lustre_handle cookie of 0.
2735 Details : class_handle_hash() generates handle cookies by incrementing
2736 global counter, and can hit 0 occasionaly (this is unlikely, but
2737 not impossible, because initial value of cookie counter is
2738 selected randonly). Value of 0 is used as a sentinel meaning
2739 "unassigned handle" --- avoid it. Also coalesce two critical
2740 sections in this function into one.
2742 Severity : enhancement
2744 Description: allow liblustre clients to delegate truncate locking to OST
2745 Details : To avoid overhead of locking, liblustre client instructs OST to
2746 take extent lock in ost_punch() on client's behalf. New connection
2747 flag is added to handle backward compatibility.
2749 Severity : enhancement
2750 Bugzilla : 4928, 7341, 9758
2751 Description: allow number of OST service threads to be specified
2752 Details : a module parameter allows the number of OST service threads
2753 to be specified via "options ost ost_num_threads={N}" in the
2754 OSS's /etc/modules.conf or /etc/modprobe.conf.
2758 Bugzilla : 6146, 9635, 9895
2759 Description: servers crash with bad pointer in target_handle_connect()
2760 Details : In rare cases when a client is reconnecting it was possible that
2761 the connection request was the last reference for that export.
2762 We would temporarily drop the export reference and get a new
2763 one, but this may have been the last reference and the export
2764 was just destroyed. Get new reference before dropping old one.
2766 Severity : enhancement
2767 Frequency : if client is started with failover MDS
2769 Description: Allow multiple MDS hostnames in the mount command
2770 Details : Try to read the configuration from all specified MDS
2771 hostnames during a client mount in case the "primary"
2774 Severity : enhancement
2776 Description: Stop sending data to evicted clients as soon as possible.
2777 Details : Check if the client we are about to send or are sending data to
2778 was evicted already. (Check is done every second of waiting,
2779 for which l_wait_event interface was extended to allow checking
2780 of exit condition at specified intervals).
2783 Frequency : rare, normally only when NFS exporting is done from client
2785 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2787 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2790 Frequency : always, if srand() is called before liblustre initialization
2792 Description: Liblustre uses system PRNG disturbing its usage by user application
2793 Details : Introduce internal to lustre fast and high-quality PRNG for
2794 lustre usage and make liblustre and some other places in generic
2795 lustre code to use it.
2797 Severity : enhancement
2798 Bugzilla : 9477, 9557, 9870
2799 Description: Verify that the MDS configuration logs are updated when xml is
2800 Details : Check if the .xml configuration logs are newer than the config
2801 logs stored on the MDS and report an error if this is the case.
2802 Request --write-conf, or allow starting with --old_conf.
2804 Severity : enhancement
2806 Description: Handle symlinks in the path when checking if Lustre is mounted.
2807 Details : Resolve intermediate symlinks when checking if a client has
2808 mounted a filesystem to avoid duplicate client mounts.
2813 Description: lconf can hit an error exception but still return success.
2814 Details : The lconf command catches the Command error exception at the top
2815 level script context and will exit with the associated exit
2816 status, but doesn't ensure that this exit status is non-zero.
2821 Description: failure of ptlrpc thread startup can cause oops
2822 Details : Starting a ptlrpc service thread can fail if there are a large
2823 number of threads or the server memory is very fragmented.
2824 Handle this without oopsing.
2827 Frequency : always, only if liblustre and non-default acceptor port was used
2829 Description: liblustre cannot connect to servers with non-default acceptor port
2830 Details : tcpnal_set_default_params() was not called and was therefore
2831 ignoring the environment varaible TCPNAL_PORT, as well as other
2832 TCPNAL_ environment variables
2837 Description: two objects could be created on the same OST for a single file
2838 Details : If an OST is down, in some cases it was possible to create two
2839 objects on a single OST for a single file. No problems other
2840 than potential performance impact and spurious error messages.
2844 Bugzilla : 5681, 9562
2845 Description: Client may oops in ll_unhash_aliases
2846 Details : Client dcache may become inconsistent in race condition.
2847 In some cases "getcwd" can fail if the current directory is
2853 Description: Inode refcounting problems in NFS export code
2854 Details : link_raw functions used to call d_instantiate without obtaining
2855 extra inode reference first.
2859 Bugzilla : 9942, 9903
2860 Description: Referencing freed requests leading to crash, memleaks with NFS.
2861 Details : We used to require that call to ll_revalidate_it was always
2862 followed by ll_lookup_it. Also with revalidate_special() it is
2863 possible to call ll_revalidate_it() twice for the same dentry
2864 even if first occurence returned success. This fix changes semantic
2865 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2866 reference on a request referred from the intent.
2867 ll_intent_release() then releases such a request.
2870 Frequency : rare, normally benchmark loads only
2872 Description: unlinked inodes were kept in memory on the client
2873 Details : If a client is repeatedly creating and unlinking files it
2874 can accumulate a lot of stale inodes in the inode slab cache.
2875 If there is no other client load running this can cause the
2876 client node to run out of memory. Instead flush old inodes
2877 from client cache that have the same inode number as a new inode.
2880 Frequency : SLES9 2.6.5 kernel and long filenames only
2881 Bugzilla : 9969, 10379
2882 Description: utime reports stale NFS file handle
2883 Details : SLES9 uses out-of-dentry names in some cases, which confused
2884 the lustre dentry revalidation. Change it to always use the
2888 Frequency : rare, unless heavy write-truncate concurrency is continuous
2889 Bugzilla : 4180, 6984, 7171, 9963, 9331
2890 Description: OST becomes very slow and/or deadlocked during object unlink
2891 Details : filter_destroy() was holding onto the parent directory lock
2892 while truncating+unlinking objects. For very large objects this
2893 may block other threads for a long time and slow overall OST
2894 responsiveness. It may also be possible to get a lock ordering
2895 deadlock in this case, or run out of journal credits because of
2896 the combined truncate+unlink. Solution is to do object truncate
2897 first in one transaction without parent lock, and then do the
2898 final unlink in a new transaction with the parent lock. This
2899 reduces the lock hold time dramatically.
2902 Frequency : rare, 2.4 kernels only
2904 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2905 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2906 before unmounting their filesystem in order to avoid deadlock.
2907 The kernel_locked() function in 2.4 kernels only checks whether
2908 the kernel lock is held, not whether it is this process that is
2909 holding it as 2.6 kernels do.
2914 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2915 Details : The client ptlrpc code may be trying to reconnect to a down
2916 server before a previous connection attempt has timed out.
2917 Increase the reconnect interval to be longer than the connection
2918 timeout interval to avoid sending duplicate connections to
2922 Frequency : echo_client brw_test command
2924 Description: fix echo_client to work with OST preallocated code
2925 Details : OST preallocation code (5137) didn't take echo_client IO path
2926 into account: echo_client calls filter methods outside of any
2927 OST thread and, hence, there is no per-thread preallocated
2928 pages and buffers to use. Solution: hijack pga pages for IO. As
2929 a byproduct, this avoids unnecessary data copying.
2933 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2934 Description: Client can oops in mdc_commit_close() after open replay
2935 Details : It was possible for the MDS to return an open request with no
2936 transaction number in mds_finish_transno() if the client was
2937 evicted, but without actually returning an error. Clients
2938 would later try to replay that open and may trip an assertion
2939 Simplify the client close codepath, and always return an error
2940 from the MDS in case the open is not successful.
2943 Frequency : rare, 2.6 OSTs only
2945 Description: OST may deadlock under high load on fragmented files
2946 Details : If there was a heavy load and highly-fragmented OST filesystems
2947 it was possible to have all the OST threads deadlock waiting on
2948 allocation of biovecs, because the biovecs were not released
2949 until the entire RPC IO was completed. Instead, release biovecs
2950 as soon as they are complete to ensure forward IO progress.
2952 Severity : enhancement
2954 Description: Support for specifying external journal device at mount
2955 Details : If an OST or MDS device is formatted with an external journal
2956 device, this device major/minor is stored in the ext3 superblock
2957 and may not be valid for failover. Allow detecting and
2958 specifying the external journal at mount time.
2963 Description: Mounting an MDS with pending unlinked files may cause oops
2964 Details : target_finish_recovery() calls mds_postrecov() which returned
2965 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2966 considers this an error and immediately begins cleaning up the
2967 lov, just after starting the mds_lov process
2969 Severity : enhancement
2971 Description: Implement 'lfs df' to report actual free space on per-OST basis
2972 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2973 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2974 usage of objects; '-h' to report in human readable format.
2976 ------------------------------------------------------------------------------
2978 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2985 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2986 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2987 at startup during orphan unlinking. Instead, skip buddy update
2988 before setup, as it will be regenerated after recovery is complete.
2993 Description: If an OST is inactive, its locks might reference stale inodes.
2994 Details : lov_change_cbdata() must iterate over all namespaces, even if
2995 they are inactive to clear inode references from the lock.
2997 Severity : enhancement
2998 Frequency : occasional, if non-standard max_dirty_mb used
3000 Description: Client will block write RPCs if not enough grant
3001 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3002 then the client will block writes while waiting for another RPC
3003 to complete instead of consuming its dirty limit. With change
3004 we get improved performance when max_dirty_mb is small.
3006 Severity : enhancement
3007 Bugzilla : 3389, 6253
3008 Description: Add support for supplementary groups on the MDS.
3009 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3010 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3011 lookups for user supplementary groups into a cache.
3015 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3016 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3017 for liblustre O_EXCL is enforced by MDS.
3021 Description: readdir never returns NULL in liblustre.
3022 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3023 dentry in struct dirent.
3027 Frequency : liblustre only, depends on application IO pattern
3028 Description: liblustre clients evicted if not contacting servers
3029 Details : Don't put liblustre clients into the ping_evictor list, so
3030 they will not be evicted by the pinger ever.
3032 Severity : enhancement
3034 Description: Add ability to evict clients by NID from MDS.
3035 Details : By echoing "nid:$NID" string into
3036 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3037 $NID would be instantly evicted from this MDS and from all active
3038 OSTs connected to it.
3042 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3043 Details : lookup_it_finish() used to query file size from OSTs that was not
3048 Description: service threads change working directory to that of init
3049 Details : Starting lustre service threads may pin the working directory
3050 of the parent thread, making that filesystem busy. Threads
3051 now change to the working directory of init to avoid this.
3055 Frequency : during shutdown only
3056 Description: shutdown with a failed MDS or OST can cause unmount to hang
3057 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3062 Frequency : 2.6 only
3063 Description: chmod/chown may include an extra supplementary group
3064 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3065 supplementary group and if none is specified this is used.
3068 Bugzilla : 5479 (6816)
3070 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3071 Details : If lookup is in progress on a file that is unlinked we might try
3072 to revalidate the inode and fail in revalidate after lookup is
3073 complete and ll_file_open() enqueues the open again but
3074 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3077 Frequency : always, if lconf --abort_recovery used
3079 Description: lconf --abort_recovery fails with 'Operation not supported'
3080 Details : lconf was attempting to abort recovery on the MDT device and not
3083 ------------------------------------------------------------------------------
3085 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3090 Frequency : rare (only unsupported configurations with a node running as an
3092 Bugzilla : 6514, 5137
3093 Description: Mounting a Lustre file system on a node running as an OST could
3095 Details : OSTs now preallocates memory needed to write out data at
3096 startup, instead of when needed, to avoid having to
3097 allocate memory in possibly low memory situations.
3098 Specifically, if the file system is mounted on on OST,
3099 memory pressure could force it to try to write out data,
3100 which it needed to allocate memory to do. Due to the low
3101 memory, it would be unable to do so and the node would
3102 become unresponsive.
3104 Severity : enhancement
3106 Description: Addition of lconf --service command line option
3107 Details : lconf now accepts a '--service <arg>' option, which is
3108 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3110 Severity : enhancement
3112 Description: Failover mode is now the default for OSTs.
3113 Details : By default, OSTs will now run in failover mode. To return to
3114 the old behaviour, add '--failout' to the lmc line for OSTs.
3116 Severity : enhancement
3118 Description: Health checks are now provided for MDS and OSTs
3119 Details : Additional detailed health check information on MSD and OSTs
3120 is now provided through the procfs health_check value.
3123 Frequency : occasional, depends on IO load
3125 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3126 after numerous create/delete cycles
3127 Details : The ext3 inode allocation policy would not allocate new inodes
3128 very well on the OSTs because there are no new directories
3129 being created. Instead we look for groups with free space if
3130 the parent directories are nearly full.
3135 Description: Network or server problems during mount may cause partially
3136 mounted clients instead of returning an error.
3137 Details : The config llog parsing code may overwrite the error return
3138 code during mount error handling, returning success instead
3144 Description: MDS can fail to allocate large reply buffers
3145 Details : After long uptimes the MDS can fail to allocate large reply
3146 buffers (e.g. zconf client mount config records) due to memory
3147 fragmentation or consumption by the buffer cache. Preallocate
3148 some large reply buffers so that these replies can be sent even
3149 under memory pressure.
3153 Frequency : rare (liblustre)
3154 Description: fsx running with liblustre complained that using truncate() to
3155 extend the file doesn't work. This patch corrects that issue.
3156 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3157 ATTR_SIZE and lsm use in llu_setattr_raw.
3161 Frequency : rare, only 2.6 kernels
3162 Description: Unusual file access patterns on the MDS may result in inode
3163 data being lost in very rare circumstances.
3164 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3165 code in the 2.6 kernel caused the inode and/or EA data not to
3166 be read from disk, causing single-file corruption.
3170 Frequency : rare, only 2.6 filesystems using extents
3171 Description: Heavy concurrent write and delete load may cause data corruption.
3172 Details : It was possible under high-load situations to have an extent
3173 metadata block in the block device cache from a just-unlinked
3174 file overwrite a newly-allocated data block. We now unmap any
3175 metadata buffers that alias just-allocated data blocks.
3179 Frequency : filesystems with default stripe_count larger than 77
3180 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3181 Details : lconf specifies an inode size of 4096 bytes when the default
3182 stripe_count is larger than 77. This conflicts with the default
3183 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3184 this case to avoid pinning too much memory for large EAs.
3186 ------------------------------------------------------------------------------
3188 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3193 Frequency : rare (extremely heavy IO load with hundreds of clients)
3195 Description: Client is evicted, gets IO error writing to file
3196 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3197 caused clients to be evicted for AST timeouts. The fixes in
3198 bug 5192 mean we no longer need to have such short AST timeouts
3199 so ldlm_timeout has been increased.
3202 Frequency : occasional during --force or --failover shutdown under load
3203 Bugzilla : 5949, 4834
3204 Description: Server oops/LBUG if stopped with --force or --failover under load
3205 Details : a collection of import/export refcount and cleanup ordering
3206 issues fixed for safer force cleanup
3209 Frequency : only filesystems larger than 120 OSTs
3210 Bugzilla : 5990, 6223
3211 Description: lfs getstripe would oops on a very large filesystem
3212 Details : lov_getconfig used kfree on vmalloc'd memory
3215 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3216 Bugzilla : 6242, 6243
3217 Description: reading from files that had been truncated to a non-zero size
3218 but never opened returned no data
3219 Details : ll_file_read() reads zeros from no-object files to EOF
3224 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3225 Details : There's little harm in aborting MDS/OSS recovery and letting it
3226 try again, so I removed the LASSERT and return an error instead.
3228 Severity : enhancement
3230 Description: New debugging infrastructure for tracking down data corruption
3231 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3232 (b) cover more of the client-side code path, and (c) try to narrow
3233 down where problems occurred
3237 Bugzilla : 3819, 4364, 4397, 6313
3238 Description: Racing close and eviction MDS could cause assertion in mds_close
3239 Details : It was possible to get multiple mfd references during close and
3240 client eviction, leading to one thread referencing a freed mfd.
3242 Severity: : enhancement
3243 Bugzilla : 3262, 6359
3244 Description: Attempts to reconnect to servers are now more aggressive.
3245 Details : This builds on the enhanced upcall-less recovery that was added
3246 in 1.4.2. When trying to reconnect to servers, clients will
3247 now try each server in the failover group every 10 seconds. By
3248 default, clients would previously try one server every 25 seconds.
3253 Description: After recovery, certain operations trigger a failed
3254 assertion on a client.
3255 Details : Failing over an mds, using lconf -d --failover, while a
3256 client was doing a readdir() call would cause the client to
3257 LBUG after recovery completed and the readdir() was resent.
3259 Severity : enhancement
3261 Description: Default groups are now added by lconf
3262 Details : You can now run lconf --group <servicename> without having to
3263 manually add groups with lmc.
3266 Frequency : occasional
3268 Description: Nodes with an elan id of 0 trigger a failed assertion
3271 Frequency : always when accessing e.g. tty/console device nodes
3273 Description: tty and some other devices nodes cannot be used on lustre
3274 Details : file's private_data field is used by device data and lustre
3275 values in there got lost. New field was added to struct file to
3276 store fs-specific private data.
3279 Frequency : when exporting Lustre via NFS
3281 Description: NFSD failed occasionally when looking up a path component
3282 Details : NFSD is looking up ".." which was broken in ext3 directories
3283 that had grown large enough to become hashed.
3286 Frequency : Clusters with multiple interfaces not on the same subnet
3288 Description: Nodes will repeatedly try to reconnect to an interface which it
3289 cannot reach and report an error to the log.
3290 Details : Extra peer list entries will be created by lconf with some peers
3291 unreachable. lconf now validates the peer before adding it.
3294 Frequency : Only if a default stripe is set on the filesystem root.
3296 Description: Setting a default stripe on the filesystem root prevented the
3297 filesystem from being remounted.
3298 Details : The client was sending extra request flags in the root getattr
3299 request and did not allocate a reply buffer for the dir EA.
3302 Frequency : occasional, higher if lots of files are accessed by one client
3303 Bugzilla : 6159, 6097
3304 Description: Client trips assertion regarding lsm mismatch/magic
3305 Details : While revalidating inodes the VFS looks up inodes with ifind()
3306 and in rare cases can find an inode that is being freed.
3307 The ll_test_inode() code will free the lsm during ifind()
3308 when it finds an existing inode and then the VFS later attaches
3309 this free lsm to a new inode.
3313 Bugzilla : 6422, 7030
3314 Description: MDS deadlock between mkdir and client eviction
3315 Details : Creating a new file via mkdir or mknod (starting a transaction
3316 and getting the ns lock) can deadlock with client eviction
3317 (gets ns lock and trying to finish a synchronous transaction).
3320 Frequency : occasional
3321 Description: While starting a server, the fsfilt_ext3 module could not be
3323 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3324 kernels. Previously, lconf would still use the ext3 name
3325 when trying to load modules. Now, it will correctly use
3326 ext3 on 2.4 and ldiskfs on 2.6.
3328 Severity : enhancement
3329 Description: The default stripe count has been changed to 1
3330 Details : The interpretation of the default stripe count (0, to lfs
3331 or lmc) has been changed to mean striping across a single
3332 OST, rather than all available. For general usage we have
3333 found a stripe count of 1 or 2 works best.
3335 Severity : enhancement
3336 Description: Add support for compiling against Cray portals.
3337 Details : Conditional compiling for some areas that are different
3341 Frequency : occasional
3342 Bugzilla : 6409, 6834
3343 Description: Creating files with an explicit stripe count may lead to
3344 a failed assertion on the MDS
3345 Details : If some OSTs are full or unavailable, creating files may
3346 trigger a failed assertion on the MDS. Now, Lustre will
3347 try to use other servers or return an error to the
3351 Frequency : occasional
3353 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3354 caused lock timeout during readahead (since 1.4.2).
3355 Details : Processes doing readahead might match a lock that hasn't been
3356 granted yet if there are overlapping and conflicting lock
3357 requests. The readahead process waits on ungranted lock
3358 (original lock is CBPENDING), while OST waits for that process
3359 to cancel CBPENDING read lock and eventually evicts client.
3361 Severity : enhancement
3363 Description: Initial enabling of flock support for clients
3364 Details : Implements fcntl advisory locking and file status functions.
3365 This feature is provided as an optional mount flag (default
3366 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3367 locking are implemented yet, and those that are are not guaranteed
3368 to be completely correct in production environments.
3369 mount -t lustre -o [flock|noflock] ...
3372 Frequency : occasional
3374 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3375 assertion in the ext3 JBD (journaling) layer.
3376 Details : The b_committed_data struct is protected by the big kernel lock
3377 in 2.4 kernels, serializing journal_commit_transaction() and
3378 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3379 access to this struct. In 2.6 kernels there is finer grained
3380 locking to improve SMP performance of the JBD layer.
3384 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3389 Description: merge and cleanup kernel patches.
3392 -----------------------------------------------------------------------------
3394 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3396 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3397 versions of Lustre. Please update all servers and clients to
3398 version 1.4.2 or later at the same time. You must also run
3399 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3400 to update the configuration logs.
3402 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3403 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3404 - swab llog records of type '0' so we get proper header size/idx (5861)
3405 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3406 - fix rename of one directory over another leaking an inode (5953)
3407 - avoid SetPageDirty on 2.6 (5981)
3408 - don't re-add just-being-destroyed locks to the waiting list (5653)
3409 - when creating new directories, inherit the parent's custom
3410 striping settings if present parent (3048)
3411 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3412 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3413 - don't hold client locks on temporary worklist from l_lru (5666)
3414 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3415 - automatically evict dead clients (5921)
3416 - Update file size properly in create+truncate+fstat case (6196)
3417 - Do not unhash mountpoint dentries, do not allow removal of
3419 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3420 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3421 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3422 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3423 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3424 - Correctly handle reads of files with no objects (6243)
3425 - lctl recover will also mark a device active if deactivate used (5933)
3427 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3428 - allow --write-conf on an MDS with different nettype than client (5619)
3429 - don't write config llogs to MDS for mounts not from that MDS (5617)
3430 - lconf should create multiple TCP connections from a client (5201)
3431 - init scripts are now turned off by default; run chkconfig --on
3432 lustre and chkconfig --on lustrefs to use them
3433 - upcalls are no longer needed for clients to recover to failover
3435 - add --abort-recovery option to lconf to abort recovery on device
3437 - add support for an arbitrary number of OSTs (3026)
3438 - Quota support protocol changes.
3439 - forward compatibility changes to wire structs (6007)
3440 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3441 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3442 - improved "lustre status" script
3443 - initialize blocksize for non-regular files (6062)
3444 - added --disable-server and --disable-client configure options (5782)
3445 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3446 - Vanilla 2.4.29 support
3447 - increase maximum number of obd devices to 520 (6242)
3448 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3449 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3450 SLES 9 2.6.5 kernels
3451 - limit stripes per file to 160 (the maximum EA size) (6093)
3453 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3456 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3457 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3458 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3459 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3460 - small ext3 extents cleanups and fixes (5733)
3461 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3462 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3463 - avoid needless client->OST connect, fix handle mismatch (5317)
3464 - fix DLM error path that led to out-of-sync client, long delays (5779)
3465 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3466 - fix several locking issues related to i_size (5492,5624,5654,5672)
3467 - don't move pending lock onto export if it is already evicted (5683)
3468 - fix kernel oops when creating .foo in unlinked directory (5548)
3469 - fix deadlock in obdfilter statistics vs. object create (5811)
3470 - use time_{before,after} to avoid timer jiffies wrap (5882)
3471 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3472 - Do not leak request if server was not able to process it (5154)
3473 - If mds_open unable to find parent dir, make that negative lookup(5154)
3474 - don't create new directories with extent-mapping (5909, 5936)
3476 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3477 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3478 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3479 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3480 - support for RHEL4 kernel on i686 (5773)
3481 - provide error messages when incompatible logs are encountered (5898)
3483 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3484 * version 1.4.0.10 (1.4.1 release candidate 1)
3486 - don't keep a lock reference when lock is not granted (4238)
3487 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3488 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3489 - drop import inflight refcount on signal_completed_replay error (5255)
3490 - unlock page after async write error during send (3677)
3491 - handle missing objects in filter_preprw_read properly (5265)
3492 - no transno return for symlink open, don't save no-trasno open (3440)
3493 - don't try to complete elan receive that already failed (4012)
3494 - free RPC server reply state on error (5406)
3495 - clean up thread from ptlrpc_start_thread() on error (5160)
3496 - readahead could read extra page into cache that wasn't ejected (5388)
3497 - prevent races in class_attach/setup/cleanup/detach (5260)
3498 - don't dereference de->d_inode after l_dput of de (5458)
3499 - use "int" for stripe value returned from lock_to_stripe (5544)
3500 - mballoc allocation and error-checking fixes in 2.6 (5504)
3501 - block device patches to fix I/O request sizes in 2.6 (5482)
3502 - look up hostnames for IB nals (5602)
3503 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3504 - don't start multiple acceptors for the same port (5277)
3505 - fix incorrect LASSERT in mds_getattr_name (5635)
3506 - export a proc file for general "ping" checking (5628)
3507 - fix "lfs check" to not block when the MDS is down (5628)
3509 - service request history (4965)
3510 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3511 - create an "evict_client" /proc entry on OSTs, like the MDS has
3512 - fix mount usage message, return errors per mount(8) (5168)
3513 - change grep [] to grep "[]" in tests so they work in more UMLs
3514 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3515 - remove ancient LOV_MAGIC_V0 EA support (5047)
3516 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3517 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3518 - allow readahead allocations to fail when low on memory (5383)
3519 - mmap locking landed again, after considerable improvement (2828)
3520 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3522 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3525 - send OST transaction number in read/write reply to free req (4966)
3526 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3527 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3528 - the watchdog thread now runs as interruptible (5246)
3529 - flock/lockf fixes (but it's still disabled, pending 5135)
3530 - don't use EXT3 constants in llite code (5094)
3531 - memory shortage at startup could cause assertion (5176)
3533 - reorganization of lov code
3534 - single portals codebase
3536 - add extents/mballoc support (5025)
3537 - direct I/O reads in the obdfilter (4048)
3538 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3540 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3543 - send OST transaction number in read/write reply to free req (4966)
3544 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3545 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3546 - the watchdog thread now runs as interruptible (5246)
3547 - handle missing objects in filter_preprw_read properly (5265)
3548 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3549 - drop import inflight refcount on signal_completed_replay error (5255)
3550 - unlock page after async write error during send (3677)
3551 - return original error code on reconstructed replies (3761)
3552 - no transno return for symlink open, don't save no-trasno open (3440)
3554 - add pid to ldlm debugging output (4922)
3555 - bump the watchdog timeouts -- we can't handle 30sec yet
3556 - extra debugging for orphan dentry/inode bug (5259)
3558 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3561 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3562 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3563 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3564 - fix NULL dereference in /proc/sys/portals/routes (4827)
3565 - allow failed mdc_close() operations to be interrupted (4561)
3566 - stop precreate on OST before MDS would time out on it (4778)
3567 - don't send partial-page writes before EOF from client (4410)
3568 - discard client grant for sub-page writes on large-page clients (4520)
3569 - don't free dentries not owned by NFS code, check generation (4806)
3570 - fix lsm leak if mds_create_objects() fails (4801)
3571 - limit debug_daemon file size, always print CERROR messages (4789)
3572 - use transno after validating reply (3892)
3573 - process timed out requests if import state changes (3754)
3574 - update mtime on OST during writes, return in glimpse (4829)
3575 - add mkfsoptions to LDAP (4679)
3576 - use ->max_readahead method instead of zapping global ra (5039)
3577 - don't interrupt __l_wait_event() during strace
3579 - add software watchdogs to catch hung threads quickly (4941)
3580 - make lustrefs init script start after nfs is mounted
3581 - fix CWARN/ERROR duplication (4930)
3582 - return async write errors to application if possible (2248)
3583 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3584 - print NAL number in %x format (4645)
3585 - update barely-supported suse-2.4.21-171 series (4842)
3586 - support for sles 9 %post scripts
3587 - support for building 2.6 kernel-source packages
3588 - support for sles km_* packages
3590 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3593 - ignore -ENOENT errors in osc_destroy (3639)
3594 - notify osc create thread that OSC is being cleaned up (4600)
3595 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3596 - reconstruct ost_handle() like mds_handle() (4657)
3597 - create a new thread to do import eviction to avoid deadlock (3969)
3598 - let lconf resolve symlinked-to devices (4629)
3599 - don't unlink "objects" from directory with default EA (4554)
3600 - hold socknal file ref over connect in case target is down (4394)
3601 - allow more than 32000 subdirectories in a single directory (3244)
3602 - fix blocks count for O_DIRECT writes (3751)
3603 - OST returns ENOSPC from object create when no space left (4539)
3604 - don't send truncate RPC if file size isn't changing (4410)
3605 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3606 - bind to privileged port in socknal and tcpnal (3689)
3608 - rate limit CERROR/CWARN console message to avoid overload (4519)
3609 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3610 - basic mmap support (3918)
3611 - kernel patch series update from b1_4 (4711)
3613 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3616 - avoid crash during MDS cleanup with OST shut down (2775)
3617 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3618 - don't use bad inodes on the MDS (3744)
3619 - dynamic object preallocation to improve recovery speed (4236)
3620 - don't hold spinlock over lock dumping or change debug flags (4401)
3621 - don't zero obd_dev when it is force cleaned (3651)
3622 - print grants to console if they go negative (4431)
3623 - "lctl deactivate" will stop automatic recovery attempts (3406)
3624 - look for existing locks in ldlm_handle_enqueue() (3764)
3625 - don't resolve lock handle twice in recovery avoiding race (4401)
3626 - revalidate should check working dir is a directory (4134)
3628 - don't always mark "slow" obdfilter messages as errors (4418)
3630 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3633 - don't close LustreDB during write_conf until it is done (3860)
3634 - fix typo in lconf for_each_profile (3821)
3635 - allow dumping logs from multiple threads at one time (3820)
3636 - don't allow multiple threads in OSC recovery (3812)
3637 - fix debug_size parameters (3864)
3638 - fix mds_postrecov to initialize import for llog ctxt (3121)
3639 - replace config semaphore with spinlock (3306)
3640 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3641 - don't allow enqueue to complete on a destroyed export (3822)
3642 - down write_lock before checking llog header bitmap (3825)
3643 - recover from lock replay timeout (3764)
3644 - up llog sem before sending rpc (3652)
3645 - reduce ns lock hold times when setting kms (3267)
3646 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3647 - fix NULL deref and obd_dev leak on setup error (3312)
3648 - replace some LBUG about llog ops with error handling (3841)
3649 - don't match INVALID dentries from d_lookup and spin (3784)
3650 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3651 - fix invalid assertion in ptlrpc_set_wait (3880)
3653 - add libwrap support for the TCP acceptor (3996)
3654 - add /proc/sys/portals/routes for non-root route listing (3994)
3655 - allow setting MDS UUID in .xml (2580)
3656 - print the stack of a process that LBUGs (4228)
3658 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3661 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3662 - make sure to unset replay flag from failed open requests (3440)
3663 - if default stripe count is 0, use OST count for inode size (3636)
3664 - update parent mtime/ctime on client for create/unlink (2611)
3665 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3666 - free recovery state on server during a forced cleanup (3571)
3667 - unregister_reply for resent reqs (3063)
3668 - loop back devices mounting and status check on 2.6 (3563)
3669 - fix resource-creation race that can provoke i_size == 0 (3513)
3670 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3671 - more debugging for page-accounting assertion (3746)
3672 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3673 - avoid deadlock after precreation failure (3758)
3674 - fix race and lock order deadlock in orphan handling (3450, 3750)
3675 - add validity checks when grabbing inodes from l_ast_data (3599)
3677 - add /proc/.../recovery_status to obdfilter (3428)
3678 - lightweight CDEBUG infrastructure, debug daemon (3668)
3679 - change default OSC RPC parameters to be better on small clusters
3680 - turn off OST read cache for files smaller than 32MB
3681 - install man pages and include them in rpms (3100)
3682 - add new init script for (un)mounting lustre filesystems (2593)
3683 - run chkconfig in %post for init scripts (3701)
3684 - drop scimac NAL (unmaintained)
3686 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3689 - clean kiobufs before and after use (3485)
3690 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3691 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3692 - change init script to fail more clearly if not run as root (1528)
3693 - allow clients to reconnect during replay (1742)
3694 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3695 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3696 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3697 - fix cond_resched (3554)
3698 - only evict unfinished clients after recovery (3515)
3699 - allow bulk resend, prevent data loss (3570)
3700 - dynamic ptlrpc request buffer allocation (2102)
3701 - don't allow unlinking open directory if it isn't empty (2904)
3702 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3703 - remove extraneous obd dereference causing LASSERT failure (3334)
3704 - don't use get_cycles() when creating temp. files on the mds (3156)
3705 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3706 - handle EEXIST for set-stripe, set proper directory name (3336)
3708 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3709 - fix ksocknal_fmb_callback() error messages (2918)
3711 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3714 - don't copy lvb into (possibly NULL) reply on error (2983)
3715 - don't deref dentry after dput, don't free lvb on error (2922)
3716 - use the kms to determine writeback rpc length (2947)
3717 - increment oti_logcookies when osc is inactive (2948)
3718 - update client's i_blocks count via lvb messages (2543)
3719 - handle intent open/close of special files properly (1557)
3720 - mount MDS with errors=remount-ro, like obdfilter (2009)
3721 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3722 - don't use cancelling-locks' kms values (2947)
3723 - use highest lock extent for kms, not last one (2925)
3724 - don't dereference ERR_PTR() dentry in error handling path (3107)
3725 - fix thread race in portals_debug_dumplog() (3122)
3726 - create lprocfs device entries at setup instead of at attach (1519)
3727 - common AST error handler, don't evict client on completion race (3145)
3728 - zero nameidata in detach_mnt in 2.6 (3118)
3729 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3730 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3731 - initialize RPC timeout timer earlier for 2.6 (3219)
3732 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3733 - print nal/nid for unknown nid (3258)
3734 - additional checks for oscc recovery before doing precreate (3284)
3735 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3736 - don't crash in mdc_close for bad permissions on open (3285)
3737 - zero i_rdev for non-device files (3147)
3738 - clear page->private before handing to FS, better assertion (3119)
3739 - tune the read pipeline (3236)
3740 - fix incorrect decref of invalidated dentry (2350)
3741 - provide read-ahead stats and refine rpc in flight stats (3328)
3742 - don't hold journal transaction open across create RPC (3313)
3743 - update atime on MDS at close time (3265)
3744 - close LDAP connection when recovering to avoid server load (3315)
3745 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3746 - don't leak open file on MDS after open resend (3325)
3747 - serialize filter_precreate and filter_destroy_precreated (3329)
3748 - loop device shouldn't call sync_dev() for nul device (3092)
3749 - clear page cache after eviction (2766)
3750 - resynchronize MDS->OST in background (2824)
3751 - refuse to mount the same filesystem twice on same mountpoint (3394)
3752 - allow llmount to create routes for mounting behind routers (3320)
3753 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3754 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3755 - fix rare problem with rename on htree directories (3417)
3757 - allow default OST striping configuration per directory (1414)
3758 - fix compilation for qswnal for 2.6 kernels (3125)
3759 - increase maximum number of MDS request buffers for large systems
3760 - change liblustreapi to be useful for external progs like lfsck (3098)
3761 - increase local configuration timeout for slow disks (3353)
3762 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3764 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3767 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3768 - don't overwrite extent policy data in reply if lock was blocked (2901)
3769 - drop filter export grants atomically with removal from device (2663)
3770 - del obd_self_export from work_list in class_disconnect_exports (2908)
3771 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3772 - swab reply message in mdc_close, other PPC fixes (2464)
3773 - fix destroying of named logs (2325)
3774 - overwrite old logs when running lconf --write_conf (2264)
3775 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3776 - fix race in target_handle_connect (2898)
3777 - mds_reint_create() should take same inode create lock (2926)
3778 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3779 - don't close files for self_export to avoid uninitialized obd (2936)
3780 - allow MDS with the same name as client node (2939)
3781 - hold dentry reference for closed log files for unlink (2325)
3782 - reserve space for all logs during transactions (2059)
3783 - don't evict page beyond end of stripe extent (2925)
3784 - don't oops on a deleted current working directory (2399)
3785 - handle hard links to targets without a parent properly (2517)
3786 - don't dereference NULL lock when racing during eviction (2867)
3787 - don't grow lock extents when lots of conflicting locks (2919)
3789 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3792 - account for cache space usage on clients to avoid data loss (974)
3793 - lfsck support in lustre kernel code (2349)
3794 - reduce journal credits needed for BRW writes (2370)
3795 - orphan handling to avoid losing space on client/server crashes
3796 - ptlrpcd can be blocked, stopping ALL progress (2477)
3797 - use lock value blocks to assist in proper KMS, faster stat (1021)
3798 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3799 - recovery for initial connections (2355)
3800 - fixes for mds_cleanup_orphans (1934)
3801 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3802 - block all file creations until orphan recovery completes (1901)
3803 - client remove rq_connection from request struct (2423)
3804 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3805 - recovery timer race (2670)
3806 - mdc_close recovey bug (2532)
3807 - ptlrpc cleanup bug (2710)
3808 - mds timeout on local locks (2588)
3809 - namespace lock held during RPCs (2431)
3810 - handle interrupted sync write properly (2503)
3811 - don't try to handle a message that hasn't been replied to (2699)
3812 - client assert failure during cleanup after abort recovery (2701)
3813 - leak mdc device after failed mount (2712)
3814 - ptlrpc_check_set allows timedout requests to complete (2714)
3815 - wait for inflight reqs when ptlrpcd finishes (2710)
3816 - make sure unregistered services are removed from the srv_list
3817 - reset bulk XID's when resending them (caught by 1138 test)
3818 - unregister_bulk after timeout
3819 - fix lconf error (2694)
3820 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3821 - readahead locks pages, leaves pending causing memory pressure (2673)
3822 - increase OST request buffers to 4096 on large machines (2729)
3823 - fix up permission of existing directories in simple_mkdir (2661)
3824 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3825 - don't assign transno to errored transactions (2742)
3826 - don't delete objects on OST if given a bogus objid from MDS (2751)
3827 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3828 - if rq_no_resend, then timeout request after recovery (2432)
3829 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3830 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3831 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3832 - MDS recovery shouldn't delete objects that it has given out (2730)
3833 - if enqueue arrives after completion, don't clobber LVB (2819)
3834 - don't unlock pages twice when trigger_group_io returns error (2814)
3835 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3836 - don't write pages to disk if there was an error (1450)
3837 - don't ping imports that have recovery disabled (2676)
3838 - take buffered bytes into account when balancing socknal conn (2817)
3839 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3840 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3841 - remove little-endian swabbing of llog records (1987)
3842 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3843 - retry reposting request buffers if they fail (1191)
3844 - grow extent at grant time to avoid granting a revoked lock (2809)
3845 - lock revoke doesn't evict page if covered by a second lock (2765)
3846 - disable VM readahead to avoid reading outside lock extents (2805)
3848 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3849 - updated kernel patches for hp-2.4.20 kernel (2681)
3851 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3854 - fix truncated write corruption (2366)
3855 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3857 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3858 - ensure len(uuid) < 37 in lmc (1171)
3859 - fix ia64 OOPS in llog_test (2255)
3860 - zero end of page at obdfilter for partial page writes (2648)
3861 - don't leave stale dentries around after renames (bug 2428)
3862 - fix timeouts when evicting a client with a single lock held (2642)
3863 - set deadline for the initial HELLO message to drain (2634)
3864 - print out dotted-quad IP addresses in the socknal (2302)
3866 - additional debugging for MDS client eviction problem (2443)
3867 - fix mkfsoptions support for osts (2603, 2604)
3869 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3872 - add series for the vanilla 2.6.0 kernel
3873 - add series for the vanilla 2.4.24 kernel
3874 - add series for a cray x86/64 UL kernel drop
3875 - fix xattr patches for the vanilla 2.4.19 series
3877 - generate true UUIDs in lmc (1171)
3878 - have portals stack dumping break in UML (2466)
3879 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3880 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3881 - wake callback waiting threads on client eviction (2460)
3882 - Add --ptldebug and --subsystem to lmc (1719)
3883 - update assertion to allow safe interrupt allocation
3884 - set rq_no_resend for cancel requests (2432)
3885 - recalculate ptlrpcd timeout after resend (2494)
3886 - call vfs_rmdir when removing pending directories (2368)
3887 - fix renaming a file to itself (2429)
3888 - lmc creates a default one-stripe lov (2454)
3889 - expand procfs space to handle large clusters (2326)
3890 - increase UML stack to avoid overflow
3891 - update lconf's list of debug and subsystem masks
3892 - fix lfs find --obd (2510)
3893 - /proc tunable for disabling filter read caching (2591)
3894 - stop rpm packages from altering slapd.conf (2301)
3895 - disable nagle in the socknal under 0conf (2578)
3896 - choose mds inode size based on stripe count (2572)
3897 - fix kernel-source rpm problems (2516)
3899 - add --disable-doc to avoid pdf generation (2421)
3900 - update documentation, tests, type-os, comments
3901 - avoid format warnings on ia64
3902 - remove the TOE NAL
3903 - tiny code cleanups by removing unused fields
3905 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3908 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3909 - remove the most blatant lies from BUILDING file (2371)
3910 - change default debug level to reasonable production setting
3911 - reduce client side cache size to reduce cache flush time
3912 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3913 - make TCP zerocopy and pinger support enabled by default (2476)
3914 - sync writes completed after process exits caused crashes (2319)
3915 - maintain correct mount count on the MDS (2356)
3916 - backout 1557, because 2316 wasn't really fixed
3917 - better file I/O statistics gathering in /proc
3918 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3919 - another kernel patch to fix zero-copy TCP function export
3920 - don't take duplicate lock when processing re-sent getattr (2420)
3921 - lctl uses obd_self_export instead of creating new conn (2353)
3922 - MDS/OST recovery case which requires object creation asserted (2425)
3923 - move lfs from /usr/sbin to /usr/bin in packages
3924 - fix race between mds_client_add and mds_client_free (2417)
3925 - use kmalloc instead of slabs in portals (2430)
3926 - don't create duplicate records when a failover MDS is present (2442)
3927 - remove unnecessary mount age check (2332)
3928 - don't remove directory inodes from locks prematurely (2451)
3929 - don't break if MDS service name is the same as hostname (2103)
3930 - fix races in client write RPC generation when cache full (2482)
3932 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3935 - remove now-unused request->rq_obd (278)
3936 - if an allocation fails, print out how much memory we've used (1933)
3937 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3938 - add the "configurable stack size" patch to most series files (1256)
3939 - ability to write large log records, for 100+ OST configs (2306)
3940 - fix NULL deref when filter_prep fails (2314)
3941 - fix operator precedence error in filter_sync
3942 - dynamic allocation of socknal TX descriptors (2315)
3943 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3944 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3945 - ensure that CWARN messages in Portals always get to the syslog
3946 - __init/__exit are not for prototype decls (ldlm_init/exit)
3947 - x86-64 compile warning fixes
3948 - fix gateway LMC keyword conflict (2318)
3949 - fix MDS lock inversions in getattr/reint paths (1844)
3950 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3951 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3952 - DLM race condition prevented some lock evictions (2328)
3953 - ENOMEM detection and retry on socknal sends (2230)
3954 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3955 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3956 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3957 - better error messages when a client is rejected during recovery (1505)
3958 - avoid cancelling locks which were never granted, after failure (2330)
3959 - fix i_sem/journal inversion in mds_client_add (2333)
3960 - fix truncate/getattr lock cycle deadlock (2334)
3961 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3962 - fix two rare exit paths which could leak an l_lock() ref (2321)
3963 - fencepost error in MDS/OST orphan recovery (2226)
3964 - make log record alignment 8 bytes (1988)
3965 - lstripe now fails when requested offset > ost_count (2237)
3966 - ensure that all kernel series have a complete list.h (1607)
3967 - fix crashes in special-file operations (2316)
3968 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3970 - allow configurable automake binary, for testing new versions
3971 - small update to the lfs documentation
3973 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3975 * fix negative export reference count in fsfilt_sync (2312)
3977 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3978 * release candidate 0.9.1
3980 - orphans are moved into the PENDING directory for possible recovery
3981 - replayed opens now open by fid for orphan/rename safety (1042)
3982 - last close of an orphan inode generates a transno (683)
3983 - chdir() and mount() now pin the directory entry (1020)
3984 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3985 - discard very old requests without processing them (1502)
3986 - remove some common, well-understood CERRORs (1505)
3987 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3988 - clear "grant" flags in OST replies until OST grant code lands (1644)
3989 - fix read performance by not clobbering i_blksize on client (1598)
3990 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3991 - make LRU size a /proc tunable, clears locks when reduced (707)
3992 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3993 - server locks take references on exports now (1558)
3994 - build fixes for 2.4.20-rh trees (1663)
3995 - return an error from lov_create if all OSCs are inactive (1751)
3996 - fix import levels when a reconnect happens without a timeout (1597)
3997 - exit early from mds_open if we get a lookup error (1749)
3998 - partial page read at EOF wouldn't wait for disk before sending (1642)
3999 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4000 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4001 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4002 - write/truncate lock inversion (1639)
4003 - Don't auto-load obdclass, portals modules during cleanup (1495)
4004 - fix timestamps from jumping to "now" (1763)
4005 - extra journal assertions (1648)
4006 - add an extra multiunlink test (1771)
4007 - fix read_record/write_record API (1776)
4008 - fix leak of offset_extent, possible incorrect i_size later (1772)
4009 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4010 - Debugging for the kqswnal_get_idle_tx problems (1820)
4011 - Allow recovery to be attempted multiple times (1536)
4012 - Write out MDS last_rcvd file after it is first created (1600)
4013 - Fix tx_descriptor leak in failed transmit situations (1827)
4014 - ext3 journaling fixes for assertion failure after IO error (1871)
4015 - class_export_put() on freed export after completion AST error (1896)
4016 - Fix revalidate looping in VFS (1322)
4017 - Don't access a freed export during MDS_REINT timeout (1521)
4018 - Add open-unlink recovery support on the MDS (1673,1764)
4019 - Return an error if no MDS data was read from last_rcvd (1946)
4020 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4021 - Don't setup a disk device that doesn't match exported UUID (317)
4022 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4023 - avoid committing NULL handle in force close
4024 - local.sh is now a one-stripe LOV configuration
4025 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4026 - fix ext3 htree duplicate directory entry corruption (1516)
4027 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4028 - update to new LOV EA format (2097)
4029 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4031 2003-06-15 Phil Schwan <phil@clusterfs.com>
4034 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4035 - per-import/export recovery handling (958, 931, 959)
4036 - multiple last-rcvd slots, for serving multiple FSes (949)
4037 - connections are again shared between multiple imp/exports (963, 964)
4038 - "umount -f" would hang if any requests needed to be sent (393, 978)
4039 - avoid pinning large req buffer by copying for queued messages (989)
4040 - add "uuid" to "lctl device" command to help upcalls (991)
4041 - "open" RPCs with transnos would confuse recovery counters (1037)
4042 - do proper endian conversion of all wire messages (288, 340, 891)
4043 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4044 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4045 - fix ptlrpc_connection leak in target_handle_connect (1174)
4046 - fix import refcounting bug in OST and MDS cleanup (1134)
4047 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4048 - fix very unlikely obd_types race condition (501)
4049 - remove osc_open hack for echo_client (1187)
4050 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4051 - a failure in read_inode2 leads to deadlock (1139)
4052 - cancel ack-locks as soon as transaction is committed (1072)
4053 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4054 - make sure to commitrw after any preprw to avoid deadlock (1162)
4055 - failing to execute a file in a lustre FS would lock inode (1203)
4056 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4057 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4058 - obd_uuid2tgt would walk off the end of the list (1255)
4059 - on IA64 the osc would give portals incorrect bulk size (1258)
4060 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4061 - fix lock inversion caused by new llite matching code (1282)
4062 - limit the number of dirty pages on a client to 10MB (1286)
4063 - timed out locks were not being corrected cancelled (1289)
4064 - fix O_DIRECT above 4GB on IA-32 (1292)
4065 * major user-visible changes
4066 - fail out/fail over policy now controlled by the upcall (993)
4068 - add OBD_PING to check server availability and failure (954)
4069 - lustre messages are now sent in sending host order (288, 340, 891)
4070 - add eadatalen to MDS getattr reply (340)
4071 - OST read replies may contain second buffer, with per-page status (593)
4073 2003-03-11 Phil Schwan <phil@clusterfs.com>
4076 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4077 - failed open()s could cause deadlock; fixed (867, 869)
4078 - stop cancelling OST locks when files are closed (481)
4079 - overlapping XID spaces caused network corruption (851, 853)
4080 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4081 - setattr_raw added, to do single-RPC, server-side setattrs
4082 - lmc/lconf syntax change for OST UUIDs
4083 - fix crashy race condition between ptlrpc_free_req and osc_close
4084 - don't use request in mdc_enqueue if we hit a timeout (889)
4085 - don't set the inode i_size for regular files from the MDS (896)
4086 - handle out of order completion AST (842)
4087 - don't LBUG if a lock request times out after receiving AST (913)
4088 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4089 - if a bad lock AST arrives, send an error instead of dropping entirely
4090 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4091 - fix leak in bulk IO when only partially completed (899, 900, 926)
4092 - fix O_DIRECT for ia64 (55)
4093 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4094 - C-z after timeout could hang a process forever; fixed (977)
4096 - client-side I/O cache (678, 924, 929, 941, 970)
4098 - READPAGE and SETATTRs which don't take server-side locks get
4101 2003-02-11 Phil Schwan <phil@clusterfs.com>
4104 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4105 - Fix fsfilt_extN_readpage() to read a full page of directory
4106 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4107 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4108 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4109 - Most intent operations were removed, in favour of a new RPC mode
4110 that does a single RPC to the server and bypasses most of the VFS
4111 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4112 - Aggressively cancel local locks on DLM servers
4113 - mds_reint_unlink sends EA to the client if it's the last nlink.
4114 client uses that EA to unlink OST objects.
4115 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4116 - recursive symlinks were fixed (439)
4117 - fixed NULL deref in DEBUG_REQ
4118 - filter_update_lastobjid no longer calls sync, which annoyed extN
4119 - fixed multi-client small-writes to a single file problem (445)
4120 - fixed mtime updates during file writes (607)
4121 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4122 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4123 - fixed Linux OST size reporting problem (444, 656)
4124 - OST now updates object mtime with writes or setattr (607, 619)
4125 - client verifies file size before zeroing page past EOF (445)
4126 - OST now writes last allocated objid to disk with allocation (108)
4127 - LOV on echo now works (409)
4129 - mds_reint_unlink sends a new buffer, with the EA included. this
4130 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4131 set if a regular file was being unlinked, and it was the last link
4132 - use PtlGet from the target for bulk writes (315)
4133 - OST now updates object mtime with writes or setattr (607, 619)
4134 - LDLM now has a grant-time callback to revalidate locked items, if
4136 - Many MDS operations were reorganized to combat race conditions
4138 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4139 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4141 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4144 - Fully reactivate OST imports after reconnection (512, others)
4145 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4146 - More graceful error handling for truncating on dead OST (515)
4147 - Don't error out unless we're actually accessing dead stripes (474)
4148 - Fix garbage sizes when stripes are missing (410)
4149 - LRU counters were broken, causing constant lock purge (433, 432)
4150 - garbage on read from stripes with failed OSTs (441)
4151 - mark OSCs as active before reconnecting during recovery (438)
4152 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4153 - lfind did not preserve OST order in output (443)
4154 - symlinks cause hung clients, incorrect data (439)
4155 - stop dereferencing request after dropping refcount (457)
4156 - don't LASSERT(spin_is_locked) on non-SMP (455)
4157 - fixes for many rename() bugs
4158 - fstat didn't correctly synchronize attributes (399)
4159 - server must handle lock cancellation during blocking AST prep (487)
4160 - bulk descriptors were free()d too soon (511)
4161 - fix paths in lconf, which would load incorrect modules (451, 507)
4162 - fix confusing lconf 'host not found' error message (386)
4163 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4164 - fix race condition in mdc_blocking_ast() for inode access (526)
4165 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4166 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4167 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4168 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4169 - fix lstripe in several ways (kernel side) (527)
4170 - fix request leak in ldlm_cli_enqueue (262)
4171 - incorrect OSC was marked inactive after OST failure
4172 - call mds_fs_cleanup before unmounting filesystem (524)
4173 - fix races between taking ns_lock and ldlm_lock_change_resource
4174 - fix races updating LOV export open file list
4175 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4176 - fix recovery NULL deref in ldlm_cli_cancel_unused
4177 - fix some DLM races by using new hash table for lock handles (419)
4178 - permit the client to specify desired inodes, at replay
4179 - duplicate requests when we queue them for replay reintegration
4180 - fix last_rcvd offset calculation
4181 - sync after each recovered transaction, so we always make progress
4182 - never, not always, ERESTART requests without transnos
4183 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4185 - skip replay if the MDS says that the client is already connected
4186 - don't check for a recovery-enabled export to match lctl's UUID
4187 - don't INC_USE_COUNT for phantom exports
4188 - don't crash when cleaning up phantom exports (567)
4189 - don't double-finish or set replay data for errored mdc_open requests
4190 - abort requests when they time out, so we don't get old replies
4191 - send/receive replies for AST messages again
4192 - if the client says that it doesn't have the lock, cancel it on the
4194 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4195 mark it as destroyed, it will all work out when decref is called
4196 - fix module use counts (22, 581)
4198 - ASTs now expect a reply (server cancels lock on error reply)
4200 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4203 - fix many simultaneous client startup (392)
4204 - fix dentry->d_it clobbering
4205 - credentials weren't being shipped for readdir/getattr operations
4206 - remove invalid assertions triggered during some concurrent MD
4208 - proper Lustre versions added (336, 389)
4209 - fix memory leak for create error case (398)
4210 - fix LOV locking bug that would get cli/srv out of sync
4211 - fix echo client over LOV (409)
4212 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4213 - fix double-O_EXCL intent crash (424)
4214 - avoid sending multiple lock CANCELs (352)
4216 - MDS can do multi-client recovery (modulo bugs in new code)
4218 - many updates, edits, cleanups
4220 2002-11-18 Phil Schwan <phil@clusterfs.com>
4223 - fix null d_it dereference (346)
4224 - fix full OST/dbench hang (333)
4225 - fix permission problem with file removal (286)
4226 - fix removal of OSCs from LOV when they fail
4227 - fix NULL deref during bulk timeout (214)
4228 - fix problems related to multiple filesystems on one MDS (241)
4229 - fixed serious subtle metadata locking bugs
4230 - free locks on clients when inodes are removed due to memory
4232 - fix inode pointer in lock data (285)
4233 - partial support for multiple MDS on a single host (241)
4234 - data locks weren't cancelled at clear_inode time (290, 311)
4235 - intent locks could lead to unbounded lock growth (205)
4236 - added a maximum lock count, an LRU list, and a flusher
4237 - fix multiple rename (365)
4238 - properly abstracted the echo client
4239 - OSC locked 1 byte too many; fixed
4240 - rewrote brw callback code:
4241 - fixed recovery bugs related to LOVs (306)
4242 - fixed too-many-pages-in-one-write crash (191)
4243 - fixed (again) crash in sync_io_timeout (214)
4244 - probably fixed callback-related race (385)
4246 - Add capability to MDS protocol
4247 - LDLM cancellations and callbacks on different portals
4249 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4252 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4253 - defer open object destruction to close time (601981, 138)
4254 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4255 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4256 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4257 - don't call obd_setattr in ll_file_release for destroyed objects
4258 * protocol change to lustre_msg: move |version| and add |flags|
4259 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4260 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4261 * added replay of create, unlink, link and rename operations during
4262 MDS failover; recovery should be much more robust now
4263 * remove failed OSCs from LOVs (only lov_create uses this so far)
4264 * the lustre-HOWTO was brought (more) up to date (582544)
4266 2002-10-23 Phil Schwan <phil@clusterfs.com>
4269 - in-use dentries weren't being reused properly (617851)
4270 - prevent multiple LDLM setup (599178)
4271 - fix LOV size calculations for truncate (617853)
4272 - fix client handling of MDS intent errors (POSIX)
4273 - fix permission bug in lovstripe.c test (624321)
4274 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4275 - truncate past end of file could corrupt data
4276 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4277 - a race in recovery could return ETIMEDOUT to apps (623947)
4278 - building outside the source directory was fixed
4279 * the lustre-HOWTO was brought (more) up to date (582544)
4280 * major progress was made on recovery functionality
4282 2002-10-10 Phil Schwan <phil@clusterfs.com>
4285 - recovery deadlock fix
4286 - rm -rf causes LBUG fix (617817)
4287 - file open by multiple tasks fix (618962)
4288 - directory permissions bugs (602707 and 620007)
4289 - journal_stop fixed with locking (611313)
4290 - O_APPEND failures resolved (618273, perhaps 614459)
4291 - lconf PATH fix (619770)
4292 - IA64 build fix (621450)
4293 - RPC buffer sizes scale with amount of memory
4295 2002-10-01 Phil Schwan <phil@clusterfs.com>
4298 - locks would be cancelled without throwing away data pages,
4299 resulting in inconsistent data (605627)
4300 - inode attributes were not always being refreshed (605627, 612449)
4301 - lconf now continues to cleanup after lctl reports an error
4302 - MDS now enforces user permissions (602707)
4303 - lprocfs cleanup fixed, but not yet enabled (614157)
4304 - fixed infinite server hang, should a client not respond to an AST
4305 - avoid going into recovery if user calls readlink() with a buffer
4306 that's too small (613941)
4307 - AST RPCs no longer require replies (614867) -- this may be changed
4308 - don't crash server if client sends an IOV that's too big (611336)
4309 - fixed lock conversion deadlock (611892)
4310 - fixed the following of symlinks (614622)
4311 * recovery: the server can remove locks from a client that dies, other
4312 clients can make progress
4313 * more extN patch fixes
4314 * compile-time configurable ptlrpc buffer allocations
4316 - collaborative read cache document
4317 - Lustre Lite Performance CDR document-in-progress
4319 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4322 - fix typo in patch-2.4.18
4324 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4327 - clear ptlrpc request each time in handle_incoming_request()
4328 - unlink of files now destroys the object on the OST
4330 2002-09-19 Peter Braam <braam@clusterfs.com>
4332 * add hard link support
4333 * change obdfile creation method
4334 * kernel patch changed
4336 2002-09-19 Peter Braam <braam@clusterfs.com>
4339 - stack overflow bug in extN fixed
4341 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4343 * documentation updates
4344 - add man pages for config tools
4345 - update tests/README to describe testing with new config tools
4346 - finish metadata API descriptions
4347 * bug fixes and cleanups
4348 - statfs workaround for 16TB limit
4349 - LOV stripe allocation improved, can stripe on subset of OSTs
4350 - LOV file size/IO offset was wrong for files > 4GB in size
4351 - object EA data was being dropped, caused files to be unreadable
4352 - memory overflow with non-LOV OST caused memory corruption
4353 - fixed regression tests to work with new config tools, obdfilter
4354 - fixed bug when directory size became larger than 1 block
4355 - fixed bug (for single client case) when PWD was deleted
4356 - invalidate local directory pages when doing intent-based ops
4357 - avoid LDLM oops when lock callback contained bad data
4359 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4361 * documentation updates
4362 * bug fixes and cleanups
4363 - configuration tools
4366 - 64-bit compile warnings
4367 - 64-bit internal statfs data
4369 * test_brw on persistent OST devices
4371 * lprocfs (disabled)
4373 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4375 * documentation updates
4376 * bug fixes and cleanups
4377 * configuration tools
4379 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4382 * many small fixes to 0_5_4
4383 * io/network handling
4384 * thinkos in MDS operations
4386 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4389 * crucial basic fixes to 0.5.3
4390 * IOR, Iozone work over Elan
4393 2002-08-07 Phil Schwan <phil@clusterfs.com>
4394 * version 0_5_3, our first alpha
4395 * we use the new Portals iovs
4396 * documentation updates
4397 * bug fixes and cleanups
4398 * small changes in the DLM wire protocol
4400 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4401 * version 0_5_1 with some initial stability,
4402 * locking on MD and file I/O.
4403 * documentation updates
4404 * several bug fixes since 0.5.0
4405 * small changes in wire protocol
4407 2002-07-18 Phil Schwan <phil@clusterfs.com>
4409 * delivered as Lustre Light Alpha
4410 * fixed a crash after handling invalid MDS requests
4411 * fixed directory pages for architectures with non-4k pages sizes
4413 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4414 * release version v0_4_4
4415 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4417 * Fixes a number of interruption problems with OST operations.
4418 * Update documentation for portals header changes
4419 * Move all wire protocol structs/defines to lustre_idl.h
4420 * Fixes symlink length bug.
4421 * Add tcpdump to repository.
4423 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4424 * release version v0_4_3
4425 * Fixes statfs for inodes on extN.
4426 * Fixes bug in runtests which would delete /etc/hosts.
4427 * Use 64-bit object IDs wherever possible (not into VFS though)
4428 Remove ost_get_info, which is unused by lustre, and out of date.
4430 2002-07-03 Peter Braam <braam@clusterfs.com>
4431 * release version v0_4_2 Fixes a lookup error (type not passed)
4432 * move forward to head of Portals
4433 * move forward to latest Lustre kernel
4435 2002-06-25 Peter Braam <braam@clusterfs.com>
4436 * release version v0_4_1. Hopefully stable on single node use.