1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.1.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
18 Description: Don't allow skipping OSTs if index has been specified.
19 Details : Don't allow skipping OSTs if index has been specified, make locking
20 in internal create lots better.
22 Severity : enhancement
24 Description: Add rhel5 support to HEAD.
28 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
29 Details : access to bitfield in structure is always rounded to long
30 and this produce problem with not atomic change any bit.
34 Description: Lustre make rpms failed.
35 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
36 builds Lustre rpms from tarball.
39 Frequency : rare on shutdown ost
41 Description: If llog cancel was not send before clean_exports phase, this can
42 produce deadlock in llog code.
43 Details : If llog thread has last reference to obd and call class_import_put
44 this produce deadlock because llog_cleanup_commit_master wait when
45 last llog_commit_thread exited, but this never success because was
46 called from llog_commit_thread.
50 Description: allow userland application know is lost one of stripes.
51 Details : fill lvb_blocks with error code on ost and return it to
52 application if error flag found.
56 Description: NULL lov_tgts causing MDS oops
57 Details : more safe checks for NULL lov_tgts for avoid oops.
59 Severity : enhancement
61 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
65 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
66 Details : If an OST is down the MDS will hang indefinitely in
67 obd_statfs_rqset() waiting for the statfs data. While for
68 MDS QOS usage of statfs, it should not stuck in waiting.
70 Severity : enhancement
72 Description: remote_acl support
73 Details : Support ACL-based permission check for remote user.
74 Support setfacl/getfacl for remote user with the utils
75 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
76 the system "{s,g}etfacl" utils.
78 Severity : enhancement
80 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
82 Severity : enhancement
84 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
88 Description: make mgs_setparam() handle fsname containing dash
89 Details : fsname containing a dash does not work with lctl conf_param
91 Severity : enhancement
93 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
95 Severity : enhancement
97 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
101 Frequency : rare, at shutdown
102 Description: access already free / zero obd_namespace.
103 Details : if client_disconnect_export was called without force flag set,
104 and exist connect request in flight, this can produce access to
105 NULL pointer (or already free pointer) when connect_interpret
106 store ocd flags in obd_namespace.
110 Frequency : only at startup
111 Description: not alloc memory with spinlock held.
112 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
113 if any spinlock held.
115 Severity : enhancement
117 Description: make lustre randomly fail allocating memory
118 Details : Make lustre randomly failed allocating memory for testing purpose.
120 Severity : enhancement
122 Description: lost problems with lov objid file
123 Details : Fixes some scability and access to not inited memory problems
124 in work with lov objdid file.
129 Description: lfs find does not continue on file error
130 Details : Continue other files processing when a file/dir is absent.
134 Description: Inconsistent usage of lustre_pack_reply()
135 Details : Standardize the usage of lustre_pack_reply() such that it
136 always generate a CERROR on failure.
139 Frequency : occasional
141 Description: MDS hang or stay in waiting lock
142 Details : If client receive lock with CBPENDING flag ldlm need send lock
143 cancel as separate rpc, to avoid situation when cancel request
144 can't processed due all i/o threads stay in wait lock.
148 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
152 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
156 Description: Update to SLES9 kernel-2.6.5-7.287.3.
160 Description: Update to RHEL4 latest kernel.
162 Severity : enhancement
164 Description: Build SLES10 patchless client fails
165 Details : The configure was broken by run ./configure with
166 --with-linux-obj=.... argument for patchless client. When the
167 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
168 can't search header adequately. Use absolute path such as
169 -I($LINUX)/include instead.
173 Description: interrupt oig_wait produce painc on resend.
174 Details : brw_redo_request can be used for resend requests from ptlrpcd and
175 private set, and this produce situation when rq_ptlrpcd_data not
176 copyed to new allocated request and triggered LBUG on assert
177 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
178 ptlrpcd set if request is changed and can be safety changed to use
183 Description: LASSERT_{REQ,REP}SWAB macros are buggy
184 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
185 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
186 these macros and replace them with their difinitions instead.
190 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
191 Details : Update which_patch & target file for SLES10 latest kernel.
194 Description: Remove client patches from SLES 10 kernel.
195 Details : This causes SLES 10 clients to behave as patchless clients
196 even on a Lustre-patched (server) kernel.
198 Severity : enhancement
200 Description: self-adjustable client's lru lists
201 Details : use adaptive algorithm for managing client cached locks lru
202 lists according to current server load, other client's work
203 pattern, memory activities, etc. Both, server and client
204 side namespaces provide number of proc tunables for controlling
207 Severity : enhancement
209 Description: light-weight GSS support
210 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
211 and reduce performance overhead.
213 Severity : enhancement
215 Description: Linux keyring support
216 Details : Support using service of Linux keyring for Lustre GSS internal
217 context refresh/cache mechanism.
221 Description: Fix errors in lfs documentation
222 Details : Fixes man pages
226 Description: don't use GFP_* in generic Lustre code.
227 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
231 Description: obdclass is limited by single OBD_ALLOC(idarray)
232 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
236 Description: warnings with build patchless client with vanila 2.6.19 and up
237 Details : change old ctl_table style and replace ctl_table/ctl_table_header
238 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
242 Frequency : Only for SLES
243 Description: sanity_quota fail test_1
244 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
245 file, which makes TSTID[2] inunique.
249 Frequency : Only for SLES9
250 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
251 Details : Update target/ChangeLog/which_patch .
255 Frequency : Only for test_10 in sanity-quota.sh
256 Description: a bug in quota test code
257 Details : checking a test flag in a wrong place
261 Frequency : Only for test_13 in sanity-quota.sh
262 Description: a bug in quota test script
263 Details : shouldn't check overfull allocated quota
267 Description: jbd statistics
268 Details : Port older jbd statistics patch for sles10
270 Severity : enhancement
272 Description: Add printing inode info into message about error in writepage.
277 Description: reply_lock_interpret crash due to race with it and lock cancel.
278 Details : Do not replay locks that are being cancelled. Do not reference
279 locks by their address during replay, just by their handle.
283 Frequency : When flocks are used.
284 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
285 Details : Flock locks might destroy just granted lock if it could be merged
286 with another existing flock, this is done in completion handler,
287 so teach ldlm_cli_enquque_fini that this is a valid case for
293 Description: Oops in read and write path when failing to allocate lock.
294 Details : Check if lock allocation failed and return error back.
298 Description: lstripe command fails for valid OST index
299 Details : The stripe offset is compared to lov->desc.ld_tgt_count
300 instead of lov->desc.ld_active_tgt_count.
304 Description: sanity.sh failed test 103
305 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
310 Description: df doesn't work properly if diskfs blocksize != 4K
311 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
315 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
316 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
320 Description: obd_health_check_timeout too short
321 Details : set obd_health_check_timeout as 1.5x of obd_timeout
325 Description: llapi_file_create() does not allow some changes
326 Details : add llapi_file_open() that allows specifying the mode and
327 open flags, and also returns an open file handle.
331 Description: merge and cleanup kernel patches.
332 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
336 Description: Add journal checksum support.(Kernel part)
337 Details : The journal checksum feature adds two new flags i.e
338 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
339 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
340 indicates that the commit block contains the checksum for
341 the blocks described by the descriptor blocks. Now commit
342 record can be sent to disk without waiting for descriptor
343 blocks to be written to disk. This behavior is controlled
344 using JBD2_FEATURE_ASYNC_COMMIT flag.
348 Description: OSS needs mutliple precreate threads
349 Details : Add ability to start more than one create thread per OSS.
353 Description: Sanity.sh test_65k failed: setstripe should have succeeded
354 Details : execute lfs setstripe on client
358 Description: mds_obd_create error creating tmp object
359 Details : When the user sets quota on root, llog will be affected and can't
360 create files and write files.
363 Frequency : Always on ia64 patchless client, and possibly others.
365 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
366 Details : This allows the patchless client to be loaded on architectures
371 Description: RedHat Update kernel for RHEL5
372 Details : Add the kernel config file for RHEL5.
376 Description: RedHat Update kernel for RHEL5
377 Details : Modify the kernel config file more closer RHEL5.
381 Description: Build failure against Centos5 (RHEL5)
382 Details : Define PAGE_SIZE when it isn't present.
386 Description: test_7 on sanity-quota.sh can't be used on separate machines
387 Details : add facet to handle it
391 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
392 Details : with NFS, the anon dentry's parent was set to itself in
393 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
394 corresponding dentry other than use rec->ur_name.
396 Severity : enhancement
398 Description: lfs setstripe enhancement
399 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
403 Description: enable data checksumming by default
404 Details : enable checksum by default, allow --disable-checksum
405 configure option and "-o nochecksum" mount option. Checksums
406 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
410 Description: lustre support for RHEL5
411 Details : Add support for RHEL5.
415 Description: Client eviction due to failover config
416 Details : after a connection loss, the lustre client should attempt to
417 reconnect to the last active server first before trying the
418 other potential connections.
422 Description: when mds and osts use different quota unit(32bit and 64bit),
423 quota will be released repeatly.
424 Details : void sending multiple quota reqs to mds, which will keep the
425 status between the reqs.
429 Description: osts not allocated evenly to files
430 Details : change the condition to increase offset_idx
433 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
435 Description: Data corruption for OSTs that are formatted larger than 2TB
437 Details : When generating the bio request for lustre file writes the
438 sector number would overflow a temporary variable before being
439 used for the IO. The data reads correctly from Lustre (which
440 will overflow in a similar manner) but other file data or
441 filesystem metadata may be corrupted in some cases.
445 Description: Tune the kernel for good SCSI performance.
446 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
447 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
452 Description: rewrite ext2-derived code in obdclass/uuid.c
453 Details : rewrite inherited code (uuid parsing code from ext2 utils)
454 from scratch preserving functionality.
458 Description: "lfs find -obd UUID" prints directories
459 Details : "lfs find -obd UUID" will return all directory names instead
460 of just file names. It is incorrect because the directories
461 do not reside on the OSTs.
465 Description: use i_size_read and i_size_write in 2.6 port
466 Details : replace inode->i_size access with i_size_read/write()
469 Frequency : when removing large files
471 Description: scheduling issue during removal of large Lustre files
472 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
473 It causes scheduling issues when removing large files (17TB in the
477 Frequency : only with liblustre clients on XT3
479 Description: evictions taking too long
480 Details : allow llrd to evict clients directly on OSTs
483 Frequency : only on ppc
485 Description: /proc/fs/lustre/devices broken on ppc
486 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
487 We should make sure the type of 'index' is loff_t and then cast
488 explicitly as needed below. Do not assign an explicitly cast
492 Frequency : only for rhel5
494 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
495 Details : Modify the target file & which_kernel.
497 Severity : enhancement
499 Description: omit set fsid for export NFS
500 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
501 as FS_REQUIRES_DEV for avoid problems with generate fsid.
505 Frequency : Always, for kernels after 2.6.16
506 Description: Fix warning idr_remove called for id=.. which is not allocated.
507 Details : Last kernels save old s_dev before kill super and not allow
508 to restore from callback - restore it before call kill_anon_super.
512 Description: buffer overruns could theoretically occur
513 Details : llapi_semantic_traverse() modifies the "path" argument by
514 appending values to the end of the origin string, and a
515 overrun may occur. Adding buffer overrun check in liblustreapi.
519 Description: Fix error on 'ls .' at the top of the Lustre mount.
520 Details : Don't revalidate dentry if it is a root dentry.
524 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
525 Details : Modify vm-tunables-rhel4.patch.
529 Description: Kernel config for 2.6.18-vanilla.
530 Details : Modify targets/2.6-vanilla.target.in.
531 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
532 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
533 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
534 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
538 Description: improve handling recoverable errors
539 Details : if request processig with error which can be recoverable on server
540 request should be resend, otherwise page released from cache and
545 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
546 Details : Modify target file & which_patch.
547 A flaw was found in the IA32 system call emulation provided
548 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
549 value could be stored in the %RAX register, which could trigger an
550 out-of-bounds system call table access. An untrusted local user
551 could exploit this flaw to run code in the kernel
552 (ie a root privilege escalation). (CVE-2007-4573).
556 Description: O_DIRECT bypasses client statistics.
557 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
558 read_ahead_stats were updated. copied the stats section from
559 osc_send_oap_rpc() into async_internal().
563 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
567 Description: change order of libsysio includes
568 Details : '#include sysio.h' should always come before '#include xtio.h'
572 Description: handle "serious error: objid * already exists" more gracefully
573 Details : If LAST_ID value on disk is smaller than the objects existing in
574 the O/0/d* directories, it indicates disk corruption and causes an
575 LBUG(). If the object is 0-length, then we should use the existing
576 object. This will help to avoid a full fsck in most cases.
578 Severity : enhancement
580 Description: adapt the lustre_config script to support the upgrade case
581 Details : Add "-u" option for lustre_config script to support upgrading 1.4
582 server targets to 1.6 in parallel.
586 Description: To avoid grant space > avaible space when the disk is almost
587 full. Without this patch you might see the error "grant XXXX >
588 available" or some LBUG about grant, when the disk is almost
590 Details : In filter_check_grant, for non_grant cache write, we should
591 check the left space by if (*left > ungranted + bytes), instead
592 of (*left > ungranted), because only we are sure the left space
593 is enough for another "bytes", then the ungrant space should be
594 increase. In client, we should update cl_avail_grant only there
595 is OBD_MD_FLGRANT in the reply.
599 Description: Update RHEL 4 kernel to fix local root privilege escalation.
600 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
601 described in CVE-2007-4573. This problem could allow untrusted
602 local users to gain root access.
605 Frequency : when using O_DIRECT and quotas
607 Description: Incorrect file ownership on O_DIRECT output files
608 Details : block usage reported by 'lfs quota' does not take into account
609 files that have been written with O_DIRECT.
614 Description: touch file failed when fs is not full
615 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
616 otherwise we can get ENOSP while fs is not full.
620 Description: parallel lock callbacks
621 Details : Instead of sending blocking and completion callbacks as separated
622 requests, adding them to a set and sending in parallel.
625 Frequency : only for Cray XT3
626 Bugzilla : 12829/13455
627 Description: Changing primary group doesn't change the group lustre assigns to
629 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
630 overridden with the primary group provided by the group upcall,
631 whereas the supplied fsgid can be trusted if it is in the list of
632 supplementary groups returned by the group upcall.
634 Severity : enhancement
636 Description: Allow masking D_WARNING, D_ERROR messages from console
637 Details : Console messages can now be disabled via lnet.printk.
641 Description: User code with malformed file open parameter crashes client node
642 Details : Before packing join_file req, all the related reference should be
643 checked carefully in case some malformed flags cause fake
644 join_file req on client.
648 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
649 Details : ldlm_completion_ast() assumes that a lock is granted when the req
650 mode is equal to the granted mode. However, it should also check
651 that LDLM_FL_CP_REQD is not set.
655 Description: Heavy nfs access might result in deadlocks
656 Details : After ELC code landed, it is now improper to enqueue any mds
657 locks under och_sem, because enqueue might want to decide to
658 cancel open locks for same inode we are holding och_sem for.
662 Description: Client eviction while running blogbench
663 Details : A lot of unlink operations with concurrent I/O can lead to a
664 deadlock causing evictions. To address the problem, the number of
665 oustanding OST_DESTROY requests is now throttled to
666 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
667 callbacks are processed in priority.
669 --------------------------------------------------------------------------------
671 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
673 * Support for kernels:
674 2.6.5-7.283 (SLES 9),
675 2.6.9-55.EL (RHEL 4),
676 2.6.16.46-0.14 (SLES 10),
677 2.6.18.8 vanilla (kernel.org)
678 * Client support for unpatched kernels:
679 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
680 2.6.16 - 2.6.22 vanilla (kernel.org)
681 * Due to recently discovered recovery problems, we do not recommend
682 using patchless RHEL 4 clients with this or any earlier release.
683 * Recommended e2fsprogs version: 1.39.cfs8
684 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
685 * Starting with this release, the ldiskfs backing filesystem required
686 by Lustre is now in its own package, lustre-ldiskfs. This package
687 should be installed. It is versioned separately from Lustre and
688 may be released separately in future.
692 Description: block reactivating mgc import until all deactivates complete
693 Details : Fix race when failing back MDT/MGS to itself (testing)
696 Frequency : at statup only
698 Description: mds_lov_synchronize race leads to various problems
699 Details : simultaneous MDT->OST connections at startup can cause the
700 sync to abort, leaving the OSC in a bad state.
702 Severity : enhancement
704 Description: add optional extra BUILD_VERSION info
705 Details : add a new environment variable (namely LUSTRE_VERS) which allows
706 to override the lustre version.
709 Frequency : 2.6.18 servers only
711 Description: ll_kern_mount() doesn't release the module reference
712 Details : The ldiskfs module reference count never drops down to 0
713 because ll_kern_mount() doesn't release the module reference.
718 Description: server LBUG when using old ost_num_threads parameter
719 Details : Accept the old ost_num_threads parameter but warn that it
720 is deprecated, and fix an off-by-one error that caused an LBUG.
725 Description: Transient SCSI error results in persistent IO issue
726 Details : iobuf->dr_error is not reinitialized to 0 between two
730 Frequency : sometimes when underlying device returns I/O errors
732 Description: OSTs not going read-only during write failures
733 Details : OSTs are not remounted read-only when the journal commit threads
734 get I/O errors because fsfilt_ext3 calls journal_start/stop()
735 instead of the ext3 wrappers.
739 Description: poor connect scaling with increasing client count
740 Details : Don't run filter_grant_sanity_check for more than 100 exports
741 to improve scaling for large numbers of clients.
744 Frequency : SLES10 only
746 Description: sanity-quota.sh quotacheck failed: rc = -22
747 Details : Quotas cannot be enabled on SLES10.
750 Frequency : liblustre clients only
752 Description: getdirentries does not give error when run on compute nodes
753 Details : getdirentries does not fail when the size specified as an argument
754 is too small to contain at least one entry
756 Severity : enhancement
758 Description: Add LNET router traceability for debug purposes
759 Details : If a checksum failure occurs with a router as part of the
760 IO path, the NID of the last router that forwarded the bulk data
761 is printed so it can be identified.
766 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
767 Details : Due to a race condition, liblustre clients were occasionally
770 Severity : enhancement
772 Description: lfs setstripe use optional parameters instead of postional
775 Severity : enhancement
777 Description: Nanosecond timestamp support for ldiskfs
778 Details : The on-disk ldiskfs filesystem has added support for nanosecond
779 resolution timestamps. There is not yet support for this at
780 the Lustre filesystem level.
783 Frequency : during server recovery
785 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
786 Details : request with rq_no_resend flag not awake l_wait_event if they get a
790 Frequency : nfs export on patchless client
792 Description: connectathon hang when test nfs export over patchless client
793 Details : Disconnected dentry cannot be found with lookup, so we do not need
794 to unhash it or make it invalid
797 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
798 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
799 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
804 Description: open req refcounting wrong on reconnect
805 Details : If reconnect happened between getting open reply from server and
806 call to mdc_set_replay_data in ll_file_open, we will schedule
807 replay for unreferenced request that we are about to free.
808 Subsequent close will crash in variety of ways.
809 Check that request is still eligible for replay in
810 mdc_set_replay_data().
815 Description: disable writes to filesystem when reading health_check file
816 Details : the default for reading the health_check proc file has changed
817 to NOT do a journal transaction and write to disk, because this
818 can cause reads of the /proc file to hang and block HA state
819 checking on a healthy but otherwise heavily loaded system. It
820 is possible to return to the previous behaviour during configure
821 with --enable-health-write.
823 Severity : enhancement
825 Description: 64-bit inode version
826 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
827 to the inode. This will be required for version-based recovery.
832 Description: MDS fails to start if a duplicate client export is detected
833 Details : in some rare cases it was possible for a client to connect to
834 an MDS multiple times. Upon recovery the MDS would detect this
835 and fail during startup. Handle this more gracefully.
837 Severity : enhancement
839 Description: Add -o localflock option to simulate old noflock
841 Details : This will achieve local-only flock/fcntl locks
847 Description: log_commit_thread vs filter_destroy race leads to crash
848 Details : Take import reference before releasing llog record semaphore
853 Description: Wrong request locking in request set processing
854 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
855 imp_delayed_list, in this place should be used imp_lock.
858 Frequency : when reconnection
860 Description: Grant Leak when osc reconnect to OST
861 Details : When osc reconnect ost, OST(filter) should check whether it
862 should grant more space to client by comparing fed_grant and
863 cl_avail_grant, and return the granted space to client instead
864 of "new granted" space, because client will call osc_init_grant
865 to update the client grant space info.
868 Frequency : when client reconnect to OST
870 Description: Grant Leak when osc do resend and replay bulk write
871 Details : When osc reconnect to OST, OST(filter)should clear grant info of
872 bulk write request, because the grant info while be sync between
873 OSC and OST when reconnect, and we should ignore the grant info
874 these of resend/replay write req.
879 Description: Grant space more than avaiable left space sometimes.
880 Details : When then OST is about to be full, if two bulk writing from
881 different clients came to OST. Accord the avaliable space of the
882 OST, the first req should be permitted, and the second one
883 should be denied by ENOSPC. But if the seconde arrived before
884 the first one is commited. The OST might wrongly permit second
885 writing, which will cause grant space > avaiable space.
888 Frequency : when client is evicted
890 Description: Grant might be wrongly erased when osc is evicted by OST
891 Details : when the import is evicted by server, it will fork another
892 thread ptlrpc_invalidate_import_thread to invalidate the
893 import, where the grant will be set to 0. While the original
894 thread will update the grant it got when connecting. So if
895 the former happened latter, the grant will be wrongly errased
896 because of this race.
901 Description: Checking Stale with correct fid
902 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
903 to check whether it is stale, because sometimes, we want the
904 enqueue happened anyway, and op_data.fid2 will not be initialized.
906 Severity : enhancement
908 Description: update patchless client
909 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
912 Frequency : only with 2.4 kernel
914 Description: random memory corruption
915 Details : size of struct ll_inode_info is to big for union inode.u and this
916 can be cause of random memory corruption.
921 Description: Memory leak in recovery
922 Details : Lov_mds_md was not free in an error handler in mds_create_object.
923 It should also check obd_fail before fsfilt_start, otherwise if
924 fsfilt_start return -EROFS,(failover mds during mds recovery).
925 then the req will return with repmsg->transno = 0 and rc = EROFS.
926 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
927 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
928 be freed no matter whether fsfilt_commit success or not.
931 Frequency : only with huge count clients
933 Description: Prevents from taking the superblock lock in llap_from_page for
935 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
936 for llap_from_page call in ll_removepage prevents from taking the
937 superblock lock for a soon died page.
942 Description: Not check open intent error before release open handle
943 Details : in some rare cases, the open intent error is not checked before
944 release open handle, which may cause
945 ASSERTION(open_req->rq_transno != 0), because it tries to release
946 the failed open handle.
951 Description: Set cat log bitmap only after create log success.
952 Details : in some rare cases, the cat log bitmap is set too early. and it
953 should be set only after create log success.
957 Description: Accessing a block bevice can re-enable I/O when Lustre is
958 tearing down a device.
959 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
964 Description: service threads may hog cpus when there are a lot of requests
965 Details : Insert cond_resched to give other threads a chance to use some CPU
970 Description: the cat log was not initialized in recovery
971 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
972 unlink log on mds will not be initialized until mds post
973 recovery. And also in mds post recovery, the unlink log will
974 initialization will be done asynchronausly, so there will be race
975 between add unlink log and unlink log initialization.
979 Description: brw_stats were being printed incorrectly
980 Details : brw_stats were being printed as log2 but all of them were not
981 recorded as log2. Also remove some code duplication arising from
982 filter_tally_{read,write}.
986 Frequency : rare, only in recovery.
987 Description: ASSERTION(req->rq_type != LI_POISON) failed
988 Details : imp_lock should be held while iterating over imp_sending_list for
989 prevent destroy request after get timeout in ptlrpc_queue_wait.
993 Description: replay-single.sh test 52 fails
994 Details : A lock's skiplist need to be cleanup when it being unlinked
995 from its resource list.
999 Description: Short directio read returns full requested size rather than
1001 Details : Direct I/O operations should return actual amount of bytes
1002 transferred rather than requested size.
1004 Severity : enhancement
1006 Description: metadata RPC reduction (e.g. for rm performance)
1007 Details : decrease the amount of synchronous RPC between clients and servers
1008 by canceling conflicing lock before the operation on the client
1009 and packing thier handles into the main operation RPC to server.
1011 Severity : enhancement
1013 Description: Async OSC create to avoid the blocking unnecessarily.
1014 Details : If a OST has no remain object, system will block on the creating
1015 when need to create a new object on this OST. Now, ways use
1016 pre-created objects when available, instead of blocking on an
1017 empty osc while others are not empty. If we must block, we block
1018 for the shortest possible period of time.
1020 Severity : enhancement
1022 Description: refine locking for avoid write wrong info into lov_objid file
1023 Details : fix possible races with add new target and write/update data in
1026 --------------------------------------------------------------------------------
1028 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1033 Frequency : on some architectures
1035 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1036 Details : Uninitialized flags sometimes cause configuration commands to
1040 Frequency : patchless clients only
1042 Description: missing __iget() symbol export
1043 Details : The __iget() symbol export is missing. To avoid the need for
1044 this on patchless clients the deathrow inode reaper is turned
1045 off, and we depend on the VM to clean up old inodes. This
1046 dependency was during via the fix for bug 12181.
1048 --------------------------------------------------------------------------------
1050 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1052 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1053 INTEROPERATE with older versions automatically. In many cases a
1054 special upgrade step is needed. Please read the
1055 user documentation before upgrading any part of a live system.
1056 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1057 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1058 * WARNING: Lustre configuration and startup changes are required with
1059 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1065 Severity : enhancement
1067 Description: MountConf
1068 Details : Lustre configuration is now managed via mkfs and mount
1069 commands instead of lmc and lconf. New obd types (MGS, MGC)
1070 are added for dynamic configuration management. See
1071 https://mail.clusterfs.com/wikis/lustre/MountConf for
1074 Severity : enhancement
1076 Description: dynamic OST addition
1077 Details : OSTs can now be added to a live filesystem
1079 Severity : enhancement
1081 Description: startup order invariance
1082 Details : MDTs and OSTs can be started in any order. Clients only
1083 require the MDT to complete startup.
1085 Severity : enhancement
1087 Description: parallel, asynchronous orphan cleanup
1088 Details : orphan cleanup is now performed in separate threads for each
1089 OST, allowing parallel non-blocking operation.
1091 Severity : enhancement
1093 Description: optimized stripe assignment
1094 Details : stripe assignments are now made based on ost space available,
1095 ost previous usage, and OSS previous usage, in order to try
1096 to optimize storage space and networking resources.
1098 Severity : enhancement
1100 Description: Permanently set tunables
1101 Details : All writable /proc/fs/lustre tunables can now be permanently
1102 set on a per-server basis, at mkfs time or on a live system.
1104 Severity : enhancement
1106 Description: Lustre message v2
1107 Details : Add lustre message format v2.
1109 Severity : enhancement
1111 Description: client OST exclusion list
1112 Details : Clients can be started with a list of OSTs that should be
1113 declared "inactive" for known non-responsive OSTs.
1117 Description: SPEC SFS validation failure on NFS v2 over lustre.
1118 Details : Changes the blocksize for regular files to be 2x RPC size,
1119 and not depend on stripe size.
1121 Severity : enhancement
1123 Description: Multiple MD RPCs in flight.
1124 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1125 To avoid overly-overloading MDS, introduce a limit on number of
1126 MDS RPCs in flight for a single client and add /proc controls
1127 to adjust this limit.
1129 Severity : enhancement
1131 Description: client read/write statistics
1132 Details : Add client read/write call usage stats for performance
1133 analysis of user processes.
1134 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1135 file access. extents_stats shows chunk size distribution.
1136 extents_stats_per_process show chunk size distribution per
1139 Severity : enhancement
1141 Description: mds statistics
1142 Details : Add detailed mds operations statistics in
1143 /proc/fs/lustre/mds/*/stats.
1147 Description: Failure of copying files with lustre special EAs.
1148 Details : Client side always return success for setxattr call for lustre
1149 special xattr (currently only "trusted.lov").
1154 Description: Refcount LNET uuids
1155 Details : The global LNET uuid list grew linearly with every startup;
1156 refcount repeated list entries instead of always adding to
1159 Severity : enhancement
1161 Description: Dynamic service threads
1162 Details : Within a small range, start extra service threads
1163 automatically when the request queue builds up.
1166 Frequency : mixed-endian client/server environments
1168 Description: mixed-endian crashes
1169 Details : The new msg_v2 system had some failures in mixed-endian
1172 Severity : enhancement
1174 Description: Easy OST removal
1175 Details : OSTs can be permanently deactivated with e.g. 'lctl
1176 conf_param lustre-OST0001.osc.active=0'
1178 Severity : enhancement
1180 Description: MGS proc entries
1181 Details : Added basic proc entries for the MGS showing what filesystems
1184 Severity : enhancement
1186 Description: provide MGS failover
1187 Details : Added config lock reacquisition after MGS server failover.
1189 Severity : enhancement
1191 Description: add Linux 2.4 support
1192 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1196 Description: a large application tries to do I/O to the same resource and dies
1197 in the middle of it.
1198 Details : Check the req->rq_arrival time after the call to
1199 ost_brw_lock_get(), but before we do anything about
1200 processing it & sending the BULK transfer request. This
1201 should help move old stale pending locks off the queue as
1202 quickly as obd_timeout.
1205 Frequency : when an incorrect nid is specified during startup
1207 Description: ptlrpc connect to non-existant node causes kernel crash
1208 Details : LNET can't be re-entered from an event callback, which
1209 happened when we expire a message after the export has been
1210 cleaned up. Instead, hand the zombie cleanup off to another
1213 Severity : enhancement
1215 Description: plain/inodebits lock performance improvement
1216 Details : Grouping plain/inodebits in granted list by their request modes
1217 and bits policy, thus improving the performance of search through
1221 Frequency : only if OST filesystem is corrupted
1223 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1224 Details : for a short time RPCs with bulk IO are in the replay list,
1225 but replay of bulk IOs is unimplemented. If the OST filesystem
1226 is corrupted due to disk cache incoherency and then replay is
1227 started it is possible to trip an assertion. Avoid putting
1228 committed RPCs into the replay list at all to avoid this issue.
1231 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1233 Description: System hang on startup
1234 Details : This bug allowed the liblustre (e.g. catamount) client to
1235 return to the app before handling all startup RPCs. This
1236 could leave the node unresponsive to lustre network traffic
1237 and manifested as a server ptllnd timeout.
1239 Severity : enhancement
1241 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1242 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1243 boolean to control whether to print peer debug info when a
1244 client's RPC times out.
1247 Frequency : only for kernels with patches from Lustre below 1.4.3
1249 Description: Remove old rdonly API
1250 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1253 Frequency : only for devices with external journals
1255 Description: Set external device read-only also
1256 Details : During a commanded failover stop, we set the disk device
1257 read-only while the server shuts down. We now also set any
1258 external journal device read-only at the same time.
1261 Frequency : when upgrading from 1.4 while trying to change parameters
1263 Description: The wrong (new) MDC name was used when setting parameters for
1264 upgraded MDT's. Also allows changing of OSC (and MDC)
1265 parameters if --writeconf is specified at tunefs upgrade time.
1268 Frequency : when setting specific ost indicies
1270 Description: QOS code breaks on skipped indicies
1271 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1272 created with --index need not be sequential.
1274 Severity : enhancement
1276 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1277 Details : The uninit_groups feature works in conjunction with the kernel
1278 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1279 up the pass1 processing of e2fsck. This is a read-only feature
1280 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1281 mount filesystems that have had this feature enabled.
1283 Severity : enhancement
1285 Description: Improve multi-block allocation algorithm to avoid fragmentation
1286 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1287 allocation locality and avoid filesystem fragmentation.
1289 ------------------------------------------------------------------------------
1291 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1293 * Support for kernels:
1294 2.4.21-47.0.1.EL (RHEL 3)
1295 2.6.5-7.283 (SLES 9)
1296 2.6.9-42.0.10.EL (RHEL 4)
1297 2.6.12.6 vanilla (kernel.org)
1298 2.6.16.27-0.9 (SLES 10)
1299 * Recommended e2fsprogs version: 1.39.cfs5
1301 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1305 Frequency : occasional, depends on client load and configuration
1306 Bugzilla : 12181, 12203
1307 Description: data loss for recently-modified files
1309 Details : In some cases it is possible that recently written or created
1310 files may not be written to disk in a timely manner (this should
1311 normally be within 30s unless client IO load is very high).
1312 The problem appears as zero-length files or files that are a
1313 multiple of 1MB in size after a client crash or client eviction
1314 that are missing data at the end of the file.
1316 This problem is more likely to be hit on clients where files are
1317 repeatedly created and unlinked in the same directory, clients
1318 have a large amount of RAM, have many CPUs, the filesystem has
1319 many OSTs, the clients are rebooted frequently, and/or the files
1320 are not accessed by other nodes after being written.
1322 The presence of the problem can be detected by looking at
1323 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1324 smaller than the second (nr_unused) then dirty files will not
1325 be flushed automatically to disk. "sync; sleep 10" should be
1326 run several times on the node before unmounting it to update
1327 Lustre (this is also safe to run on nodes without this problem).
1329 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1330 that can cause this same problem, so customers using that kernel
1331 also need to update the kernel in addition to Lustre. In order
1332 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1334 It is normal that files written just before a client crash (less
1335 than 30s) may not yet have been flushed to disk, even for local
1339 Frequency : frequent on thin XT3 nodes
1341 Description: UUID collision on thin XT3 Linux nodes
1342 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1343 randomly, since we relied on an insufficiently-seeded PRNG.
1348 Description: OSS hangs after "All ost request buffers busy"
1349 Details : A deadlock between quota and journal operations caused OSS
1350 hangs after printing "All ost request buffers busy."
1353 Frequency : always on liblustre builds
1355 Description: Cleanup compiler warnings on liblustre
1358 Frequency : always on liblustre builds on XT3
1360 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1364 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1365 > 32000 subdirectories
1366 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1367 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1368 subdirectory count crosses 32000. This will aid e2fsck to
1369 correctly handle more than 32000 subdirectories.
1372 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1374 Description: System hang on startup
1375 Details : This bug allowed the liblustre (e.g. catamount) client to
1376 return to the app before handling all startup RPCs. This
1377 could leave the node unresponsive to lustre network traffic
1378 and manifested as a server ptllnd timeout.
1380 Severity : enhancement
1382 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1383 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1384 boolean to control whether to print peer debug info when a
1385 client's RPC times out.
1390 Description: make O_SYNC working on 2.6 kernels
1391 Details : 2.6 kernels use different method for mark pages for write,
1392 so need add a code to lustre for O_SYNC work.
1397 Description: Failure to close file and release space on NFS
1398 Details : Put inode details into lock acquired in ll_intent_file_open.
1399 Use mdc_intent_lock in ll_intent_open to properly
1400 detect all kind of errors unhandled by mdc_enqueue.
1405 Description: proc file read during shutdown sometimes raced obd removal,
1407 Details : Add lock to prevent obd access after proc file removal.
1410 Frequency : Only for files larger than 4GB on 32-bit clients.
1412 Description: improperly doing page alignment of locks
1413 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1414 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1419 Description: RPCs being resent when they shouldn't be
1420 Details : Some RPCs that should not be resent are being resent. This
1421 can cause inconsistencies in the RPC state machine. Do not
1422 resend such requests.
1425 Frequency : rare, only with NFS export
1427 Description: Crash on NFS re-export node
1428 Details : under very unusual load conditions an assertion is hit in
1429 ll_intent_file_open()
1432 Frequency : only if OST filesystem is corrupted
1434 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1435 Details : for a short time RPCs with bulk IO are in the replay list,
1436 but replay of bulk IOs is unimplemented. If the OST filesystem
1437 is corrupted due to disk cache incoherency and then replay is
1438 started it is possible to trip an assertion. Avoid putting
1439 committed RPCs into the replay list at all to avoid this issue.
1444 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1445 Details : Large single O_DIRECT read and write calls can fail to allocate
1446 a sufficiently large buffer to process the request. In case of
1447 allocation failure the allocation is retried with a smaller
1448 buffer and broken into smaller requests.
1450 Severity : enhancement
1452 Description: Add -o localflock option to simulate old noflock behaviour.
1453 Details : This will achieve local-only flock/fcntl locks coherentness.
1458 Description: versioning check is incomplete
1459 Details : Checking the version difference of client vs. server, report
1460 error if the gap is too big.
1465 Description: add support PG_writeback bit
1466 Details : add support for PG_writeback bit for Lustre, for more carefull
1467 work with page cache in 2.6 kernel. This also fix some deadlocks
1468 and remove hack for work O_SYNC with 2.6 kernel.
1470 Severity : enhancement
1472 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1473 Details : The uninit_groups feature works in conjunction with the kernel
1474 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1475 up the pass1 processing of e2fsck. This is a read-only feature
1476 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1477 mount filesystems that have had this feature enabled.
1479 Severity : enhancement
1481 Description: Improve multi-block allocation algorithm to avoid fragmentation
1482 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1483 allocation locality and avoid filesystem fragmentation.
1485 ------------------------------------------------------------------------------
1487 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1489 * Support for kernels:
1490 2.6.9-42.0.3.EL (RHEL 4)
1491 2.6.5-7.276 (SLES 9)
1492 2.4.21-47.0.1.EL (RHEL 3)
1493 2.6.12.6 vanilla (kernel.org)
1494 2.6.16.21-0.8 (SLES10)
1495 * Recommended e2fsprogs version: 1.39.cfs2-0
1497 * The backwards-compatible /proc/sys/portals symlink has been removed
1498 in this release. Before upgrading, please ensure that you change
1499 any configuration scripts or /etc/sysctl.conf files that access
1500 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1501 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1502 in advance of the upgrade on any system running Lustre 1.4.6 or
1503 newer, since /proc/sys/lnet was added in that version.
1504 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1508 Frequency : only when quota is used
1510 Description: avoid scanning export list for quota master
1511 Details : Change the algorithms to avoid scanning export list in order
1512 to improve the efficiency.
1515 Frequency : MDS failover only, very rarely
1517 Description: "went back in time" messages on mds failover
1518 Details : The greatest transno may be lost when the current operation
1519 finishes with an error (transno==0) and the client's last_rcvd
1520 record is over-written. Save the greatest transno in the
1521 mds_last_transno for this case.
1524 Frequency : always for specific kernels and striping counts
1526 Description: client may get "Matching packet too big" without ACL support
1527 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1528 when trying to access files in certain configurations. The
1529 clients should in fact be denied when mounting because they do
1530 not understand ACLs.
1533 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1535 Description: many clients connecting with IO in progress causes connect timeouts
1536 Details : Avoid synchronous journal commits to avoid delays caused by many
1537 clients connecting/disconnecting when bulk IO is in progress.
1538 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1539 OST_IO_PORTAL to avoid delays behind potentially many pending
1543 Frequency : occasionally with multiple writers to a single file
1545 Description: shared writes to file may result in wrong size reported by stat()
1546 Details : Allow growing of kms when extent lock is cancelled
1549 Frequency : always with random mmap IO to multi-striped file
1551 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1552 Details : Only if the hole is at the end of OST object so that kms is too
1553 small. Fix is to increase kms accordingly in ll_nopage.
1556 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1558 Description: writes to a missing object would leak memory on the OST
1559 Details : If there is an inconsistency between the MDS and OST filesystems,
1560 such that the MDS references an object that doesn't exist, writes
1561 to that object will leak memory due to incorrect cleanup in the
1562 error handling path, eventually running out of memory on the OST.
1567 Description: Creating too long symlink causes lustre errors
1568 Details : Check symlink and name lengths before sending requests to MDS.
1571 Frequency : only if flock is enabled (not on by default)
1573 Description: posix locks not released on fd closure on 2.6.9+
1574 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1575 kernels, this caused such locks not to be released on fd close and
1576 then assertions on fs unmount about still used locks.
1579 Frequency : MDS failover only, very rarely
1581 Description: clients may get ASSERTION(granted_lock != NULL)
1582 Details : When request was taking a long time, and a client was resending
1583 a getattr by name lock request. The were multiple lock requests
1584 with the same client lock handle and
1585 mds_getattr_name->fixup_handle_for_resent_request found one of the
1586 lock handles but later failed with ASSERTION(granted_lock != NULL).
1591 Description: handle->h_buffer_credits > 0, assertion failure
1592 Details : h_buffer_credits is zero after truncate, causing assertion
1593 failure. This patch extends the transaction or creates a new
1597 Frequency : NFS re-export or patchless client
1598 Bugzilla : 11179, 10796
1599 Description: Crash on NFS re-export node (__d_move)
1600 Details : We do not want to hash the dentry if we don't have a lock.
1601 But if this dentry is later used in d_move, we'd hit uninitialised
1602 list head d_hash, so we just do this to init d_hash field but
1603 leave dentry unhashed.
1606 Frequency : NFS re-export or patchless client
1608 Description: NFS exports has problem with symbolic link
1609 Details : lustre client didn't properly install dentry when re-exported
1610 to NFS or running patchless client.
1613 Frequency : NFS re-export or patchless client
1615 Description: Various nfs/patchless fixes.
1616 Details : fixes reuse disconected alias for lookup process - this fixes
1617 warning "find_exported_dentry: npd != pd",
1618 fix permission error with open files at nfs.
1619 fix apply umask when do revalidate.
1622 Frequency : occasional
1624 Description: Crash on NFS re-export node
1625 Details : calling clear_page() on the wrong pointer triggered oops in
1626 generic_mapping_read().
1629 Frequency : rarely, using O_DIRECT IO
1631 Description: unaligned directio crashes client with LASSERT
1632 Details : check for unaligned buffers before trying any requests.
1635 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1637 Description: stale data returned from RAID cache
1638 Details : If only a small amount of IO is done to the RAID device before
1639 reading it again it is possible to get stale data from the RAID
1640 cache instead of reading it from disk.
1643 Frequency : always for sles10 kernel
1645 Description: sles10 support
1646 Details : ll_follow_link: compile fixes and using of nd_set_link
1647 under newer kernels.
1650 Frequency : depends on arch, kernel and compiler version, always on sles10
1653 Description: recursive or deep enough symlinks cause stack overflow
1654 Details : getting rid of large stack-allocated variable in
1658 Frequency : depends on hardware
1660 Description: lustre write performance loss in the SLES10 kernel
1661 Details : the performance loss is caused by using of write barriers in the
1662 ext3 code. The SLES10 kernel turns barrier support on by
1663 default. The fix is to undo that change for ldiskfs.
1665 ------------------------------------------------------------------------------
1667 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1669 * Support for kernels:
1670 2.6.9-42.0.3EL (RHEL 4)
1671 2.6.5-7.276 (SLES 9)
1672 2.4.21-47.0.1.EL (RHEL 3)
1673 2.6.12.6 vanilla (kernel.org)
1677 Frequency : quota enabled and large files being deleted
1679 Description: releasing more than 4GB of quota at once hangs OST
1680 Details : If a user deletes more than 4GB of files on a single OST it
1681 will cause the OST to spin in an infinite loop. Release
1682 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1687 Description: statfs data retrieved from /proc may be stale or zero
1688 Details : When reading per-device statfs data from /proc, in the
1689 {kbytes,files}_{total,free,avail} files, it may appear
1690 as zero or be out of date.
1693 Frequency : systems with MD RAID1 external journal devices
1695 Description: lconf's call to blkid is confused by RAID1 journal devices
1696 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1697 of returning all block devices that match the journal UUID.
1700 Frequency : always, for aggregate stripe size over 4GB
1702 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1703 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1704 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1707 Frequency : always if "lfs find" used on a local file/directory
1709 Description: "lfs find" segfaults if used on a local file/directory
1710 Details : The case where a directory component was not specified wasn't
1711 handled correctly. Handle this properly.
1714 Frequency : always on ppc64
1716 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1717 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1718 perform endianness conversion.
1721 Frequency : rarely (truncate to non-zero file size after write under load)
1722 Bugzilla : 10730, 10687
1723 Description: Files padded with zeros to next 4K multiple
1724 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1725 it is possible that files that are truncated to a non-zero size
1726 immediately after being written are filled with zero bytes beyond
1727 the truncated size. No file data is lost.
1729 Severity : enhancement
1731 Description: Allow recovery/failover for liblustre clients.
1732 Details : liblustre clients were unaware of failover configurations until
1735 Severity : enhancement
1737 Description: user file locks should fail when not mounting with flock option
1738 Details : Set up an error-returning stub in ll_file_operations.lock field
1739 to prevent incorrect behaviour when client is mounted without
1740 flock option. Also, set up properly f_op->flock field for
1744 Frequency : always on ia64
1746 Description: "lfs df" loops on printing out MDS statfs information
1747 Details : The obd_ioctl_data was not initialized and in some systems
1748 this caused a failure during the ioctl that did not return
1749 an error. Initialize the struct and return an error on failure.
1752 Frequency : SLES 9 only
1754 Description: Error of copying files with lustre special EAs as root
1755 Details : Client side always return success for setxattr call for lustre
1756 special xattr (currently only "trusted.lov").
1759 Frequency : rarely on clusters with both ia64+i386 clients
1761 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1762 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1763 clients are doing concurrent writes to the same file it is
1764 possible that the ia64 clients may LASSERT because the OST
1765 extent locks are not PAGE_SIZE aligned. Ensure that grown
1766 locks are always aligned on the request boundary.
1769 Frequency : specific use, occasional
1771 Description: Overwriting in use executable truncates on-disk binary image
1772 Details : If one node attempts to overwrite an executable in use by
1773 another node, we now correctly return ETXTBSY instead of
1774 truncating the file.
1776 Severity : enhancement
1778 Description: Async OSC create to avoid the blocking unnecessarily.
1779 Details : If a OST has no remain object, system will block on the creating
1780 when need to create a new object on this OST. Now, ways use
1781 pre-created objects when available, instead of blocking on an
1782 empty osc while others are not empty. If we must block, we block
1783 for the shortest possible period of time.
1788 Description: chmod on Lustre root is propagated to other clients
1789 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1790 invalid by the follow_mount time.
1793 Frequency : liblustre clients only
1795 Description: Race in 'instant cancel' lock handling could lead to such locks
1796 never to be granted in case of SMP MDS
1797 Details : Do not destroy not yet granted but cbpending locks in
1801 Frequency : replay/resend of open
1803 Description: non null lock assetion failure in mds_intent_policy
1804 Details : Trying to replay/resend lockless open requests resulted in
1805 mds_open() returning 0 with no lock. Now it sets a flag if
1806 a lock is going to be returned.
1808 Severity : enhancement
1810 Description: Checksum enhancements
1811 Details : New checksum enhancements allow for resending RPCs that failed
1814 Severity : enhancement
1816 Description: Tunables on number of dirty pages in cacche
1817 Details : Allow to set limit on number of dirty pages cached.
1822 Description: client crash on unmount - lock still has references
1823 Details : In some error handling cases it was possible to leak a lock
1824 reference on a client while accessing a file. This was not
1825 harmful to the client during operation, but would cause the
1826 client to crash when the filesystem is unmounted.
1829 Frequency : specific case, rare
1831 Description: ETXTBSY on mds though file not in use by client
1832 Details : ETXTBSY is no longer incorrectly returned when attempting to
1833 chmod or chown a directory that the user previously tried to
1834 execute or a currently-executing binary.
1837 Frequency : extremely rare except on liblustre-based clients
1839 Description: Lustre space not freed when files are deleted
1840 Details : Clean up open-unlinked files after client eviction. Previously
1841 the unlink was skipped and the files remained as orphans.
1846 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1847 Details : In some cases it was possible to send a blocking callback to
1848 a client doing a glimpse, even though that client didn't get
1849 a lock granted. When the glimpse lock is cancelled on the OST
1850 the freed lock is left on the waiting list and corrupted the list.
1853 Frequency : all core dumps
1855 Description: Broke core dumps to lustre
1856 Details : Negative dentry may be unhashed if parent does not have UPDATE
1857 lock, but some callers, e.g. do_coredump, expect dentry to be
1858 hashed after successful create, hash it in ll_create_it.
1860 ------------------------------------------------------------------------------
1862 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1864 * Support for kernels:
1865 2.6.9-42.0.2.EL (RHEL 4)
1866 2.6.5-7.276 (SLES 9)
1867 2.4.21-40.EL (RHEL 3)
1868 2.6.12.6 vanilla (kernel.org)
1872 Frequency : always on RHEL 3
1874 Description: Number of open files grows over time
1875 Details : The number of open files grows over time, whether or not
1876 Lustre is started. This was due to a filp leak introduced
1877 by one of our kernel patches.
1879 ------------------------------------------------------------------------------
1881 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1883 * Support for kernels:
1884 2.6.9-42.EL (RHEL 4)
1885 2.6.5-7.276 (SLES 9)
1886 2.4.21-40.EL (RHEL 3)
1887 2.6.12.6 vanilla (kernel.org)
1892 Bugzilla : 5719, 9635, 9792, 9684
1893 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1894 Details : If a server is under heavy load and cannot reply to new
1895 connection requests before the client resends the (re)connect,
1896 the connection handling code can behave badly if two service
1897 threads are concurrently handing separate (re)connections from
1898 the same client. Add better locking to the connection handling
1899 code, and ensure that only a single connection will be processed
1900 for a given client UUID, even if the lock is dropped.
1902 Severity : enhancement
1904 Description: add TCP zero-copy support to kernel
1905 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1906 sends if the hardware supports scatter-gather and checksumming.
1907 This allows socklnd to do client-write and server-read more
1908 efficiently and reduce CPU utilization from skbuf copying.
1911 Frequency : only if NFS exporting from client
1913 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1914 Details : If setting attributes on a file created under NFS that had
1915 never been opened it would be possible to oops the client
1916 if the file had no objects.
1919 Frequency : always for liblustre
1921 Description: liblustre client does MDS+OSTs setattr RPC for each write
1922 Details : When doing a write from a liblustre client, the client
1923 incorrectly issued an RPC to the MDS and each OST the file was
1924 striped over in order to update the timestamps. When writing
1925 with small chunks and many clients this could overwhelm the MDS
1926 with RPCs. In all cases it would slow down the write because
1927 these RPCs are unnecessary.
1929 Severity : enhancement
1931 Description: allow number of MDS service threads to be changed at module load
1932 Details : It is now possible to change the number of MDS service threads
1933 running. Adding "options mds mds_num_threads={N}" to the MDS's
1934 /etc/modprobe.conf will set the number of threads for the next
1935 time Lustre is restarted (assuming the "mds" module is also
1936 reloaded at that time). The default number of threads will
1937 stay the same, 32 for most systems.
1942 Description: OST crash if filesystem is unformatted or corrupt
1943 Details : If an OST is started on a device that has never been formatted
1944 or if the filesystem is corrupt and cannot even mount then the
1945 error handling cleanup routines would dereference a NULL pointer.
1950 Description: NULL pointer deref in llap_from_page.
1951 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1952 incorrect) mapping if the page was truncated/reclaimed while it was
1953 searched for. Check for this condition and skip such pages when
1954 doing readahead. Introduce extra check to llap_from_page() to
1955 verify page->mapping->host is non-NULL (so page is not anonymous).
1958 Frequency : Sometimes when using sys_sendfile
1960 Description: "page not covered by a lock" warnings from ll_readpage
1961 Details : sendfile called ll_readpage without right page locks present.
1962 Now we introduced ll_file_sendfile that does necessary locking
1963 around call to generic_file_sendfile() much like we do in
1967 Frequency : with certain MDS communication failures at client mount time
1969 Description: NULL pointer deref after failed client mount
1970 Details : a client connection request may delayed by the network layer
1971 and not be sent until after the PTLRPC layer has timed out the
1972 request. If the client fails the mount immediately it will try
1973 to clean up before the network times out the request. Add a
1974 reference from the request import to the obd device and delay
1975 the cleanup until the network drops the request.
1978 Frequency : occasionally during client (re)connect
1980 Description: assertion failure during client (re)connect
1981 Details : processing a client connection request may be delayed by the
1982 client or server longer than the client connect timeout. This
1983 causes the client to resend the connection request. If the
1984 original connection request is replied in this interval, the
1985 client may trip an assertion failure in ptlrpc_connect_interpret()
1986 which thought it would be the only running connect process.
1989 Frequency : only with obd_echo servers and clients that are rebooted
1991 Description: kernel BUG accessing uninitialized data structure
1992 Details : When running an obd_echo server it did not start the ping_evictor
1993 thread, and when a client was evicted an uninitialized data
1994 structure was accessed. Start the ping_evictor in the RPC
1995 service startup instead of the OBD startup.
1997 Severity : enhancement
1998 Bugzilla : 10193 (patchless)
1999 Description: Remove dependency on various unexported kernel interfaces.
2000 Details : No longer need reparent_to_init, exit_mm, exit_files,
2001 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2004 Frequency : rare (only users of deprecated and unsupported LDAP config)
2006 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2007 Details : LDAP apparently contains 'lustreName' attributes instead of
2008 'name'. A simple remapping of the name is sufficient.
2011 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2013 Description: waiting_locks_callback trips kernel BUG if client is evicted
2014 Details : Running with the dump_on_timeout debug flag turned on makes
2015 it possible that the waiting_locks_callback() can try to dump
2016 the Lustre kernel debug logs from an interrupt handler. Defer
2017 this log dumping to the expired_lock_main() thread.
2019 Severity : enhancement
2021 Description: Support NFS exporting on 2.6 kernels.
2022 Details : Implement non-rawops metadata methods for NFS server to use without
2023 changing NFS server code.
2026 Frequency : very rare (synthetic metadata workload only)
2028 Description: two racing renames might cause an MDS thread to deadlock
2029 Details : Running the "racer" program may cause one MDS thread to rename
2030 a file from being the source of a rename to being the target of
2031 a rename at exactly the same time that another thread is doing
2032 so, and the second thread has already enqueued these locks after
2033 doing a lookup of the target and is trying to relock them in
2034 order. Ensure that we don't try to re-lock the same resource.
2037 Frequency : only very large systems with liblustre clients
2039 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2040 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2041 all OSTs in parallel. This allows the eviction of stale liblustre
2042 clients to proceed much faster than if they were done in series,
2043 and also offers similar improvements for other set_info RPCs.
2048 Description: excessive CPU usage during initial read phase on client
2049 Details : During the initial read phase on a client, it would agressively
2050 retry readahead on the file, consuming too much CPU and impacting
2051 performance (since 1.4.5.8). Improve the readahead algorithm
2052 to avoid this, and also improve some other common cases (read
2053 of small files in particular, where "small" is files smaller than
2054 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2059 Description: MDS crash when receiving packet with unknown intent.
2060 Details : Do not LBUG in unknown intent case, just return -EFAULT
2062 Severity : enhancement
2063 Bugzilla : 9293, 9385
2064 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2065 Details : Do not serialize getattr (non-intent version) and statfs.
2068 Frequency : occasional, when OST network is overloaded/intermittent
2070 Description: client evicted by OST after bulk IO timeout
2071 Details : If a client sends a bulk IO request (read or write) the OST
2072 may evict the client if it is unresposive to its data GET/PUT
2073 request. This is incorrect if the network is overloaded (takes
2074 too long to transfer the RPC data) or dropped the OST GET/PUT
2075 request. There is no need to evict the client at all, since
2076 the pinger and/or lock callbacks will handle this, and the
2077 client can restart the bulk request.
2080 Frequency : Always when mmapping file with no objects
2082 Description: client crashes when mmapping file with no objects
2083 Details : Check that we actually have objects in a file before doing any
2084 operations on objects in ll_vm_open, ll_vm_close and
2090 Description: Request leak when working with deleted CWD
2091 Details : Introduce advanced request refcount tracking for requests
2092 referenced from lustre intent.
2094 Severity : Enhancement
2096 Description: Cache open file handles on client.
2097 Details : MDS now will return special lock along with openhandle, if
2098 requested and client is allowed to hold openhandle, even if unused,
2099 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2100 closing files for every read/write openration.
2102 Severity : Enhancement
2104 Description: Cache open negative dentries on client when possible.
2105 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2106 negative dentries on lock revocation.
2111 Description: Remounting a client read-only wasn't possible with a zconf mount
2112 Details : It wasn't possible to remount a client read-only with llmount.
2114 Severity : enhancement
2115 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2116 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2117 MPICH is included to add Lustre-specific ADIO interfaces.
2118 This is based closely on the UFS ADIO layer and only differs
2119 in file creation, in order to allow the OST striping to be set.
2120 This is user-contributed code and not supported by CFS.
2125 Description: extended inode attributes (immutable, append-only) work improperly
2126 when 2.4 and 2.6 kernels are used on client/server or vice versa
2127 Details : Introduce kernel-independent values for these flags.
2129 Severity : enhancement
2132 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2133 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2134 granularity. Now, Fractional MB granularity is be supported,
2135 this is very useful for low memory system.
2137 Severity : enhancement
2139 Description: Getattr by fid
2140 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2141 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2142 have locks and avoids getattr rpc requests for it on every stat.
2145 Frequency : Always, for filesystems larger than 2TB
2147 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2148 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2149 filesystems larger than 2TB to be mounted with mballoc enabled.
2152 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2154 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2155 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2156 filesystems larger than 2TB is mounted then the kernel will
2157 silently corrupt the start of the filesystem. CONFIG_LBD is
2158 enabled for all CFS-supported kernels, but the possibility of
2159 this happening with a modified kernel config exists.
2161 Severity : enhancement
2163 Description: add client O_DIRECT support for 2.6 kernels
2164 Details : It is now possible to do O_DIRECT reads and writes to files
2165 in the Lustre client mountpoint on 2.6 kernel clients.
2167 Severity : enhancement
2169 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2170 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2171 parallel, not waiting for response from every OST before sending
2172 a rpc to the next OST.
2177 Description: setattr vs write race when updating file timestamps
2178 Details : Client processes that update a file timestamp into the past
2179 right after writing to the file (e.g. tar) it is possible that
2180 the updated file modification time can be reset to the current
2181 time due to a race between processing the setattr and write RPC.
2183 Severity : enhancement
2185 Description: Bring 'lfs find' closer in line with regular Linux find.
2186 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2187 -print0 options and obtains all the needed info through the lustre
2190 Severity : enhancement
2192 Description: support up to 1024 configured devices on one node
2193 Details : change obd_dev array from statically allocated to dynamically
2194 allocated structs as they are first used to reduce memory usage
2199 Description: Flush dirty partially truncated pages during truncate
2200 Details : Immediatelly flush partially truncated pages in filter_setattr,
2201 this way we completely avoid having any pages in page cache on OST
2202 and can retire ugly workarounds during writes to flush such pages.
2207 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2208 Details : protect inode from truncation within vfs_unlink() context
2209 just take a reference before calling vfs_unlink() and release it
2210 when parent's i_sem is free.
2215 Description: last_id value checked outside lock on OST caused LASSERT failure
2216 Details : If there were multiple MDS->OST object precreate requests in
2217 flight, it was possible that the OST's last object id was checked
2218 outside a lock and incorrectly tripped an assertion. Move checks
2219 inside locks, and discard old precreate requests.
2222 Frequency : always, if extents are used on OSTs
2224 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2225 Details : OSTs using the extents format would not zero the high 16 bits of
2226 the index physical block number. This is not a problem for any
2227 OST filesystems smaller than 16TB, and no kernels support ext3
2228 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2229 new/modified files) and can be fixed for existing filesystems
2230 with e2fsprogs-1.39-cfs1.
2235 Description: import connection selection may be incorrect if timer wraps
2236 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2237 import connections to be ignored if the 32-bit jiffies counter
2238 wraps. Use a 64-bit jiffies counter.
2241 Frequency : very large clusters immediately after boot
2243 Description: LNET request buffers exhausted under heavy short-term load
2244 Details : If a large number of client requests are generated on a service
2245 that has previously never seen so many requests it is possible
2246 that the request buffer growth cannot keep up with the spike in
2247 demand. Instead of dropping incoming requests, they are held in
2248 the LND until the RPC service can accept more requests.
2251 Frequency : Sometimes during replay
2253 Description: Assertion failure in ll_local_open after replay.
2254 Details : If replay happened on an open request reply before we were able
2255 to set replay handler, reply will become not swabbed tripping the
2256 assertion in ll_local_open. Now we set the handler right after
2257 recognising of open request
2260 Frequency : very rare
2262 Description: kernel reports "badness in vsnprintf"
2263 Details : Reading from the "recovery_status" /proc file in small chunks
2264 may cause a negative length in lprocfs_obd_rd_recovery_status()
2265 call to vsnprintf() (which is otherwise harmless). Exit early
2266 if there is no more space in the output buffer.
2268 Severity : enhancement
2270 Description: clear OBD RPC statistics by writing to them
2271 Details : It is now possible to clear the OBD RPC statistics by writing
2272 to the "stats" file.
2277 Description: Client mtime is not the same on different clients after utimes
2278 Details : In some cases, the client was using the utimes() syscall on
2279 a file cached on another node. The clients now validate the
2280 ctime from the MDS + OSTs to determine which one is right.
2285 Description: Inability to activate failout mode
2286 Details : lconf script incorrectly assumed that in pythong string's numeric
2287 value is used in comparisons.
2290 Frequency : always with multiple stripes per file
2292 Description: Inefficient object allocation for mutli-stripe files
2293 Details : When selecting which OSTs to stripe files over, for files with
2294 a stripe count that divides evenly into the number of OSTs,
2295 the MDS is always picking the same starting OST for each file.
2296 Return the OST selection heuristic to the original design.
2301 Description: mount failures may take full timeout to return an error
2302 Details : Under some heavy load conditions it is possible that a
2303 failed mount can wait for the full obd_timeout interval,
2304 possibly several minutes, before reporting an error.
2305 Instead return an error as soon as the status is known.
2307 Frequency : quota enabled and large files being deleted
2309 Description: releasing more than 4GB of quota at once hangs OST
2310 Details : If a user deletes more than 4GB of files on a single OST it
2311 will cause the OST to spin in an infinite loop. Release
2312 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2317 Description: statfs data retrieved from /proc may be stale or zero
2318 Details : When reading per-device statfs data from /proc, in the
2319 {kbytes,files}_{total,free,avail} files, it may appear
2320 as zero or be out of date.
2323 Frequency : systems with MD RAID1 external journal devices
2325 Description: lconf's call to blkid is confused by RAID1 journal devices
2326 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2327 of returning all block devices that match the journal UUID.
2330 Frequency : always, for aggregate stripe size over 4GB
2332 Description: assertion fails when trying to use 4GB stripe size
2333 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2334 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2337 Frequency : always on ppc64
2339 Description: the first write on an ext3 filesystem with mballoc got stuck
2340 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2341 perform endianness conversion.
2343 ------------------------------------------------------------------------------
2345 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2347 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2348 INTEROPERATE with older versions automatically. Please read the
2349 user documentation before upgrading any part of a live system.
2350 * WARNING: Lustre networking configuration changes are required with
2351 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2354 * Support for kernels:
2355 2.6.9-22.0.2.EL (RHEL 4)
2356 2.6.5-7.244 (SLES 9)
2357 2.6.12.6 vanilla (kernel.org)
2360 Severity : enhancement
2361 Bugzilla : 7981/8208
2362 Description: Introduced Lustre Networking (LNET)
2363 Details : LNET is new networking infrastructure for Lustre, it includes
2364 a reorganized network configuration mode (see the user
2365 documentation for full details) as well as support for routing
2366 between different network fabrics. Lustre Networking Devices
2367 (LNDs) for the supported network fabrics have also been
2368 created for this new infrastructure.
2370 Severity : enhancement
2371 Description: Introduced Access control lists
2372 Details : clients can set ACLs on files and directories in order to have
2373 more fine-grained permissions than the standard Unix UGO+RWX.
2374 The MDS must be started with the "-o acl" mount option.
2376 Severity : enhancement
2377 Description: Introduced filesystem quotas
2378 Details : Administrators may now establish per-user quotas on the
2381 Severity : enhancement
2383 Description: Configuration change for the XT3
2384 The PTLLND is now used to run Lustre over Portals on the XT3
2385 The configure option(s) --with-cray-portals are no longer used.
2386 Rather --with-portals=<path-to-portals-includes> is used to
2387 enable building on the XT3. In addition to enable XT3 specific
2388 features the option --enable-cray-xt3 must be used.
2393 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2394 Details : A race between allocating a new llcd and re-getting the llcd_lock
2395 allowed another thread to grab newly-allocated llcd.
2397 Severity : enhancement
2399 Description: 2.6 OST async journal commit and locking fix to improve performance
2400 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2401 2.6 kernels are now async as they already were in 2.4 kernels so
2402 that they can commit concurrently with the network bulk transfer.
2403 For block-allocated files the filter allocation semaphore is held
2404 to avoid filesystem fragmentation during allocation. BKL lock
2405 removed for 2.6 xattr operations where it is no longer needed.
2410 Description: lconf incorrectly determined whether two IP networks could talk
2411 Details : In some more complicated routing and multiple-network
2412 configurations, lconf will avoid trying to make a network
2413 connection to a disjoint part of the IP space. It was doing the
2414 math incorrectly for one set of cases.
2419 Description: Fix for potential infinite loop processing records in an llog.
2420 Details : If an llog record is corrupted/zeroed, it is possible to loop
2421 forever in llog_process(). Validate the llog record length
2422 and skip the remainder of the block on error.
2425 Frequency : occasional (liblustre only)
2427 Description: liblustre could not open files whose last component is a symlink
2428 Details : sysio_path_walk() would incorrectly pass the open intent to
2429 intermediate path components.
2432 Frequency : rare (liblustre only with non-standard tuning)
2433 Bugzilla : 7201 (7350)
2434 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2435 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2436 after it has been released, only do this for referenced locks
2438 Severity : enhancement
2440 Description: specifying an (invalid) directory default stripe_size of -1
2441 would reset the directory default striping
2442 Details : stripe_size -1 was used internally to signal directory stripe
2443 removal, now use "all default" to signal dir stripe removal
2444 as a directory striping of "all default" is not useful
2447 Frequency : common for large clusters running liblustre clients
2449 Description: doing an ls when liblustre clients are running is slow
2450 Details : sending a glimpse AST to a liblustre client waits for every AST
2451 to time out, as liblustre clients will not respond. Since they
2452 cannot cache data we refresh the OST lock LVB from disk instead.
2454 Severity : enhancement
2456 Description: doing an ls at the same time as file IO can be slow
2457 Details : enqueue and other "small" requests can be blocked behind many
2458 large IO requests. Create a new OST IO portal for non-IO
2459 requests so they can be processed faster.
2462 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2464 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2465 created file with mode 000
2466 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2467 allow an ftruncate() as a non-root user to a file with mode 000.
2468 Lustre now respects this flag to disable mode checks when
2469 truncating a file owned by the user
2472 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2475 Description: Revoking locks from clients that went dead or catatonic might take
2477 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2478 cancellation of such locks instant on servers without waiting for
2479 any reply from clients. Clients drops these locks when cancel
2480 notification from server is received without replying.
2483 Frequency : liblustre-only, when liblustre client dies or becomes busy
2485 Description: Doing ls on Linux clients can take a long time with active
2487 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2488 granting such locks to it in the first place if possible. Locks
2489 are taken by proxy on the OST during the read or write and
2490 dropped immediately afterward. Add connect flags handling, do
2491 not grant locks to liblustre clients for glimpse ASTs.
2493 Severity : enhancement
2495 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2496 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2497 sequential reads and behaves badly with applications doing random
2498 reads. Improve it by reading ahead at least read region, and
2499 avoiding excessive large RPC for small reads.
2501 Severity : enhancement
2503 Description: Creating more than 1000 files for a single job may cause a load
2504 imbalance on the OSTs if there are also a large number of OSTs.
2505 Details : qos_prep_create() uses an OST index reseed value that is an
2506 even multiple of the number of available OSTs so that if the
2507 reseed happens in the middle of the object allocation it will
2508 still utilize the OSTs as uniformly as possible.
2513 Description: OST or MDS may oops in ping_evictor_main()
2514 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2515 but doesn't restart at beginning of obd_exports_timed list
2518 Severity : enhancement
2520 Description: improve by-nid export eviction on the MDS and OST
2521 Details : allow multiple exports with the same NID to be evicted at one
2522 time without re-searching the exports list.
2525 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2527 Description: MDS may oops in groups_free()
2528 Details : in rare race conditions a newly allocated group_info struct is
2529 freed again, and this can be NULL. The 2.4 compatibility code
2530 for groups_free() checked for a NULL pointer, but 2.6 did not.
2533 Frequency : common for liblustre clients doing little filesystem IO
2534 Bugzilla : 9352, 7313
2535 Description: server may evict liblustre clients accessing contended locks
2536 Details : if a client is granted a lock or receives a completion AST
2537 with a blocking AST already set it would not reply to the AST
2538 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2541 Frequency : lfs setstripe, only systems with more than 160 OSTs
2543 Description: unable to set striping with a starting offset beyond OST 160
2544 Details : llapi_create_file() incorrectly limited the starting stripe
2545 index to the maximum single-file stripe count.
2548 Frequency : LDAP users only
2550 Description: lconf did not handle in-kernel recovery with LDAP properly
2551 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2553 Severity : enhancement
2555 Description: bind OST threads to NUMA nodes to improve performance
2556 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2557 node and do their allocations there to localize memory access
2559 Severity : enhancement
2561 Description: llmount can determine client NID directly from Myrinet (GM)
2562 Details : the client NID code from gmnalnid was moved directly into
2563 llmount, removing the need to use this or specifying the
2564 client NID explicitly when mounting GM clients with zeroconf
2567 Frequency : if client is started with down MDS
2569 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2570 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2571 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2572 ptlrpc_import_delayed_req() because we didn't check if the
2573 request was interrupted, and we also didn't break out of the
2574 event loop if there was no timeout
2579 Description: data loss during non-page-aligned writes to a single file from
2580 both multiple nodes and multiple threads on one node at same time
2581 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2582 inconsistency led to false short-reads, that were cached and later
2583 used by ->prepare_write() to fill in partially written page,
2584 leading to data loss.
2587 Frequency : always, if lconf --abort_recovery used
2589 Description: lconf --abort_recovery fails with 'Operation not supported'
2590 Details : lconf was attempting to abort recovery on the MDT device and not
2593 Severity : enhancement
2595 Description: remove cleanup logs
2596 Details : replace lconf-generated cleanup logs with lustre internal
2597 cleanup routines. Eliminates the need for client-cleanup and
2600 Severity : enhancement
2602 Description: add support for EAs (user and system) on lustre filesystems
2603 Details : it is now possible to store extended attributes in the Lustre
2604 client filesystem, and with the user_xattr mount option it
2605 is possible to allow users to store EAs on their files also
2607 Severity : enhancement
2609 Description: Add possibility (config option) to show minimal available OST free
2611 Details : When compiled with --enable-mindf configure option, statfs(2)
2612 (and so, df) will return least minimal free space available from
2613 all OSTs as amount of free space on FS, instead of summary of
2614 free spaces of all OSTs.
2616 Severity : enhancement
2618 Description: do not expand extent locks acquired on OST-side
2619 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2620 by server: they are not cached anyway.
2623 Frequency : when mmap is used/binaries executed from Lustre
2625 Description: Unmmap pages before throwing them away from read cache.
2626 Details : llap_shrink cache now attempts to unmap pages before discarding
2627 them (if unmapping failed - do not discard). SLES9 kernel has
2628 extra checks that trigger if this unmapping is not done first.
2633 Description: lconf didn't resolve symlinks before checking to see whether a
2634 given mountpoint was already in use
2637 Frequency : when migrating failover services
2638 Bugzilla : 6395, 9514
2639 Description: When migrating a subset of services from a node (e.g. failback
2640 from a failover service node) the remaining services would
2641 time out and evict clients.
2642 Details : lconf --force (implied by --failover) sets the global obd_timeout
2643 to 5 seconds in order to quickly disconnect, but this caused
2644 other RPCs to time out too quickly. Do not change the global
2645 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2647 Severity : enhancement
2648 Frequency : if MDS is started with down OST
2649 Bugzilla : 9439,5706
2650 Description: Allow startup/shutdown of an MDS without depending on the
2651 availability of the OSTs.
2652 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2653 Add appropriate locking and lov-osc refcounts for safe
2654 cleaning. Add osc abort_inflight calls in case the
2655 synchronize never started.
2658 Frequency : occasional (Cray XT3 only)
2660 Description: root not authorized to access files in CRAY_PORTALS environment
2661 Details : The client process capabilities were not honoured on the MDS in
2662 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2663 been accessed by an authorized user then root was able to access
2664 the file on the local client also. The root user capabilities
2665 are now allowed on the MDS, as this environment has secure UID.
2668 Frequency : occasional
2670 Description: ldiskfs "too long searching" message happens too often
2671 Details : A debugging message (otherwise harmless) prints too often on
2672 the OST console. This has been reduced to only happen when
2673 there are fragmentation problems on the filesystem.
2678 Description: Division by zero in statfs when all OSCs are inactive
2679 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2680 lov_statfs divides by value returned by lov_get_stripecnt().
2684 Bugzilla : 9489, 3273
2685 Description: First write from each client to each OST was only 4kB in size,
2686 to initialize client writeback cache, which caused sub-optimal
2687 RPCs and poor layout on disk for the first writen file.
2688 Details : Clients now request an initial cache grant at (re)connect time
2689 and so that they can start streaming writes to the cache right
2690 away and always do full-sized RPCs if there is enough data.
2691 If the OST is rebooted the client also re-establishes its grant
2692 so that client cached writes will be honoured under the grant.
2697 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2698 Details : Now I/O RPCs go to different portal number and (presumably) fast
2699 lock requests (and glimses) and other RPCs get their own service
2700 threads pool that should be able to service those RPCs
2703 Severity : enhancement
2705 Description: Ability to exchange lustre version between client and servers and
2706 issue warnings at client side if client is too old. Also for
2707 liblustre clients there is ability to refuse connection of too old
2709 Details : New 'version' field is added to connect data structure that is
2710 filled with version info. That info is later checked by server and
2714 Frequency : rare, liblustre only.
2715 Bugzilla : 9296, 9581
2716 Description: Two simultaneous writes from liblustre at offset within same page
2717 might proceed at the same time overwriting eachother with stale
2719 Details : I/O lock withing llu_file_prwv was released too early, before data
2720 actually was hitting the wire. Extended lock-holding time until
2721 server acknowledges receiving data.
2724 Frequency : extremely rare. Never observed in practice.
2726 Description: avoid generating lustre_handle cookie of 0.
2727 Details : class_handle_hash() generates handle cookies by incrementing
2728 global counter, and can hit 0 occasionaly (this is unlikely, but
2729 not impossible, because initial value of cookie counter is
2730 selected randonly). Value of 0 is used as a sentinel meaning
2731 "unassigned handle" --- avoid it. Also coalesce two critical
2732 sections in this function into one.
2734 Severity : enhancement
2736 Description: allow liblustre clients to delegate truncate locking to OST
2737 Details : To avoid overhead of locking, liblustre client instructs OST to
2738 take extent lock in ost_punch() on client's behalf. New connection
2739 flag is added to handle backward compatibility.
2741 Severity : enhancement
2742 Bugzilla : 4928, 7341, 9758
2743 Description: allow number of OST service threads to be specified
2744 Details : a module parameter allows the number of OST service threads
2745 to be specified via "options ost ost_num_threads={N}" in the
2746 OSS's /etc/modules.conf or /etc/modprobe.conf.
2750 Bugzilla : 6146, 9635, 9895
2751 Description: servers crash with bad pointer in target_handle_connect()
2752 Details : In rare cases when a client is reconnecting it was possible that
2753 the connection request was the last reference for that export.
2754 We would temporarily drop the export reference and get a new
2755 one, but this may have been the last reference and the export
2756 was just destroyed. Get new reference before dropping old one.
2758 Severity : enhancement
2759 Frequency : if client is started with failover MDS
2761 Description: Allow multiple MDS hostnames in the mount command
2762 Details : Try to read the configuration from all specified MDS
2763 hostnames during a client mount in case the "primary"
2766 Severity : enhancement
2768 Description: Stop sending data to evicted clients as soon as possible.
2769 Details : Check if the client we are about to send or are sending data to
2770 was evicted already. (Check is done every second of waiting,
2771 for which l_wait_event interface was extended to allow checking
2772 of exit condition at specified intervals).
2775 Frequency : rare, normally only when NFS exporting is done from client
2777 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2779 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2782 Frequency : always, if srand() is called before liblustre initialization
2784 Description: Liblustre uses system PRNG disturbing its usage by user application
2785 Details : Introduce internal to lustre fast and high-quality PRNG for
2786 lustre usage and make liblustre and some other places in generic
2787 lustre code to use it.
2789 Severity : enhancement
2790 Bugzilla : 9477, 9557, 9870
2791 Description: Verify that the MDS configuration logs are updated when xml is
2792 Details : Check if the .xml configuration logs are newer than the config
2793 logs stored on the MDS and report an error if this is the case.
2794 Request --write-conf, or allow starting with --old_conf.
2796 Severity : enhancement
2798 Description: Handle symlinks in the path when checking if Lustre is mounted.
2799 Details : Resolve intermediate symlinks when checking if a client has
2800 mounted a filesystem to avoid duplicate client mounts.
2805 Description: lconf can hit an error exception but still return success.
2806 Details : The lconf command catches the Command error exception at the top
2807 level script context and will exit with the associated exit
2808 status, but doesn't ensure that this exit status is non-zero.
2813 Description: failure of ptlrpc thread startup can cause oops
2814 Details : Starting a ptlrpc service thread can fail if there are a large
2815 number of threads or the server memory is very fragmented.
2816 Handle this without oopsing.
2819 Frequency : always, only if liblustre and non-default acceptor port was used
2821 Description: liblustre cannot connect to servers with non-default acceptor port
2822 Details : tcpnal_set_default_params() was not called and was therefore
2823 ignoring the environment varaible TCPNAL_PORT, as well as other
2824 TCPNAL_ environment variables
2829 Description: two objects could be created on the same OST for a single file
2830 Details : If an OST is down, in some cases it was possible to create two
2831 objects on a single OST for a single file. No problems other
2832 than potential performance impact and spurious error messages.
2836 Bugzilla : 5681, 9562
2837 Description: Client may oops in ll_unhash_aliases
2838 Details : Client dcache may become inconsistent in race condition.
2839 In some cases "getcwd" can fail if the current directory is
2845 Description: Inode refcounting problems in NFS export code
2846 Details : link_raw functions used to call d_instantiate without obtaining
2847 extra inode reference first.
2851 Bugzilla : 9942, 9903
2852 Description: Referencing freed requests leading to crash, memleaks with NFS.
2853 Details : We used to require that call to ll_revalidate_it was always
2854 followed by ll_lookup_it. Also with revalidate_special() it is
2855 possible to call ll_revalidate_it() twice for the same dentry
2856 even if first occurence returned success. This fix changes semantic
2857 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2858 reference on a request referred from the intent.
2859 ll_intent_release() then releases such a request.
2862 Frequency : rare, normally benchmark loads only
2864 Description: unlinked inodes were kept in memory on the client
2865 Details : If a client is repeatedly creating and unlinking files it
2866 can accumulate a lot of stale inodes in the inode slab cache.
2867 If there is no other client load running this can cause the
2868 client node to run out of memory. Instead flush old inodes
2869 from client cache that have the same inode number as a new inode.
2872 Frequency : SLES9 2.6.5 kernel and long filenames only
2873 Bugzilla : 9969, 10379
2874 Description: utime reports stale NFS file handle
2875 Details : SLES9 uses out-of-dentry names in some cases, which confused
2876 the lustre dentry revalidation. Change it to always use the
2880 Frequency : rare, unless heavy write-truncate concurrency is continuous
2881 Bugzilla : 4180, 6984, 7171, 9963, 9331
2882 Description: OST becomes very slow and/or deadlocked during object unlink
2883 Details : filter_destroy() was holding onto the parent directory lock
2884 while truncating+unlinking objects. For very large objects this
2885 may block other threads for a long time and slow overall OST
2886 responsiveness. It may also be possible to get a lock ordering
2887 deadlock in this case, or run out of journal credits because of
2888 the combined truncate+unlink. Solution is to do object truncate
2889 first in one transaction without parent lock, and then do the
2890 final unlink in a new transaction with the parent lock. This
2891 reduces the lock hold time dramatically.
2894 Frequency : rare, 2.4 kernels only
2896 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2897 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2898 before unmounting their filesystem in order to avoid deadlock.
2899 The kernel_locked() function in 2.4 kernels only checks whether
2900 the kernel lock is held, not whether it is this process that is
2901 holding it as 2.6 kernels do.
2906 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2907 Details : The client ptlrpc code may be trying to reconnect to a down
2908 server before a previous connection attempt has timed out.
2909 Increase the reconnect interval to be longer than the connection
2910 timeout interval to avoid sending duplicate connections to
2914 Frequency : echo_client brw_test command
2916 Description: fix echo_client to work with OST preallocated code
2917 Details : OST preallocation code (5137) didn't take echo_client IO path
2918 into account: echo_client calls filter methods outside of any
2919 OST thread and, hence, there is no per-thread preallocated
2920 pages and buffers to use. Solution: hijack pga pages for IO. As
2921 a byproduct, this avoids unnecessary data copying.
2925 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2926 Description: Client can oops in mdc_commit_close() after open replay
2927 Details : It was possible for the MDS to return an open request with no
2928 transaction number in mds_finish_transno() if the client was
2929 evicted, but without actually returning an error. Clients
2930 would later try to replay that open and may trip an assertion
2931 Simplify the client close codepath, and always return an error
2932 from the MDS in case the open is not successful.
2935 Frequency : rare, 2.6 OSTs only
2937 Description: OST may deadlock under high load on fragmented files
2938 Details : If there was a heavy load and highly-fragmented OST filesystems
2939 it was possible to have all the OST threads deadlock waiting on
2940 allocation of biovecs, because the biovecs were not released
2941 until the entire RPC IO was completed. Instead, release biovecs
2942 as soon as they are complete to ensure forward IO progress.
2944 Severity : enhancement
2946 Description: Support for specifying external journal device at mount
2947 Details : If an OST or MDS device is formatted with an external journal
2948 device, this device major/minor is stored in the ext3 superblock
2949 and may not be valid for failover. Allow detecting and
2950 specifying the external journal at mount time.
2955 Description: Mounting an MDS with pending unlinked files may cause oops
2956 Details : target_finish_recovery() calls mds_postrecov() which returned
2957 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2958 considers this an error and immediately begins cleaning up the
2959 lov, just after starting the mds_lov process
2961 Severity : enhancement
2963 Description: Implement 'lfs df' to report actual free space on per-OST basis
2964 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2965 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2966 usage of objects; '-h' to report in human readable format.
2968 ------------------------------------------------------------------------------
2970 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2977 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2978 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2979 at startup during orphan unlinking. Instead, skip buddy update
2980 before setup, as it will be regenerated after recovery is complete.
2985 Description: If an OST is inactive, its locks might reference stale inodes.
2986 Details : lov_change_cbdata() must iterate over all namespaces, even if
2987 they are inactive to clear inode references from the lock.
2989 Severity : enhancement
2990 Frequency : occasional, if non-standard max_dirty_mb used
2992 Description: Client will block write RPCs if not enough grant
2993 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2994 then the client will block writes while waiting for another RPC
2995 to complete instead of consuming its dirty limit. With change
2996 we get improved performance when max_dirty_mb is small.
2998 Severity : enhancement
2999 Bugzilla : 3389, 6253
3000 Description: Add support for supplementary groups on the MDS.
3001 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3002 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3003 lookups for user supplementary groups into a cache.
3007 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3008 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3009 for liblustre O_EXCL is enforced by MDS.
3013 Description: readdir never returns NULL in liblustre.
3014 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3015 dentry in struct dirent.
3019 Frequency : liblustre only, depends on application IO pattern
3020 Description: liblustre clients evicted if not contacting servers
3021 Details : Don't put liblustre clients into the ping_evictor list, so
3022 they will not be evicted by the pinger ever.
3024 Severity : enhancement
3026 Description: Add ability to evict clients by NID from MDS.
3027 Details : By echoing "nid:$NID" string into
3028 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3029 $NID would be instantly evicted from this MDS and from all active
3030 OSTs connected to it.
3034 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3035 Details : lookup_it_finish() used to query file size from OSTs that was not
3040 Description: service threads change working directory to that of init
3041 Details : Starting lustre service threads may pin the working directory
3042 of the parent thread, making that filesystem busy. Threads
3043 now change to the working directory of init to avoid this.
3047 Frequency : during shutdown only
3048 Description: shutdown with a failed MDS or OST can cause unmount to hang
3049 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3054 Frequency : 2.6 only
3055 Description: chmod/chown may include an extra supplementary group
3056 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3057 supplementary group and if none is specified this is used.
3060 Bugzilla : 5479 (6816)
3062 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3063 Details : If lookup is in progress on a file that is unlinked we might try
3064 to revalidate the inode and fail in revalidate after lookup is
3065 complete and ll_file_open() enqueues the open again but
3066 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3069 Frequency : always, if lconf --abort_recovery used
3071 Description: lconf --abort_recovery fails with 'Operation not supported'
3072 Details : lconf was attempting to abort recovery on the MDT device and not
3075 ------------------------------------------------------------------------------
3077 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3082 Frequency : rare (only unsupported configurations with a node running as an
3084 Bugzilla : 6514, 5137
3085 Description: Mounting a Lustre file system on a node running as an OST could
3087 Details : OSTs now preallocates memory needed to write out data at
3088 startup, instead of when needed, to avoid having to
3089 allocate memory in possibly low memory situations.
3090 Specifically, if the file system is mounted on on OST,
3091 memory pressure could force it to try to write out data,
3092 which it needed to allocate memory to do. Due to the low
3093 memory, it would be unable to do so and the node would
3094 become unresponsive.
3096 Severity : enhancement
3098 Description: Addition of lconf --service command line option
3099 Details : lconf now accepts a '--service <arg>' option, which is
3100 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3102 Severity : enhancement
3104 Description: Failover mode is now the default for OSTs.
3105 Details : By default, OSTs will now run in failover mode. To return to
3106 the old behaviour, add '--failout' to the lmc line for OSTs.
3108 Severity : enhancement
3110 Description: Health checks are now provided for MDS and OSTs
3111 Details : Additional detailed health check information on MSD and OSTs
3112 is now provided through the procfs health_check value.
3115 Frequency : occasional, depends on IO load
3117 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3118 after numerous create/delete cycles
3119 Details : The ext3 inode allocation policy would not allocate new inodes
3120 very well on the OSTs because there are no new directories
3121 being created. Instead we look for groups with free space if
3122 the parent directories are nearly full.
3127 Description: Network or server problems during mount may cause partially
3128 mounted clients instead of returning an error.
3129 Details : The config llog parsing code may overwrite the error return
3130 code during mount error handling, returning success instead
3136 Description: MDS can fail to allocate large reply buffers
3137 Details : After long uptimes the MDS can fail to allocate large reply
3138 buffers (e.g. zconf client mount config records) due to memory
3139 fragmentation or consumption by the buffer cache. Preallocate
3140 some large reply buffers so that these replies can be sent even
3141 under memory pressure.
3145 Frequency : rare (liblustre)
3146 Description: fsx running with liblustre complained that using truncate() to
3147 extend the file doesn't work. This patch corrects that issue.
3148 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3149 ATTR_SIZE and lsm use in llu_setattr_raw.
3153 Frequency : rare, only 2.6 kernels
3154 Description: Unusual file access patterns on the MDS may result in inode
3155 data being lost in very rare circumstances.
3156 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3157 code in the 2.6 kernel caused the inode and/or EA data not to
3158 be read from disk, causing single-file corruption.
3162 Frequency : rare, only 2.6 filesystems using extents
3163 Description: Heavy concurrent write and delete load may cause data corruption.
3164 Details : It was possible under high-load situations to have an extent
3165 metadata block in the block device cache from a just-unlinked
3166 file overwrite a newly-allocated data block. We now unmap any
3167 metadata buffers that alias just-allocated data blocks.
3171 Frequency : filesystems with default stripe_count larger than 77
3172 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3173 Details : lconf specifies an inode size of 4096 bytes when the default
3174 stripe_count is larger than 77. This conflicts with the default
3175 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3176 this case to avoid pinning too much memory for large EAs.
3178 ------------------------------------------------------------------------------
3180 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3185 Frequency : rare (extremely heavy IO load with hundreds of clients)
3187 Description: Client is evicted, gets IO error writing to file
3188 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3189 caused clients to be evicted for AST timeouts. The fixes in
3190 bug 5192 mean we no longer need to have such short AST timeouts
3191 so ldlm_timeout has been increased.
3194 Frequency : occasional during --force or --failover shutdown under load
3195 Bugzilla : 5949, 4834
3196 Description: Server oops/LBUG if stopped with --force or --failover under load
3197 Details : a collection of import/export refcount and cleanup ordering
3198 issues fixed for safer force cleanup
3201 Frequency : only filesystems larger than 120 OSTs
3202 Bugzilla : 5990, 6223
3203 Description: lfs getstripe would oops on a very large filesystem
3204 Details : lov_getconfig used kfree on vmalloc'd memory
3207 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3208 Bugzilla : 6242, 6243
3209 Description: reading from files that had been truncated to a non-zero size
3210 but never opened returned no data
3211 Details : ll_file_read() reads zeros from no-object files to EOF
3216 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3217 Details : There's little harm in aborting MDS/OSS recovery and letting it
3218 try again, so I removed the LASSERT and return an error instead.
3220 Severity : enhancement
3222 Description: New debugging infrastructure for tracking down data corruption
3223 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3224 (b) cover more of the client-side code path, and (c) try to narrow
3225 down where problems occurred
3229 Bugzilla : 3819, 4364, 4397, 6313
3230 Description: Racing close and eviction MDS could cause assertion in mds_close
3231 Details : It was possible to get multiple mfd references during close and
3232 client eviction, leading to one thread referencing a freed mfd.
3234 Severity: : enhancement
3235 Bugzilla : 3262, 6359
3236 Description: Attempts to reconnect to servers are now more aggressive.
3237 Details : This builds on the enhanced upcall-less recovery that was added
3238 in 1.4.2. When trying to reconnect to servers, clients will
3239 now try each server in the failover group every 10 seconds. By
3240 default, clients would previously try one server every 25 seconds.
3245 Description: After recovery, certain operations trigger a failed
3246 assertion on a client.
3247 Details : Failing over an mds, using lconf -d --failover, while a
3248 client was doing a readdir() call would cause the client to
3249 LBUG after recovery completed and the readdir() was resent.
3251 Severity : enhancement
3253 Description: Default groups are now added by lconf
3254 Details : You can now run lconf --group <servicename> without having to
3255 manually add groups with lmc.
3258 Frequency : occasional
3260 Description: Nodes with an elan id of 0 trigger a failed assertion
3263 Frequency : always when accessing e.g. tty/console device nodes
3265 Description: tty and some other devices nodes cannot be used on lustre
3266 Details : file's private_data field is used by device data and lustre
3267 values in there got lost. New field was added to struct file to
3268 store fs-specific private data.
3271 Frequency : when exporting Lustre via NFS
3273 Description: NFSD failed occasionally when looking up a path component
3274 Details : NFSD is looking up ".." which was broken in ext3 directories
3275 that had grown large enough to become hashed.
3278 Frequency : Clusters with multiple interfaces not on the same subnet
3280 Description: Nodes will repeatedly try to reconnect to an interface which it
3281 cannot reach and report an error to the log.
3282 Details : Extra peer list entries will be created by lconf with some peers
3283 unreachable. lconf now validates the peer before adding it.
3286 Frequency : Only if a default stripe is set on the filesystem root.
3288 Description: Setting a default stripe on the filesystem root prevented the
3289 filesystem from being remounted.
3290 Details : The client was sending extra request flags in the root getattr
3291 request and did not allocate a reply buffer for the dir EA.
3294 Frequency : occasional, higher if lots of files are accessed by one client
3295 Bugzilla : 6159, 6097
3296 Description: Client trips assertion regarding lsm mismatch/magic
3297 Details : While revalidating inodes the VFS looks up inodes with ifind()
3298 and in rare cases can find an inode that is being freed.
3299 The ll_test_inode() code will free the lsm during ifind()
3300 when it finds an existing inode and then the VFS later attaches
3301 this free lsm to a new inode.
3305 Bugzilla : 6422, 7030
3306 Description: MDS deadlock between mkdir and client eviction
3307 Details : Creating a new file via mkdir or mknod (starting a transaction
3308 and getting the ns lock) can deadlock with client eviction
3309 (gets ns lock and trying to finish a synchronous transaction).
3312 Frequency : occasional
3313 Description: While starting a server, the fsfilt_ext3 module could not be
3315 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3316 kernels. Previously, lconf would still use the ext3 name
3317 when trying to load modules. Now, it will correctly use
3318 ext3 on 2.4 and ldiskfs on 2.6.
3320 Severity : enhancement
3321 Description: The default stripe count has been changed to 1
3322 Details : The interpretation of the default stripe count (0, to lfs
3323 or lmc) has been changed to mean striping across a single
3324 OST, rather than all available. For general usage we have
3325 found a stripe count of 1 or 2 works best.
3327 Severity : enhancement
3328 Description: Add support for compiling against Cray portals.
3329 Details : Conditional compiling for some areas that are different
3333 Frequency : occasional
3334 Bugzilla : 6409, 6834
3335 Description: Creating files with an explicit stripe count may lead to
3336 a failed assertion on the MDS
3337 Details : If some OSTs are full or unavailable, creating files may
3338 trigger a failed assertion on the MDS. Now, Lustre will
3339 try to use other servers or return an error to the
3343 Frequency : occasional
3345 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3346 caused lock timeout during readahead (since 1.4.2).
3347 Details : Processes doing readahead might match a lock that hasn't been
3348 granted yet if there are overlapping and conflicting lock
3349 requests. The readahead process waits on ungranted lock
3350 (original lock is CBPENDING), while OST waits for that process
3351 to cancel CBPENDING read lock and eventually evicts client.
3353 Severity : enhancement
3355 Description: Initial enabling of flock support for clients
3356 Details : Implements fcntl advisory locking and file status functions.
3357 This feature is provided as an optional mount flag (default
3358 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3359 locking are implemented yet, and those that are are not guaranteed
3360 to be completely correct in production environments.
3361 mount -t lustre -o [flock|noflock] ...
3364 Frequency : occasional
3366 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3367 assertion in the ext3 JBD (journaling) layer.
3368 Details : The b_committed_data struct is protected by the big kernel lock
3369 in 2.4 kernels, serializing journal_commit_transaction() and
3370 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3371 access to this struct. In 2.6 kernels there is finer grained
3372 locking to improve SMP performance of the JBD layer.
3376 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3381 Description: merge and cleanup kernel patches.
3384 -----------------------------------------------------------------------------
3386 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3388 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3389 versions of Lustre. Please update all servers and clients to
3390 version 1.4.2 or later at the same time. You must also run
3391 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3392 to update the configuration logs.
3394 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3395 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3396 - swab llog records of type '0' so we get proper header size/idx (5861)
3397 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3398 - fix rename of one directory over another leaking an inode (5953)
3399 - avoid SetPageDirty on 2.6 (5981)
3400 - don't re-add just-being-destroyed locks to the waiting list (5653)
3401 - when creating new directories, inherit the parent's custom
3402 striping settings if present parent (3048)
3403 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3404 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3405 - don't hold client locks on temporary worklist from l_lru (5666)
3406 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3407 - automatically evict dead clients (5921)
3408 - Update file size properly in create+truncate+fstat case (6196)
3409 - Do not unhash mountpoint dentries, do not allow removal of
3411 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3412 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3413 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3414 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3415 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3416 - Correctly handle reads of files with no objects (6243)
3417 - lctl recover will also mark a device active if deactivate used (5933)
3419 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3420 - allow --write-conf on an MDS with different nettype than client (5619)
3421 - don't write config llogs to MDS for mounts not from that MDS (5617)
3422 - lconf should create multiple TCP connections from a client (5201)
3423 - init scripts are now turned off by default; run chkconfig --on
3424 lustre and chkconfig --on lustrefs to use them
3425 - upcalls are no longer needed for clients to recover to failover
3427 - add --abort-recovery option to lconf to abort recovery on device
3429 - add support for an arbitrary number of OSTs (3026)
3430 - Quota support protocol changes.
3431 - forward compatibility changes to wire structs (6007)
3432 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3433 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3434 - improved "lustre status" script
3435 - initialize blocksize for non-regular files (6062)
3436 - added --disable-server and --disable-client configure options (5782)
3437 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3438 - Vanilla 2.4.29 support
3439 - increase maximum number of obd devices to 520 (6242)
3440 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3441 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3442 SLES 9 2.6.5 kernels
3443 - limit stripes per file to 160 (the maximum EA size) (6093)
3445 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3448 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3449 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3450 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3451 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3452 - small ext3 extents cleanups and fixes (5733)
3453 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3454 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3455 - avoid needless client->OST connect, fix handle mismatch (5317)
3456 - fix DLM error path that led to out-of-sync client, long delays (5779)
3457 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3458 - fix several locking issues related to i_size (5492,5624,5654,5672)
3459 - don't move pending lock onto export if it is already evicted (5683)
3460 - fix kernel oops when creating .foo in unlinked directory (5548)
3461 - fix deadlock in obdfilter statistics vs. object create (5811)
3462 - use time_{before,after} to avoid timer jiffies wrap (5882)
3463 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3464 - Do not leak request if server was not able to process it (5154)
3465 - If mds_open unable to find parent dir, make that negative lookup(5154)
3466 - don't create new directories with extent-mapping (5909, 5936)
3468 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3469 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3470 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3471 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3472 - support for RHEL4 kernel on i686 (5773)
3473 - provide error messages when incompatible logs are encountered (5898)
3475 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3476 * version 1.4.0.10 (1.4.1 release candidate 1)
3478 - don't keep a lock reference when lock is not granted (4238)
3479 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3480 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3481 - drop import inflight refcount on signal_completed_replay error (5255)
3482 - unlock page after async write error during send (3677)
3483 - handle missing objects in filter_preprw_read properly (5265)
3484 - no transno return for symlink open, don't save no-trasno open (3440)
3485 - don't try to complete elan receive that already failed (4012)
3486 - free RPC server reply state on error (5406)
3487 - clean up thread from ptlrpc_start_thread() on error (5160)
3488 - readahead could read extra page into cache that wasn't ejected (5388)
3489 - prevent races in class_attach/setup/cleanup/detach (5260)
3490 - don't dereference de->d_inode after l_dput of de (5458)
3491 - use "int" for stripe value returned from lock_to_stripe (5544)
3492 - mballoc allocation and error-checking fixes in 2.6 (5504)
3493 - block device patches to fix I/O request sizes in 2.6 (5482)
3494 - look up hostnames for IB nals (5602)
3495 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3496 - don't start multiple acceptors for the same port (5277)
3497 - fix incorrect LASSERT in mds_getattr_name (5635)
3498 - export a proc file for general "ping" checking (5628)
3499 - fix "lfs check" to not block when the MDS is down (5628)
3501 - service request history (4965)
3502 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3503 - create an "evict_client" /proc entry on OSTs, like the MDS has
3504 - fix mount usage message, return errors per mount(8) (5168)
3505 - change grep [] to grep "[]" in tests so they work in more UMLs
3506 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3507 - remove ancient LOV_MAGIC_V0 EA support (5047)
3508 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3509 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3510 - allow readahead allocations to fail when low on memory (5383)
3511 - mmap locking landed again, after considerable improvement (2828)
3512 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3514 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3517 - send OST transaction number in read/write reply to free req (4966)
3518 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3519 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3520 - the watchdog thread now runs as interruptible (5246)
3521 - flock/lockf fixes (but it's still disabled, pending 5135)
3522 - don't use EXT3 constants in llite code (5094)
3523 - memory shortage at startup could cause assertion (5176)
3525 - reorganization of lov code
3526 - single portals codebase
3528 - add extents/mballoc support (5025)
3529 - direct I/O reads in the obdfilter (4048)
3530 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3532 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3535 - send OST transaction number in read/write reply to free req (4966)
3536 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3537 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3538 - the watchdog thread now runs as interruptible (5246)
3539 - handle missing objects in filter_preprw_read properly (5265)
3540 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3541 - drop import inflight refcount on signal_completed_replay error (5255)
3542 - unlock page after async write error during send (3677)
3543 - return original error code on reconstructed replies (3761)
3544 - no transno return for symlink open, don't save no-trasno open (3440)
3546 - add pid to ldlm debugging output (4922)
3547 - bump the watchdog timeouts -- we can't handle 30sec yet
3548 - extra debugging for orphan dentry/inode bug (5259)
3550 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3553 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3554 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3555 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3556 - fix NULL dereference in /proc/sys/portals/routes (4827)
3557 - allow failed mdc_close() operations to be interrupted (4561)
3558 - stop precreate on OST before MDS would time out on it (4778)
3559 - don't send partial-page writes before EOF from client (4410)
3560 - discard client grant for sub-page writes on large-page clients (4520)
3561 - don't free dentries not owned by NFS code, check generation (4806)
3562 - fix lsm leak if mds_create_objects() fails (4801)
3563 - limit debug_daemon file size, always print CERROR messages (4789)
3564 - use transno after validating reply (3892)
3565 - process timed out requests if import state changes (3754)
3566 - update mtime on OST during writes, return in glimpse (4829)
3567 - add mkfsoptions to LDAP (4679)
3568 - use ->max_readahead method instead of zapping global ra (5039)
3569 - don't interrupt __l_wait_event() during strace
3571 - add software watchdogs to catch hung threads quickly (4941)
3572 - make lustrefs init script start after nfs is mounted
3573 - fix CWARN/ERROR duplication (4930)
3574 - return async write errors to application if possible (2248)
3575 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3576 - print NAL number in %x format (4645)
3577 - update barely-supported suse-2.4.21-171 series (4842)
3578 - support for sles 9 %post scripts
3579 - support for building 2.6 kernel-source packages
3580 - support for sles km_* packages
3582 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3585 - ignore -ENOENT errors in osc_destroy (3639)
3586 - notify osc create thread that OSC is being cleaned up (4600)
3587 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3588 - reconstruct ost_handle() like mds_handle() (4657)
3589 - create a new thread to do import eviction to avoid deadlock (3969)
3590 - let lconf resolve symlinked-to devices (4629)
3591 - don't unlink "objects" from directory with default EA (4554)
3592 - hold socknal file ref over connect in case target is down (4394)
3593 - allow more than 32000 subdirectories in a single directory (3244)
3594 - fix blocks count for O_DIRECT writes (3751)
3595 - OST returns ENOSPC from object create when no space left (4539)
3596 - don't send truncate RPC if file size isn't changing (4410)
3597 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3598 - bind to privileged port in socknal and tcpnal (3689)
3600 - rate limit CERROR/CWARN console message to avoid overload (4519)
3601 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3602 - basic mmap support (3918)
3603 - kernel patch series update from b1_4 (4711)
3605 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3608 - avoid crash during MDS cleanup with OST shut down (2775)
3609 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3610 - don't use bad inodes on the MDS (3744)
3611 - dynamic object preallocation to improve recovery speed (4236)
3612 - don't hold spinlock over lock dumping or change debug flags (4401)
3613 - don't zero obd_dev when it is force cleaned (3651)
3614 - print grants to console if they go negative (4431)
3615 - "lctl deactivate" will stop automatic recovery attempts (3406)
3616 - look for existing locks in ldlm_handle_enqueue() (3764)
3617 - don't resolve lock handle twice in recovery avoiding race (4401)
3618 - revalidate should check working dir is a directory (4134)
3620 - don't always mark "slow" obdfilter messages as errors (4418)
3622 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3625 - don't close LustreDB during write_conf until it is done (3860)
3626 - fix typo in lconf for_each_profile (3821)
3627 - allow dumping logs from multiple threads at one time (3820)
3628 - don't allow multiple threads in OSC recovery (3812)
3629 - fix debug_size parameters (3864)
3630 - fix mds_postrecov to initialize import for llog ctxt (3121)
3631 - replace config semaphore with spinlock (3306)
3632 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3633 - don't allow enqueue to complete on a destroyed export (3822)
3634 - down write_lock before checking llog header bitmap (3825)
3635 - recover from lock replay timeout (3764)
3636 - up llog sem before sending rpc (3652)
3637 - reduce ns lock hold times when setting kms (3267)
3638 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3639 - fix NULL deref and obd_dev leak on setup error (3312)
3640 - replace some LBUG about llog ops with error handling (3841)
3641 - don't match INVALID dentries from d_lookup and spin (3784)
3642 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3643 - fix invalid assertion in ptlrpc_set_wait (3880)
3645 - add libwrap support for the TCP acceptor (3996)
3646 - add /proc/sys/portals/routes for non-root route listing (3994)
3647 - allow setting MDS UUID in .xml (2580)
3648 - print the stack of a process that LBUGs (4228)
3650 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3653 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3654 - make sure to unset replay flag from failed open requests (3440)
3655 - if default stripe count is 0, use OST count for inode size (3636)
3656 - update parent mtime/ctime on client for create/unlink (2611)
3657 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3658 - free recovery state on server during a forced cleanup (3571)
3659 - unregister_reply for resent reqs (3063)
3660 - loop back devices mounting and status check on 2.6 (3563)
3661 - fix resource-creation race that can provoke i_size == 0 (3513)
3662 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3663 - more debugging for page-accounting assertion (3746)
3664 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3665 - avoid deadlock after precreation failure (3758)
3666 - fix race and lock order deadlock in orphan handling (3450, 3750)
3667 - add validity checks when grabbing inodes from l_ast_data (3599)
3669 - add /proc/.../recovery_status to obdfilter (3428)
3670 - lightweight CDEBUG infrastructure, debug daemon (3668)
3671 - change default OSC RPC parameters to be better on small clusters
3672 - turn off OST read cache for files smaller than 32MB
3673 - install man pages and include them in rpms (3100)
3674 - add new init script for (un)mounting lustre filesystems (2593)
3675 - run chkconfig in %post for init scripts (3701)
3676 - drop scimac NAL (unmaintained)
3678 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3681 - clean kiobufs before and after use (3485)
3682 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3683 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3684 - change init script to fail more clearly if not run as root (1528)
3685 - allow clients to reconnect during replay (1742)
3686 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3687 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3688 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3689 - fix cond_resched (3554)
3690 - only evict unfinished clients after recovery (3515)
3691 - allow bulk resend, prevent data loss (3570)
3692 - dynamic ptlrpc request buffer allocation (2102)
3693 - don't allow unlinking open directory if it isn't empty (2904)
3694 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3695 - remove extraneous obd dereference causing LASSERT failure (3334)
3696 - don't use get_cycles() when creating temp. files on the mds (3156)
3697 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3698 - handle EEXIST for set-stripe, set proper directory name (3336)
3700 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3701 - fix ksocknal_fmb_callback() error messages (2918)
3703 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3706 - don't copy lvb into (possibly NULL) reply on error (2983)
3707 - don't deref dentry after dput, don't free lvb on error (2922)
3708 - use the kms to determine writeback rpc length (2947)
3709 - increment oti_logcookies when osc is inactive (2948)
3710 - update client's i_blocks count via lvb messages (2543)
3711 - handle intent open/close of special files properly (1557)
3712 - mount MDS with errors=remount-ro, like obdfilter (2009)
3713 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3714 - don't use cancelling-locks' kms values (2947)
3715 - use highest lock extent for kms, not last one (2925)
3716 - don't dereference ERR_PTR() dentry in error handling path (3107)
3717 - fix thread race in portals_debug_dumplog() (3122)
3718 - create lprocfs device entries at setup instead of at attach (1519)
3719 - common AST error handler, don't evict client on completion race (3145)
3720 - zero nameidata in detach_mnt in 2.6 (3118)
3721 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3722 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3723 - initialize RPC timeout timer earlier for 2.6 (3219)
3724 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3725 - print nal/nid for unknown nid (3258)
3726 - additional checks for oscc recovery before doing precreate (3284)
3727 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3728 - don't crash in mdc_close for bad permissions on open (3285)
3729 - zero i_rdev for non-device files (3147)
3730 - clear page->private before handing to FS, better assertion (3119)
3731 - tune the read pipeline (3236)
3732 - fix incorrect decref of invalidated dentry (2350)
3733 - provide read-ahead stats and refine rpc in flight stats (3328)
3734 - don't hold journal transaction open across create RPC (3313)
3735 - update atime on MDS at close time (3265)
3736 - close LDAP connection when recovering to avoid server load (3315)
3737 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3738 - don't leak open file on MDS after open resend (3325)
3739 - serialize filter_precreate and filter_destroy_precreated (3329)
3740 - loop device shouldn't call sync_dev() for nul device (3092)
3741 - clear page cache after eviction (2766)
3742 - resynchronize MDS->OST in background (2824)
3743 - refuse to mount the same filesystem twice on same mountpoint (3394)
3744 - allow llmount to create routes for mounting behind routers (3320)
3745 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3746 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3747 - fix rare problem with rename on htree directories (3417)
3749 - allow default OST striping configuration per directory (1414)
3750 - fix compilation for qswnal for 2.6 kernels (3125)
3751 - increase maximum number of MDS request buffers for large systems
3752 - change liblustreapi to be useful for external progs like lfsck (3098)
3753 - increase local configuration timeout for slow disks (3353)
3754 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3756 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3759 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3760 - don't overwrite extent policy data in reply if lock was blocked (2901)
3761 - drop filter export grants atomically with removal from device (2663)
3762 - del obd_self_export from work_list in class_disconnect_exports (2908)
3763 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3764 - swab reply message in mdc_close, other PPC fixes (2464)
3765 - fix destroying of named logs (2325)
3766 - overwrite old logs when running lconf --write_conf (2264)
3767 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3768 - fix race in target_handle_connect (2898)
3769 - mds_reint_create() should take same inode create lock (2926)
3770 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3771 - don't close files for self_export to avoid uninitialized obd (2936)
3772 - allow MDS with the same name as client node (2939)
3773 - hold dentry reference for closed log files for unlink (2325)
3774 - reserve space for all logs during transactions (2059)
3775 - don't evict page beyond end of stripe extent (2925)
3776 - don't oops on a deleted current working directory (2399)
3777 - handle hard links to targets without a parent properly (2517)
3778 - don't dereference NULL lock when racing during eviction (2867)
3779 - don't grow lock extents when lots of conflicting locks (2919)
3781 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3784 - account for cache space usage on clients to avoid data loss (974)
3785 - lfsck support in lustre kernel code (2349)
3786 - reduce journal credits needed for BRW writes (2370)
3787 - orphan handling to avoid losing space on client/server crashes
3788 - ptlrpcd can be blocked, stopping ALL progress (2477)
3789 - use lock value blocks to assist in proper KMS, faster stat (1021)
3790 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3791 - recovery for initial connections (2355)
3792 - fixes for mds_cleanup_orphans (1934)
3793 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3794 - block all file creations until orphan recovery completes (1901)
3795 - client remove rq_connection from request struct (2423)
3796 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3797 - recovery timer race (2670)
3798 - mdc_close recovey bug (2532)
3799 - ptlrpc cleanup bug (2710)
3800 - mds timeout on local locks (2588)
3801 - namespace lock held during RPCs (2431)
3802 - handle interrupted sync write properly (2503)
3803 - don't try to handle a message that hasn't been replied to (2699)
3804 - client assert failure during cleanup after abort recovery (2701)
3805 - leak mdc device after failed mount (2712)
3806 - ptlrpc_check_set allows timedout requests to complete (2714)
3807 - wait for inflight reqs when ptlrpcd finishes (2710)
3808 - make sure unregistered services are removed from the srv_list
3809 - reset bulk XID's when resending them (caught by 1138 test)
3810 - unregister_bulk after timeout
3811 - fix lconf error (2694)
3812 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3813 - readahead locks pages, leaves pending causing memory pressure (2673)
3814 - increase OST request buffers to 4096 on large machines (2729)
3815 - fix up permission of existing directories in simple_mkdir (2661)
3816 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3817 - don't assign transno to errored transactions (2742)
3818 - don't delete objects on OST if given a bogus objid from MDS (2751)
3819 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3820 - if rq_no_resend, then timeout request after recovery (2432)
3821 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3822 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3823 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3824 - MDS recovery shouldn't delete objects that it has given out (2730)
3825 - if enqueue arrives after completion, don't clobber LVB (2819)
3826 - don't unlock pages twice when trigger_group_io returns error (2814)
3827 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3828 - don't write pages to disk if there was an error (1450)
3829 - don't ping imports that have recovery disabled (2676)
3830 - take buffered bytes into account when balancing socknal conn (2817)
3831 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3832 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3833 - remove little-endian swabbing of llog records (1987)
3834 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3835 - retry reposting request buffers if they fail (1191)
3836 - grow extent at grant time to avoid granting a revoked lock (2809)
3837 - lock revoke doesn't evict page if covered by a second lock (2765)
3838 - disable VM readahead to avoid reading outside lock extents (2805)
3840 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3841 - updated kernel patches for hp-2.4.20 kernel (2681)
3843 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3846 - fix truncated write corruption (2366)
3847 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3849 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3850 - ensure len(uuid) < 37 in lmc (1171)
3851 - fix ia64 OOPS in llog_test (2255)
3852 - zero end of page at obdfilter for partial page writes (2648)
3853 - don't leave stale dentries around after renames (bug 2428)
3854 - fix timeouts when evicting a client with a single lock held (2642)
3855 - set deadline for the initial HELLO message to drain (2634)
3856 - print out dotted-quad IP addresses in the socknal (2302)
3858 - additional debugging for MDS client eviction problem (2443)
3859 - fix mkfsoptions support for osts (2603, 2604)
3861 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3864 - add series for the vanilla 2.6.0 kernel
3865 - add series for the vanilla 2.4.24 kernel
3866 - add series for a cray x86/64 UL kernel drop
3867 - fix xattr patches for the vanilla 2.4.19 series
3869 - generate true UUIDs in lmc (1171)
3870 - have portals stack dumping break in UML (2466)
3871 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3872 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3873 - wake callback waiting threads on client eviction (2460)
3874 - Add --ptldebug and --subsystem to lmc (1719)
3875 - update assertion to allow safe interrupt allocation
3876 - set rq_no_resend for cancel requests (2432)
3877 - recalculate ptlrpcd timeout after resend (2494)
3878 - call vfs_rmdir when removing pending directories (2368)
3879 - fix renaming a file to itself (2429)
3880 - lmc creates a default one-stripe lov (2454)
3881 - expand procfs space to handle large clusters (2326)
3882 - increase UML stack to avoid overflow
3883 - update lconf's list of debug and subsystem masks
3884 - fix lfs find --obd (2510)
3885 - /proc tunable for disabling filter read caching (2591)
3886 - stop rpm packages from altering slapd.conf (2301)
3887 - disable nagle in the socknal under 0conf (2578)
3888 - choose mds inode size based on stripe count (2572)
3889 - fix kernel-source rpm problems (2516)
3891 - add --disable-doc to avoid pdf generation (2421)
3892 - update documentation, tests, type-os, comments
3893 - avoid format warnings on ia64
3894 - remove the TOE NAL
3895 - tiny code cleanups by removing unused fields
3897 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3900 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3901 - remove the most blatant lies from BUILDING file (2371)
3902 - change default debug level to reasonable production setting
3903 - reduce client side cache size to reduce cache flush time
3904 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3905 - make TCP zerocopy and pinger support enabled by default (2476)
3906 - sync writes completed after process exits caused crashes (2319)
3907 - maintain correct mount count on the MDS (2356)
3908 - backout 1557, because 2316 wasn't really fixed
3909 - better file I/O statistics gathering in /proc
3910 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3911 - another kernel patch to fix zero-copy TCP function export
3912 - don't take duplicate lock when processing re-sent getattr (2420)
3913 - lctl uses obd_self_export instead of creating new conn (2353)
3914 - MDS/OST recovery case which requires object creation asserted (2425)
3915 - move lfs from /usr/sbin to /usr/bin in packages
3916 - fix race between mds_client_add and mds_client_free (2417)
3917 - use kmalloc instead of slabs in portals (2430)
3918 - don't create duplicate records when a failover MDS is present (2442)
3919 - remove unnecessary mount age check (2332)
3920 - don't remove directory inodes from locks prematurely (2451)
3921 - don't break if MDS service name is the same as hostname (2103)
3922 - fix races in client write RPC generation when cache full (2482)
3924 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3927 - remove now-unused request->rq_obd (278)
3928 - if an allocation fails, print out how much memory we've used (1933)
3929 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3930 - add the "configurable stack size" patch to most series files (1256)
3931 - ability to write large log records, for 100+ OST configs (2306)
3932 - fix NULL deref when filter_prep fails (2314)
3933 - fix operator precedence error in filter_sync
3934 - dynamic allocation of socknal TX descriptors (2315)
3935 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3936 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3937 - ensure that CWARN messages in Portals always get to the syslog
3938 - __init/__exit are not for prototype decls (ldlm_init/exit)
3939 - x86-64 compile warning fixes
3940 - fix gateway LMC keyword conflict (2318)
3941 - fix MDS lock inversions in getattr/reint paths (1844)
3942 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3943 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3944 - DLM race condition prevented some lock evictions (2328)
3945 - ENOMEM detection and retry on socknal sends (2230)
3946 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3947 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3948 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3949 - better error messages when a client is rejected during recovery (1505)
3950 - avoid cancelling locks which were never granted, after failure (2330)
3951 - fix i_sem/journal inversion in mds_client_add (2333)
3952 - fix truncate/getattr lock cycle deadlock (2334)
3953 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3954 - fix two rare exit paths which could leak an l_lock() ref (2321)
3955 - fencepost error in MDS/OST orphan recovery (2226)
3956 - make log record alignment 8 bytes (1988)
3957 - lstripe now fails when requested offset > ost_count (2237)
3958 - ensure that all kernel series have a complete list.h (1607)
3959 - fix crashes in special-file operations (2316)
3960 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3962 - allow configurable automake binary, for testing new versions
3963 - small update to the lfs documentation
3965 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3967 * fix negative export reference count in fsfilt_sync (2312)
3969 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3970 * release candidate 0.9.1
3972 - orphans are moved into the PENDING directory for possible recovery
3973 - replayed opens now open by fid for orphan/rename safety (1042)
3974 - last close of an orphan inode generates a transno (683)
3975 - chdir() and mount() now pin the directory entry (1020)
3976 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3977 - discard very old requests without processing them (1502)
3978 - remove some common, well-understood CERRORs (1505)
3979 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3980 - clear "grant" flags in OST replies until OST grant code lands (1644)
3981 - fix read performance by not clobbering i_blksize on client (1598)
3982 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3983 - make LRU size a /proc tunable, clears locks when reduced (707)
3984 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3985 - server locks take references on exports now (1558)
3986 - build fixes for 2.4.20-rh trees (1663)
3987 - return an error from lov_create if all OSCs are inactive (1751)
3988 - fix import levels when a reconnect happens without a timeout (1597)
3989 - exit early from mds_open if we get a lookup error (1749)
3990 - partial page read at EOF wouldn't wait for disk before sending (1642)
3991 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3992 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3993 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3994 - write/truncate lock inversion (1639)
3995 - Don't auto-load obdclass, portals modules during cleanup (1495)
3996 - fix timestamps from jumping to "now" (1763)
3997 - extra journal assertions (1648)
3998 - add an extra multiunlink test (1771)
3999 - fix read_record/write_record API (1776)
4000 - fix leak of offset_extent, possible incorrect i_size later (1772)
4001 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4002 - Debugging for the kqswnal_get_idle_tx problems (1820)
4003 - Allow recovery to be attempted multiple times (1536)
4004 - Write out MDS last_rcvd file after it is first created (1600)
4005 - Fix tx_descriptor leak in failed transmit situations (1827)
4006 - ext3 journaling fixes for assertion failure after IO error (1871)
4007 - class_export_put() on freed export after completion AST error (1896)
4008 - Fix revalidate looping in VFS (1322)
4009 - Don't access a freed export during MDS_REINT timeout (1521)
4010 - Add open-unlink recovery support on the MDS (1673,1764)
4011 - Return an error if no MDS data was read from last_rcvd (1946)
4012 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4013 - Don't setup a disk device that doesn't match exported UUID (317)
4014 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4015 - avoid committing NULL handle in force close
4016 - local.sh is now a one-stripe LOV configuration
4017 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4018 - fix ext3 htree duplicate directory entry corruption (1516)
4019 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4020 - update to new LOV EA format (2097)
4021 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4023 2003-06-15 Phil Schwan <phil@clusterfs.com>
4026 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4027 - per-import/export recovery handling (958, 931, 959)
4028 - multiple last-rcvd slots, for serving multiple FSes (949)
4029 - connections are again shared between multiple imp/exports (963, 964)
4030 - "umount -f" would hang if any requests needed to be sent (393, 978)
4031 - avoid pinning large req buffer by copying for queued messages (989)
4032 - add "uuid" to "lctl device" command to help upcalls (991)
4033 - "open" RPCs with transnos would confuse recovery counters (1037)
4034 - do proper endian conversion of all wire messages (288, 340, 891)
4035 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4036 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4037 - fix ptlrpc_connection leak in target_handle_connect (1174)
4038 - fix import refcounting bug in OST and MDS cleanup (1134)
4039 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4040 - fix very unlikely obd_types race condition (501)
4041 - remove osc_open hack for echo_client (1187)
4042 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4043 - a failure in read_inode2 leads to deadlock (1139)
4044 - cancel ack-locks as soon as transaction is committed (1072)
4045 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4046 - make sure to commitrw after any preprw to avoid deadlock (1162)
4047 - failing to execute a file in a lustre FS would lock inode (1203)
4048 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4049 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4050 - obd_uuid2tgt would walk off the end of the list (1255)
4051 - on IA64 the osc would give portals incorrect bulk size (1258)
4052 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4053 - fix lock inversion caused by new llite matching code (1282)
4054 - limit the number of dirty pages on a client to 10MB (1286)
4055 - timed out locks were not being corrected cancelled (1289)
4056 - fix O_DIRECT above 4GB on IA-32 (1292)
4057 * major user-visible changes
4058 - fail out/fail over policy now controlled by the upcall (993)
4060 - add OBD_PING to check server availability and failure (954)
4061 - lustre messages are now sent in sending host order (288, 340, 891)
4062 - add eadatalen to MDS getattr reply (340)
4063 - OST read replies may contain second buffer, with per-page status (593)
4065 2003-03-11 Phil Schwan <phil@clusterfs.com>
4068 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4069 - failed open()s could cause deadlock; fixed (867, 869)
4070 - stop cancelling OST locks when files are closed (481)
4071 - overlapping XID spaces caused network corruption (851, 853)
4072 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4073 - setattr_raw added, to do single-RPC, server-side setattrs
4074 - lmc/lconf syntax change for OST UUIDs
4075 - fix crashy race condition between ptlrpc_free_req and osc_close
4076 - don't use request in mdc_enqueue if we hit a timeout (889)
4077 - don't set the inode i_size for regular files from the MDS (896)
4078 - handle out of order completion AST (842)
4079 - don't LBUG if a lock request times out after receiving AST (913)
4080 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4081 - if a bad lock AST arrives, send an error instead of dropping entirely
4082 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4083 - fix leak in bulk IO when only partially completed (899, 900, 926)
4084 - fix O_DIRECT for ia64 (55)
4085 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4086 - C-z after timeout could hang a process forever; fixed (977)
4088 - client-side I/O cache (678, 924, 929, 941, 970)
4090 - READPAGE and SETATTRs which don't take server-side locks get
4093 2003-02-11 Phil Schwan <phil@clusterfs.com>
4096 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4097 - Fix fsfilt_extN_readpage() to read a full page of directory
4098 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4099 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4100 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4101 - Most intent operations were removed, in favour of a new RPC mode
4102 that does a single RPC to the server and bypasses most of the VFS
4103 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4104 - Aggressively cancel local locks on DLM servers
4105 - mds_reint_unlink sends EA to the client if it's the last nlink.
4106 client uses that EA to unlink OST objects.
4107 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4108 - recursive symlinks were fixed (439)
4109 - fixed NULL deref in DEBUG_REQ
4110 - filter_update_lastobjid no longer calls sync, which annoyed extN
4111 - fixed multi-client small-writes to a single file problem (445)
4112 - fixed mtime updates during file writes (607)
4113 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4114 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4115 - fixed Linux OST size reporting problem (444, 656)
4116 - OST now updates object mtime with writes or setattr (607, 619)
4117 - client verifies file size before zeroing page past EOF (445)
4118 - OST now writes last allocated objid to disk with allocation (108)
4119 - LOV on echo now works (409)
4121 - mds_reint_unlink sends a new buffer, with the EA included. this
4122 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4123 set if a regular file was being unlinked, and it was the last link
4124 - use PtlGet from the target for bulk writes (315)
4125 - OST now updates object mtime with writes or setattr (607, 619)
4126 - LDLM now has a grant-time callback to revalidate locked items, if
4128 - Many MDS operations were reorganized to combat race conditions
4130 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4131 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4133 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4136 - Fully reactivate OST imports after reconnection (512, others)
4137 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4138 - More graceful error handling for truncating on dead OST (515)
4139 - Don't error out unless we're actually accessing dead stripes (474)
4140 - Fix garbage sizes when stripes are missing (410)
4141 - LRU counters were broken, causing constant lock purge (433, 432)
4142 - garbage on read from stripes with failed OSTs (441)
4143 - mark OSCs as active before reconnecting during recovery (438)
4144 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4145 - lfind did not preserve OST order in output (443)
4146 - symlinks cause hung clients, incorrect data (439)
4147 - stop dereferencing request after dropping refcount (457)
4148 - don't LASSERT(spin_is_locked) on non-SMP (455)
4149 - fixes for many rename() bugs
4150 - fstat didn't correctly synchronize attributes (399)
4151 - server must handle lock cancellation during blocking AST prep (487)
4152 - bulk descriptors were free()d too soon (511)
4153 - fix paths in lconf, which would load incorrect modules (451, 507)
4154 - fix confusing lconf 'host not found' error message (386)
4155 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4156 - fix race condition in mdc_blocking_ast() for inode access (526)
4157 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4158 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4159 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4160 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4161 - fix lstripe in several ways (kernel side) (527)
4162 - fix request leak in ldlm_cli_enqueue (262)
4163 - incorrect OSC was marked inactive after OST failure
4164 - call mds_fs_cleanup before unmounting filesystem (524)
4165 - fix races between taking ns_lock and ldlm_lock_change_resource
4166 - fix races updating LOV export open file list
4167 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4168 - fix recovery NULL deref in ldlm_cli_cancel_unused
4169 - fix some DLM races by using new hash table for lock handles (419)
4170 - permit the client to specify desired inodes, at replay
4171 - duplicate requests when we queue them for replay reintegration
4172 - fix last_rcvd offset calculation
4173 - sync after each recovered transaction, so we always make progress
4174 - never, not always, ERESTART requests without transnos
4175 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4177 - skip replay if the MDS says that the client is already connected
4178 - don't check for a recovery-enabled export to match lctl's UUID
4179 - don't INC_USE_COUNT for phantom exports
4180 - don't crash when cleaning up phantom exports (567)
4181 - don't double-finish or set replay data for errored mdc_open requests
4182 - abort requests when they time out, so we don't get old replies
4183 - send/receive replies for AST messages again
4184 - if the client says that it doesn't have the lock, cancel it on the
4186 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4187 mark it as destroyed, it will all work out when decref is called
4188 - fix module use counts (22, 581)
4190 - ASTs now expect a reply (server cancels lock on error reply)
4192 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4195 - fix many simultaneous client startup (392)
4196 - fix dentry->d_it clobbering
4197 - credentials weren't being shipped for readdir/getattr operations
4198 - remove invalid assertions triggered during some concurrent MD
4200 - proper Lustre versions added (336, 389)
4201 - fix memory leak for create error case (398)
4202 - fix LOV locking bug that would get cli/srv out of sync
4203 - fix echo client over LOV (409)
4204 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4205 - fix double-O_EXCL intent crash (424)
4206 - avoid sending multiple lock CANCELs (352)
4208 - MDS can do multi-client recovery (modulo bugs in new code)
4210 - many updates, edits, cleanups
4212 2002-11-18 Phil Schwan <phil@clusterfs.com>
4215 - fix null d_it dereference (346)
4216 - fix full OST/dbench hang (333)
4217 - fix permission problem with file removal (286)
4218 - fix removal of OSCs from LOV when they fail
4219 - fix NULL deref during bulk timeout (214)
4220 - fix problems related to multiple filesystems on one MDS (241)
4221 - fixed serious subtle metadata locking bugs
4222 - free locks on clients when inodes are removed due to memory
4224 - fix inode pointer in lock data (285)
4225 - partial support for multiple MDS on a single host (241)
4226 - data locks weren't cancelled at clear_inode time (290, 311)
4227 - intent locks could lead to unbounded lock growth (205)
4228 - added a maximum lock count, an LRU list, and a flusher
4229 - fix multiple rename (365)
4230 - properly abstracted the echo client
4231 - OSC locked 1 byte too many; fixed
4232 - rewrote brw callback code:
4233 - fixed recovery bugs related to LOVs (306)
4234 - fixed too-many-pages-in-one-write crash (191)
4235 - fixed (again) crash in sync_io_timeout (214)
4236 - probably fixed callback-related race (385)
4238 - Add capability to MDS protocol
4239 - LDLM cancellations and callbacks on different portals
4241 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4244 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4245 - defer open object destruction to close time (601981, 138)
4246 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4247 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4248 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4249 - don't call obd_setattr in ll_file_release for destroyed objects
4250 * protocol change to lustre_msg: move |version| and add |flags|
4251 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4252 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4253 * added replay of create, unlink, link and rename operations during
4254 MDS failover; recovery should be much more robust now
4255 * remove failed OSCs from LOVs (only lov_create uses this so far)
4256 * the lustre-HOWTO was brought (more) up to date (582544)
4258 2002-10-23 Phil Schwan <phil@clusterfs.com>
4261 - in-use dentries weren't being reused properly (617851)
4262 - prevent multiple LDLM setup (599178)
4263 - fix LOV size calculations for truncate (617853)
4264 - fix client handling of MDS intent errors (POSIX)
4265 - fix permission bug in lovstripe.c test (624321)
4266 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4267 - truncate past end of file could corrupt data
4268 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4269 - a race in recovery could return ETIMEDOUT to apps (623947)
4270 - building outside the source directory was fixed
4271 * the lustre-HOWTO was brought (more) up to date (582544)
4272 * major progress was made on recovery functionality
4274 2002-10-10 Phil Schwan <phil@clusterfs.com>
4277 - recovery deadlock fix
4278 - rm -rf causes LBUG fix (617817)
4279 - file open by multiple tasks fix (618962)
4280 - directory permissions bugs (602707 and 620007)
4281 - journal_stop fixed with locking (611313)
4282 - O_APPEND failures resolved (618273, perhaps 614459)
4283 - lconf PATH fix (619770)
4284 - IA64 build fix (621450)
4285 - RPC buffer sizes scale with amount of memory
4287 2002-10-01 Phil Schwan <phil@clusterfs.com>
4290 - locks would be cancelled without throwing away data pages,
4291 resulting in inconsistent data (605627)
4292 - inode attributes were not always being refreshed (605627, 612449)
4293 - lconf now continues to cleanup after lctl reports an error
4294 - MDS now enforces user permissions (602707)
4295 - lprocfs cleanup fixed, but not yet enabled (614157)
4296 - fixed infinite server hang, should a client not respond to an AST
4297 - avoid going into recovery if user calls readlink() with a buffer
4298 that's too small (613941)
4299 - AST RPCs no longer require replies (614867) -- this may be changed
4300 - don't crash server if client sends an IOV that's too big (611336)
4301 - fixed lock conversion deadlock (611892)
4302 - fixed the following of symlinks (614622)
4303 * recovery: the server can remove locks from a client that dies, other
4304 clients can make progress
4305 * more extN patch fixes
4306 * compile-time configurable ptlrpc buffer allocations
4308 - collaborative read cache document
4309 - Lustre Lite Performance CDR document-in-progress
4311 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4314 - fix typo in patch-2.4.18
4316 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4319 - clear ptlrpc request each time in handle_incoming_request()
4320 - unlink of files now destroys the object on the OST
4322 2002-09-19 Peter Braam <braam@clusterfs.com>
4324 * add hard link support
4325 * change obdfile creation method
4326 * kernel patch changed
4328 2002-09-19 Peter Braam <braam@clusterfs.com>
4331 - stack overflow bug in extN fixed
4333 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4335 * documentation updates
4336 - add man pages for config tools
4337 - update tests/README to describe testing with new config tools
4338 - finish metadata API descriptions
4339 * bug fixes and cleanups
4340 - statfs workaround for 16TB limit
4341 - LOV stripe allocation improved, can stripe on subset of OSTs
4342 - LOV file size/IO offset was wrong for files > 4GB in size
4343 - object EA data was being dropped, caused files to be unreadable
4344 - memory overflow with non-LOV OST caused memory corruption
4345 - fixed regression tests to work with new config tools, obdfilter
4346 - fixed bug when directory size became larger than 1 block
4347 - fixed bug (for single client case) when PWD was deleted
4348 - invalidate local directory pages when doing intent-based ops
4349 - avoid LDLM oops when lock callback contained bad data
4351 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4353 * documentation updates
4354 * bug fixes and cleanups
4355 - configuration tools
4358 - 64-bit compile warnings
4359 - 64-bit internal statfs data
4361 * test_brw on persistent OST devices
4363 * lprocfs (disabled)
4365 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4367 * documentation updates
4368 * bug fixes and cleanups
4369 * configuration tools
4371 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4374 * many small fixes to 0_5_4
4375 * io/network handling
4376 * thinkos in MDS operations
4378 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4381 * crucial basic fixes to 0.5.3
4382 * IOR, Iozone work over Elan
4385 2002-08-07 Phil Schwan <phil@clusterfs.com>
4386 * version 0_5_3, our first alpha
4387 * we use the new Portals iovs
4388 * documentation updates
4389 * bug fixes and cleanups
4390 * small changes in the DLM wire protocol
4392 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4393 * version 0_5_1 with some initial stability,
4394 * locking on MD and file I/O.
4395 * documentation updates
4396 * several bug fixes since 0.5.0
4397 * small changes in wire protocol
4399 2002-07-18 Phil Schwan <phil@clusterfs.com>
4401 * delivered as Lustre Light Alpha
4402 * fixed a crash after handling invalid MDS requests
4403 * fixed directory pages for architectures with non-4k pages sizes
4405 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4406 * release version v0_4_4
4407 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4409 * Fixes a number of interruption problems with OST operations.
4410 * Update documentation for portals header changes
4411 * Move all wire protocol structs/defines to lustre_idl.h
4412 * Fixes symlink length bug.
4413 * Add tcpdump to repository.
4415 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4416 * release version v0_4_3
4417 * Fixes statfs for inodes on extN.
4418 * Fixes bug in runtests which would delete /etc/hosts.
4419 * Use 64-bit object IDs wherever possible (not into VFS though)
4420 Remove ost_get_info, which is unused by lustre, and out of date.
4422 2002-07-03 Peter Braam <braam@clusterfs.com>
4423 * release version v0_4_2 Fixes a lookup error (type not passed)
4424 * move forward to head of Portals
4425 * move forward to latest Lustre kernel
4427 2002-06-25 Peter Braam <braam@clusterfs.com>
4428 * release version v0_4_1. Hopefully stable on single node use.