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).
15 Severity : enhancement
17 Description: Add rhel5 support to HEAD.
21 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
22 Details : access to bitfield in structure is always rounded to long
23 and this produce problem with not atomic change any bit.
27 Description: Lustre make rpms failed.
28 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
29 builds Lustre rpms from tarball.
32 Frequency : rare on shutdown ost
34 Description: If llog cancel was not send before clean_exports phase, this can
35 produce deadlock in llog code.
36 Details : If llog thread has last reference to obd and call class_import_put
37 this produce deadlock because llog_cleanup_commit_master wait when
38 last llog_commit_thread exited, but this never success because was
39 called from llog_commit_thread.
43 Description: allow userland application know is lost one of stripes.
44 Details : fill lvb_blocks with error code on ost and return it to
45 application if error flag found.
49 Description: NULL lov_tgts causing MDS oops
50 Details : more safe checks for NULL lov_tgts for avoid oops.
52 Severity : enhancement
54 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
58 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
59 Details : If an OST is down the MDS will hang indefinitely in
60 obd_statfs_rqset() waiting for the statfs data. While for
61 MDS QOS usage of statfs, it should not stuck in waiting.
63 Severity : enhancement
65 Description: remote_acl support
66 Details : Support ACL-based permission check for remote user.
67 Support setfacl/getfacl for remote user with the utils
68 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
69 the system "{s,g}etfacl" utils.
71 Severity : enhancement
73 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
75 Severity : enhancement
77 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
81 Description: make mgs_setparam() handle fsname containing dash
82 Details : fsname containing a dash does not work with lctl conf_param
84 Severity : enhancement
86 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
88 Severity : enhancement
90 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
94 Frequency : rare, at shutdown
95 Description: access already free / zero obd_namespace.
96 Details : if client_disconnect_export was called without force flag set,
97 and exist connect request in flight, this can produce access to
98 NULL pointer (or already free pointer) when connect_interpret
99 store ocd flags in obd_namespace.
103 Frequency : only at startup
104 Description: not alloc memory with spinlock held.
105 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
106 if any spinlock held.
108 Severity : enhancement
110 Description: make lustre randomly fail allocating memory
111 Details : Make lustre randomly failed allocating memory for testing purpose.
113 Severity : enhancement
115 Description: lost problems with lov objid file
116 Details : Fixes some scability and access to not inited memory problems
117 in work with lov objdid file.
122 Description: lfs find does not continue on file error
123 Details : Continue other files processing when a file/dir is absent.
127 Description: Inconsistent usage of lustre_pack_reply()
128 Details : Standardize the usage of lustre_pack_reply() such that it
129 always generate a CERROR on failure.
132 Frequency : occasional
134 Description: MDS hang or stay in waiting lock
135 Details : If client receive lock with CBPENDING flag ldlm need send lock
136 cancel as separate rpc, to avoid situation when cancel request
137 can't processed due all i/o threads stay in wait lock.
141 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
145 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
149 Description: Update to SLES9 kernel-2.6.5-7.287.3.
153 Description: Update to RHEL4 latest kernel.
155 Severity : enhancement
157 Description: Build SLES10 patchless client fails
158 Details : The configure was broken by run ./configure with
159 --with-linux-obj=.... argument for patchless client. When the
160 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
161 can't search header adequately. Use absolute path such as
162 -I($LINUX)/include instead.
166 Description: interrupt oig_wait produce painc on resend.
167 Details : brw_redo_request can be used for resend requests from ptlrpcd and
168 private set, and this produce situation when rq_ptlrpcd_data not
169 copyed to new allocated request and triggered LBUG on assert
170 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
171 ptlrpcd set if request is changed and can be safety changed to use
176 Description: LASSERT_{REQ,REP}SWAB macros are buggy
177 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
178 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
179 these macros and replace them with their difinitions instead.
183 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
184 Details : Update which_patch & target file for SLES10 latest kernel.
187 Description: Remove client patches from SLES 10 kernel.
188 Details : This causes SLES 10 clients to behave as patchless clients
189 even on a Lustre-patched (server) kernel.
191 Severity : enhancement
193 Description: self-adjustable client's lru lists
194 Details : use adaptive algorithm for managing client cached locks lru
195 lists according to current server load, other client's work
196 pattern, memory activities, etc. Both, server and client
197 side namespaces provide number of proc tunables for controlling
200 Severity : enhancement
202 Description: light-weight GSS support
203 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
204 and reduce performance overhead.
206 Severity : enhancement
208 Description: Linux keyring support
209 Details : Support using service of Linux keyring for Lustre GSS internal
210 context refresh/cache mechanism.
214 Description: Fix errors in lfs documentation
215 Details : Fixes man pages
219 Description: don't use GFP_* in generic Lustre code.
220 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
224 Description: obdclass is limited by single OBD_ALLOC(idarray)
225 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
229 Description: warnings with build patchless client with vanila 2.6.19 and up
230 Details : change old ctl_table style and replace ctl_table/ctl_table_header
231 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
235 Frequency : Only for SLES
236 Description: sanity_quota fail test_1
237 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
238 file, which makes TSTID[2] inunique.
242 Frequency : Only for SLES9
243 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
244 Details : Update target/ChangeLog/which_patch .
248 Frequency : Only for test_10 in sanity-quota.sh
249 Description: a bug in quota test code
250 Details : checking a test flag in a wrong place
254 Frequency : Only for test_13 in sanity-quota.sh
255 Description: a bug in quota test script
256 Details : shouldn't check overfull allocated quota
260 Description: jbd statistics
261 Details : Port older jbd statistics patch for sles10
263 Severity : enhancement
265 Description: Add printing inode info into message about error in writepage.
270 Description: reply_lock_interpret crash due to race with it and lock cancel.
271 Details : Do not replay locks that are being cancelled. Do not reference
272 locks by their address during replay, just by their handle.
276 Frequency : When flocks are used.
277 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
278 Details : Flock locks might destroy just granted lock if it could be merged
279 with another existing flock, this is done in completion handler,
280 so teach ldlm_cli_enquque_fini that this is a valid case for
286 Description: Oops in read and write path when failing to allocate lock.
287 Details : Check if lock allocation failed and return error back.
291 Description: lstripe command fails for valid OST index
292 Details : The stripe offset is compared to lov->desc.ld_tgt_count
293 instead of lov->desc.ld_active_tgt_count.
297 Description: sanity.sh failed test 103
298 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
303 Description: df doesn't work properly if diskfs blocksize != 4K
304 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
308 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
309 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
313 Description: obd_health_check_timeout too short
314 Details : set obd_health_check_timeout as 1.5x of obd_timeout
318 Description: llapi_file_create() does not allow some changes
319 Details : add llapi_file_open() that allows specifying the mode and
320 open flags, and also returns an open file handle.
324 Description: merge and cleanup kernel patches.
325 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
329 Description: Add journal checksum support.(Kernel part)
330 Details : The journal checksum feature adds two new flags i.e
331 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
332 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
333 indicates that the commit block contains the checksum for
334 the blocks described by the descriptor blocks. Now commit
335 record can be sent to disk without waiting for descriptor
336 blocks to be written to disk. This behavior is controlled
337 using JBD2_FEATURE_ASYNC_COMMIT flag.
341 Description: OSS needs mutliple precreate threads
342 Details : Add ability to start more than one create thread per OSS.
346 Description: Sanity.sh test_65k failed: setstripe should have succeeded
347 Details : execute lfs setstripe on client
351 Description: mds_obd_create error creating tmp object
352 Details : When the user sets quota on root, llog will be affected and can't
353 create files and write files.
356 Frequency : Always on ia64 patchless client, and possibly others.
358 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
359 Details : This allows the patchless client to be loaded on architectures
364 Description: RedHat Update kernel for RHEL5
365 Details : Add the kernel config file for RHEL5.
369 Description: RedHat Update kernel for RHEL5
370 Details : Modify the kernel config file more closer RHEL5.
374 Description: Build failure against Centos5 (RHEL5)
375 Details : Define PAGE_SIZE when it isn't present.
379 Description: test_7 on sanity-quota.sh can't be used on separate machines
380 Details : add facet to handle it
384 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
385 Details : with NFS, the anon dentry's parent was set to itself in
386 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
387 corresponding dentry other than use rec->ur_name.
389 Severity : enhancement
391 Description: lfs setstripe enhancement
392 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
396 Description: enable data checksumming by default
397 Details : enable checksum by default, allow --disable-checksum
398 configure option and "-o nochecksum" mount option. Checksums
399 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
403 Description: lustre support for RHEL5
404 Details : Add support for RHEL5.
408 Description: Client eviction due to failover config
409 Details : after a connection loss, the lustre client should attempt to
410 reconnect to the last active server first before trying the
411 other potential connections.
415 Description: when mds and osts use different quota unit(32bit and 64bit),
416 quota will be released repeatly.
417 Details : void sending multiple quota reqs to mds, which will keep the
418 status between the reqs.
422 Description: osts not allocated evenly to files
423 Details : change the condition to increase offset_idx
426 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
428 Description: Data corruption for OSTs that are formatted larger than 2TB
430 Details : When generating the bio request for lustre file writes the
431 sector number would overflow a temporary variable before being
432 used for the IO. The data reads correctly from Lustre (which
433 will overflow in a similar manner) but other file data or
434 filesystem metadata may be corrupted in some cases.
438 Description: Tune the kernel for good SCSI performance.
439 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
440 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
445 Description: rewrite ext2-derived code in obdclass/uuid.c
446 Details : rewrite inherited code (uuid parsing code from ext2 utils)
447 from scratch preserving functionality.
451 Description: "lfs find -obd UUID" prints directories
452 Details : "lfs find -obd UUID" will return all directory names instead
453 of just file names. It is incorrect because the directories
454 do not reside on the OSTs.
458 Description: use i_size_read and i_size_write in 2.6 port
459 Details : replace inode->i_size access with i_size_read/write()
462 Frequency : when removing large files
464 Description: scheduling issue during removal of large Lustre files
465 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
466 It causes scheduling issues when removing large files (17TB in the
470 Frequency : only with liblustre clients on XT3
472 Description: evictions taking too long
473 Details : allow llrd to evict clients directly on OSTs
476 Frequency : only on ppc
478 Description: /proc/fs/lustre/devices broken on ppc
479 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
480 We should make sure the type of 'index' is loff_t and then cast
481 explicitly as needed below. Do not assign an explicitly cast
485 Frequency : only for rhel5
487 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
488 Details : Modify the target file & which_kernel.
490 Severity : enhancement
492 Description: omit set fsid for export NFS
493 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
494 as FS_REQUIRES_DEV for avoid problems with generate fsid.
498 Frequency : Always, for kernels after 2.6.16
499 Description: Fix warning idr_remove called for id=.. which is not allocated.
500 Details : Last kernels save old s_dev before kill super and not allow
501 to restore from callback - restore it before call kill_anon_super.
505 Description: buffer overruns could theoretically occur
506 Details : llapi_semantic_traverse() modifies the "path" argument by
507 appending values to the end of the origin string, and a
508 overrun may occur. Adding buffer overrun check in liblustreapi.
512 Description: Fix error on 'ls .' at the top of the Lustre mount.
513 Details : Don't revalidate dentry if it is a root dentry.
517 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
518 Details : Modify vm-tunables-rhel4.patch.
522 Description: Kernel config for 2.6.18-vanilla.
523 Details : Modify targets/2.6-vanilla.target.in.
524 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
525 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
526 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
527 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
531 Description: improve handling recoverable errors
532 Details : if request processig with error which can be recoverable on server
533 request should be resend, otherwise page released from cache and
538 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
539 Details : Modify target file & which_patch.
540 A flaw was found in the IA32 system call emulation provided
541 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
542 value could be stored in the %RAX register, which could trigger an
543 out-of-bounds system call table access. An untrusted local user
544 could exploit this flaw to run code in the kernel
545 (ie a root privilege escalation). (CVE-2007-4573).
549 Description: O_DIRECT bypasses client statistics.
550 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
551 read_ahead_stats were updated. copied the stats section from
552 osc_send_oap_rpc() into async_internal().
556 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
560 Description: change order of libsysio includes
561 Details : '#include sysio.h' should always come before '#include xtio.h'
565 Description: handle "serious error: objid * already exists" more gracefully
566 Details : If LAST_ID value on disk is smaller than the objects existing in
567 the O/0/d* directories, it indicates disk corruption and causes an
568 LBUG(). If the object is 0-length, then we should use the existing
569 object. This will help to avoid a full fsck in most cases.
571 Severity : enhancement
573 Description: adapt the lustre_config script to support the upgrade case
574 Details : Add "-u" option for lustre_config script to support upgrading 1.4
575 server targets to 1.6 in parallel.
579 Description: To avoid grant space > avaible space when the disk is almost
580 full. Without this patch you might see the error "grant XXXX >
581 available" or some LBUG about grant, when the disk is almost
583 Details : In filter_check_grant, for non_grant cache write, we should
584 check the left space by if (*left > ungranted + bytes), instead
585 of (*left > ungranted), because only we are sure the left space
586 is enough for another "bytes", then the ungrant space should be
587 increase. In client, we should update cl_avail_grant only there
588 is OBD_MD_FLGRANT in the reply.
592 Description: Update RHEL 4 kernel to fix local root privilege escalation.
593 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
594 described in CVE-2007-4573. This problem could allow untrusted
595 local users to gain root access.
598 Frequency : when using O_DIRECT and quotas
600 Description: Incorrect file ownership on O_DIRECT output files
601 Details : block usage reported by 'lfs quota' does not take into account
602 files that have been written with O_DIRECT.
607 Description: touch file failed when fs is not full
608 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
609 otherwise we can get ENOSP while fs is not full.
613 Description: parallel lock callbacks
614 Details : Instead of sending blocking and completion callbacks as separated
615 requests, adding them to a set and sending in parallel.
618 Frequency : only for Cray XT3
619 Bugzilla : 12829/13455
620 Description: Changing primary group doesn't change the group lustre assigns to
622 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
623 overridden with the primary group provided by the group upcall,
624 whereas the supplied fsgid can be trusted if it is in the list of
625 supplementary groups returned by the group upcall.
627 Severity : enhancement
629 Description: Allow masking D_WARNING, D_ERROR messages from console
630 Details : Console messages can now be disabled via lnet.printk.
634 Description: User code with malformed file open parameter crashes client node
635 Details : Before packing join_file req, all the related reference should be
636 checked carefully in case some malformed flags cause fake
637 join_file req on client.
641 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
642 Details : ldlm_completion_ast() assumes that a lock is granted when the req
643 mode is equal to the granted mode. However, it should also check
644 that LDLM_FL_CP_REQD is not set.
648 Description: Heavy nfs access might result in deadlocks
649 Details : After ELC code landed, it is now improper to enqueue any mds
650 locks under och_sem, because enqueue might want to decide to
651 cancel open locks for same inode we are holding och_sem for.
655 Description: Client eviction while running blogbench
656 Details : A lot of unlink operations with concurrent I/O can lead to a
657 deadlock causing evictions. To address the problem, the number of
658 oustanding OST_DESTROY requests is now throttled to
659 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
660 callbacks are processed in priority.
662 --------------------------------------------------------------------------------
664 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
666 * Support for kernels:
667 2.6.5-7.283 (SLES 9),
668 2.6.9-55.EL (RHEL 4),
669 2.6.16.46-0.14 (SLES 10),
670 2.6.18.8 vanilla (kernel.org)
671 * Client support for unpatched kernels:
672 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
673 2.6.16 - 2.6.22 vanilla (kernel.org)
674 * Due to recently discovered recovery problems, we do not recommend
675 using patchless RHEL 4 clients with this or any earlier release.
676 * Recommended e2fsprogs version: 1.39.cfs8
677 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
678 * Starting with this release, the ldiskfs backing filesystem required
679 by Lustre is now in its own package, lustre-ldiskfs. This package
680 should be installed. It is versioned separately from Lustre and
681 may be released separately in future.
685 Description: block reactivating mgc import until all deactivates complete
686 Details : Fix race when failing back MDT/MGS to itself (testing)
689 Frequency : at statup only
691 Description: mds_lov_synchronize race leads to various problems
692 Details : simultaneous MDT->OST connections at startup can cause the
693 sync to abort, leaving the OSC in a bad state.
695 Severity : enhancement
697 Description: add optional extra BUILD_VERSION info
698 Details : add a new environment variable (namely LUSTRE_VERS) which allows
699 to override the lustre version.
702 Frequency : 2.6.18 servers only
704 Description: ll_kern_mount() doesn't release the module reference
705 Details : The ldiskfs module reference count never drops down to 0
706 because ll_kern_mount() doesn't release the module reference.
711 Description: server LBUG when using old ost_num_threads parameter
712 Details : Accept the old ost_num_threads parameter but warn that it
713 is deprecated, and fix an off-by-one error that caused an LBUG.
718 Description: Transient SCSI error results in persistent IO issue
719 Details : iobuf->dr_error is not reinitialized to 0 between two
723 Frequency : sometimes when underlying device returns I/O errors
725 Description: OSTs not going read-only during write failures
726 Details : OSTs are not remounted read-only when the journal commit threads
727 get I/O errors because fsfilt_ext3 calls journal_start/stop()
728 instead of the ext3 wrappers.
732 Description: poor connect scaling with increasing client count
733 Details : Don't run filter_grant_sanity_check for more than 100 exports
734 to improve scaling for large numbers of clients.
737 Frequency : SLES10 only
739 Description: sanity-quota.sh quotacheck failed: rc = -22
740 Details : Quotas cannot be enabled on SLES10.
743 Frequency : liblustre clients only
745 Description: getdirentries does not give error when run on compute nodes
746 Details : getdirentries does not fail when the size specified as an argument
747 is too small to contain at least one entry
749 Severity : enhancement
751 Description: Add LNET router traceability for debug purposes
752 Details : If a checksum failure occurs with a router as part of the
753 IO path, the NID of the last router that forwarded the bulk data
754 is printed so it can be identified.
759 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
760 Details : Due to a race condition, liblustre clients were occasionally
763 Severity : enhancement
765 Description: lfs setstripe use optional parameters instead of postional
768 Severity : enhancement
770 Description: Nanosecond timestamp support for ldiskfs
771 Details : The on-disk ldiskfs filesystem has added support for nanosecond
772 resolution timestamps. There is not yet support for this at
773 the Lustre filesystem level.
776 Frequency : during server recovery
778 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
779 Details : request with rq_no_resend flag not awake l_wait_event if they get a
783 Frequency : nfs export on patchless client
785 Description: connectathon hang when test nfs export over patchless client
786 Details : Disconnected dentry cannot be found with lookup, so we do not need
787 to unhash it or make it invalid
790 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
791 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
792 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
797 Description: open req refcounting wrong on reconnect
798 Details : If reconnect happened between getting open reply from server and
799 call to mdc_set_replay_data in ll_file_open, we will schedule
800 replay for unreferenced request that we are about to free.
801 Subsequent close will crash in variety of ways.
802 Check that request is still eligible for replay in
803 mdc_set_replay_data().
808 Description: disable writes to filesystem when reading health_check file
809 Details : the default for reading the health_check proc file has changed
810 to NOT do a journal transaction and write to disk, because this
811 can cause reads of the /proc file to hang and block HA state
812 checking on a healthy but otherwise heavily loaded system. It
813 is possible to return to the previous behaviour during configure
814 with --enable-health-write.
816 Severity : enhancement
818 Description: 64-bit inode version
819 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
820 to the inode. This will be required for version-based recovery.
825 Description: MDS fails to start if a duplicate client export is detected
826 Details : in some rare cases it was possible for a client to connect to
827 an MDS multiple times. Upon recovery the MDS would detect this
828 and fail during startup. Handle this more gracefully.
830 Severity : enhancement
832 Description: Add -o localflock option to simulate old noflock
834 Details : This will achieve local-only flock/fcntl locks
840 Description: log_commit_thread vs filter_destroy race leads to crash
841 Details : Take import reference before releasing llog record semaphore
846 Description: Wrong request locking in request set processing
847 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
848 imp_delayed_list, in this place should be used imp_lock.
851 Frequency : when reconnection
853 Description: Grant Leak when osc reconnect to OST
854 Details : When osc reconnect ost, OST(filter) should check whether it
855 should grant more space to client by comparing fed_grant and
856 cl_avail_grant, and return the granted space to client instead
857 of "new granted" space, because client will call osc_init_grant
858 to update the client grant space info.
861 Frequency : when client reconnect to OST
863 Description: Grant Leak when osc do resend and replay bulk write
864 Details : When osc reconnect to OST, OST(filter)should clear grant info of
865 bulk write request, because the grant info while be sync between
866 OSC and OST when reconnect, and we should ignore the grant info
867 these of resend/replay write req.
872 Description: Grant space more than avaiable left space sometimes.
873 Details : When then OST is about to be full, if two bulk writing from
874 different clients came to OST. Accord the avaliable space of the
875 OST, the first req should be permitted, and the second one
876 should be denied by ENOSPC. But if the seconde arrived before
877 the first one is commited. The OST might wrongly permit second
878 writing, which will cause grant space > avaiable space.
881 Frequency : when client is evicted
883 Description: Grant might be wrongly erased when osc is evicted by OST
884 Details : when the import is evicted by server, it will fork another
885 thread ptlrpc_invalidate_import_thread to invalidate the
886 import, where the grant will be set to 0. While the original
887 thread will update the grant it got when connecting. So if
888 the former happened latter, the grant will be wrongly errased
889 because of this race.
894 Description: Checking Stale with correct fid
895 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
896 to check whether it is stale, because sometimes, we want the
897 enqueue happened anyway, and op_data.fid2 will not be initialized.
899 Severity : enhancement
901 Description: update patchless client
902 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
905 Frequency : only with 2.4 kernel
907 Description: random memory corruption
908 Details : size of struct ll_inode_info is to big for union inode.u and this
909 can be cause of random memory corruption.
914 Description: Memory leak in recovery
915 Details : Lov_mds_md was not free in an error handler in mds_create_object.
916 It should also check obd_fail before fsfilt_start, otherwise if
917 fsfilt_start return -EROFS,(failover mds during mds recovery).
918 then the req will return with repmsg->transno = 0 and rc = EROFS.
919 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
920 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
921 be freed no matter whether fsfilt_commit success or not.
924 Frequency : only with huge count clients
926 Description: Prevents from taking the superblock lock in llap_from_page for
928 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
929 for llap_from_page call in ll_removepage prevents from taking the
930 superblock lock for a soon died page.
935 Description: Not check open intent error before release open handle
936 Details : in some rare cases, the open intent error is not checked before
937 release open handle, which may cause
938 ASSERTION(open_req->rq_transno != 0), because it tries to release
939 the failed open handle.
944 Description: Set cat log bitmap only after create log success.
945 Details : in some rare cases, the cat log bitmap is set too early. and it
946 should be set only after create log success.
950 Description: Accessing a block bevice can re-enable I/O when Lustre is
951 tearing down a device.
952 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
957 Description: service threads may hog cpus when there are a lot of requests
958 Details : Insert cond_resched to give other threads a chance to use some CPU
963 Description: the cat log was not initialized in recovery
964 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
965 unlink log on mds will not be initialized until mds post
966 recovery. And also in mds post recovery, the unlink log will
967 initialization will be done asynchronausly, so there will be race
968 between add unlink log and unlink log initialization.
972 Description: brw_stats were being printed incorrectly
973 Details : brw_stats were being printed as log2 but all of them were not
974 recorded as log2. Also remove some code duplication arising from
975 filter_tally_{read,write}.
979 Frequency : rare, only in recovery.
980 Description: ASSERTION(req->rq_type != LI_POISON) failed
981 Details : imp_lock should be held while iterating over imp_sending_list for
982 prevent destroy request after get timeout in ptlrpc_queue_wait.
986 Description: replay-single.sh test 52 fails
987 Details : A lock's skiplist need to be cleanup when it being unlinked
988 from its resource list.
992 Description: Short directio read returns full requested size rather than
994 Details : Direct I/O operations should return actual amount of bytes
995 transferred rather than requested size.
997 Severity : enhancement
999 Description: metadata RPC reduction (e.g. for rm performance)
1000 Details : decrease the amount of synchronous RPC between clients and servers
1001 by canceling conflicing lock before the operation on the client
1002 and packing thier handles into the main operation RPC to server.
1004 Severity : enhancement
1006 Description: Async OSC create to avoid the blocking unnecessarily.
1007 Details : If a OST has no remain object, system will block on the creating
1008 when need to create a new object on this OST. Now, ways use
1009 pre-created objects when available, instead of blocking on an
1010 empty osc while others are not empty. If we must block, we block
1011 for the shortest possible period of time.
1013 Severity : enhancement
1015 Description: refine locking for avoid write wrong info into lov_objid file
1016 Details : fix possible races with add new target and write/update data in
1019 --------------------------------------------------------------------------------
1021 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1026 Frequency : on some architectures
1028 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1029 Details : Uninitialized flags sometimes cause configuration commands to
1033 Frequency : patchless clients only
1035 Description: missing __iget() symbol export
1036 Details : The __iget() symbol export is missing. To avoid the need for
1037 this on patchless clients the deathrow inode reaper is turned
1038 off, and we depend on the VM to clean up old inodes. This
1039 dependency was during via the fix for bug 12181.
1041 --------------------------------------------------------------------------------
1043 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1045 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1046 INTEROPERATE with older versions automatically. In many cases a
1047 special upgrade step is needed. Please read the
1048 user documentation before upgrading any part of a live system.
1049 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1050 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1051 * WARNING: Lustre configuration and startup changes are required with
1052 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1058 Severity : enhancement
1060 Description: MountConf
1061 Details : Lustre configuration is now managed via mkfs and mount
1062 commands instead of lmc and lconf. New obd types (MGS, MGC)
1063 are added for dynamic configuration management. See
1064 https://mail.clusterfs.com/wikis/lustre/MountConf for
1067 Severity : enhancement
1069 Description: dynamic OST addition
1070 Details : OSTs can now be added to a live filesystem
1072 Severity : enhancement
1074 Description: startup order invariance
1075 Details : MDTs and OSTs can be started in any order. Clients only
1076 require the MDT to complete startup.
1078 Severity : enhancement
1080 Description: parallel, asynchronous orphan cleanup
1081 Details : orphan cleanup is now performed in separate threads for each
1082 OST, allowing parallel non-blocking operation.
1084 Severity : enhancement
1086 Description: optimized stripe assignment
1087 Details : stripe assignments are now made based on ost space available,
1088 ost previous usage, and OSS previous usage, in order to try
1089 to optimize storage space and networking resources.
1091 Severity : enhancement
1093 Description: Permanently set tunables
1094 Details : All writable /proc/fs/lustre tunables can now be permanently
1095 set on a per-server basis, at mkfs time or on a live system.
1097 Severity : enhancement
1099 Description: Lustre message v2
1100 Details : Add lustre message format v2.
1102 Severity : enhancement
1104 Description: client OST exclusion list
1105 Details : Clients can be started with a list of OSTs that should be
1106 declared "inactive" for known non-responsive OSTs.
1110 Description: SPEC SFS validation failure on NFS v2 over lustre.
1111 Details : Changes the blocksize for regular files to be 2x RPC size,
1112 and not depend on stripe size.
1114 Severity : enhancement
1116 Description: Multiple MD RPCs in flight.
1117 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1118 To avoid overly-overloading MDS, introduce a limit on number of
1119 MDS RPCs in flight for a single client and add /proc controls
1120 to adjust this limit.
1122 Severity : enhancement
1124 Description: client read/write statistics
1125 Details : Add client read/write call usage stats for performance
1126 analysis of user processes.
1127 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1128 file access. extents_stats shows chunk size distribution.
1129 extents_stats_per_process show chunk size distribution per
1132 Severity : enhancement
1134 Description: mds statistics
1135 Details : Add detailed mds operations statistics in
1136 /proc/fs/lustre/mds/*/stats.
1140 Description: Failure of copying files with lustre special EAs.
1141 Details : Client side always return success for setxattr call for lustre
1142 special xattr (currently only "trusted.lov").
1147 Description: Refcount LNET uuids
1148 Details : The global LNET uuid list grew linearly with every startup;
1149 refcount repeated list entries instead of always adding to
1152 Severity : enhancement
1154 Description: Dynamic service threads
1155 Details : Within a small range, start extra service threads
1156 automatically when the request queue builds up.
1159 Frequency : mixed-endian client/server environments
1161 Description: mixed-endian crashes
1162 Details : The new msg_v2 system had some failures in mixed-endian
1165 Severity : enhancement
1167 Description: Easy OST removal
1168 Details : OSTs can be permanently deactivated with e.g. 'lctl
1169 conf_param lustre-OST0001.osc.active=0'
1171 Severity : enhancement
1173 Description: MGS proc entries
1174 Details : Added basic proc entries for the MGS showing what filesystems
1177 Severity : enhancement
1179 Description: provide MGS failover
1180 Details : Added config lock reacquisition after MGS server failover.
1182 Severity : enhancement
1184 Description: add Linux 2.4 support
1185 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1189 Description: a large application tries to do I/O to the same resource and dies
1190 in the middle of it.
1191 Details : Check the req->rq_arrival time after the call to
1192 ost_brw_lock_get(), but before we do anything about
1193 processing it & sending the BULK transfer request. This
1194 should help move old stale pending locks off the queue as
1195 quickly as obd_timeout.
1198 Frequency : when an incorrect nid is specified during startup
1200 Description: ptlrpc connect to non-existant node causes kernel crash
1201 Details : LNET can't be re-entered from an event callback, which
1202 happened when we expire a message after the export has been
1203 cleaned up. Instead, hand the zombie cleanup off to another
1206 Severity : enhancement
1208 Description: plain/inodebits lock performance improvement
1209 Details : Grouping plain/inodebits in granted list by their request modes
1210 and bits policy, thus improving the performance of search through
1214 Frequency : only if OST filesystem is corrupted
1216 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1217 Details : for a short time RPCs with bulk IO are in the replay list,
1218 but replay of bulk IOs is unimplemented. If the OST filesystem
1219 is corrupted due to disk cache incoherency and then replay is
1220 started it is possible to trip an assertion. Avoid putting
1221 committed RPCs into the replay list at all to avoid this issue.
1224 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1226 Description: System hang on startup
1227 Details : This bug allowed the liblustre (e.g. catamount) client to
1228 return to the app before handling all startup RPCs. This
1229 could leave the node unresponsive to lustre network traffic
1230 and manifested as a server ptllnd timeout.
1232 Severity : enhancement
1234 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1235 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1236 boolean to control whether to print peer debug info when a
1237 client's RPC times out.
1240 Frequency : only for kernels with patches from Lustre below 1.4.3
1242 Description: Remove old rdonly API
1243 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1246 Frequency : only for devices with external journals
1248 Description: Set external device read-only also
1249 Details : During a commanded failover stop, we set the disk device
1250 read-only while the server shuts down. We now also set any
1251 external journal device read-only at the same time.
1254 Frequency : when upgrading from 1.4 while trying to change parameters
1256 Description: The wrong (new) MDC name was used when setting parameters for
1257 upgraded MDT's. Also allows changing of OSC (and MDC)
1258 parameters if --writeconf is specified at tunefs upgrade time.
1261 Frequency : when setting specific ost indicies
1263 Description: QOS code breaks on skipped indicies
1264 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1265 created with --index need not be sequential.
1267 Severity : enhancement
1269 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1270 Details : The uninit_groups feature works in conjunction with the kernel
1271 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1272 up the pass1 processing of e2fsck. This is a read-only feature
1273 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1274 mount filesystems that have had this feature enabled.
1276 Severity : enhancement
1278 Description: Improve multi-block allocation algorithm to avoid fragmentation
1279 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1280 allocation locality and avoid filesystem fragmentation.
1282 ------------------------------------------------------------------------------
1284 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1286 * Support for kernels:
1287 2.4.21-47.0.1.EL (RHEL 3)
1288 2.6.5-7.283 (SLES 9)
1289 2.6.9-42.0.10.EL (RHEL 4)
1290 2.6.12.6 vanilla (kernel.org)
1291 2.6.16.27-0.9 (SLES 10)
1292 * Recommended e2fsprogs version: 1.39.cfs5
1294 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1298 Frequency : occasional, depends on client load and configuration
1299 Bugzilla : 12181, 12203
1300 Description: data loss for recently-modified files
1302 Details : In some cases it is possible that recently written or created
1303 files may not be written to disk in a timely manner (this should
1304 normally be within 30s unless client IO load is very high).
1305 The problem appears as zero-length files or files that are a
1306 multiple of 1MB in size after a client crash or client eviction
1307 that are missing data at the end of the file.
1309 This problem is more likely to be hit on clients where files are
1310 repeatedly created and unlinked in the same directory, clients
1311 have a large amount of RAM, have many CPUs, the filesystem has
1312 many OSTs, the clients are rebooted frequently, and/or the files
1313 are not accessed by other nodes after being written.
1315 The presence of the problem can be detected by looking at
1316 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1317 smaller than the second (nr_unused) then dirty files will not
1318 be flushed automatically to disk. "sync; sleep 10" should be
1319 run several times on the node before unmounting it to update
1320 Lustre (this is also safe to run on nodes without this problem).
1322 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1323 that can cause this same problem, so customers using that kernel
1324 also need to update the kernel in addition to Lustre. In order
1325 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1327 It is normal that files written just before a client crash (less
1328 than 30s) may not yet have been flushed to disk, even for local
1332 Frequency : frequent on thin XT3 nodes
1334 Description: UUID collision on thin XT3 Linux nodes
1335 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1336 randomly, since we relied on an insufficiently-seeded PRNG.
1341 Description: OSS hangs after "All ost request buffers busy"
1342 Details : A deadlock between quota and journal operations caused OSS
1343 hangs after printing "All ost request buffers busy."
1346 Frequency : always on liblustre builds
1348 Description: Cleanup compiler warnings on liblustre
1351 Frequency : always on liblustre builds on XT3
1353 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1357 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1358 > 32000 subdirectories
1359 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1360 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1361 subdirectory count crosses 32000. This will aid e2fsck to
1362 correctly handle more than 32000 subdirectories.
1365 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1367 Description: System hang on startup
1368 Details : This bug allowed the liblustre (e.g. catamount) client to
1369 return to the app before handling all startup RPCs. This
1370 could leave the node unresponsive to lustre network traffic
1371 and manifested as a server ptllnd timeout.
1373 Severity : enhancement
1375 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1376 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1377 boolean to control whether to print peer debug info when a
1378 client's RPC times out.
1383 Description: make O_SYNC working on 2.6 kernels
1384 Details : 2.6 kernels use different method for mark pages for write,
1385 so need add a code to lustre for O_SYNC work.
1390 Description: Failure to close file and release space on NFS
1391 Details : Put inode details into lock acquired in ll_intent_file_open.
1392 Use mdc_intent_lock in ll_intent_open to properly
1393 detect all kind of errors unhandled by mdc_enqueue.
1398 Description: proc file read during shutdown sometimes raced obd removal,
1400 Details : Add lock to prevent obd access after proc file removal.
1403 Frequency : Only for files larger than 4GB on 32-bit clients.
1405 Description: improperly doing page alignment of locks
1406 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1407 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1412 Description: RPCs being resent when they shouldn't be
1413 Details : Some RPCs that should not be resent are being resent. This
1414 can cause inconsistencies in the RPC state machine. Do not
1415 resend such requests.
1418 Frequency : rare, only with NFS export
1420 Description: Crash on NFS re-export node
1421 Details : under very unusual load conditions an assertion is hit in
1422 ll_intent_file_open()
1425 Frequency : only if OST filesystem is corrupted
1427 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1428 Details : for a short time RPCs with bulk IO are in the replay list,
1429 but replay of bulk IOs is unimplemented. If the OST filesystem
1430 is corrupted due to disk cache incoherency and then replay is
1431 started it is possible to trip an assertion. Avoid putting
1432 committed RPCs into the replay list at all to avoid this issue.
1437 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1438 Details : Large single O_DIRECT read and write calls can fail to allocate
1439 a sufficiently large buffer to process the request. In case of
1440 allocation failure the allocation is retried with a smaller
1441 buffer and broken into smaller requests.
1443 Severity : enhancement
1445 Description: Add -o localflock option to simulate old noflock behaviour.
1446 Details : This will achieve local-only flock/fcntl locks coherentness.
1451 Description: versioning check is incomplete
1452 Details : Checking the version difference of client vs. server, report
1453 error if the gap is too big.
1458 Description: add support PG_writeback bit
1459 Details : add support for PG_writeback bit for Lustre, for more carefull
1460 work with page cache in 2.6 kernel. This also fix some deadlocks
1461 and remove hack for work O_SYNC with 2.6 kernel.
1463 Severity : enhancement
1465 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1466 Details : The uninit_groups feature works in conjunction with the kernel
1467 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1468 up the pass1 processing of e2fsck. This is a read-only feature
1469 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1470 mount filesystems that have had this feature enabled.
1472 Severity : enhancement
1474 Description: Improve multi-block allocation algorithm to avoid fragmentation
1475 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1476 allocation locality and avoid filesystem fragmentation.
1478 ------------------------------------------------------------------------------
1480 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1482 * Support for kernels:
1483 2.6.9-42.0.3.EL (RHEL 4)
1484 2.6.5-7.276 (SLES 9)
1485 2.4.21-47.0.1.EL (RHEL 3)
1486 2.6.12.6 vanilla (kernel.org)
1487 2.6.16.21-0.8 (SLES10)
1488 * Recommended e2fsprogs version: 1.39.cfs2-0
1490 * The backwards-compatible /proc/sys/portals symlink has been removed
1491 in this release. Before upgrading, please ensure that you change
1492 any configuration scripts or /etc/sysctl.conf files that access
1493 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1494 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1495 in advance of the upgrade on any system running Lustre 1.4.6 or
1496 newer, since /proc/sys/lnet was added in that version.
1497 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1501 Frequency : only when quota is used
1503 Description: avoid scanning export list for quota master
1504 Details : Change the algorithms to avoid scanning export list in order
1505 to improve the efficiency.
1508 Frequency : MDS failover only, very rarely
1510 Description: "went back in time" messages on mds failover
1511 Details : The greatest transno may be lost when the current operation
1512 finishes with an error (transno==0) and the client's last_rcvd
1513 record is over-written. Save the greatest transno in the
1514 mds_last_transno for this case.
1517 Frequency : always for specific kernels and striping counts
1519 Description: client may get "Matching packet too big" without ACL support
1520 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1521 when trying to access files in certain configurations. The
1522 clients should in fact be denied when mounting because they do
1523 not understand ACLs.
1526 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1528 Description: many clients connecting with IO in progress causes connect timeouts
1529 Details : Avoid synchronous journal commits to avoid delays caused by many
1530 clients connecting/disconnecting when bulk IO is in progress.
1531 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1532 OST_IO_PORTAL to avoid delays behind potentially many pending
1536 Frequency : occasionally with multiple writers to a single file
1538 Description: shared writes to file may result in wrong size reported by stat()
1539 Details : Allow growing of kms when extent lock is cancelled
1542 Frequency : always with random mmap IO to multi-striped file
1544 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1545 Details : Only if the hole is at the end of OST object so that kms is too
1546 small. Fix is to increase kms accordingly in ll_nopage.
1549 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1551 Description: writes to a missing object would leak memory on the OST
1552 Details : If there is an inconsistency between the MDS and OST filesystems,
1553 such that the MDS references an object that doesn't exist, writes
1554 to that object will leak memory due to incorrect cleanup in the
1555 error handling path, eventually running out of memory on the OST.
1560 Description: Creating too long symlink causes lustre errors
1561 Details : Check symlink and name lengths before sending requests to MDS.
1564 Frequency : only if flock is enabled (not on by default)
1566 Description: posix locks not released on fd closure on 2.6.9+
1567 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1568 kernels, this caused such locks not to be released on fd close and
1569 then assertions on fs unmount about still used locks.
1572 Frequency : MDS failover only, very rarely
1574 Description: clients may get ASSERTION(granted_lock != NULL)
1575 Details : When request was taking a long time, and a client was resending
1576 a getattr by name lock request. The were multiple lock requests
1577 with the same client lock handle and
1578 mds_getattr_name->fixup_handle_for_resent_request found one of the
1579 lock handles but later failed with ASSERTION(granted_lock != NULL).
1584 Description: handle->h_buffer_credits > 0, assertion failure
1585 Details : h_buffer_credits is zero after truncate, causing assertion
1586 failure. This patch extends the transaction or creates a new
1590 Frequency : NFS re-export or patchless client
1591 Bugzilla : 11179, 10796
1592 Description: Crash on NFS re-export node (__d_move)
1593 Details : We do not want to hash the dentry if we don't have a lock.
1594 But if this dentry is later used in d_move, we'd hit uninitialised
1595 list head d_hash, so we just do this to init d_hash field but
1596 leave dentry unhashed.
1599 Frequency : NFS re-export or patchless client
1601 Description: NFS exports has problem with symbolic link
1602 Details : lustre client didn't properly install dentry when re-exported
1603 to NFS or running patchless client.
1606 Frequency : NFS re-export or patchless client
1608 Description: Various nfs/patchless fixes.
1609 Details : fixes reuse disconected alias for lookup process - this fixes
1610 warning "find_exported_dentry: npd != pd",
1611 fix permission error with open files at nfs.
1612 fix apply umask when do revalidate.
1615 Frequency : occasional
1617 Description: Crash on NFS re-export node
1618 Details : calling clear_page() on the wrong pointer triggered oops in
1619 generic_mapping_read().
1622 Frequency : rarely, using O_DIRECT IO
1624 Description: unaligned directio crashes client with LASSERT
1625 Details : check for unaligned buffers before trying any requests.
1628 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1630 Description: stale data returned from RAID cache
1631 Details : If only a small amount of IO is done to the RAID device before
1632 reading it again it is possible to get stale data from the RAID
1633 cache instead of reading it from disk.
1636 Frequency : always for sles10 kernel
1638 Description: sles10 support
1639 Details : ll_follow_link: compile fixes and using of nd_set_link
1640 under newer kernels.
1643 Frequency : depends on arch, kernel and compiler version, always on sles10
1646 Description: recursive or deep enough symlinks cause stack overflow
1647 Details : getting rid of large stack-allocated variable in
1651 Frequency : depends on hardware
1653 Description: lustre write performance loss in the SLES10 kernel
1654 Details : the performance loss is caused by using of write barriers in the
1655 ext3 code. The SLES10 kernel turns barrier support on by
1656 default. The fix is to undo that change for ldiskfs.
1658 ------------------------------------------------------------------------------
1660 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1662 * Support for kernels:
1663 2.6.9-42.0.3EL (RHEL 4)
1664 2.6.5-7.276 (SLES 9)
1665 2.4.21-47.0.1.EL (RHEL 3)
1666 2.6.12.6 vanilla (kernel.org)
1670 Frequency : quota enabled and large files being deleted
1672 Description: releasing more than 4GB of quota at once hangs OST
1673 Details : If a user deletes more than 4GB of files on a single OST it
1674 will cause the OST to spin in an infinite loop. Release
1675 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1680 Description: statfs data retrieved from /proc may be stale or zero
1681 Details : When reading per-device statfs data from /proc, in the
1682 {kbytes,files}_{total,free,avail} files, it may appear
1683 as zero or be out of date.
1686 Frequency : systems with MD RAID1 external journal devices
1688 Description: lconf's call to blkid is confused by RAID1 journal devices
1689 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1690 of returning all block devices that match the journal UUID.
1693 Frequency : always, for aggregate stripe size over 4GB
1695 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1696 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1697 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1700 Frequency : always if "lfs find" used on a local file/directory
1702 Description: "lfs find" segfaults if used on a local file/directory
1703 Details : The case where a directory component was not specified wasn't
1704 handled correctly. Handle this properly.
1707 Frequency : always on ppc64
1709 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1710 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1711 perform endianness conversion.
1714 Frequency : rarely (truncate to non-zero file size after write under load)
1715 Bugzilla : 10730, 10687
1716 Description: Files padded with zeros to next 4K multiple
1717 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1718 it is possible that files that are truncated to a non-zero size
1719 immediately after being written are filled with zero bytes beyond
1720 the truncated size. No file data is lost.
1722 Severity : enhancement
1724 Description: Allow recovery/failover for liblustre clients.
1725 Details : liblustre clients were unaware of failover configurations until
1728 Severity : enhancement
1730 Description: user file locks should fail when not mounting with flock option
1731 Details : Set up an error-returning stub in ll_file_operations.lock field
1732 to prevent incorrect behaviour when client is mounted without
1733 flock option. Also, set up properly f_op->flock field for
1737 Frequency : always on ia64
1739 Description: "lfs df" loops on printing out MDS statfs information
1740 Details : The obd_ioctl_data was not initialized and in some systems
1741 this caused a failure during the ioctl that did not return
1742 an error. Initialize the struct and return an error on failure.
1745 Frequency : SLES 9 only
1747 Description: Error of copying files with lustre special EAs as root
1748 Details : Client side always return success for setxattr call for lustre
1749 special xattr (currently only "trusted.lov").
1752 Frequency : rarely on clusters with both ia64+i386 clients
1754 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1755 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1756 clients are doing concurrent writes to the same file it is
1757 possible that the ia64 clients may LASSERT because the OST
1758 extent locks are not PAGE_SIZE aligned. Ensure that grown
1759 locks are always aligned on the request boundary.
1762 Frequency : specific use, occasional
1764 Description: Overwriting in use executable truncates on-disk binary image
1765 Details : If one node attempts to overwrite an executable in use by
1766 another node, we now correctly return ETXTBSY instead of
1767 truncating the file.
1769 Severity : enhancement
1771 Description: Async OSC create to avoid the blocking unnecessarily.
1772 Details : If a OST has no remain object, system will block on the creating
1773 when need to create a new object on this OST. Now, ways use
1774 pre-created objects when available, instead of blocking on an
1775 empty osc while others are not empty. If we must block, we block
1776 for the shortest possible period of time.
1781 Description: chmod on Lustre root is propagated to other clients
1782 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1783 invalid by the follow_mount time.
1786 Frequency : liblustre clients only
1788 Description: Race in 'instant cancel' lock handling could lead to such locks
1789 never to be granted in case of SMP MDS
1790 Details : Do not destroy not yet granted but cbpending locks in
1794 Frequency : replay/resend of open
1796 Description: non null lock assetion failure in mds_intent_policy
1797 Details : Trying to replay/resend lockless open requests resulted in
1798 mds_open() returning 0 with no lock. Now it sets a flag if
1799 a lock is going to be returned.
1801 Severity : enhancement
1803 Description: Checksum enhancements
1804 Details : New checksum enhancements allow for resending RPCs that failed
1807 Severity : enhancement
1809 Description: Tunables on number of dirty pages in cacche
1810 Details : Allow to set limit on number of dirty pages cached.
1815 Description: client crash on unmount - lock still has references
1816 Details : In some error handling cases it was possible to leak a lock
1817 reference on a client while accessing a file. This was not
1818 harmful to the client during operation, but would cause the
1819 client to crash when the filesystem is unmounted.
1822 Frequency : specific case, rare
1824 Description: ETXTBSY on mds though file not in use by client
1825 Details : ETXTBSY is no longer incorrectly returned when attempting to
1826 chmod or chown a directory that the user previously tried to
1827 execute or a currently-executing binary.
1830 Frequency : extremely rare except on liblustre-based clients
1832 Description: Lustre space not freed when files are deleted
1833 Details : Clean up open-unlinked files after client eviction. Previously
1834 the unlink was skipped and the files remained as orphans.
1839 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1840 Details : In some cases it was possible to send a blocking callback to
1841 a client doing a glimpse, even though that client didn't get
1842 a lock granted. When the glimpse lock is cancelled on the OST
1843 the freed lock is left on the waiting list and corrupted the list.
1846 Frequency : all core dumps
1848 Description: Broke core dumps to lustre
1849 Details : Negative dentry may be unhashed if parent does not have UPDATE
1850 lock, but some callers, e.g. do_coredump, expect dentry to be
1851 hashed after successful create, hash it in ll_create_it.
1853 ------------------------------------------------------------------------------
1855 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1857 * Support for kernels:
1858 2.6.9-42.0.2.EL (RHEL 4)
1859 2.6.5-7.276 (SLES 9)
1860 2.4.21-40.EL (RHEL 3)
1861 2.6.12.6 vanilla (kernel.org)
1865 Frequency : always on RHEL 3
1867 Description: Number of open files grows over time
1868 Details : The number of open files grows over time, whether or not
1869 Lustre is started. This was due to a filp leak introduced
1870 by one of our kernel patches.
1872 ------------------------------------------------------------------------------
1874 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1876 * Support for kernels:
1877 2.6.9-42.EL (RHEL 4)
1878 2.6.5-7.276 (SLES 9)
1879 2.4.21-40.EL (RHEL 3)
1880 2.6.12.6 vanilla (kernel.org)
1885 Bugzilla : 5719, 9635, 9792, 9684
1886 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1887 Details : If a server is under heavy load and cannot reply to new
1888 connection requests before the client resends the (re)connect,
1889 the connection handling code can behave badly if two service
1890 threads are concurrently handing separate (re)connections from
1891 the same client. Add better locking to the connection handling
1892 code, and ensure that only a single connection will be processed
1893 for a given client UUID, even if the lock is dropped.
1895 Severity : enhancement
1897 Description: add TCP zero-copy support to kernel
1898 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1899 sends if the hardware supports scatter-gather and checksumming.
1900 This allows socklnd to do client-write and server-read more
1901 efficiently and reduce CPU utilization from skbuf copying.
1904 Frequency : only if NFS exporting from client
1906 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1907 Details : If setting attributes on a file created under NFS that had
1908 never been opened it would be possible to oops the client
1909 if the file had no objects.
1912 Frequency : always for liblustre
1914 Description: liblustre client does MDS+OSTs setattr RPC for each write
1915 Details : When doing a write from a liblustre client, the client
1916 incorrectly issued an RPC to the MDS and each OST the file was
1917 striped over in order to update the timestamps. When writing
1918 with small chunks and many clients this could overwhelm the MDS
1919 with RPCs. In all cases it would slow down the write because
1920 these RPCs are unnecessary.
1922 Severity : enhancement
1924 Description: allow number of MDS service threads to be changed at module load
1925 Details : It is now possible to change the number of MDS service threads
1926 running. Adding "options mds mds_num_threads={N}" to the MDS's
1927 /etc/modprobe.conf will set the number of threads for the next
1928 time Lustre is restarted (assuming the "mds" module is also
1929 reloaded at that time). The default number of threads will
1930 stay the same, 32 for most systems.
1935 Description: OST crash if filesystem is unformatted or corrupt
1936 Details : If an OST is started on a device that has never been formatted
1937 or if the filesystem is corrupt and cannot even mount then the
1938 error handling cleanup routines would dereference a NULL pointer.
1943 Description: NULL pointer deref in llap_from_page.
1944 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1945 incorrect) mapping if the page was truncated/reclaimed while it was
1946 searched for. Check for this condition and skip such pages when
1947 doing readahead. Introduce extra check to llap_from_page() to
1948 verify page->mapping->host is non-NULL (so page is not anonymous).
1951 Frequency : Sometimes when using sys_sendfile
1953 Description: "page not covered by a lock" warnings from ll_readpage
1954 Details : sendfile called ll_readpage without right page locks present.
1955 Now we introduced ll_file_sendfile that does necessary locking
1956 around call to generic_file_sendfile() much like we do in
1960 Frequency : with certain MDS communication failures at client mount time
1962 Description: NULL pointer deref after failed client mount
1963 Details : a client connection request may delayed by the network layer
1964 and not be sent until after the PTLRPC layer has timed out the
1965 request. If the client fails the mount immediately it will try
1966 to clean up before the network times out the request. Add a
1967 reference from the request import to the obd device and delay
1968 the cleanup until the network drops the request.
1971 Frequency : occasionally during client (re)connect
1973 Description: assertion failure during client (re)connect
1974 Details : processing a client connection request may be delayed by the
1975 client or server longer than the client connect timeout. This
1976 causes the client to resend the connection request. If the
1977 original connection request is replied in this interval, the
1978 client may trip an assertion failure in ptlrpc_connect_interpret()
1979 which thought it would be the only running connect process.
1982 Frequency : only with obd_echo servers and clients that are rebooted
1984 Description: kernel BUG accessing uninitialized data structure
1985 Details : When running an obd_echo server it did not start the ping_evictor
1986 thread, and when a client was evicted an uninitialized data
1987 structure was accessed. Start the ping_evictor in the RPC
1988 service startup instead of the OBD startup.
1990 Severity : enhancement
1991 Bugzilla : 10193 (patchless)
1992 Description: Remove dependency on various unexported kernel interfaces.
1993 Details : No longer need reparent_to_init, exit_mm, exit_files,
1994 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1997 Frequency : rare (only users of deprecated and unsupported LDAP config)
1999 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2000 Details : LDAP apparently contains 'lustreName' attributes instead of
2001 'name'. A simple remapping of the name is sufficient.
2004 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2006 Description: waiting_locks_callback trips kernel BUG if client is evicted
2007 Details : Running with the dump_on_timeout debug flag turned on makes
2008 it possible that the waiting_locks_callback() can try to dump
2009 the Lustre kernel debug logs from an interrupt handler. Defer
2010 this log dumping to the expired_lock_main() thread.
2012 Severity : enhancement
2014 Description: Support NFS exporting on 2.6 kernels.
2015 Details : Implement non-rawops metadata methods for NFS server to use without
2016 changing NFS server code.
2019 Frequency : very rare (synthetic metadata workload only)
2021 Description: two racing renames might cause an MDS thread to deadlock
2022 Details : Running the "racer" program may cause one MDS thread to rename
2023 a file from being the source of a rename to being the target of
2024 a rename at exactly the same time that another thread is doing
2025 so, and the second thread has already enqueued these locks after
2026 doing a lookup of the target and is trying to relock them in
2027 order. Ensure that we don't try to re-lock the same resource.
2030 Frequency : only very large systems with liblustre clients
2032 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2033 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2034 all OSTs in parallel. This allows the eviction of stale liblustre
2035 clients to proceed much faster than if they were done in series,
2036 and also offers similar improvements for other set_info RPCs.
2041 Description: excessive CPU usage during initial read phase on client
2042 Details : During the initial read phase on a client, it would agressively
2043 retry readahead on the file, consuming too much CPU and impacting
2044 performance (since 1.4.5.8). Improve the readahead algorithm
2045 to avoid this, and also improve some other common cases (read
2046 of small files in particular, where "small" is files smaller than
2047 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2052 Description: MDS crash when receiving packet with unknown intent.
2053 Details : Do not LBUG in unknown intent case, just return -EFAULT
2055 Severity : enhancement
2056 Bugzilla : 9293, 9385
2057 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2058 Details : Do not serialize getattr (non-intent version) and statfs.
2061 Frequency : occasional, when OST network is overloaded/intermittent
2063 Description: client evicted by OST after bulk IO timeout
2064 Details : If a client sends a bulk IO request (read or write) the OST
2065 may evict the client if it is unresposive to its data GET/PUT
2066 request. This is incorrect if the network is overloaded (takes
2067 too long to transfer the RPC data) or dropped the OST GET/PUT
2068 request. There is no need to evict the client at all, since
2069 the pinger and/or lock callbacks will handle this, and the
2070 client can restart the bulk request.
2073 Frequency : Always when mmapping file with no objects
2075 Description: client crashes when mmapping file with no objects
2076 Details : Check that we actually have objects in a file before doing any
2077 operations on objects in ll_vm_open, ll_vm_close and
2083 Description: Request leak when working with deleted CWD
2084 Details : Introduce advanced request refcount tracking for requests
2085 referenced from lustre intent.
2087 Severity : Enhancement
2089 Description: Cache open file handles on client.
2090 Details : MDS now will return special lock along with openhandle, if
2091 requested and client is allowed to hold openhandle, even if unused,
2092 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2093 closing files for every read/write openration.
2095 Severity : Enhancement
2097 Description: Cache open negative dentries on client when possible.
2098 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2099 negative dentries on lock revocation.
2104 Description: Remounting a client read-only wasn't possible with a zconf mount
2105 Details : It wasn't possible to remount a client read-only with llmount.
2107 Severity : enhancement
2108 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2109 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2110 MPICH is included to add Lustre-specific ADIO interfaces.
2111 This is based closely on the UFS ADIO layer and only differs
2112 in file creation, in order to allow the OST striping to be set.
2113 This is user-contributed code and not supported by CFS.
2118 Description: extended inode attributes (immutable, append-only) work improperly
2119 when 2.4 and 2.6 kernels are used on client/server or vice versa
2120 Details : Introduce kernel-independent values for these flags.
2122 Severity : enhancement
2125 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2126 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2127 granularity. Now, Fractional MB granularity is be supported,
2128 this is very useful for low memory system.
2130 Severity : enhancement
2132 Description: Getattr by fid
2133 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2134 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2135 have locks and avoids getattr rpc requests for it on every stat.
2138 Frequency : Always, for filesystems larger than 2TB
2140 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2141 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2142 filesystems larger than 2TB to be mounted with mballoc enabled.
2145 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2147 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2148 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2149 filesystems larger than 2TB is mounted then the kernel will
2150 silently corrupt the start of the filesystem. CONFIG_LBD is
2151 enabled for all CFS-supported kernels, but the possibility of
2152 this happening with a modified kernel config exists.
2154 Severity : enhancement
2156 Description: add client O_DIRECT support for 2.6 kernels
2157 Details : It is now possible to do O_DIRECT reads and writes to files
2158 in the Lustre client mountpoint on 2.6 kernel clients.
2160 Severity : enhancement
2162 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2163 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2164 parallel, not waiting for response from every OST before sending
2165 a rpc to the next OST.
2170 Description: setattr vs write race when updating file timestamps
2171 Details : Client processes that update a file timestamp into the past
2172 right after writing to the file (e.g. tar) it is possible that
2173 the updated file modification time can be reset to the current
2174 time due to a race between processing the setattr and write RPC.
2176 Severity : enhancement
2178 Description: Bring 'lfs find' closer in line with regular Linux find.
2179 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2180 -print0 options and obtains all the needed info through the lustre
2183 Severity : enhancement
2185 Description: support up to 1024 configured devices on one node
2186 Details : change obd_dev array from statically allocated to dynamically
2187 allocated structs as they are first used to reduce memory usage
2192 Description: Flush dirty partially truncated pages during truncate
2193 Details : Immediatelly flush partially truncated pages in filter_setattr,
2194 this way we completely avoid having any pages in page cache on OST
2195 and can retire ugly workarounds during writes to flush such pages.
2200 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2201 Details : protect inode from truncation within vfs_unlink() context
2202 just take a reference before calling vfs_unlink() and release it
2203 when parent's i_sem is free.
2208 Description: last_id value checked outside lock on OST caused LASSERT failure
2209 Details : If there were multiple MDS->OST object precreate requests in
2210 flight, it was possible that the OST's last object id was checked
2211 outside a lock and incorrectly tripped an assertion. Move checks
2212 inside locks, and discard old precreate requests.
2215 Frequency : always, if extents are used on OSTs
2217 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2218 Details : OSTs using the extents format would not zero the high 16 bits of
2219 the index physical block number. This is not a problem for any
2220 OST filesystems smaller than 16TB, and no kernels support ext3
2221 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2222 new/modified files) and can be fixed for existing filesystems
2223 with e2fsprogs-1.39-cfs1.
2228 Description: import connection selection may be incorrect if timer wraps
2229 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2230 import connections to be ignored if the 32-bit jiffies counter
2231 wraps. Use a 64-bit jiffies counter.
2234 Frequency : very large clusters immediately after boot
2236 Description: LNET request buffers exhausted under heavy short-term load
2237 Details : If a large number of client requests are generated on a service
2238 that has previously never seen so many requests it is possible
2239 that the request buffer growth cannot keep up with the spike in
2240 demand. Instead of dropping incoming requests, they are held in
2241 the LND until the RPC service can accept more requests.
2244 Frequency : Sometimes during replay
2246 Description: Assertion failure in ll_local_open after replay.
2247 Details : If replay happened on an open request reply before we were able
2248 to set replay handler, reply will become not swabbed tripping the
2249 assertion in ll_local_open. Now we set the handler right after
2250 recognising of open request
2253 Frequency : very rare
2255 Description: kernel reports "badness in vsnprintf"
2256 Details : Reading from the "recovery_status" /proc file in small chunks
2257 may cause a negative length in lprocfs_obd_rd_recovery_status()
2258 call to vsnprintf() (which is otherwise harmless). Exit early
2259 if there is no more space in the output buffer.
2261 Severity : enhancement
2263 Description: clear OBD RPC statistics by writing to them
2264 Details : It is now possible to clear the OBD RPC statistics by writing
2265 to the "stats" file.
2270 Description: Client mtime is not the same on different clients after utimes
2271 Details : In some cases, the client was using the utimes() syscall on
2272 a file cached on another node. The clients now validate the
2273 ctime from the MDS + OSTs to determine which one is right.
2278 Description: Inability to activate failout mode
2279 Details : lconf script incorrectly assumed that in pythong string's numeric
2280 value is used in comparisons.
2283 Frequency : always with multiple stripes per file
2285 Description: Inefficient object allocation for mutli-stripe files
2286 Details : When selecting which OSTs to stripe files over, for files with
2287 a stripe count that divides evenly into the number of OSTs,
2288 the MDS is always picking the same starting OST for each file.
2289 Return the OST selection heuristic to the original design.
2294 Description: mount failures may take full timeout to return an error
2295 Details : Under some heavy load conditions it is possible that a
2296 failed mount can wait for the full obd_timeout interval,
2297 possibly several minutes, before reporting an error.
2298 Instead return an error as soon as the status is known.
2300 Frequency : quota enabled and large files being deleted
2302 Description: releasing more than 4GB of quota at once hangs OST
2303 Details : If a user deletes more than 4GB of files on a single OST it
2304 will cause the OST to spin in an infinite loop. Release
2305 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2310 Description: statfs data retrieved from /proc may be stale or zero
2311 Details : When reading per-device statfs data from /proc, in the
2312 {kbytes,files}_{total,free,avail} files, it may appear
2313 as zero or be out of date.
2316 Frequency : systems with MD RAID1 external journal devices
2318 Description: lconf's call to blkid is confused by RAID1 journal devices
2319 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2320 of returning all block devices that match the journal UUID.
2323 Frequency : always, for aggregate stripe size over 4GB
2325 Description: assertion fails when trying to use 4GB stripe size
2326 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2327 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2330 Frequency : always on ppc64
2332 Description: the first write on an ext3 filesystem with mballoc got stuck
2333 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2334 perform endianness conversion.
2336 ------------------------------------------------------------------------------
2338 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2340 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2341 INTEROPERATE with older versions automatically. Please read the
2342 user documentation before upgrading any part of a live system.
2343 * WARNING: Lustre networking configuration changes are required with
2344 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2347 * Support for kernels:
2348 2.6.9-22.0.2.EL (RHEL 4)
2349 2.6.5-7.244 (SLES 9)
2350 2.6.12.6 vanilla (kernel.org)
2353 Severity : enhancement
2354 Bugzilla : 7981/8208
2355 Description: Introduced Lustre Networking (LNET)
2356 Details : LNET is new networking infrastructure for Lustre, it includes
2357 a reorganized network configuration mode (see the user
2358 documentation for full details) as well as support for routing
2359 between different network fabrics. Lustre Networking Devices
2360 (LNDs) for the supported network fabrics have also been
2361 created for this new infrastructure.
2363 Severity : enhancement
2364 Description: Introduced Access control lists
2365 Details : clients can set ACLs on files and directories in order to have
2366 more fine-grained permissions than the standard Unix UGO+RWX.
2367 The MDS must be started with the "-o acl" mount option.
2369 Severity : enhancement
2370 Description: Introduced filesystem quotas
2371 Details : Administrators may now establish per-user quotas on the
2374 Severity : enhancement
2376 Description: Configuration change for the XT3
2377 The PTLLND is now used to run Lustre over Portals on the XT3
2378 The configure option(s) --with-cray-portals are no longer used.
2379 Rather --with-portals=<path-to-portals-includes> is used to
2380 enable building on the XT3. In addition to enable XT3 specific
2381 features the option --enable-cray-xt3 must be used.
2386 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2387 Details : A race between allocating a new llcd and re-getting the llcd_lock
2388 allowed another thread to grab newly-allocated llcd.
2390 Severity : enhancement
2392 Description: 2.6 OST async journal commit and locking fix to improve performance
2393 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2394 2.6 kernels are now async as they already were in 2.4 kernels so
2395 that they can commit concurrently with the network bulk transfer.
2396 For block-allocated files the filter allocation semaphore is held
2397 to avoid filesystem fragmentation during allocation. BKL lock
2398 removed for 2.6 xattr operations where it is no longer needed.
2403 Description: lconf incorrectly determined whether two IP networks could talk
2404 Details : In some more complicated routing and multiple-network
2405 configurations, lconf will avoid trying to make a network
2406 connection to a disjoint part of the IP space. It was doing the
2407 math incorrectly for one set of cases.
2412 Description: Fix for potential infinite loop processing records in an llog.
2413 Details : If an llog record is corrupted/zeroed, it is possible to loop
2414 forever in llog_process(). Validate the llog record length
2415 and skip the remainder of the block on error.
2418 Frequency : occasional (liblustre only)
2420 Description: liblustre could not open files whose last component is a symlink
2421 Details : sysio_path_walk() would incorrectly pass the open intent to
2422 intermediate path components.
2425 Frequency : rare (liblustre only with non-standard tuning)
2426 Bugzilla : 7201 (7350)
2427 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2428 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2429 after it has been released, only do this for referenced locks
2431 Severity : enhancement
2433 Description: specifying an (invalid) directory default stripe_size of -1
2434 would reset the directory default striping
2435 Details : stripe_size -1 was used internally to signal directory stripe
2436 removal, now use "all default" to signal dir stripe removal
2437 as a directory striping of "all default" is not useful
2440 Frequency : common for large clusters running liblustre clients
2442 Description: doing an ls when liblustre clients are running is slow
2443 Details : sending a glimpse AST to a liblustre client waits for every AST
2444 to time out, as liblustre clients will not respond. Since they
2445 cannot cache data we refresh the OST lock LVB from disk instead.
2447 Severity : enhancement
2449 Description: doing an ls at the same time as file IO can be slow
2450 Details : enqueue and other "small" requests can be blocked behind many
2451 large IO requests. Create a new OST IO portal for non-IO
2452 requests so they can be processed faster.
2455 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2457 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2458 created file with mode 000
2459 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2460 allow an ftruncate() as a non-root user to a file with mode 000.
2461 Lustre now respects this flag to disable mode checks when
2462 truncating a file owned by the user
2465 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2468 Description: Revoking locks from clients that went dead or catatonic might take
2470 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2471 cancellation of such locks instant on servers without waiting for
2472 any reply from clients. Clients drops these locks when cancel
2473 notification from server is received without replying.
2476 Frequency : liblustre-only, when liblustre client dies or becomes busy
2478 Description: Doing ls on Linux clients can take a long time with active
2480 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2481 granting such locks to it in the first place if possible. Locks
2482 are taken by proxy on the OST during the read or write and
2483 dropped immediately afterward. Add connect flags handling, do
2484 not grant locks to liblustre clients for glimpse ASTs.
2486 Severity : enhancement
2488 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2489 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2490 sequential reads and behaves badly with applications doing random
2491 reads. Improve it by reading ahead at least read region, and
2492 avoiding excessive large RPC for small reads.
2494 Severity : enhancement
2496 Description: Creating more than 1000 files for a single job may cause a load
2497 imbalance on the OSTs if there are also a large number of OSTs.
2498 Details : qos_prep_create() uses an OST index reseed value that is an
2499 even multiple of the number of available OSTs so that if the
2500 reseed happens in the middle of the object allocation it will
2501 still utilize the OSTs as uniformly as possible.
2506 Description: OST or MDS may oops in ping_evictor_main()
2507 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2508 but doesn't restart at beginning of obd_exports_timed list
2511 Severity : enhancement
2513 Description: improve by-nid export eviction on the MDS and OST
2514 Details : allow multiple exports with the same NID to be evicted at one
2515 time without re-searching the exports list.
2518 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2520 Description: MDS may oops in groups_free()
2521 Details : in rare race conditions a newly allocated group_info struct is
2522 freed again, and this can be NULL. The 2.4 compatibility code
2523 for groups_free() checked for a NULL pointer, but 2.6 did not.
2526 Frequency : common for liblustre clients doing little filesystem IO
2527 Bugzilla : 9352, 7313
2528 Description: server may evict liblustre clients accessing contended locks
2529 Details : if a client is granted a lock or receives a completion AST
2530 with a blocking AST already set it would not reply to the AST
2531 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2534 Frequency : lfs setstripe, only systems with more than 160 OSTs
2536 Description: unable to set striping with a starting offset beyond OST 160
2537 Details : llapi_create_file() incorrectly limited the starting stripe
2538 index to the maximum single-file stripe count.
2541 Frequency : LDAP users only
2543 Description: lconf did not handle in-kernel recovery with LDAP properly
2544 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2546 Severity : enhancement
2548 Description: bind OST threads to NUMA nodes to improve performance
2549 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2550 node and do their allocations there to localize memory access
2552 Severity : enhancement
2554 Description: llmount can determine client NID directly from Myrinet (GM)
2555 Details : the client NID code from gmnalnid was moved directly into
2556 llmount, removing the need to use this or specifying the
2557 client NID explicitly when mounting GM clients with zeroconf
2560 Frequency : if client is started with down MDS
2562 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2563 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2564 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2565 ptlrpc_import_delayed_req() because we didn't check if the
2566 request was interrupted, and we also didn't break out of the
2567 event loop if there was no timeout
2572 Description: data loss during non-page-aligned writes to a single file from
2573 both multiple nodes and multiple threads on one node at same time
2574 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2575 inconsistency led to false short-reads, that were cached and later
2576 used by ->prepare_write() to fill in partially written page,
2577 leading to data loss.
2580 Frequency : always, if lconf --abort_recovery used
2582 Description: lconf --abort_recovery fails with 'Operation not supported'
2583 Details : lconf was attempting to abort recovery on the MDT device and not
2586 Severity : enhancement
2588 Description: remove cleanup logs
2589 Details : replace lconf-generated cleanup logs with lustre internal
2590 cleanup routines. Eliminates the need for client-cleanup and
2593 Severity : enhancement
2595 Description: add support for EAs (user and system) on lustre filesystems
2596 Details : it is now possible to store extended attributes in the Lustre
2597 client filesystem, and with the user_xattr mount option it
2598 is possible to allow users to store EAs on their files also
2600 Severity : enhancement
2602 Description: Add possibility (config option) to show minimal available OST free
2604 Details : When compiled with --enable-mindf configure option, statfs(2)
2605 (and so, df) will return least minimal free space available from
2606 all OSTs as amount of free space on FS, instead of summary of
2607 free spaces of all OSTs.
2609 Severity : enhancement
2611 Description: do not expand extent locks acquired on OST-side
2612 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2613 by server: they are not cached anyway.
2616 Frequency : when mmap is used/binaries executed from Lustre
2618 Description: Unmmap pages before throwing them away from read cache.
2619 Details : llap_shrink cache now attempts to unmap pages before discarding
2620 them (if unmapping failed - do not discard). SLES9 kernel has
2621 extra checks that trigger if this unmapping is not done first.
2626 Description: lconf didn't resolve symlinks before checking to see whether a
2627 given mountpoint was already in use
2630 Frequency : when migrating failover services
2631 Bugzilla : 6395, 9514
2632 Description: When migrating a subset of services from a node (e.g. failback
2633 from a failover service node) the remaining services would
2634 time out and evict clients.
2635 Details : lconf --force (implied by --failover) sets the global obd_timeout
2636 to 5 seconds in order to quickly disconnect, but this caused
2637 other RPCs to time out too quickly. Do not change the global
2638 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2640 Severity : enhancement
2641 Frequency : if MDS is started with down OST
2642 Bugzilla : 9439,5706
2643 Description: Allow startup/shutdown of an MDS without depending on the
2644 availability of the OSTs.
2645 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2646 Add appropriate locking and lov-osc refcounts for safe
2647 cleaning. Add osc abort_inflight calls in case the
2648 synchronize never started.
2651 Frequency : occasional (Cray XT3 only)
2653 Description: root not authorized to access files in CRAY_PORTALS environment
2654 Details : The client process capabilities were not honoured on the MDS in
2655 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2656 been accessed by an authorized user then root was able to access
2657 the file on the local client also. The root user capabilities
2658 are now allowed on the MDS, as this environment has secure UID.
2661 Frequency : occasional
2663 Description: ldiskfs "too long searching" message happens too often
2664 Details : A debugging message (otherwise harmless) prints too often on
2665 the OST console. This has been reduced to only happen when
2666 there are fragmentation problems on the filesystem.
2671 Description: Division by zero in statfs when all OSCs are inactive
2672 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2673 lov_statfs divides by value returned by lov_get_stripecnt().
2677 Bugzilla : 9489, 3273
2678 Description: First write from each client to each OST was only 4kB in size,
2679 to initialize client writeback cache, which caused sub-optimal
2680 RPCs and poor layout on disk for the first writen file.
2681 Details : Clients now request an initial cache grant at (re)connect time
2682 and so that they can start streaming writes to the cache right
2683 away and always do full-sized RPCs if there is enough data.
2684 If the OST is rebooted the client also re-establishes its grant
2685 so that client cached writes will be honoured under the grant.
2690 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2691 Details : Now I/O RPCs go to different portal number and (presumably) fast
2692 lock requests (and glimses) and other RPCs get their own service
2693 threads pool that should be able to service those RPCs
2696 Severity : enhancement
2698 Description: Ability to exchange lustre version between client and servers and
2699 issue warnings at client side if client is too old. Also for
2700 liblustre clients there is ability to refuse connection of too old
2702 Details : New 'version' field is added to connect data structure that is
2703 filled with version info. That info is later checked by server and
2707 Frequency : rare, liblustre only.
2708 Bugzilla : 9296, 9581
2709 Description: Two simultaneous writes from liblustre at offset within same page
2710 might proceed at the same time overwriting eachother with stale
2712 Details : I/O lock withing llu_file_prwv was released too early, before data
2713 actually was hitting the wire. Extended lock-holding time until
2714 server acknowledges receiving data.
2717 Frequency : extremely rare. Never observed in practice.
2719 Description: avoid generating lustre_handle cookie of 0.
2720 Details : class_handle_hash() generates handle cookies by incrementing
2721 global counter, and can hit 0 occasionaly (this is unlikely, but
2722 not impossible, because initial value of cookie counter is
2723 selected randonly). Value of 0 is used as a sentinel meaning
2724 "unassigned handle" --- avoid it. Also coalesce two critical
2725 sections in this function into one.
2727 Severity : enhancement
2729 Description: allow liblustre clients to delegate truncate locking to OST
2730 Details : To avoid overhead of locking, liblustre client instructs OST to
2731 take extent lock in ost_punch() on client's behalf. New connection
2732 flag is added to handle backward compatibility.
2734 Severity : enhancement
2735 Bugzilla : 4928, 7341, 9758
2736 Description: allow number of OST service threads to be specified
2737 Details : a module parameter allows the number of OST service threads
2738 to be specified via "options ost ost_num_threads={N}" in the
2739 OSS's /etc/modules.conf or /etc/modprobe.conf.
2743 Bugzilla : 6146, 9635, 9895
2744 Description: servers crash with bad pointer in target_handle_connect()
2745 Details : In rare cases when a client is reconnecting it was possible that
2746 the connection request was the last reference for that export.
2747 We would temporarily drop the export reference and get a new
2748 one, but this may have been the last reference and the export
2749 was just destroyed. Get new reference before dropping old one.
2751 Severity : enhancement
2752 Frequency : if client is started with failover MDS
2754 Description: Allow multiple MDS hostnames in the mount command
2755 Details : Try to read the configuration from all specified MDS
2756 hostnames during a client mount in case the "primary"
2759 Severity : enhancement
2761 Description: Stop sending data to evicted clients as soon as possible.
2762 Details : Check if the client we are about to send or are sending data to
2763 was evicted already. (Check is done every second of waiting,
2764 for which l_wait_event interface was extended to allow checking
2765 of exit condition at specified intervals).
2768 Frequency : rare, normally only when NFS exporting is done from client
2770 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2772 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2775 Frequency : always, if srand() is called before liblustre initialization
2777 Description: Liblustre uses system PRNG disturbing its usage by user application
2778 Details : Introduce internal to lustre fast and high-quality PRNG for
2779 lustre usage and make liblustre and some other places in generic
2780 lustre code to use it.
2782 Severity : enhancement
2783 Bugzilla : 9477, 9557, 9870
2784 Description: Verify that the MDS configuration logs are updated when xml is
2785 Details : Check if the .xml configuration logs are newer than the config
2786 logs stored on the MDS and report an error if this is the case.
2787 Request --write-conf, or allow starting with --old_conf.
2789 Severity : enhancement
2791 Description: Handle symlinks in the path when checking if Lustre is mounted.
2792 Details : Resolve intermediate symlinks when checking if a client has
2793 mounted a filesystem to avoid duplicate client mounts.
2798 Description: lconf can hit an error exception but still return success.
2799 Details : The lconf command catches the Command error exception at the top
2800 level script context and will exit with the associated exit
2801 status, but doesn't ensure that this exit status is non-zero.
2806 Description: failure of ptlrpc thread startup can cause oops
2807 Details : Starting a ptlrpc service thread can fail if there are a large
2808 number of threads or the server memory is very fragmented.
2809 Handle this without oopsing.
2812 Frequency : always, only if liblustre and non-default acceptor port was used
2814 Description: liblustre cannot connect to servers with non-default acceptor port
2815 Details : tcpnal_set_default_params() was not called and was therefore
2816 ignoring the environment varaible TCPNAL_PORT, as well as other
2817 TCPNAL_ environment variables
2822 Description: two objects could be created on the same OST for a single file
2823 Details : If an OST is down, in some cases it was possible to create two
2824 objects on a single OST for a single file. No problems other
2825 than potential performance impact and spurious error messages.
2829 Bugzilla : 5681, 9562
2830 Description: Client may oops in ll_unhash_aliases
2831 Details : Client dcache may become inconsistent in race condition.
2832 In some cases "getcwd" can fail if the current directory is
2838 Description: Inode refcounting problems in NFS export code
2839 Details : link_raw functions used to call d_instantiate without obtaining
2840 extra inode reference first.
2844 Bugzilla : 9942, 9903
2845 Description: Referencing freed requests leading to crash, memleaks with NFS.
2846 Details : We used to require that call to ll_revalidate_it was always
2847 followed by ll_lookup_it. Also with revalidate_special() it is
2848 possible to call ll_revalidate_it() twice for the same dentry
2849 even if first occurence returned success. This fix changes semantic
2850 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2851 reference on a request referred from the intent.
2852 ll_intent_release() then releases such a request.
2855 Frequency : rare, normally benchmark loads only
2857 Description: unlinked inodes were kept in memory on the client
2858 Details : If a client is repeatedly creating and unlinking files it
2859 can accumulate a lot of stale inodes in the inode slab cache.
2860 If there is no other client load running this can cause the
2861 client node to run out of memory. Instead flush old inodes
2862 from client cache that have the same inode number as a new inode.
2865 Frequency : SLES9 2.6.5 kernel and long filenames only
2866 Bugzilla : 9969, 10379
2867 Description: utime reports stale NFS file handle
2868 Details : SLES9 uses out-of-dentry names in some cases, which confused
2869 the lustre dentry revalidation. Change it to always use the
2873 Frequency : rare, unless heavy write-truncate concurrency is continuous
2874 Bugzilla : 4180, 6984, 7171, 9963, 9331
2875 Description: OST becomes very slow and/or deadlocked during object unlink
2876 Details : filter_destroy() was holding onto the parent directory lock
2877 while truncating+unlinking objects. For very large objects this
2878 may block other threads for a long time and slow overall OST
2879 responsiveness. It may also be possible to get a lock ordering
2880 deadlock in this case, or run out of journal credits because of
2881 the combined truncate+unlink. Solution is to do object truncate
2882 first in one transaction without parent lock, and then do the
2883 final unlink in a new transaction with the parent lock. This
2884 reduces the lock hold time dramatically.
2887 Frequency : rare, 2.4 kernels only
2889 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2890 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2891 before unmounting their filesystem in order to avoid deadlock.
2892 The kernel_locked() function in 2.4 kernels only checks whether
2893 the kernel lock is held, not whether it is this process that is
2894 holding it as 2.6 kernels do.
2899 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2900 Details : The client ptlrpc code may be trying to reconnect to a down
2901 server before a previous connection attempt has timed out.
2902 Increase the reconnect interval to be longer than the connection
2903 timeout interval to avoid sending duplicate connections to
2907 Frequency : echo_client brw_test command
2909 Description: fix echo_client to work with OST preallocated code
2910 Details : OST preallocation code (5137) didn't take echo_client IO path
2911 into account: echo_client calls filter methods outside of any
2912 OST thread and, hence, there is no per-thread preallocated
2913 pages and buffers to use. Solution: hijack pga pages for IO. As
2914 a byproduct, this avoids unnecessary data copying.
2918 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2919 Description: Client can oops in mdc_commit_close() after open replay
2920 Details : It was possible for the MDS to return an open request with no
2921 transaction number in mds_finish_transno() if the client was
2922 evicted, but without actually returning an error. Clients
2923 would later try to replay that open and may trip an assertion
2924 Simplify the client close codepath, and always return an error
2925 from the MDS in case the open is not successful.
2928 Frequency : rare, 2.6 OSTs only
2930 Description: OST may deadlock under high load on fragmented files
2931 Details : If there was a heavy load and highly-fragmented OST filesystems
2932 it was possible to have all the OST threads deadlock waiting on
2933 allocation of biovecs, because the biovecs were not released
2934 until the entire RPC IO was completed. Instead, release biovecs
2935 as soon as they are complete to ensure forward IO progress.
2937 Severity : enhancement
2939 Description: Support for specifying external journal device at mount
2940 Details : If an OST or MDS device is formatted with an external journal
2941 device, this device major/minor is stored in the ext3 superblock
2942 and may not be valid for failover. Allow detecting and
2943 specifying the external journal at mount time.
2948 Description: Mounting an MDS with pending unlinked files may cause oops
2949 Details : target_finish_recovery() calls mds_postrecov() which returned
2950 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2951 considers this an error and immediately begins cleaning up the
2952 lov, just after starting the mds_lov process
2954 Severity : enhancement
2956 Description: Implement 'lfs df' to report actual free space on per-OST basis
2957 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2958 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2959 usage of objects; '-h' to report in human readable format.
2961 ------------------------------------------------------------------------------
2963 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2970 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2971 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2972 at startup during orphan unlinking. Instead, skip buddy update
2973 before setup, as it will be regenerated after recovery is complete.
2978 Description: If an OST is inactive, its locks might reference stale inodes.
2979 Details : lov_change_cbdata() must iterate over all namespaces, even if
2980 they are inactive to clear inode references from the lock.
2982 Severity : enhancement
2983 Frequency : occasional, if non-standard max_dirty_mb used
2985 Description: Client will block write RPCs if not enough grant
2986 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2987 then the client will block writes while waiting for another RPC
2988 to complete instead of consuming its dirty limit. With change
2989 we get improved performance when max_dirty_mb is small.
2991 Severity : enhancement
2992 Bugzilla : 3389, 6253
2993 Description: Add support for supplementary groups on the MDS.
2994 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2995 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2996 lookups for user supplementary groups into a cache.
3000 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3001 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3002 for liblustre O_EXCL is enforced by MDS.
3006 Description: readdir never returns NULL in liblustre.
3007 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3008 dentry in struct dirent.
3012 Frequency : liblustre only, depends on application IO pattern
3013 Description: liblustre clients evicted if not contacting servers
3014 Details : Don't put liblustre clients into the ping_evictor list, so
3015 they will not be evicted by the pinger ever.
3017 Severity : enhancement
3019 Description: Add ability to evict clients by NID from MDS.
3020 Details : By echoing "nid:$NID" string into
3021 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3022 $NID would be instantly evicted from this MDS and from all active
3023 OSTs connected to it.
3027 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3028 Details : lookup_it_finish() used to query file size from OSTs that was not
3033 Description: service threads change working directory to that of init
3034 Details : Starting lustre service threads may pin the working directory
3035 of the parent thread, making that filesystem busy. Threads
3036 now change to the working directory of init to avoid this.
3040 Frequency : during shutdown only
3041 Description: shutdown with a failed MDS or OST can cause unmount to hang
3042 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3047 Frequency : 2.6 only
3048 Description: chmod/chown may include an extra supplementary group
3049 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3050 supplementary group and if none is specified this is used.
3053 Bugzilla : 5479 (6816)
3055 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3056 Details : If lookup is in progress on a file that is unlinked we might try
3057 to revalidate the inode and fail in revalidate after lookup is
3058 complete and ll_file_open() enqueues the open again but
3059 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3062 Frequency : always, if lconf --abort_recovery used
3064 Description: lconf --abort_recovery fails with 'Operation not supported'
3065 Details : lconf was attempting to abort recovery on the MDT device and not
3068 ------------------------------------------------------------------------------
3070 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3075 Frequency : rare (only unsupported configurations with a node running as an
3077 Bugzilla : 6514, 5137
3078 Description: Mounting a Lustre file system on a node running as an OST could
3080 Details : OSTs now preallocates memory needed to write out data at
3081 startup, instead of when needed, to avoid having to
3082 allocate memory in possibly low memory situations.
3083 Specifically, if the file system is mounted on on OST,
3084 memory pressure could force it to try to write out data,
3085 which it needed to allocate memory to do. Due to the low
3086 memory, it would be unable to do so and the node would
3087 become unresponsive.
3089 Severity : enhancement
3091 Description: Addition of lconf --service command line option
3092 Details : lconf now accepts a '--service <arg>' option, which is
3093 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3095 Severity : enhancement
3097 Description: Failover mode is now the default for OSTs.
3098 Details : By default, OSTs will now run in failover mode. To return to
3099 the old behaviour, add '--failout' to the lmc line for OSTs.
3101 Severity : enhancement
3103 Description: Health checks are now provided for MDS and OSTs
3104 Details : Additional detailed health check information on MSD and OSTs
3105 is now provided through the procfs health_check value.
3108 Frequency : occasional, depends on IO load
3110 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3111 after numerous create/delete cycles
3112 Details : The ext3 inode allocation policy would not allocate new inodes
3113 very well on the OSTs because there are no new directories
3114 being created. Instead we look for groups with free space if
3115 the parent directories are nearly full.
3120 Description: Network or server problems during mount may cause partially
3121 mounted clients instead of returning an error.
3122 Details : The config llog parsing code may overwrite the error return
3123 code during mount error handling, returning success instead
3129 Description: MDS can fail to allocate large reply buffers
3130 Details : After long uptimes the MDS can fail to allocate large reply
3131 buffers (e.g. zconf client mount config records) due to memory
3132 fragmentation or consumption by the buffer cache. Preallocate
3133 some large reply buffers so that these replies can be sent even
3134 under memory pressure.
3138 Frequency : rare (liblustre)
3139 Description: fsx running with liblustre complained that using truncate() to
3140 extend the file doesn't work. This patch corrects that issue.
3141 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3142 ATTR_SIZE and lsm use in llu_setattr_raw.
3146 Frequency : rare, only 2.6 kernels
3147 Description: Unusual file access patterns on the MDS may result in inode
3148 data being lost in very rare circumstances.
3149 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3150 code in the 2.6 kernel caused the inode and/or EA data not to
3151 be read from disk, causing single-file corruption.
3155 Frequency : rare, only 2.6 filesystems using extents
3156 Description: Heavy concurrent write and delete load may cause data corruption.
3157 Details : It was possible under high-load situations to have an extent
3158 metadata block in the block device cache from a just-unlinked
3159 file overwrite a newly-allocated data block. We now unmap any
3160 metadata buffers that alias just-allocated data blocks.
3164 Frequency : filesystems with default stripe_count larger than 77
3165 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3166 Details : lconf specifies an inode size of 4096 bytes when the default
3167 stripe_count is larger than 77. This conflicts with the default
3168 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3169 this case to avoid pinning too much memory for large EAs.
3171 ------------------------------------------------------------------------------
3173 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3178 Frequency : rare (extremely heavy IO load with hundreds of clients)
3180 Description: Client is evicted, gets IO error writing to file
3181 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3182 caused clients to be evicted for AST timeouts. The fixes in
3183 bug 5192 mean we no longer need to have such short AST timeouts
3184 so ldlm_timeout has been increased.
3187 Frequency : occasional during --force or --failover shutdown under load
3188 Bugzilla : 5949, 4834
3189 Description: Server oops/LBUG if stopped with --force or --failover under load
3190 Details : a collection of import/export refcount and cleanup ordering
3191 issues fixed for safer force cleanup
3194 Frequency : only filesystems larger than 120 OSTs
3195 Bugzilla : 5990, 6223
3196 Description: lfs getstripe would oops on a very large filesystem
3197 Details : lov_getconfig used kfree on vmalloc'd memory
3200 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3201 Bugzilla : 6242, 6243
3202 Description: reading from files that had been truncated to a non-zero size
3203 but never opened returned no data
3204 Details : ll_file_read() reads zeros from no-object files to EOF
3209 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3210 Details : There's little harm in aborting MDS/OSS recovery and letting it
3211 try again, so I removed the LASSERT and return an error instead.
3213 Severity : enhancement
3215 Description: New debugging infrastructure for tracking down data corruption
3216 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3217 (b) cover more of the client-side code path, and (c) try to narrow
3218 down where problems occurred
3222 Bugzilla : 3819, 4364, 4397, 6313
3223 Description: Racing close and eviction MDS could cause assertion in mds_close
3224 Details : It was possible to get multiple mfd references during close and
3225 client eviction, leading to one thread referencing a freed mfd.
3227 Severity: : enhancement
3228 Bugzilla : 3262, 6359
3229 Description: Attempts to reconnect to servers are now more aggressive.
3230 Details : This builds on the enhanced upcall-less recovery that was added
3231 in 1.4.2. When trying to reconnect to servers, clients will
3232 now try each server in the failover group every 10 seconds. By
3233 default, clients would previously try one server every 25 seconds.
3238 Description: After recovery, certain operations trigger a failed
3239 assertion on a client.
3240 Details : Failing over an mds, using lconf -d --failover, while a
3241 client was doing a readdir() call would cause the client to
3242 LBUG after recovery completed and the readdir() was resent.
3244 Severity : enhancement
3246 Description: Default groups are now added by lconf
3247 Details : You can now run lconf --group <servicename> without having to
3248 manually add groups with lmc.
3251 Frequency : occasional
3253 Description: Nodes with an elan id of 0 trigger a failed assertion
3256 Frequency : always when accessing e.g. tty/console device nodes
3258 Description: tty and some other devices nodes cannot be used on lustre
3259 Details : file's private_data field is used by device data and lustre
3260 values in there got lost. New field was added to struct file to
3261 store fs-specific private data.
3264 Frequency : when exporting Lustre via NFS
3266 Description: NFSD failed occasionally when looking up a path component
3267 Details : NFSD is looking up ".." which was broken in ext3 directories
3268 that had grown large enough to become hashed.
3271 Frequency : Clusters with multiple interfaces not on the same subnet
3273 Description: Nodes will repeatedly try to reconnect to an interface which it
3274 cannot reach and report an error to the log.
3275 Details : Extra peer list entries will be created by lconf with some peers
3276 unreachable. lconf now validates the peer before adding it.
3279 Frequency : Only if a default stripe is set on the filesystem root.
3281 Description: Setting a default stripe on the filesystem root prevented the
3282 filesystem from being remounted.
3283 Details : The client was sending extra request flags in the root getattr
3284 request and did not allocate a reply buffer for the dir EA.
3287 Frequency : occasional, higher if lots of files are accessed by one client
3288 Bugzilla : 6159, 6097
3289 Description: Client trips assertion regarding lsm mismatch/magic
3290 Details : While revalidating inodes the VFS looks up inodes with ifind()
3291 and in rare cases can find an inode that is being freed.
3292 The ll_test_inode() code will free the lsm during ifind()
3293 when it finds an existing inode and then the VFS later attaches
3294 this free lsm to a new inode.
3298 Bugzilla : 6422, 7030
3299 Description: MDS deadlock between mkdir and client eviction
3300 Details : Creating a new file via mkdir or mknod (starting a transaction
3301 and getting the ns lock) can deadlock with client eviction
3302 (gets ns lock and trying to finish a synchronous transaction).
3305 Frequency : occasional
3306 Description: While starting a server, the fsfilt_ext3 module could not be
3308 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3309 kernels. Previously, lconf would still use the ext3 name
3310 when trying to load modules. Now, it will correctly use
3311 ext3 on 2.4 and ldiskfs on 2.6.
3313 Severity : enhancement
3314 Description: The default stripe count has been changed to 1
3315 Details : The interpretation of the default stripe count (0, to lfs
3316 or lmc) has been changed to mean striping across a single
3317 OST, rather than all available. For general usage we have
3318 found a stripe count of 1 or 2 works best.
3320 Severity : enhancement
3321 Description: Add support for compiling against Cray portals.
3322 Details : Conditional compiling for some areas that are different
3326 Frequency : occasional
3327 Bugzilla : 6409, 6834
3328 Description: Creating files with an explicit stripe count may lead to
3329 a failed assertion on the MDS
3330 Details : If some OSTs are full or unavailable, creating files may
3331 trigger a failed assertion on the MDS. Now, Lustre will
3332 try to use other servers or return an error to the
3336 Frequency : occasional
3338 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3339 caused lock timeout during readahead (since 1.4.2).
3340 Details : Processes doing readahead might match a lock that hasn't been
3341 granted yet if there are overlapping and conflicting lock
3342 requests. The readahead process waits on ungranted lock
3343 (original lock is CBPENDING), while OST waits for that process
3344 to cancel CBPENDING read lock and eventually evicts client.
3346 Severity : enhancement
3348 Description: Initial enabling of flock support for clients
3349 Details : Implements fcntl advisory locking and file status functions.
3350 This feature is provided as an optional mount flag (default
3351 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3352 locking are implemented yet, and those that are are not guaranteed
3353 to be completely correct in production environments.
3354 mount -t lustre -o [flock|noflock] ...
3357 Frequency : occasional
3359 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3360 assertion in the ext3 JBD (journaling) layer.
3361 Details : The b_committed_data struct is protected by the big kernel lock
3362 in 2.4 kernels, serializing journal_commit_transaction() and
3363 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3364 access to this struct. In 2.6 kernels there is finer grained
3365 locking to improve SMP performance of the JBD layer.
3369 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3374 Description: merge and cleanup kernel patches.
3377 -----------------------------------------------------------------------------
3379 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3381 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3382 versions of Lustre. Please update all servers and clients to
3383 version 1.4.2 or later at the same time. You must also run
3384 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3385 to update the configuration logs.
3387 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3388 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3389 - swab llog records of type '0' so we get proper header size/idx (5861)
3390 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3391 - fix rename of one directory over another leaking an inode (5953)
3392 - avoid SetPageDirty on 2.6 (5981)
3393 - don't re-add just-being-destroyed locks to the waiting list (5653)
3394 - when creating new directories, inherit the parent's custom
3395 striping settings if present parent (3048)
3396 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3397 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3398 - don't hold client locks on temporary worklist from l_lru (5666)
3399 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3400 - automatically evict dead clients (5921)
3401 - Update file size properly in create+truncate+fstat case (6196)
3402 - Do not unhash mountpoint dentries, do not allow removal of
3404 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3405 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3406 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3407 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3408 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3409 - Correctly handle reads of files with no objects (6243)
3410 - lctl recover will also mark a device active if deactivate used (5933)
3412 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3413 - allow --write-conf on an MDS with different nettype than client (5619)
3414 - don't write config llogs to MDS for mounts not from that MDS (5617)
3415 - lconf should create multiple TCP connections from a client (5201)
3416 - init scripts are now turned off by default; run chkconfig --on
3417 lustre and chkconfig --on lustrefs to use them
3418 - upcalls are no longer needed for clients to recover to failover
3420 - add --abort-recovery option to lconf to abort recovery on device
3422 - add support for an arbitrary number of OSTs (3026)
3423 - Quota support protocol changes.
3424 - forward compatibility changes to wire structs (6007)
3425 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3426 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3427 - improved "lustre status" script
3428 - initialize blocksize for non-regular files (6062)
3429 - added --disable-server and --disable-client configure options (5782)
3430 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3431 - Vanilla 2.4.29 support
3432 - increase maximum number of obd devices to 520 (6242)
3433 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3434 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3435 SLES 9 2.6.5 kernels
3436 - limit stripes per file to 160 (the maximum EA size) (6093)
3438 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3441 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3442 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3443 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3444 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3445 - small ext3 extents cleanups and fixes (5733)
3446 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3447 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3448 - avoid needless client->OST connect, fix handle mismatch (5317)
3449 - fix DLM error path that led to out-of-sync client, long delays (5779)
3450 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3451 - fix several locking issues related to i_size (5492,5624,5654,5672)
3452 - don't move pending lock onto export if it is already evicted (5683)
3453 - fix kernel oops when creating .foo in unlinked directory (5548)
3454 - fix deadlock in obdfilter statistics vs. object create (5811)
3455 - use time_{before,after} to avoid timer jiffies wrap (5882)
3456 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3457 - Do not leak request if server was not able to process it (5154)
3458 - If mds_open unable to find parent dir, make that negative lookup(5154)
3459 - don't create new directories with extent-mapping (5909, 5936)
3461 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3462 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3463 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3464 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3465 - support for RHEL4 kernel on i686 (5773)
3466 - provide error messages when incompatible logs are encountered (5898)
3468 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3469 * version 1.4.0.10 (1.4.1 release candidate 1)
3471 - don't keep a lock reference when lock is not granted (4238)
3472 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3473 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3474 - drop import inflight refcount on signal_completed_replay error (5255)
3475 - unlock page after async write error during send (3677)
3476 - handle missing objects in filter_preprw_read properly (5265)
3477 - no transno return for symlink open, don't save no-trasno open (3440)
3478 - don't try to complete elan receive that already failed (4012)
3479 - free RPC server reply state on error (5406)
3480 - clean up thread from ptlrpc_start_thread() on error (5160)
3481 - readahead could read extra page into cache that wasn't ejected (5388)
3482 - prevent races in class_attach/setup/cleanup/detach (5260)
3483 - don't dereference de->d_inode after l_dput of de (5458)
3484 - use "int" for stripe value returned from lock_to_stripe (5544)
3485 - mballoc allocation and error-checking fixes in 2.6 (5504)
3486 - block device patches to fix I/O request sizes in 2.6 (5482)
3487 - look up hostnames for IB nals (5602)
3488 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3489 - don't start multiple acceptors for the same port (5277)
3490 - fix incorrect LASSERT in mds_getattr_name (5635)
3491 - export a proc file for general "ping" checking (5628)
3492 - fix "lfs check" to not block when the MDS is down (5628)
3494 - service request history (4965)
3495 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3496 - create an "evict_client" /proc entry on OSTs, like the MDS has
3497 - fix mount usage message, return errors per mount(8) (5168)
3498 - change grep [] to grep "[]" in tests so they work in more UMLs
3499 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3500 - remove ancient LOV_MAGIC_V0 EA support (5047)
3501 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3502 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3503 - allow readahead allocations to fail when low on memory (5383)
3504 - mmap locking landed again, after considerable improvement (2828)
3505 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3507 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3510 - send OST transaction number in read/write reply to free req (4966)
3511 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3512 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3513 - the watchdog thread now runs as interruptible (5246)
3514 - flock/lockf fixes (but it's still disabled, pending 5135)
3515 - don't use EXT3 constants in llite code (5094)
3516 - memory shortage at startup could cause assertion (5176)
3518 - reorganization of lov code
3519 - single portals codebase
3521 - add extents/mballoc support (5025)
3522 - direct I/O reads in the obdfilter (4048)
3523 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3525 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3528 - send OST transaction number in read/write reply to free req (4966)
3529 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3530 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3531 - the watchdog thread now runs as interruptible (5246)
3532 - handle missing objects in filter_preprw_read properly (5265)
3533 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3534 - drop import inflight refcount on signal_completed_replay error (5255)
3535 - unlock page after async write error during send (3677)
3536 - return original error code on reconstructed replies (3761)
3537 - no transno return for symlink open, don't save no-trasno open (3440)
3539 - add pid to ldlm debugging output (4922)
3540 - bump the watchdog timeouts -- we can't handle 30sec yet
3541 - extra debugging for orphan dentry/inode bug (5259)
3543 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3546 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3547 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3548 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3549 - fix NULL dereference in /proc/sys/portals/routes (4827)
3550 - allow failed mdc_close() operations to be interrupted (4561)
3551 - stop precreate on OST before MDS would time out on it (4778)
3552 - don't send partial-page writes before EOF from client (4410)
3553 - discard client grant for sub-page writes on large-page clients (4520)
3554 - don't free dentries not owned by NFS code, check generation (4806)
3555 - fix lsm leak if mds_create_objects() fails (4801)
3556 - limit debug_daemon file size, always print CERROR messages (4789)
3557 - use transno after validating reply (3892)
3558 - process timed out requests if import state changes (3754)
3559 - update mtime on OST during writes, return in glimpse (4829)
3560 - add mkfsoptions to LDAP (4679)
3561 - use ->max_readahead method instead of zapping global ra (5039)
3562 - don't interrupt __l_wait_event() during strace
3564 - add software watchdogs to catch hung threads quickly (4941)
3565 - make lustrefs init script start after nfs is mounted
3566 - fix CWARN/ERROR duplication (4930)
3567 - return async write errors to application if possible (2248)
3568 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3569 - print NAL number in %x format (4645)
3570 - update barely-supported suse-2.4.21-171 series (4842)
3571 - support for sles 9 %post scripts
3572 - support for building 2.6 kernel-source packages
3573 - support for sles km_* packages
3575 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3578 - ignore -ENOENT errors in osc_destroy (3639)
3579 - notify osc create thread that OSC is being cleaned up (4600)
3580 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3581 - reconstruct ost_handle() like mds_handle() (4657)
3582 - create a new thread to do import eviction to avoid deadlock (3969)
3583 - let lconf resolve symlinked-to devices (4629)
3584 - don't unlink "objects" from directory with default EA (4554)
3585 - hold socknal file ref over connect in case target is down (4394)
3586 - allow more than 32000 subdirectories in a single directory (3244)
3587 - fix blocks count for O_DIRECT writes (3751)
3588 - OST returns ENOSPC from object create when no space left (4539)
3589 - don't send truncate RPC if file size isn't changing (4410)
3590 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3591 - bind to privileged port in socknal and tcpnal (3689)
3593 - rate limit CERROR/CWARN console message to avoid overload (4519)
3594 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3595 - basic mmap support (3918)
3596 - kernel patch series update from b1_4 (4711)
3598 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3601 - avoid crash during MDS cleanup with OST shut down (2775)
3602 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3603 - don't use bad inodes on the MDS (3744)
3604 - dynamic object preallocation to improve recovery speed (4236)
3605 - don't hold spinlock over lock dumping or change debug flags (4401)
3606 - don't zero obd_dev when it is force cleaned (3651)
3607 - print grants to console if they go negative (4431)
3608 - "lctl deactivate" will stop automatic recovery attempts (3406)
3609 - look for existing locks in ldlm_handle_enqueue() (3764)
3610 - don't resolve lock handle twice in recovery avoiding race (4401)
3611 - revalidate should check working dir is a directory (4134)
3613 - don't always mark "slow" obdfilter messages as errors (4418)
3615 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3618 - don't close LustreDB during write_conf until it is done (3860)
3619 - fix typo in lconf for_each_profile (3821)
3620 - allow dumping logs from multiple threads at one time (3820)
3621 - don't allow multiple threads in OSC recovery (3812)
3622 - fix debug_size parameters (3864)
3623 - fix mds_postrecov to initialize import for llog ctxt (3121)
3624 - replace config semaphore with spinlock (3306)
3625 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3626 - don't allow enqueue to complete on a destroyed export (3822)
3627 - down write_lock before checking llog header bitmap (3825)
3628 - recover from lock replay timeout (3764)
3629 - up llog sem before sending rpc (3652)
3630 - reduce ns lock hold times when setting kms (3267)
3631 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3632 - fix NULL deref and obd_dev leak on setup error (3312)
3633 - replace some LBUG about llog ops with error handling (3841)
3634 - don't match INVALID dentries from d_lookup and spin (3784)
3635 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3636 - fix invalid assertion in ptlrpc_set_wait (3880)
3638 - add libwrap support for the TCP acceptor (3996)
3639 - add /proc/sys/portals/routes for non-root route listing (3994)
3640 - allow setting MDS UUID in .xml (2580)
3641 - print the stack of a process that LBUGs (4228)
3643 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3646 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3647 - make sure to unset replay flag from failed open requests (3440)
3648 - if default stripe count is 0, use OST count for inode size (3636)
3649 - update parent mtime/ctime on client for create/unlink (2611)
3650 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3651 - free recovery state on server during a forced cleanup (3571)
3652 - unregister_reply for resent reqs (3063)
3653 - loop back devices mounting and status check on 2.6 (3563)
3654 - fix resource-creation race that can provoke i_size == 0 (3513)
3655 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3656 - more debugging for page-accounting assertion (3746)
3657 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3658 - avoid deadlock after precreation failure (3758)
3659 - fix race and lock order deadlock in orphan handling (3450, 3750)
3660 - add validity checks when grabbing inodes from l_ast_data (3599)
3662 - add /proc/.../recovery_status to obdfilter (3428)
3663 - lightweight CDEBUG infrastructure, debug daemon (3668)
3664 - change default OSC RPC parameters to be better on small clusters
3665 - turn off OST read cache for files smaller than 32MB
3666 - install man pages and include them in rpms (3100)
3667 - add new init script for (un)mounting lustre filesystems (2593)
3668 - run chkconfig in %post for init scripts (3701)
3669 - drop scimac NAL (unmaintained)
3671 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3674 - clean kiobufs before and after use (3485)
3675 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3676 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3677 - change init script to fail more clearly if not run as root (1528)
3678 - allow clients to reconnect during replay (1742)
3679 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3680 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3681 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3682 - fix cond_resched (3554)
3683 - only evict unfinished clients after recovery (3515)
3684 - allow bulk resend, prevent data loss (3570)
3685 - dynamic ptlrpc request buffer allocation (2102)
3686 - don't allow unlinking open directory if it isn't empty (2904)
3687 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3688 - remove extraneous obd dereference causing LASSERT failure (3334)
3689 - don't use get_cycles() when creating temp. files on the mds (3156)
3690 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3691 - handle EEXIST for set-stripe, set proper directory name (3336)
3693 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3694 - fix ksocknal_fmb_callback() error messages (2918)
3696 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3699 - don't copy lvb into (possibly NULL) reply on error (2983)
3700 - don't deref dentry after dput, don't free lvb on error (2922)
3701 - use the kms to determine writeback rpc length (2947)
3702 - increment oti_logcookies when osc is inactive (2948)
3703 - update client's i_blocks count via lvb messages (2543)
3704 - handle intent open/close of special files properly (1557)
3705 - mount MDS with errors=remount-ro, like obdfilter (2009)
3706 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3707 - don't use cancelling-locks' kms values (2947)
3708 - use highest lock extent for kms, not last one (2925)
3709 - don't dereference ERR_PTR() dentry in error handling path (3107)
3710 - fix thread race in portals_debug_dumplog() (3122)
3711 - create lprocfs device entries at setup instead of at attach (1519)
3712 - common AST error handler, don't evict client on completion race (3145)
3713 - zero nameidata in detach_mnt in 2.6 (3118)
3714 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3715 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3716 - initialize RPC timeout timer earlier for 2.6 (3219)
3717 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3718 - print nal/nid for unknown nid (3258)
3719 - additional checks for oscc recovery before doing precreate (3284)
3720 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3721 - don't crash in mdc_close for bad permissions on open (3285)
3722 - zero i_rdev for non-device files (3147)
3723 - clear page->private before handing to FS, better assertion (3119)
3724 - tune the read pipeline (3236)
3725 - fix incorrect decref of invalidated dentry (2350)
3726 - provide read-ahead stats and refine rpc in flight stats (3328)
3727 - don't hold journal transaction open across create RPC (3313)
3728 - update atime on MDS at close time (3265)
3729 - close LDAP connection when recovering to avoid server load (3315)
3730 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3731 - don't leak open file on MDS after open resend (3325)
3732 - serialize filter_precreate and filter_destroy_precreated (3329)
3733 - loop device shouldn't call sync_dev() for nul device (3092)
3734 - clear page cache after eviction (2766)
3735 - resynchronize MDS->OST in background (2824)
3736 - refuse to mount the same filesystem twice on same mountpoint (3394)
3737 - allow llmount to create routes for mounting behind routers (3320)
3738 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3739 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3740 - fix rare problem with rename on htree directories (3417)
3742 - allow default OST striping configuration per directory (1414)
3743 - fix compilation for qswnal for 2.6 kernels (3125)
3744 - increase maximum number of MDS request buffers for large systems
3745 - change liblustreapi to be useful for external progs like lfsck (3098)
3746 - increase local configuration timeout for slow disks (3353)
3747 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3749 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3752 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3753 - don't overwrite extent policy data in reply if lock was blocked (2901)
3754 - drop filter export grants atomically with removal from device (2663)
3755 - del obd_self_export from work_list in class_disconnect_exports (2908)
3756 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3757 - swab reply message in mdc_close, other PPC fixes (2464)
3758 - fix destroying of named logs (2325)
3759 - overwrite old logs when running lconf --write_conf (2264)
3760 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3761 - fix race in target_handle_connect (2898)
3762 - mds_reint_create() should take same inode create lock (2926)
3763 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3764 - don't close files for self_export to avoid uninitialized obd (2936)
3765 - allow MDS with the same name as client node (2939)
3766 - hold dentry reference for closed log files for unlink (2325)
3767 - reserve space for all logs during transactions (2059)
3768 - don't evict page beyond end of stripe extent (2925)
3769 - don't oops on a deleted current working directory (2399)
3770 - handle hard links to targets without a parent properly (2517)
3771 - don't dereference NULL lock when racing during eviction (2867)
3772 - don't grow lock extents when lots of conflicting locks (2919)
3774 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3777 - account for cache space usage on clients to avoid data loss (974)
3778 - lfsck support in lustre kernel code (2349)
3779 - reduce journal credits needed for BRW writes (2370)
3780 - orphan handling to avoid losing space on client/server crashes
3781 - ptlrpcd can be blocked, stopping ALL progress (2477)
3782 - use lock value blocks to assist in proper KMS, faster stat (1021)
3783 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3784 - recovery for initial connections (2355)
3785 - fixes for mds_cleanup_orphans (1934)
3786 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3787 - block all file creations until orphan recovery completes (1901)
3788 - client remove rq_connection from request struct (2423)
3789 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3790 - recovery timer race (2670)
3791 - mdc_close recovey bug (2532)
3792 - ptlrpc cleanup bug (2710)
3793 - mds timeout on local locks (2588)
3794 - namespace lock held during RPCs (2431)
3795 - handle interrupted sync write properly (2503)
3796 - don't try to handle a message that hasn't been replied to (2699)
3797 - client assert failure during cleanup after abort recovery (2701)
3798 - leak mdc device after failed mount (2712)
3799 - ptlrpc_check_set allows timedout requests to complete (2714)
3800 - wait for inflight reqs when ptlrpcd finishes (2710)
3801 - make sure unregistered services are removed from the srv_list
3802 - reset bulk XID's when resending them (caught by 1138 test)
3803 - unregister_bulk after timeout
3804 - fix lconf error (2694)
3805 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3806 - readahead locks pages, leaves pending causing memory pressure (2673)
3807 - increase OST request buffers to 4096 on large machines (2729)
3808 - fix up permission of existing directories in simple_mkdir (2661)
3809 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3810 - don't assign transno to errored transactions (2742)
3811 - don't delete objects on OST if given a bogus objid from MDS (2751)
3812 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3813 - if rq_no_resend, then timeout request after recovery (2432)
3814 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3815 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3816 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3817 - MDS recovery shouldn't delete objects that it has given out (2730)
3818 - if enqueue arrives after completion, don't clobber LVB (2819)
3819 - don't unlock pages twice when trigger_group_io returns error (2814)
3820 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3821 - don't write pages to disk if there was an error (1450)
3822 - don't ping imports that have recovery disabled (2676)
3823 - take buffered bytes into account when balancing socknal conn (2817)
3824 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3825 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3826 - remove little-endian swabbing of llog records (1987)
3827 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3828 - retry reposting request buffers if they fail (1191)
3829 - grow extent at grant time to avoid granting a revoked lock (2809)
3830 - lock revoke doesn't evict page if covered by a second lock (2765)
3831 - disable VM readahead to avoid reading outside lock extents (2805)
3833 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3834 - updated kernel patches for hp-2.4.20 kernel (2681)
3836 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3839 - fix truncated write corruption (2366)
3840 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3842 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3843 - ensure len(uuid) < 37 in lmc (1171)
3844 - fix ia64 OOPS in llog_test (2255)
3845 - zero end of page at obdfilter for partial page writes (2648)
3846 - don't leave stale dentries around after renames (bug 2428)
3847 - fix timeouts when evicting a client with a single lock held (2642)
3848 - set deadline for the initial HELLO message to drain (2634)
3849 - print out dotted-quad IP addresses in the socknal (2302)
3851 - additional debugging for MDS client eviction problem (2443)
3852 - fix mkfsoptions support for osts (2603, 2604)
3854 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3857 - add series for the vanilla 2.6.0 kernel
3858 - add series for the vanilla 2.4.24 kernel
3859 - add series for a cray x86/64 UL kernel drop
3860 - fix xattr patches for the vanilla 2.4.19 series
3862 - generate true UUIDs in lmc (1171)
3863 - have portals stack dumping break in UML (2466)
3864 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3865 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3866 - wake callback waiting threads on client eviction (2460)
3867 - Add --ptldebug and --subsystem to lmc (1719)
3868 - update assertion to allow safe interrupt allocation
3869 - set rq_no_resend for cancel requests (2432)
3870 - recalculate ptlrpcd timeout after resend (2494)
3871 - call vfs_rmdir when removing pending directories (2368)
3872 - fix renaming a file to itself (2429)
3873 - lmc creates a default one-stripe lov (2454)
3874 - expand procfs space to handle large clusters (2326)
3875 - increase UML stack to avoid overflow
3876 - update lconf's list of debug and subsystem masks
3877 - fix lfs find --obd (2510)
3878 - /proc tunable for disabling filter read caching (2591)
3879 - stop rpm packages from altering slapd.conf (2301)
3880 - disable nagle in the socknal under 0conf (2578)
3881 - choose mds inode size based on stripe count (2572)
3882 - fix kernel-source rpm problems (2516)
3884 - add --disable-doc to avoid pdf generation (2421)
3885 - update documentation, tests, type-os, comments
3886 - avoid format warnings on ia64
3887 - remove the TOE NAL
3888 - tiny code cleanups by removing unused fields
3890 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3893 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3894 - remove the most blatant lies from BUILDING file (2371)
3895 - change default debug level to reasonable production setting
3896 - reduce client side cache size to reduce cache flush time
3897 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3898 - make TCP zerocopy and pinger support enabled by default (2476)
3899 - sync writes completed after process exits caused crashes (2319)
3900 - maintain correct mount count on the MDS (2356)
3901 - backout 1557, because 2316 wasn't really fixed
3902 - better file I/O statistics gathering in /proc
3903 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3904 - another kernel patch to fix zero-copy TCP function export
3905 - don't take duplicate lock when processing re-sent getattr (2420)
3906 - lctl uses obd_self_export instead of creating new conn (2353)
3907 - MDS/OST recovery case which requires object creation asserted (2425)
3908 - move lfs from /usr/sbin to /usr/bin in packages
3909 - fix race between mds_client_add and mds_client_free (2417)
3910 - use kmalloc instead of slabs in portals (2430)
3911 - don't create duplicate records when a failover MDS is present (2442)
3912 - remove unnecessary mount age check (2332)
3913 - don't remove directory inodes from locks prematurely (2451)
3914 - don't break if MDS service name is the same as hostname (2103)
3915 - fix races in client write RPC generation when cache full (2482)
3917 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3920 - remove now-unused request->rq_obd (278)
3921 - if an allocation fails, print out how much memory we've used (1933)
3922 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3923 - add the "configurable stack size" patch to most series files (1256)
3924 - ability to write large log records, for 100+ OST configs (2306)
3925 - fix NULL deref when filter_prep fails (2314)
3926 - fix operator precedence error in filter_sync
3927 - dynamic allocation of socknal TX descriptors (2315)
3928 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3929 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3930 - ensure that CWARN messages in Portals always get to the syslog
3931 - __init/__exit are not for prototype decls (ldlm_init/exit)
3932 - x86-64 compile warning fixes
3933 - fix gateway LMC keyword conflict (2318)
3934 - fix MDS lock inversions in getattr/reint paths (1844)
3935 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3936 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3937 - DLM race condition prevented some lock evictions (2328)
3938 - ENOMEM detection and retry on socknal sends (2230)
3939 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3940 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3941 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3942 - better error messages when a client is rejected during recovery (1505)
3943 - avoid cancelling locks which were never granted, after failure (2330)
3944 - fix i_sem/journal inversion in mds_client_add (2333)
3945 - fix truncate/getattr lock cycle deadlock (2334)
3946 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3947 - fix two rare exit paths which could leak an l_lock() ref (2321)
3948 - fencepost error in MDS/OST orphan recovery (2226)
3949 - make log record alignment 8 bytes (1988)
3950 - lstripe now fails when requested offset > ost_count (2237)
3951 - ensure that all kernel series have a complete list.h (1607)
3952 - fix crashes in special-file operations (2316)
3953 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3955 - allow configurable automake binary, for testing new versions
3956 - small update to the lfs documentation
3958 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3960 * fix negative export reference count in fsfilt_sync (2312)
3962 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3963 * release candidate 0.9.1
3965 - orphans are moved into the PENDING directory for possible recovery
3966 - replayed opens now open by fid for orphan/rename safety (1042)
3967 - last close of an orphan inode generates a transno (683)
3968 - chdir() and mount() now pin the directory entry (1020)
3969 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3970 - discard very old requests without processing them (1502)
3971 - remove some common, well-understood CERRORs (1505)
3972 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3973 - clear "grant" flags in OST replies until OST grant code lands (1644)
3974 - fix read performance by not clobbering i_blksize on client (1598)
3975 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3976 - make LRU size a /proc tunable, clears locks when reduced (707)
3977 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3978 - server locks take references on exports now (1558)
3979 - build fixes for 2.4.20-rh trees (1663)
3980 - return an error from lov_create if all OSCs are inactive (1751)
3981 - fix import levels when a reconnect happens without a timeout (1597)
3982 - exit early from mds_open if we get a lookup error (1749)
3983 - partial page read at EOF wouldn't wait for disk before sending (1642)
3984 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3985 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3986 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3987 - write/truncate lock inversion (1639)
3988 - Don't auto-load obdclass, portals modules during cleanup (1495)
3989 - fix timestamps from jumping to "now" (1763)
3990 - extra journal assertions (1648)
3991 - add an extra multiunlink test (1771)
3992 - fix read_record/write_record API (1776)
3993 - fix leak of offset_extent, possible incorrect i_size later (1772)
3994 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3995 - Debugging for the kqswnal_get_idle_tx problems (1820)
3996 - Allow recovery to be attempted multiple times (1536)
3997 - Write out MDS last_rcvd file after it is first created (1600)
3998 - Fix tx_descriptor leak in failed transmit situations (1827)
3999 - ext3 journaling fixes for assertion failure after IO error (1871)
4000 - class_export_put() on freed export after completion AST error (1896)
4001 - Fix revalidate looping in VFS (1322)
4002 - Don't access a freed export during MDS_REINT timeout (1521)
4003 - Add open-unlink recovery support on the MDS (1673,1764)
4004 - Return an error if no MDS data was read from last_rcvd (1946)
4005 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4006 - Don't setup a disk device that doesn't match exported UUID (317)
4007 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4008 - avoid committing NULL handle in force close
4009 - local.sh is now a one-stripe LOV configuration
4010 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4011 - fix ext3 htree duplicate directory entry corruption (1516)
4012 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4013 - update to new LOV EA format (2097)
4014 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4016 2003-06-15 Phil Schwan <phil@clusterfs.com>
4019 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4020 - per-import/export recovery handling (958, 931, 959)
4021 - multiple last-rcvd slots, for serving multiple FSes (949)
4022 - connections are again shared between multiple imp/exports (963, 964)
4023 - "umount -f" would hang if any requests needed to be sent (393, 978)
4024 - avoid pinning large req buffer by copying for queued messages (989)
4025 - add "uuid" to "lctl device" command to help upcalls (991)
4026 - "open" RPCs with transnos would confuse recovery counters (1037)
4027 - do proper endian conversion of all wire messages (288, 340, 891)
4028 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4029 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4030 - fix ptlrpc_connection leak in target_handle_connect (1174)
4031 - fix import refcounting bug in OST and MDS cleanup (1134)
4032 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4033 - fix very unlikely obd_types race condition (501)
4034 - remove osc_open hack for echo_client (1187)
4035 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4036 - a failure in read_inode2 leads to deadlock (1139)
4037 - cancel ack-locks as soon as transaction is committed (1072)
4038 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4039 - make sure to commitrw after any preprw to avoid deadlock (1162)
4040 - failing to execute a file in a lustre FS would lock inode (1203)
4041 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4042 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4043 - obd_uuid2tgt would walk off the end of the list (1255)
4044 - on IA64 the osc would give portals incorrect bulk size (1258)
4045 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4046 - fix lock inversion caused by new llite matching code (1282)
4047 - limit the number of dirty pages on a client to 10MB (1286)
4048 - timed out locks were not being corrected cancelled (1289)
4049 - fix O_DIRECT above 4GB on IA-32 (1292)
4050 * major user-visible changes
4051 - fail out/fail over policy now controlled by the upcall (993)
4053 - add OBD_PING to check server availability and failure (954)
4054 - lustre messages are now sent in sending host order (288, 340, 891)
4055 - add eadatalen to MDS getattr reply (340)
4056 - OST read replies may contain second buffer, with per-page status (593)
4058 2003-03-11 Phil Schwan <phil@clusterfs.com>
4061 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4062 - failed open()s could cause deadlock; fixed (867, 869)
4063 - stop cancelling OST locks when files are closed (481)
4064 - overlapping XID spaces caused network corruption (851, 853)
4065 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4066 - setattr_raw added, to do single-RPC, server-side setattrs
4067 - lmc/lconf syntax change for OST UUIDs
4068 - fix crashy race condition between ptlrpc_free_req and osc_close
4069 - don't use request in mdc_enqueue if we hit a timeout (889)
4070 - don't set the inode i_size for regular files from the MDS (896)
4071 - handle out of order completion AST (842)
4072 - don't LBUG if a lock request times out after receiving AST (913)
4073 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4074 - if a bad lock AST arrives, send an error instead of dropping entirely
4075 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4076 - fix leak in bulk IO when only partially completed (899, 900, 926)
4077 - fix O_DIRECT for ia64 (55)
4078 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4079 - C-z after timeout could hang a process forever; fixed (977)
4081 - client-side I/O cache (678, 924, 929, 941, 970)
4083 - READPAGE and SETATTRs which don't take server-side locks get
4086 2003-02-11 Phil Schwan <phil@clusterfs.com>
4089 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4090 - Fix fsfilt_extN_readpage() to read a full page of directory
4091 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4092 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4093 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4094 - Most intent operations were removed, in favour of a new RPC mode
4095 that does a single RPC to the server and bypasses most of the VFS
4096 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4097 - Aggressively cancel local locks on DLM servers
4098 - mds_reint_unlink sends EA to the client if it's the last nlink.
4099 client uses that EA to unlink OST objects.
4100 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4101 - recursive symlinks were fixed (439)
4102 - fixed NULL deref in DEBUG_REQ
4103 - filter_update_lastobjid no longer calls sync, which annoyed extN
4104 - fixed multi-client small-writes to a single file problem (445)
4105 - fixed mtime updates during file writes (607)
4106 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4107 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4108 - fixed Linux OST size reporting problem (444, 656)
4109 - OST now updates object mtime with writes or setattr (607, 619)
4110 - client verifies file size before zeroing page past EOF (445)
4111 - OST now writes last allocated objid to disk with allocation (108)
4112 - LOV on echo now works (409)
4114 - mds_reint_unlink sends a new buffer, with the EA included. this
4115 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4116 set if a regular file was being unlinked, and it was the last link
4117 - use PtlGet from the target for bulk writes (315)
4118 - OST now updates object mtime with writes or setattr (607, 619)
4119 - LDLM now has a grant-time callback to revalidate locked items, if
4121 - Many MDS operations were reorganized to combat race conditions
4123 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4124 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4126 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4129 - Fully reactivate OST imports after reconnection (512, others)
4130 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4131 - More graceful error handling for truncating on dead OST (515)
4132 - Don't error out unless we're actually accessing dead stripes (474)
4133 - Fix garbage sizes when stripes are missing (410)
4134 - LRU counters were broken, causing constant lock purge (433, 432)
4135 - garbage on read from stripes with failed OSTs (441)
4136 - mark OSCs as active before reconnecting during recovery (438)
4137 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4138 - lfind did not preserve OST order in output (443)
4139 - symlinks cause hung clients, incorrect data (439)
4140 - stop dereferencing request after dropping refcount (457)
4141 - don't LASSERT(spin_is_locked) on non-SMP (455)
4142 - fixes for many rename() bugs
4143 - fstat didn't correctly synchronize attributes (399)
4144 - server must handle lock cancellation during blocking AST prep (487)
4145 - bulk descriptors were free()d too soon (511)
4146 - fix paths in lconf, which would load incorrect modules (451, 507)
4147 - fix confusing lconf 'host not found' error message (386)
4148 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4149 - fix race condition in mdc_blocking_ast() for inode access (526)
4150 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4151 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4152 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4153 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4154 - fix lstripe in several ways (kernel side) (527)
4155 - fix request leak in ldlm_cli_enqueue (262)
4156 - incorrect OSC was marked inactive after OST failure
4157 - call mds_fs_cleanup before unmounting filesystem (524)
4158 - fix races between taking ns_lock and ldlm_lock_change_resource
4159 - fix races updating LOV export open file list
4160 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4161 - fix recovery NULL deref in ldlm_cli_cancel_unused
4162 - fix some DLM races by using new hash table for lock handles (419)
4163 - permit the client to specify desired inodes, at replay
4164 - duplicate requests when we queue them for replay reintegration
4165 - fix last_rcvd offset calculation
4166 - sync after each recovered transaction, so we always make progress
4167 - never, not always, ERESTART requests without transnos
4168 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4170 - skip replay if the MDS says that the client is already connected
4171 - don't check for a recovery-enabled export to match lctl's UUID
4172 - don't INC_USE_COUNT for phantom exports
4173 - don't crash when cleaning up phantom exports (567)
4174 - don't double-finish or set replay data for errored mdc_open requests
4175 - abort requests when they time out, so we don't get old replies
4176 - send/receive replies for AST messages again
4177 - if the client says that it doesn't have the lock, cancel it on the
4179 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4180 mark it as destroyed, it will all work out when decref is called
4181 - fix module use counts (22, 581)
4183 - ASTs now expect a reply (server cancels lock on error reply)
4185 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4188 - fix many simultaneous client startup (392)
4189 - fix dentry->d_it clobbering
4190 - credentials weren't being shipped for readdir/getattr operations
4191 - remove invalid assertions triggered during some concurrent MD
4193 - proper Lustre versions added (336, 389)
4194 - fix memory leak for create error case (398)
4195 - fix LOV locking bug that would get cli/srv out of sync
4196 - fix echo client over LOV (409)
4197 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4198 - fix double-O_EXCL intent crash (424)
4199 - avoid sending multiple lock CANCELs (352)
4201 - MDS can do multi-client recovery (modulo bugs in new code)
4203 - many updates, edits, cleanups
4205 2002-11-18 Phil Schwan <phil@clusterfs.com>
4208 - fix null d_it dereference (346)
4209 - fix full OST/dbench hang (333)
4210 - fix permission problem with file removal (286)
4211 - fix removal of OSCs from LOV when they fail
4212 - fix NULL deref during bulk timeout (214)
4213 - fix problems related to multiple filesystems on one MDS (241)
4214 - fixed serious subtle metadata locking bugs
4215 - free locks on clients when inodes are removed due to memory
4217 - fix inode pointer in lock data (285)
4218 - partial support for multiple MDS on a single host (241)
4219 - data locks weren't cancelled at clear_inode time (290, 311)
4220 - intent locks could lead to unbounded lock growth (205)
4221 - added a maximum lock count, an LRU list, and a flusher
4222 - fix multiple rename (365)
4223 - properly abstracted the echo client
4224 - OSC locked 1 byte too many; fixed
4225 - rewrote brw callback code:
4226 - fixed recovery bugs related to LOVs (306)
4227 - fixed too-many-pages-in-one-write crash (191)
4228 - fixed (again) crash in sync_io_timeout (214)
4229 - probably fixed callback-related race (385)
4231 - Add capability to MDS protocol
4232 - LDLM cancellations and callbacks on different portals
4234 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4237 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4238 - defer open object destruction to close time (601981, 138)
4239 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4240 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4241 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4242 - don't call obd_setattr in ll_file_release for destroyed objects
4243 * protocol change to lustre_msg: move |version| and add |flags|
4244 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4245 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4246 * added replay of create, unlink, link and rename operations during
4247 MDS failover; recovery should be much more robust now
4248 * remove failed OSCs from LOVs (only lov_create uses this so far)
4249 * the lustre-HOWTO was brought (more) up to date (582544)
4251 2002-10-23 Phil Schwan <phil@clusterfs.com>
4254 - in-use dentries weren't being reused properly (617851)
4255 - prevent multiple LDLM setup (599178)
4256 - fix LOV size calculations for truncate (617853)
4257 - fix client handling of MDS intent errors (POSIX)
4258 - fix permission bug in lovstripe.c test (624321)
4259 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4260 - truncate past end of file could corrupt data
4261 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4262 - a race in recovery could return ETIMEDOUT to apps (623947)
4263 - building outside the source directory was fixed
4264 * the lustre-HOWTO was brought (more) up to date (582544)
4265 * major progress was made on recovery functionality
4267 2002-10-10 Phil Schwan <phil@clusterfs.com>
4270 - recovery deadlock fix
4271 - rm -rf causes LBUG fix (617817)
4272 - file open by multiple tasks fix (618962)
4273 - directory permissions bugs (602707 and 620007)
4274 - journal_stop fixed with locking (611313)
4275 - O_APPEND failures resolved (618273, perhaps 614459)
4276 - lconf PATH fix (619770)
4277 - IA64 build fix (621450)
4278 - RPC buffer sizes scale with amount of memory
4280 2002-10-01 Phil Schwan <phil@clusterfs.com>
4283 - locks would be cancelled without throwing away data pages,
4284 resulting in inconsistent data (605627)
4285 - inode attributes were not always being refreshed (605627, 612449)
4286 - lconf now continues to cleanup after lctl reports an error
4287 - MDS now enforces user permissions (602707)
4288 - lprocfs cleanup fixed, but not yet enabled (614157)
4289 - fixed infinite server hang, should a client not respond to an AST
4290 - avoid going into recovery if user calls readlink() with a buffer
4291 that's too small (613941)
4292 - AST RPCs no longer require replies (614867) -- this may be changed
4293 - don't crash server if client sends an IOV that's too big (611336)
4294 - fixed lock conversion deadlock (611892)
4295 - fixed the following of symlinks (614622)
4296 * recovery: the server can remove locks from a client that dies, other
4297 clients can make progress
4298 * more extN patch fixes
4299 * compile-time configurable ptlrpc buffer allocations
4301 - collaborative read cache document
4302 - Lustre Lite Performance CDR document-in-progress
4304 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4307 - fix typo in patch-2.4.18
4309 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4312 - clear ptlrpc request each time in handle_incoming_request()
4313 - unlink of files now destroys the object on the OST
4315 2002-09-19 Peter Braam <braam@clusterfs.com>
4317 * add hard link support
4318 * change obdfile creation method
4319 * kernel patch changed
4321 2002-09-19 Peter Braam <braam@clusterfs.com>
4324 - stack overflow bug in extN fixed
4326 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4328 * documentation updates
4329 - add man pages for config tools
4330 - update tests/README to describe testing with new config tools
4331 - finish metadata API descriptions
4332 * bug fixes and cleanups
4333 - statfs workaround for 16TB limit
4334 - LOV stripe allocation improved, can stripe on subset of OSTs
4335 - LOV file size/IO offset was wrong for files > 4GB in size
4336 - object EA data was being dropped, caused files to be unreadable
4337 - memory overflow with non-LOV OST caused memory corruption
4338 - fixed regression tests to work with new config tools, obdfilter
4339 - fixed bug when directory size became larger than 1 block
4340 - fixed bug (for single client case) when PWD was deleted
4341 - invalidate local directory pages when doing intent-based ops
4342 - avoid LDLM oops when lock callback contained bad data
4344 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4346 * documentation updates
4347 * bug fixes and cleanups
4348 - configuration tools
4351 - 64-bit compile warnings
4352 - 64-bit internal statfs data
4354 * test_brw on persistent OST devices
4356 * lprocfs (disabled)
4358 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4360 * documentation updates
4361 * bug fixes and cleanups
4362 * configuration tools
4364 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4367 * many small fixes to 0_5_4
4368 * io/network handling
4369 * thinkos in MDS operations
4371 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4374 * crucial basic fixes to 0.5.3
4375 * IOR, Iozone work over Elan
4378 2002-08-07 Phil Schwan <phil@clusterfs.com>
4379 * version 0_5_3, our first alpha
4380 * we use the new Portals iovs
4381 * documentation updates
4382 * bug fixes and cleanups
4383 * small changes in the DLM wire protocol
4385 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4386 * version 0_5_1 with some initial stability,
4387 * locking on MD and file I/O.
4388 * documentation updates
4389 * several bug fixes since 0.5.0
4390 * small changes in wire protocol
4392 2002-07-18 Phil Schwan <phil@clusterfs.com>
4394 * delivered as Lustre Light Alpha
4395 * fixed a crash after handling invalid MDS requests
4396 * fixed directory pages for architectures with non-4k pages sizes
4398 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4399 * release version v0_4_4
4400 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4402 * Fixes a number of interruption problems with OST operations.
4403 * Update documentation for portals header changes
4404 * Move all wire protocol structs/defines to lustre_idl.h
4405 * Fixes symlink length bug.
4406 * Add tcpdump to repository.
4408 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4409 * release version v0_4_3
4410 * Fixes statfs for inodes on extN.
4411 * Fixes bug in runtests which would delete /etc/hosts.
4412 * Use 64-bit object IDs wherever possible (not into VFS though)
4413 Remove ost_get_info, which is unused by lustre, and out of date.
4415 2002-07-03 Peter Braam <braam@clusterfs.com>
4416 * release version v0_4_2 Fixes a lookup error (type not passed)
4417 * move forward to head of Portals
4418 * move forward to latest Lustre kernel
4420 2002-06-25 Peter Braam <braam@clusterfs.com>
4421 * release version v0_4_1. Hopefully stable on single node use.