1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.5 (SLES 10),
6 2.6.18-53.1.13.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
16 Frequency : occasional
18 Description: Correctly check stale fid, not start epoch if ost not support SOM
19 Details : open with flag O_CREATE need set old fid in op_fid3 because op_fid2
20 overwrited with new generated fid, but mds can anwer with one of these
21 two fids and both is not stale. setattr incorectly start epoch and
22 assume will be called done_writeting, but without SOM done_writing
26 Frequency : rare, depends on device drivers and load
28 Description: MDS or OSS nodes crash due to stack overflow
29 Details : Code changes in 1.8.0 increased the stack usage of some functions.
30 In some cases, in conjunction with device drivers that use a lot
31 of stack the MDS (or possibly OSS) service threads could overflow
32 the stack. One change which was identified to consume additional
33 stack has been reworked to avoid the extra stack usage.
36 Frequency : occasional
38 Description: Do not fail import if osc_interpret_create gets -EAGAIN
39 Details : If osc_interpret_create got -EAGAIN it immediately exits and
40 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
41 oscc_has_objects and see OSC has no objests and call
42 oscc_internal_create to resend create request.
44 Severity : enhancement
46 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
48 Severity : enhancement
50 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
55 Description: Fix replay if there is an un-replied request and open
56 Details : In some cases, older replay request will revert the
57 mcd->mcd_last_xid on MDS which is used to record the client's
60 Severity : enhancement
62 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
64 Severity : enhancement
66 Description: Add rhel5 support to HEAD.
68 Serverity : enhancement
70 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
75 Description: Don't allow skipping OSTs if index has been specified.
76 Details : Don't allow skipping OSTs if index has been specified, make locking
77 in internal create lots better.
81 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
82 Details : access to bitfield in structure is always rounded to long
83 and this produce problem with not atomic change any bit.
87 Description: Lustre make rpms failed.
88 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
89 builds Lustre rpms from tarball.
92 Frequency : rare on shutdown ost
94 Description: If llog cancel was not send before clean_exports phase, this can
95 produce deadlock in llog code.
96 Details : If llog thread has last reference to obd and call class_import_put
97 this produce deadlock because llog_cleanup_commit_master wait when
98 last llog_commit_thread exited, but this never success because was
99 called from llog_commit_thread.
103 Description: allow userland application know is lost one of stripes.
104 Details : fill lvb_blocks with error code on ost and return it to
105 application if error flag found.
109 Description: NULL lov_tgts causing MDS oops
110 Details : more safe checks for NULL lov_tgts for avoid oops.
112 Severity : enhancement
114 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
118 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
119 Details : If an OST is down the MDS will hang indefinitely in
120 obd_statfs_rqset() waiting for the statfs data. While for
121 MDS QOS usage of statfs, it should not stuck in waiting.
123 Severity : enhancement
125 Description: remote_acl support
126 Details : Support ACL-based permission check for remote user.
127 Support setfacl/getfacl for remote user with the utils
128 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
129 the system "{s,g}etfacl" utils.
131 Severity : enhancement
133 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
135 Severity : enhancement
137 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
141 Description: make mgs_setparam() handle fsname containing dash
142 Details : fsname containing a dash does not work with lctl conf_param
144 Severity : enhancement
146 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
148 Severity : enhancement
150 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
154 Frequency : rare, at shutdown
155 Description: access already free / zero obd_namespace.
156 Details : if client_disconnect_export was called without force flag set,
157 and exist connect request in flight, this can produce access to
158 NULL pointer (or already free pointer) when connect_interpret
159 store ocd flags in obd_namespace.
163 Frequency : only at startup
164 Description: not alloc memory with spinlock held.
165 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
166 if any spinlock held.
168 Severity : enhancement
170 Description: make lustre randomly fail allocating memory
171 Details : Make lustre randomly failed allocating memory for testing purpose.
173 Severity : enhancement
175 Description: lost problems with lov objid file
176 Details : Fixes some scability and access to not inited memory problems
177 in work with lov objdid file.
182 Description: lfs find does not continue on file error
183 Details : Continue other files processing when a file/dir is absent.
187 Description: Inconsistent usage of lustre_pack_reply()
188 Details : Standardize the usage of lustre_pack_reply() such that it
189 always generate a CERROR on failure.
192 Frequency : occasional
194 Description: MDS hang or stay in waiting lock
195 Details : If client receive lock with CBPENDING flag ldlm need send lock
196 cancel as separate rpc, to avoid situation when cancel request
197 can't processed due all i/o threads stay in wait lock.
201 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
205 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
209 Description: Update to SLES9 kernel-2.6.5-7.287.3.
213 Description: Update to RHEL4 latest kernel.
215 Severity : enhancement
217 Description: Build SLES10 patchless client fails
218 Details : The configure was broken by run ./configure with
219 --with-linux-obj=.... argument for patchless client. When the
220 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
221 can't search header adequately. Use absolute path such as
222 -I($LINUX)/include instead.
226 Description: interrupt oig_wait produce painc on resend.
227 Details : brw_redo_request can be used for resend requests from ptlrpcd and
228 private set, and this produce situation when rq_ptlrpcd_data not
229 copyed to new allocated request and triggered LBUG on assert
230 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
231 ptlrpcd set if request is changed and can be safety changed to use
236 Description: LASSERT_{REQ,REP}SWAB macros are buggy
237 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
238 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
239 these macros and replace them with their difinitions instead.
243 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
244 Details : Update which_patch & target file for SLES10 latest kernel.
247 Description: Remove client patches from SLES 10 kernel.
248 Details : This causes SLES 10 clients to behave as patchless clients
249 even on a Lustre-patched (server) kernel.
251 Severity : enhancement
253 Description: self-adjustable client's lru lists
254 Details : use adaptive algorithm for managing client cached locks lru
255 lists according to current server load, other client's work
256 pattern, memory activities, etc. Both, server and client
257 side namespaces provide number of proc tunables for controlling
260 Severity : enhancement
262 Description: light-weight GSS support
263 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
264 and reduce performance overhead.
266 Severity : enhancement
268 Description: Linux keyring support
269 Details : Support using service of Linux keyring for Lustre GSS internal
270 context refresh/cache mechanism.
274 Description: Fix errors in lfs documentation
275 Details : Fixes man pages
279 Description: don't use GFP_* in generic Lustre code.
280 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
284 Description: obdclass is limited by single OBD_ALLOC(idarray)
285 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
289 Description: warnings with build patchless client with vanila 2.6.19 and up
290 Details : change old ctl_table style and replace ctl_table/ctl_table_header
291 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
295 Frequency : Only for SLES
296 Description: sanity_quota fail test_1
297 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
298 file, which makes TSTID[2] inunique.
302 Frequency : Only for SLES9
303 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
304 Details : Update target/ChangeLog/which_patch .
308 Frequency : Only for test_10 in sanity-quota.sh
309 Description: a bug in quota test code
310 Details : checking a test flag in a wrong place
314 Frequency : Only for test_13 in sanity-quota.sh
315 Description: a bug in quota test script
316 Details : shouldn't check overfull allocated quota
320 Description: jbd statistics
321 Details : Port older jbd statistics patch for sles10
323 Severity : enhancement
325 Description: Add printing inode info into message about error in writepage.
330 Description: reply_lock_interpret crash due to race with it and lock cancel.
331 Details : Do not replay locks that are being cancelled. Do not reference
332 locks by their address during replay, just by their handle.
336 Frequency : When flocks are used.
337 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
338 Details : Flock locks might destroy just granted lock if it could be merged
339 with another existing flock, this is done in completion handler,
340 so teach ldlm_cli_enquque_fini that this is a valid case for
346 Description: Oops in read and write path when failing to allocate lock.
347 Details : Check if lock allocation failed and return error back.
351 Description: lstripe command fails for valid OST index
352 Details : The stripe offset is compared to lov->desc.ld_tgt_count
353 instead of lov->desc.ld_active_tgt_count.
357 Description: sanity.sh failed test 103
358 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
363 Description: df doesn't work properly if diskfs blocksize != 4K
364 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
368 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
369 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
373 Description: obd_health_check_timeout too short
374 Details : set obd_health_check_timeout as 1.5x of obd_timeout
378 Description: llapi_file_create() does not allow some changes
379 Details : add llapi_file_open() that allows specifying the mode and
380 open flags, and also returns an open file handle.
384 Description: merge and cleanup kernel patches.
385 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
389 Description: Add journal checksum support.(Kernel part)
390 Details : The journal checksum feature adds two new flags i.e
391 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
392 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
393 indicates that the commit block contains the checksum for
394 the blocks described by the descriptor blocks. Now commit
395 record can be sent to disk without waiting for descriptor
396 blocks to be written to disk. This behavior is controlled
397 using JBD2_FEATURE_ASYNC_COMMIT flag.
401 Description: OSS needs mutliple precreate threads
402 Details : Add ability to start more than one create thread per OSS.
406 Description: Sanity.sh test_65k failed: setstripe should have succeeded
407 Details : execute lfs setstripe on client
411 Description: mds_obd_create error creating tmp object
412 Details : When the user sets quota on root, llog will be affected and can't
413 create files and write files.
416 Frequency : Always on ia64 patchless client, and possibly others.
418 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
419 Details : This allows the patchless client to be loaded on architectures
424 Description: RedHat Update kernel for RHEL5
425 Details : Add the kernel config file for RHEL5.
429 Description: RedHat Update kernel for RHEL5
430 Details : Modify the kernel config file more closer RHEL5.
434 Description: Build failure against Centos5 (RHEL5)
435 Details : Define PAGE_SIZE when it isn't present.
439 Description: test_7 on sanity-quota.sh can't be used on separate machines
440 Details : add facet to handle it
444 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
445 Details : with NFS, the anon dentry's parent was set to itself in
446 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
447 corresponding dentry other than use rec->ur_name.
449 Severity : enhancement
451 Description: lfs setstripe enhancement
452 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
456 Description: enable data checksumming by default
457 Details : enable checksum by default, allow --disable-checksum
458 configure option and "-o nochecksum" mount option. Checksums
459 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
463 Description: lustre support for RHEL5
464 Details : Add support for RHEL5.
468 Description: Client eviction due to failover config
469 Details : after a connection loss, the lustre client should attempt to
470 reconnect to the last active server first before trying the
471 other potential connections.
475 Description: when mds and osts use different quota unit(32bit and 64bit),
476 quota will be released repeatly.
477 Details : void sending multiple quota reqs to mds, which will keep the
478 status between the reqs.
482 Description: osts not allocated evenly to files
483 Details : change the condition to increase offset_idx
486 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
488 Description: Data corruption for OSTs that are formatted larger than 2TB
490 Details : When generating the bio request for lustre file writes the
491 sector number would overflow a temporary variable before being
492 used for the IO. The data reads correctly from Lustre (which
493 will overflow in a similar manner) but other file data or
494 filesystem metadata may be corrupted in some cases.
498 Description: Tune the kernel for good SCSI performance.
499 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
500 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
505 Description: rewrite ext2-derived code in obdclass/uuid.c
506 Details : rewrite inherited code (uuid parsing code from ext2 utils)
507 from scratch preserving functionality.
511 Description: "lfs find -obd UUID" prints directories
512 Details : "lfs find -obd UUID" will return all directory names instead
513 of just file names. It is incorrect because the directories
514 do not reside on the OSTs.
518 Description: use i_size_read and i_size_write in 2.6 port
519 Details : replace inode->i_size access with i_size_read/write()
522 Frequency : when removing large files
524 Description: scheduling issue during removal of large Lustre files
525 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
526 It causes scheduling issues when removing large files (17TB in the
530 Frequency : only with liblustre clients on XT3
532 Description: evictions taking too long
533 Details : allow llrd to evict clients directly on OSTs
536 Frequency : only on ppc
538 Description: /proc/fs/lustre/devices broken on ppc
539 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
540 We should make sure the type of 'index' is loff_t and then cast
541 explicitly as needed below. Do not assign an explicitly cast
545 Frequency : only for rhel5
547 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
548 Details : Modify the target file & which_kernel.
550 Severity : enhancement
552 Description: omit set fsid for export NFS
553 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
554 as FS_REQUIRES_DEV for avoid problems with generate fsid.
558 Frequency : Always, for kernels after 2.6.16
559 Description: Fix warning idr_remove called for id=.. which is not allocated.
560 Details : Last kernels save old s_dev before kill super and not allow
561 to restore from callback - restore it before call kill_anon_super.
565 Description: buffer overruns could theoretically occur
566 Details : llapi_semantic_traverse() modifies the "path" argument by
567 appending values to the end of the origin string, and a
568 overrun may occur. Adding buffer overrun check in liblustreapi.
572 Description: Fix error on 'ls .' at the top of the Lustre mount.
573 Details : Don't revalidate dentry if it is a root dentry.
577 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
578 Details : Modify vm-tunables-rhel4.patch.
582 Description: Kernel config for 2.6.18-vanilla.
583 Details : Modify targets/2.6-vanilla.target.in.
584 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
585 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
586 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
587 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
591 Description: improve handling recoverable errors
592 Details : if request processig with error which can be recoverable on server
593 request should be resend, otherwise page released from cache and
598 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
599 Details : Modify target file & which_patch.
600 A flaw was found in the IA32 system call emulation provided
601 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
602 value could be stored in the %RAX register, which could trigger an
603 out-of-bounds system call table access. An untrusted local user
604 could exploit this flaw to run code in the kernel
605 (ie a root privilege escalation). (CVE-2007-4573).
609 Description: O_DIRECT bypasses client statistics.
610 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
611 read_ahead_stats were updated. copied the stats section from
612 osc_send_oap_rpc() into async_internal().
616 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
620 Description: change order of libsysio includes
621 Details : '#include sysio.h' should always come before '#include xtio.h'
625 Description: handle "serious error: objid * already exists" more gracefully
626 Details : If LAST_ID value on disk is smaller than the objects existing in
627 the O/0/d* directories, it indicates disk corruption and causes an
628 LBUG(). If the object is 0-length, then we should use the existing
629 object. This will help to avoid a full fsck in most cases.
631 Severity : enhancement
633 Description: adapt the lustre_config script to support the upgrade case
634 Details : Add "-u" option for lustre_config script to support upgrading 1.4
635 server targets to 1.6 in parallel.
639 Description: To avoid grant space > avaible space when the disk is almost
640 full. Without this patch you might see the error "grant XXXX >
641 available" or some LBUG about grant, when the disk is almost
643 Details : In filter_check_grant, for non_grant cache write, we should
644 check the left space by if (*left > ungranted + bytes), instead
645 of (*left > ungranted), because only we are sure the left space
646 is enough for another "bytes", then the ungrant space should be
647 increase. In client, we should update cl_avail_grant only there
648 is OBD_MD_FLGRANT in the reply.
652 Description: Update RHEL 4 kernel to fix local root privilege escalation.
653 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
654 described in CVE-2007-4573. This problem could allow untrusted
655 local users to gain root access.
658 Frequency : when using O_DIRECT and quotas
660 Description: Incorrect file ownership on O_DIRECT output files
661 Details : block usage reported by 'lfs quota' does not take into account
662 files that have been written with O_DIRECT.
667 Description: touch file failed when fs is not full
668 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
669 otherwise we can get ENOSP while fs is not full.
673 Description: parallel lock callbacks
674 Details : Instead of sending blocking and completion callbacks as separated
675 requests, adding them to a set and sending in parallel.
678 Frequency : only for Cray XT3
679 Bugzilla : 12829/13455
680 Description: Changing primary group doesn't change the group lustre assigns to
682 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
683 overridden with the primary group provided by the group upcall,
684 whereas the supplied fsgid can be trusted if it is in the list of
685 supplementary groups returned by the group upcall.
687 Severity : enhancement
689 Description: Allow masking D_WARNING, D_ERROR messages from console
690 Details : Console messages can now be disabled via lnet.printk.
694 Description: User code with malformed file open parameter crashes client node
695 Details : Before packing join_file req, all the related reference should be
696 checked carefully in case some malformed flags cause fake
697 join_file req on client.
701 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
702 Details : ldlm_completion_ast() assumes that a lock is granted when the req
703 mode is equal to the granted mode. However, it should also check
704 that LDLM_FL_CP_REQD is not set.
708 Description: Heavy nfs access might result in deadlocks
709 Details : After ELC code landed, it is now improper to enqueue any mds
710 locks under och_sem, because enqueue might want to decide to
711 cancel open locks for same inode we are holding och_sem for.
715 Description: Client eviction while running blogbench
716 Details : A lot of unlink operations with concurrent I/O can lead to a
717 deadlock causing evictions. To address the problem, the number of
718 oustanding OST_DESTROY requests is now throttled to
719 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
720 callbacks are processed in priority.
724 Description: enable ACLs on MDS by default
725 Details : ACLs must be enabled on MDS by default.
728 Frequency : PPC/PPC64 only
730 Description: conflicts between asm-ppc64/types.h and lustre_types.h
731 Details : fix duplicated definitions between asm-ppc64/types.h and
732 lustre_types.h on PPC.
736 Description: better handle error messages in extents code
738 --------------------------------------------------------------------------------
740 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
742 * Support for kernels:
743 2.6.5-7.283 (SLES 9),
744 2.6.9-55.EL (RHEL 4),
745 2.6.16.46-0.14 (SLES 10),
746 2.6.18.8 vanilla (kernel.org)
747 * Client support for unpatched kernels:
748 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
749 2.6.16 - 2.6.22 vanilla (kernel.org)
750 * Due to recently discovered recovery problems, we do not recommend
751 using patchless RHEL 4 clients with this or any earlier release.
752 * Recommended e2fsprogs version: 1.39.cfs8
753 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
754 * Starting with this release, the ldiskfs backing filesystem required
755 by Lustre is now in its own package, lustre-ldiskfs. This package
756 should be installed. It is versioned separately from Lustre and
757 may be released separately in future.
761 Description: block reactivating mgc import until all deactivates complete
762 Details : Fix race when failing back MDT/MGS to itself (testing)
765 Frequency : at statup only
767 Description: mds_lov_synchronize race leads to various problems
768 Details : simultaneous MDT->OST connections at startup can cause the
769 sync to abort, leaving the OSC in a bad state.
771 Severity : enhancement
773 Description: add optional extra BUILD_VERSION info
774 Details : add a new environment variable (namely LUSTRE_VERS) which allows
775 to override the lustre version.
778 Frequency : 2.6.18 servers only
780 Description: ll_kern_mount() doesn't release the module reference
781 Details : The ldiskfs module reference count never drops down to 0
782 because ll_kern_mount() doesn't release the module reference.
787 Description: server LBUG when using old ost_num_threads parameter
788 Details : Accept the old ost_num_threads parameter but warn that it
789 is deprecated, and fix an off-by-one error that caused an LBUG.
794 Description: Transient SCSI error results in persistent IO issue
795 Details : iobuf->dr_error is not reinitialized to 0 between two
799 Frequency : sometimes when underlying device returns I/O errors
801 Description: OSTs not going read-only during write failures
802 Details : OSTs are not remounted read-only when the journal commit threads
803 get I/O errors because fsfilt_ext3 calls journal_start/stop()
804 instead of the ext3 wrappers.
808 Description: poor connect scaling with increasing client count
809 Details : Don't run filter_grant_sanity_check for more than 100 exports
810 to improve scaling for large numbers of clients.
813 Frequency : SLES10 only
815 Description: sanity-quota.sh quotacheck failed: rc = -22
816 Details : Quotas cannot be enabled on SLES10.
819 Frequency : liblustre clients only
821 Description: getdirentries does not give error when run on compute nodes
822 Details : getdirentries does not fail when the size specified as an argument
823 is too small to contain at least one entry
825 Severity : enhancement
827 Description: Add LNET router traceability for debug purposes
828 Details : If a checksum failure occurs with a router as part of the
829 IO path, the NID of the last router that forwarded the bulk data
830 is printed so it can be identified.
835 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
836 Details : Due to a race condition, liblustre clients were occasionally
839 Severity : enhancement
841 Description: lfs setstripe use optional parameters instead of postional
844 Severity : enhancement
846 Description: Nanosecond timestamp support for ldiskfs
847 Details : The on-disk ldiskfs filesystem has added support for nanosecond
848 resolution timestamps. There is not yet support for this at
849 the Lustre filesystem level.
852 Frequency : during server recovery
854 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
855 Details : request with rq_no_resend flag not awake l_wait_event if they get a
859 Frequency : nfs export on patchless client
861 Description: connectathon hang when test nfs export over patchless client
862 Details : Disconnected dentry cannot be found with lookup, so we do not need
863 to unhash it or make it invalid
866 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
867 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
868 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
873 Description: open req refcounting wrong on reconnect
874 Details : If reconnect happened between getting open reply from server and
875 call to mdc_set_replay_data in ll_file_open, we will schedule
876 replay for unreferenced request that we are about to free.
877 Subsequent close will crash in variety of ways.
878 Check that request is still eligible for replay in
879 mdc_set_replay_data().
884 Description: disable writes to filesystem when reading health_check file
885 Details : the default for reading the health_check proc file has changed
886 to NOT do a journal transaction and write to disk, because this
887 can cause reads of the /proc file to hang and block HA state
888 checking on a healthy but otherwise heavily loaded system. It
889 is possible to return to the previous behaviour during configure
890 with --enable-health-write.
892 Severity : enhancement
894 Description: 64-bit inode version
895 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
896 to the inode. This will be required for version-based recovery.
901 Description: MDS fails to start if a duplicate client export is detected
902 Details : in some rare cases it was possible for a client to connect to
903 an MDS multiple times. Upon recovery the MDS would detect this
904 and fail during startup. Handle this more gracefully.
906 Severity : enhancement
908 Description: Add -o localflock option to simulate old noflock
910 Details : This will achieve local-only flock/fcntl locks
916 Description: log_commit_thread vs filter_destroy race leads to crash
917 Details : Take import reference before releasing llog record semaphore
922 Description: Wrong request locking in request set processing
923 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
924 imp_delayed_list, in this place should be used imp_lock.
927 Frequency : when reconnection
929 Description: Grant Leak when osc reconnect to OST
930 Details : When osc reconnect ost, OST(filter) should check whether it
931 should grant more space to client by comparing fed_grant and
932 cl_avail_grant, and return the granted space to client instead
933 of "new granted" space, because client will call osc_init_grant
934 to update the client grant space info.
937 Frequency : when client reconnect to OST
939 Description: Grant Leak when osc do resend and replay bulk write
940 Details : When osc reconnect to OST, OST(filter)should clear grant info of
941 bulk write request, because the grant info while be sync between
942 OSC and OST when reconnect, and we should ignore the grant info
943 these of resend/replay write req.
948 Description: Grant space more than avaiable left space sometimes.
949 Details : When then OST is about to be full, if two bulk writing from
950 different clients came to OST. Accord the avaliable space of the
951 OST, the first req should be permitted, and the second one
952 should be denied by ENOSPC. But if the seconde arrived before
953 the first one is commited. The OST might wrongly permit second
954 writing, which will cause grant space > avaiable space.
957 Frequency : when client is evicted
959 Description: Grant might be wrongly erased when osc is evicted by OST
960 Details : when the import is evicted by server, it will fork another
961 thread ptlrpc_invalidate_import_thread to invalidate the
962 import, where the grant will be set to 0. While the original
963 thread will update the grant it got when connecting. So if
964 the former happened latter, the grant will be wrongly errased
965 because of this race.
970 Description: Checking Stale with correct fid
971 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
972 to check whether it is stale, because sometimes, we want the
973 enqueue happened anyway, and op_data.fid2 will not be initialized.
975 Severity : enhancement
977 Description: update patchless client
978 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
981 Frequency : only with 2.4 kernel
983 Description: random memory corruption
984 Details : size of struct ll_inode_info is to big for union inode.u and this
985 can be cause of random memory corruption.
990 Description: Memory leak in recovery
991 Details : Lov_mds_md was not free in an error handler in mds_create_object.
992 It should also check obd_fail before fsfilt_start, otherwise if
993 fsfilt_start return -EROFS,(failover mds during mds recovery).
994 then the req will return with repmsg->transno = 0 and rc = EROFS.
995 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
996 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
997 be freed no matter whether fsfilt_commit success or not.
1000 Frequency : only with huge count clients
1002 Description: Prevents from taking the superblock lock in llap_from_page for
1004 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1005 for llap_from_page call in ll_removepage prevents from taking the
1006 superblock lock for a soon died page.
1011 Description: Not check open intent error before release open handle
1012 Details : in some rare cases, the open intent error is not checked before
1013 release open handle, which may cause
1014 ASSERTION(open_req->rq_transno != 0), because it tries to release
1015 the failed open handle.
1020 Description: Set cat log bitmap only after create log success.
1021 Details : in some rare cases, the cat log bitmap is set too early. and it
1022 should be set only after create log success.
1026 Description: Accessing a block bevice can re-enable I/O when Lustre is
1027 tearing down a device.
1028 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1033 Description: service threads may hog cpus when there are a lot of requests
1034 Details : Insert cond_resched to give other threads a chance to use some CPU
1039 Description: the cat log was not initialized in recovery
1040 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1041 unlink log on mds will not be initialized until mds post
1042 recovery. And also in mds post recovery, the unlink log will
1043 initialization will be done asynchronausly, so there will be race
1044 between add unlink log and unlink log initialization.
1048 Description: brw_stats were being printed incorrectly
1049 Details : brw_stats were being printed as log2 but all of them were not
1050 recorded as log2. Also remove some code duplication arising from
1051 filter_tally_{read,write}.
1055 Frequency : rare, only in recovery.
1056 Description: ASSERTION(req->rq_type != LI_POISON) failed
1057 Details : imp_lock should be held while iterating over imp_sending_list for
1058 prevent destroy request after get timeout in ptlrpc_queue_wait.
1062 Description: replay-single.sh test 52 fails
1063 Details : A lock's skiplist need to be cleanup when it being unlinked
1064 from its resource list.
1068 Description: Short directio read returns full requested size rather than
1070 Details : Direct I/O operations should return actual amount of bytes
1071 transferred rather than requested size.
1073 Severity : enhancement
1075 Description: metadata RPC reduction (e.g. for rm performance)
1076 Details : decrease the amount of synchronous RPC between clients and servers
1077 by canceling conflicing lock before the operation on the client
1078 and packing thier handles into the main operation RPC to server.
1080 Severity : enhancement
1082 Description: Async OSC create to avoid the blocking unnecessarily.
1083 Details : If a OST has no remain object, system will block on the creating
1084 when need to create a new object on this OST. Now, ways use
1085 pre-created objects when available, instead of blocking on an
1086 empty osc while others are not empty. If we must block, we block
1087 for the shortest possible period of time.
1089 Severity : enhancement
1091 Description: refine locking for avoid write wrong info into lov_objid file
1092 Details : fix possible races with add new target and write/update data in
1095 --------------------------------------------------------------------------------
1097 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1102 Frequency : on some architectures
1104 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1105 Details : Uninitialized flags sometimes cause configuration commands to
1109 Frequency : patchless clients only
1111 Description: missing __iget() symbol export
1112 Details : The __iget() symbol export is missing. To avoid the need for
1113 this on patchless clients the deathrow inode reaper is turned
1114 off, and we depend on the VM to clean up old inodes. This
1115 dependency was during via the fix for bug 12181.
1117 --------------------------------------------------------------------------------
1119 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1121 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1122 INTEROPERATE with older versions automatically. In many cases a
1123 special upgrade step is needed. Please read the
1124 user documentation before upgrading any part of a live system.
1125 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1126 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1127 * WARNING: Lustre configuration and startup changes are required with
1128 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1134 Severity : enhancement
1136 Description: MountConf
1137 Details : Lustre configuration is now managed via mkfs and mount
1138 commands instead of lmc and lconf. New obd types (MGS, MGC)
1139 are added for dynamic configuration management. See
1140 https://mail.clusterfs.com/wikis/lustre/MountConf for
1143 Severity : enhancement
1145 Description: dynamic OST addition
1146 Details : OSTs can now be added to a live filesystem
1148 Severity : enhancement
1150 Description: startup order invariance
1151 Details : MDTs and OSTs can be started in any order. Clients only
1152 require the MDT to complete startup.
1154 Severity : enhancement
1156 Description: parallel, asynchronous orphan cleanup
1157 Details : orphan cleanup is now performed in separate threads for each
1158 OST, allowing parallel non-blocking operation.
1160 Severity : enhancement
1162 Description: optimized stripe assignment
1163 Details : stripe assignments are now made based on ost space available,
1164 ost previous usage, and OSS previous usage, in order to try
1165 to optimize storage space and networking resources.
1167 Severity : enhancement
1169 Description: Permanently set tunables
1170 Details : All writable /proc/fs/lustre tunables can now be permanently
1171 set on a per-server basis, at mkfs time or on a live system.
1173 Severity : enhancement
1175 Description: Lustre message v2
1176 Details : Add lustre message format v2.
1178 Severity : enhancement
1180 Description: client OST exclusion list
1181 Details : Clients can be started with a list of OSTs that should be
1182 declared "inactive" for known non-responsive OSTs.
1186 Description: SPEC SFS validation failure on NFS v2 over lustre.
1187 Details : Changes the blocksize for regular files to be 2x RPC size,
1188 and not depend on stripe size.
1190 Severity : enhancement
1192 Description: Multiple MD RPCs in flight.
1193 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1194 To avoid overly-overloading MDS, introduce a limit on number of
1195 MDS RPCs in flight for a single client and add /proc controls
1196 to adjust this limit.
1198 Severity : enhancement
1200 Description: client read/write statistics
1201 Details : Add client read/write call usage stats for performance
1202 analysis of user processes.
1203 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1204 file access. extents_stats shows chunk size distribution.
1205 extents_stats_per_process show chunk size distribution per
1208 Severity : enhancement
1210 Description: mds statistics
1211 Details : Add detailed mds operations statistics in
1212 /proc/fs/lustre/mds/*/stats.
1216 Description: Failure of copying files with lustre special EAs.
1217 Details : Client side always return success for setxattr call for lustre
1218 special xattr (currently only "trusted.lov").
1223 Description: Refcount LNET uuids
1224 Details : The global LNET uuid list grew linearly with every startup;
1225 refcount repeated list entries instead of always adding to
1228 Severity : enhancement
1230 Description: Dynamic service threads
1231 Details : Within a small range, start extra service threads
1232 automatically when the request queue builds up.
1235 Frequency : mixed-endian client/server environments
1237 Description: mixed-endian crashes
1238 Details : The new msg_v2 system had some failures in mixed-endian
1241 Severity : enhancement
1243 Description: Easy OST removal
1244 Details : OSTs can be permanently deactivated with e.g. 'lctl
1245 conf_param lustre-OST0001.osc.active=0'
1247 Severity : enhancement
1249 Description: MGS proc entries
1250 Details : Added basic proc entries for the MGS showing what filesystems
1253 Severity : enhancement
1255 Description: provide MGS failover
1256 Details : Added config lock reacquisition after MGS server failover.
1258 Severity : enhancement
1260 Description: add Linux 2.4 support
1261 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1265 Description: a large application tries to do I/O to the same resource and dies
1266 in the middle of it.
1267 Details : Check the req->rq_arrival time after the call to
1268 ost_brw_lock_get(), but before we do anything about
1269 processing it & sending the BULK transfer request. This
1270 should help move old stale pending locks off the queue as
1271 quickly as obd_timeout.
1274 Frequency : when an incorrect nid is specified during startup
1276 Description: ptlrpc connect to non-existant node causes kernel crash
1277 Details : LNET can't be re-entered from an event callback, which
1278 happened when we expire a message after the export has been
1279 cleaned up. Instead, hand the zombie cleanup off to another
1282 Severity : enhancement
1284 Description: plain/inodebits lock performance improvement
1285 Details : Grouping plain/inodebits in granted list by their request modes
1286 and bits policy, thus improving the performance of search through
1290 Frequency : only if OST filesystem is corrupted
1292 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1293 Details : for a short time RPCs with bulk IO are in the replay list,
1294 but replay of bulk IOs is unimplemented. If the OST filesystem
1295 is corrupted due to disk cache incoherency and then replay is
1296 started it is possible to trip an assertion. Avoid putting
1297 committed RPCs into the replay list at all to avoid this issue.
1300 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1302 Description: System hang on startup
1303 Details : This bug allowed the liblustre (e.g. catamount) client to
1304 return to the app before handling all startup RPCs. This
1305 could leave the node unresponsive to lustre network traffic
1306 and manifested as a server ptllnd timeout.
1308 Severity : enhancement
1310 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1311 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1312 boolean to control whether to print peer debug info when a
1313 client's RPC times out.
1316 Frequency : only for kernels with patches from Lustre below 1.4.3
1318 Description: Remove old rdonly API
1319 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1322 Frequency : only for devices with external journals
1324 Description: Set external device read-only also
1325 Details : During a commanded failover stop, we set the disk device
1326 read-only while the server shuts down. We now also set any
1327 external journal device read-only at the same time.
1330 Frequency : when upgrading from 1.4 while trying to change parameters
1332 Description: The wrong (new) MDC name was used when setting parameters for
1333 upgraded MDT's. Also allows changing of OSC (and MDC)
1334 parameters if --writeconf is specified at tunefs upgrade time.
1337 Frequency : when setting specific ost indicies
1339 Description: QOS code breaks on skipped indicies
1340 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1341 created with --index need not be sequential.
1343 Severity : enhancement
1345 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1346 Details : The uninit_groups feature works in conjunction with the kernel
1347 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1348 up the pass1 processing of e2fsck. This is a read-only feature
1349 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1350 mount filesystems that have had this feature enabled.
1352 Severity : enhancement
1354 Description: Improve multi-block allocation algorithm to avoid fragmentation
1355 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1356 allocation locality and avoid filesystem fragmentation.
1358 ------------------------------------------------------------------------------
1360 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1362 * Support for kernels:
1363 2.4.21-47.0.1.EL (RHEL 3)
1364 2.6.5-7.283 (SLES 9)
1365 2.6.9-42.0.10.EL (RHEL 4)
1366 2.6.12.6 vanilla (kernel.org)
1367 2.6.16.27-0.9 (SLES 10)
1368 * Recommended e2fsprogs version: 1.39.cfs5
1370 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1374 Frequency : occasional, depends on client load and configuration
1375 Bugzilla : 12181, 12203
1376 Description: data loss for recently-modified files
1378 Details : In some cases it is possible that recently written or created
1379 files may not be written to disk in a timely manner (this should
1380 normally be within 30s unless client IO load is very high).
1381 The problem appears as zero-length files or files that are a
1382 multiple of 1MB in size after a client crash or client eviction
1383 that are missing data at the end of the file.
1385 This problem is more likely to be hit on clients where files are
1386 repeatedly created and unlinked in the same directory, clients
1387 have a large amount of RAM, have many CPUs, the filesystem has
1388 many OSTs, the clients are rebooted frequently, and/or the files
1389 are not accessed by other nodes after being written.
1391 The presence of the problem can be detected by looking at
1392 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1393 smaller than the second (nr_unused) then dirty files will not
1394 be flushed automatically to disk. "sync; sleep 10" should be
1395 run several times on the node before unmounting it to update
1396 Lustre (this is also safe to run on nodes without this problem).
1398 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1399 that can cause this same problem, so customers using that kernel
1400 also need to update the kernel in addition to Lustre. In order
1401 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1403 It is normal that files written just before a client crash (less
1404 than 30s) may not yet have been flushed to disk, even for local
1408 Frequency : frequent on thin XT3 nodes
1410 Description: UUID collision on thin XT3 Linux nodes
1411 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1412 randomly, since we relied on an insufficiently-seeded PRNG.
1417 Description: OSS hangs after "All ost request buffers busy"
1418 Details : A deadlock between quota and journal operations caused OSS
1419 hangs after printing "All ost request buffers busy."
1422 Frequency : always on liblustre builds
1424 Description: Cleanup compiler warnings on liblustre
1427 Frequency : always on liblustre builds on XT3
1429 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1433 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1434 > 32000 subdirectories
1435 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1436 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1437 subdirectory count crosses 32000. This will aid e2fsck to
1438 correctly handle more than 32000 subdirectories.
1441 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1443 Description: System hang on startup
1444 Details : This bug allowed the liblustre (e.g. catamount) client to
1445 return to the app before handling all startup RPCs. This
1446 could leave the node unresponsive to lustre network traffic
1447 and manifested as a server ptllnd timeout.
1449 Severity : enhancement
1451 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1452 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1453 boolean to control whether to print peer debug info when a
1454 client's RPC times out.
1459 Description: make O_SYNC working on 2.6 kernels
1460 Details : 2.6 kernels use different method for mark pages for write,
1461 so need add a code to lustre for O_SYNC work.
1466 Description: Failure to close file and release space on NFS
1467 Details : Put inode details into lock acquired in ll_intent_file_open.
1468 Use mdc_intent_lock in ll_intent_open to properly
1469 detect all kind of errors unhandled by mdc_enqueue.
1474 Description: proc file read during shutdown sometimes raced obd removal,
1476 Details : Add lock to prevent obd access after proc file removal.
1479 Frequency : Only for files larger than 4GB on 32-bit clients.
1481 Description: improperly doing page alignment of locks
1482 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1483 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1488 Description: RPCs being resent when they shouldn't be
1489 Details : Some RPCs that should not be resent are being resent. This
1490 can cause inconsistencies in the RPC state machine. Do not
1491 resend such requests.
1494 Frequency : rare, only with NFS export
1496 Description: Crash on NFS re-export node
1497 Details : under very unusual load conditions an assertion is hit in
1498 ll_intent_file_open()
1501 Frequency : only if OST filesystem is corrupted
1503 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1504 Details : for a short time RPCs with bulk IO are in the replay list,
1505 but replay of bulk IOs is unimplemented. If the OST filesystem
1506 is corrupted due to disk cache incoherency and then replay is
1507 started it is possible to trip an assertion. Avoid putting
1508 committed RPCs into the replay list at all to avoid this issue.
1513 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1514 Details : Large single O_DIRECT read and write calls can fail to allocate
1515 a sufficiently large buffer to process the request. In case of
1516 allocation failure the allocation is retried with a smaller
1517 buffer and broken into smaller requests.
1519 Severity : enhancement
1521 Description: Add -o localflock option to simulate old noflock behaviour.
1522 Details : This will achieve local-only flock/fcntl locks coherentness.
1527 Description: versioning check is incomplete
1528 Details : Checking the version difference of client vs. server, report
1529 error if the gap is too big.
1534 Description: add support PG_writeback bit
1535 Details : add support for PG_writeback bit for Lustre, for more carefull
1536 work with page cache in 2.6 kernel. This also fix some deadlocks
1537 and remove hack for work O_SYNC with 2.6 kernel.
1539 Severity : enhancement
1541 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1542 Details : The uninit_groups feature works in conjunction with the kernel
1543 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1544 up the pass1 processing of e2fsck. This is a read-only feature
1545 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1546 mount filesystems that have had this feature enabled.
1548 Severity : enhancement
1550 Description: Improve multi-block allocation algorithm to avoid fragmentation
1551 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1552 allocation locality and avoid filesystem fragmentation.
1554 ------------------------------------------------------------------------------
1556 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1558 * Support for kernels:
1559 2.6.9-42.0.3.EL (RHEL 4)
1560 2.6.5-7.276 (SLES 9)
1561 2.4.21-47.0.1.EL (RHEL 3)
1562 2.6.12.6 vanilla (kernel.org)
1563 2.6.16.21-0.8 (SLES10)
1564 * Recommended e2fsprogs version: 1.39.cfs2-0
1566 * The backwards-compatible /proc/sys/portals symlink has been removed
1567 in this release. Before upgrading, please ensure that you change
1568 any configuration scripts or /etc/sysctl.conf files that access
1569 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1570 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1571 in advance of the upgrade on any system running Lustre 1.4.6 or
1572 newer, since /proc/sys/lnet was added in that version.
1573 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1577 Frequency : only when quota is used
1579 Description: avoid scanning export list for quota master
1580 Details : Change the algorithms to avoid scanning export list in order
1581 to improve the efficiency.
1584 Frequency : MDS failover only, very rarely
1586 Description: "went back in time" messages on mds failover
1587 Details : The greatest transno may be lost when the current operation
1588 finishes with an error (transno==0) and the client's last_rcvd
1589 record is over-written. Save the greatest transno in the
1590 mds_last_transno for this case.
1593 Frequency : always for specific kernels and striping counts
1595 Description: client may get "Matching packet too big" without ACL support
1596 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1597 when trying to access files in certain configurations. The
1598 clients should in fact be denied when mounting because they do
1599 not understand ACLs.
1602 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1604 Description: many clients connecting with IO in progress causes connect timeouts
1605 Details : Avoid synchronous journal commits to avoid delays caused by many
1606 clients connecting/disconnecting when bulk IO is in progress.
1607 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1608 OST_IO_PORTAL to avoid delays behind potentially many pending
1612 Frequency : occasionally with multiple writers to a single file
1614 Description: shared writes to file may result in wrong size reported by stat()
1615 Details : Allow growing of kms when extent lock is cancelled
1618 Frequency : always with random mmap IO to multi-striped file
1620 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1621 Details : Only if the hole is at the end of OST object so that kms is too
1622 small. Fix is to increase kms accordingly in ll_nopage.
1625 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1627 Description: writes to a missing object would leak memory on the OST
1628 Details : If there is an inconsistency between the MDS and OST filesystems,
1629 such that the MDS references an object that doesn't exist, writes
1630 to that object will leak memory due to incorrect cleanup in the
1631 error handling path, eventually running out of memory on the OST.
1636 Description: Creating too long symlink causes lustre errors
1637 Details : Check symlink and name lengths before sending requests to MDS.
1640 Frequency : only if flock is enabled (not on by default)
1642 Description: posix locks not released on fd closure on 2.6.9+
1643 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1644 kernels, this caused such locks not to be released on fd close and
1645 then assertions on fs unmount about still used locks.
1648 Frequency : MDS failover only, very rarely
1650 Description: clients may get ASSERTION(granted_lock != NULL)
1651 Details : When request was taking a long time, and a client was resending
1652 a getattr by name lock request. The were multiple lock requests
1653 with the same client lock handle and
1654 mds_getattr_name->fixup_handle_for_resent_request found one of the
1655 lock handles but later failed with ASSERTION(granted_lock != NULL).
1660 Description: handle->h_buffer_credits > 0, assertion failure
1661 Details : h_buffer_credits is zero after truncate, causing assertion
1662 failure. This patch extends the transaction or creates a new
1666 Frequency : NFS re-export or patchless client
1667 Bugzilla : 11179, 10796
1668 Description: Crash on NFS re-export node (__d_move)
1669 Details : We do not want to hash the dentry if we don't have a lock.
1670 But if this dentry is later used in d_move, we'd hit uninitialised
1671 list head d_hash, so we just do this to init d_hash field but
1672 leave dentry unhashed.
1675 Frequency : NFS re-export or patchless client
1677 Description: NFS exports has problem with symbolic link
1678 Details : lustre client didn't properly install dentry when re-exported
1679 to NFS or running patchless client.
1682 Frequency : NFS re-export or patchless client
1684 Description: Various nfs/patchless fixes.
1685 Details : fixes reuse disconected alias for lookup process - this fixes
1686 warning "find_exported_dentry: npd != pd",
1687 fix permission error with open files at nfs.
1688 fix apply umask when do revalidate.
1691 Frequency : occasional
1693 Description: Crash on NFS re-export node
1694 Details : calling clear_page() on the wrong pointer triggered oops in
1695 generic_mapping_read().
1698 Frequency : rarely, using O_DIRECT IO
1700 Description: unaligned directio crashes client with LASSERT
1701 Details : check for unaligned buffers before trying any requests.
1704 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1706 Description: stale data returned from RAID cache
1707 Details : If only a small amount of IO is done to the RAID device before
1708 reading it again it is possible to get stale data from the RAID
1709 cache instead of reading it from disk.
1712 Frequency : always for sles10 kernel
1714 Description: sles10 support
1715 Details : ll_follow_link: compile fixes and using of nd_set_link
1716 under newer kernels.
1719 Frequency : depends on arch, kernel and compiler version, always on sles10
1722 Description: recursive or deep enough symlinks cause stack overflow
1723 Details : getting rid of large stack-allocated variable in
1727 Frequency : depends on hardware
1729 Description: lustre write performance loss in the SLES10 kernel
1730 Details : the performance loss is caused by using of write barriers in the
1731 ext3 code. The SLES10 kernel turns barrier support on by
1732 default. The fix is to undo that change for ldiskfs.
1734 ------------------------------------------------------------------------------
1736 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1738 * Support for kernels:
1739 2.6.9-42.0.3EL (RHEL 4)
1740 2.6.5-7.276 (SLES 9)
1741 2.4.21-47.0.1.EL (RHEL 3)
1742 2.6.12.6 vanilla (kernel.org)
1746 Frequency : quota enabled and large files being deleted
1748 Description: releasing more than 4GB of quota at once hangs OST
1749 Details : If a user deletes more than 4GB of files on a single OST it
1750 will cause the OST to spin in an infinite loop. Release
1751 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1756 Description: statfs data retrieved from /proc may be stale or zero
1757 Details : When reading per-device statfs data from /proc, in the
1758 {kbytes,files}_{total,free,avail} files, it may appear
1759 as zero or be out of date.
1762 Frequency : systems with MD RAID1 external journal devices
1764 Description: lconf's call to blkid is confused by RAID1 journal devices
1765 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1766 of returning all block devices that match the journal UUID.
1769 Frequency : always, for aggregate stripe size over 4GB
1771 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1772 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1773 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1776 Frequency : always if "lfs find" used on a local file/directory
1778 Description: "lfs find" segfaults if used on a local file/directory
1779 Details : The case where a directory component was not specified wasn't
1780 handled correctly. Handle this properly.
1783 Frequency : always on ppc64
1785 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1786 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1787 perform endianness conversion.
1790 Frequency : rarely (truncate to non-zero file size after write under load)
1791 Bugzilla : 10730, 10687
1792 Description: Files padded with zeros to next 4K multiple
1793 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1794 it is possible that files that are truncated to a non-zero size
1795 immediately after being written are filled with zero bytes beyond
1796 the truncated size. No file data is lost.
1798 Severity : enhancement
1800 Description: Allow recovery/failover for liblustre clients.
1801 Details : liblustre clients were unaware of failover configurations until
1804 Severity : enhancement
1806 Description: user file locks should fail when not mounting with flock option
1807 Details : Set up an error-returning stub in ll_file_operations.lock field
1808 to prevent incorrect behaviour when client is mounted without
1809 flock option. Also, set up properly f_op->flock field for
1813 Frequency : always on ia64
1815 Description: "lfs df" loops on printing out MDS statfs information
1816 Details : The obd_ioctl_data was not initialized and in some systems
1817 this caused a failure during the ioctl that did not return
1818 an error. Initialize the struct and return an error on failure.
1821 Frequency : SLES 9 only
1823 Description: Error of copying files with lustre special EAs as root
1824 Details : Client side always return success for setxattr call for lustre
1825 special xattr (currently only "trusted.lov").
1828 Frequency : rarely on clusters with both ia64+i386 clients
1830 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1831 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1832 clients are doing concurrent writes to the same file it is
1833 possible that the ia64 clients may LASSERT because the OST
1834 extent locks are not PAGE_SIZE aligned. Ensure that grown
1835 locks are always aligned on the request boundary.
1838 Frequency : specific use, occasional
1840 Description: Overwriting in use executable truncates on-disk binary image
1841 Details : If one node attempts to overwrite an executable in use by
1842 another node, we now correctly return ETXTBSY instead of
1843 truncating the file.
1845 Severity : enhancement
1847 Description: Async OSC create to avoid the blocking unnecessarily.
1848 Details : If a OST has no remain object, system will block on the creating
1849 when need to create a new object on this OST. Now, ways use
1850 pre-created objects when available, instead of blocking on an
1851 empty osc while others are not empty. If we must block, we block
1852 for the shortest possible period of time.
1857 Description: chmod on Lustre root is propagated to other clients
1858 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1859 invalid by the follow_mount time.
1862 Frequency : liblustre clients only
1864 Description: Race in 'instant cancel' lock handling could lead to such locks
1865 never to be granted in case of SMP MDS
1866 Details : Do not destroy not yet granted but cbpending locks in
1870 Frequency : replay/resend of open
1872 Description: non null lock assetion failure in mds_intent_policy
1873 Details : Trying to replay/resend lockless open requests resulted in
1874 mds_open() returning 0 with no lock. Now it sets a flag if
1875 a lock is going to be returned.
1877 Severity : enhancement
1879 Description: Checksum enhancements
1880 Details : New checksum enhancements allow for resending RPCs that failed
1883 Severity : enhancement
1885 Description: Tunables on number of dirty pages in cacche
1886 Details : Allow to set limit on number of dirty pages cached.
1891 Description: client crash on unmount - lock still has references
1892 Details : In some error handling cases it was possible to leak a lock
1893 reference on a client while accessing a file. This was not
1894 harmful to the client during operation, but would cause the
1895 client to crash when the filesystem is unmounted.
1898 Frequency : specific case, rare
1900 Description: ETXTBSY on mds though file not in use by client
1901 Details : ETXTBSY is no longer incorrectly returned when attempting to
1902 chmod or chown a directory that the user previously tried to
1903 execute or a currently-executing binary.
1906 Frequency : extremely rare except on liblustre-based clients
1908 Description: Lustre space not freed when files are deleted
1909 Details : Clean up open-unlinked files after client eviction. Previously
1910 the unlink was skipped and the files remained as orphans.
1915 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1916 Details : In some cases it was possible to send a blocking callback to
1917 a client doing a glimpse, even though that client didn't get
1918 a lock granted. When the glimpse lock is cancelled on the OST
1919 the freed lock is left on the waiting list and corrupted the list.
1922 Frequency : all core dumps
1924 Description: Broke core dumps to lustre
1925 Details : Negative dentry may be unhashed if parent does not have UPDATE
1926 lock, but some callers, e.g. do_coredump, expect dentry to be
1927 hashed after successful create, hash it in ll_create_it.
1929 ------------------------------------------------------------------------------
1931 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1933 * Support for kernels:
1934 2.6.9-42.0.2.EL (RHEL 4)
1935 2.6.5-7.276 (SLES 9)
1936 2.4.21-40.EL (RHEL 3)
1937 2.6.12.6 vanilla (kernel.org)
1941 Frequency : always on RHEL 3
1943 Description: Number of open files grows over time
1944 Details : The number of open files grows over time, whether or not
1945 Lustre is started. This was due to a filp leak introduced
1946 by one of our kernel patches.
1948 ------------------------------------------------------------------------------
1950 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1952 * Support for kernels:
1953 2.6.9-42.EL (RHEL 4)
1954 2.6.5-7.276 (SLES 9)
1955 2.4.21-40.EL (RHEL 3)
1956 2.6.12.6 vanilla (kernel.org)
1961 Bugzilla : 5719, 9635, 9792, 9684
1962 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1963 Details : If a server is under heavy load and cannot reply to new
1964 connection requests before the client resends the (re)connect,
1965 the connection handling code can behave badly if two service
1966 threads are concurrently handing separate (re)connections from
1967 the same client. Add better locking to the connection handling
1968 code, and ensure that only a single connection will be processed
1969 for a given client UUID, even if the lock is dropped.
1971 Severity : enhancement
1973 Description: add TCP zero-copy support to kernel
1974 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1975 sends if the hardware supports scatter-gather and checksumming.
1976 This allows socklnd to do client-write and server-read more
1977 efficiently and reduce CPU utilization from skbuf copying.
1980 Frequency : only if NFS exporting from client
1982 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1983 Details : If setting attributes on a file created under NFS that had
1984 never been opened it would be possible to oops the client
1985 if the file had no objects.
1988 Frequency : always for liblustre
1990 Description: liblustre client does MDS+OSTs setattr RPC for each write
1991 Details : When doing a write from a liblustre client, the client
1992 incorrectly issued an RPC to the MDS and each OST the file was
1993 striped over in order to update the timestamps. When writing
1994 with small chunks and many clients this could overwhelm the MDS
1995 with RPCs. In all cases it would slow down the write because
1996 these RPCs are unnecessary.
1998 Severity : enhancement
2000 Description: allow number of MDS service threads to be changed at module load
2001 Details : It is now possible to change the number of MDS service threads
2002 running. Adding "options mds mds_num_threads={N}" to the MDS's
2003 /etc/modprobe.conf will set the number of threads for the next
2004 time Lustre is restarted (assuming the "mds" module is also
2005 reloaded at that time). The default number of threads will
2006 stay the same, 32 for most systems.
2011 Description: OST crash if filesystem is unformatted or corrupt
2012 Details : If an OST is started on a device that has never been formatted
2013 or if the filesystem is corrupt and cannot even mount then the
2014 error handling cleanup routines would dereference a NULL pointer.
2019 Description: NULL pointer deref in llap_from_page.
2020 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2021 incorrect) mapping if the page was truncated/reclaimed while it was
2022 searched for. Check for this condition and skip such pages when
2023 doing readahead. Introduce extra check to llap_from_page() to
2024 verify page->mapping->host is non-NULL (so page is not anonymous).
2027 Frequency : Sometimes when using sys_sendfile
2029 Description: "page not covered by a lock" warnings from ll_readpage
2030 Details : sendfile called ll_readpage without right page locks present.
2031 Now we introduced ll_file_sendfile that does necessary locking
2032 around call to generic_file_sendfile() much like we do in
2036 Frequency : with certain MDS communication failures at client mount time
2038 Description: NULL pointer deref after failed client mount
2039 Details : a client connection request may delayed by the network layer
2040 and not be sent until after the PTLRPC layer has timed out the
2041 request. If the client fails the mount immediately it will try
2042 to clean up before the network times out the request. Add a
2043 reference from the request import to the obd device and delay
2044 the cleanup until the network drops the request.
2047 Frequency : occasionally during client (re)connect
2049 Description: assertion failure during client (re)connect
2050 Details : processing a client connection request may be delayed by the
2051 client or server longer than the client connect timeout. This
2052 causes the client to resend the connection request. If the
2053 original connection request is replied in this interval, the
2054 client may trip an assertion failure in ptlrpc_connect_interpret()
2055 which thought it would be the only running connect process.
2058 Frequency : only with obd_echo servers and clients that are rebooted
2060 Description: kernel BUG accessing uninitialized data structure
2061 Details : When running an obd_echo server it did not start the ping_evictor
2062 thread, and when a client was evicted an uninitialized data
2063 structure was accessed. Start the ping_evictor in the RPC
2064 service startup instead of the OBD startup.
2066 Severity : enhancement
2067 Bugzilla : 10193 (patchless)
2068 Description: Remove dependency on various unexported kernel interfaces.
2069 Details : No longer need reparent_to_init, exit_mm, exit_files,
2070 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2073 Frequency : rare (only users of deprecated and unsupported LDAP config)
2075 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2076 Details : LDAP apparently contains 'lustreName' attributes instead of
2077 'name'. A simple remapping of the name is sufficient.
2080 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2082 Description: waiting_locks_callback trips kernel BUG if client is evicted
2083 Details : Running with the dump_on_timeout debug flag turned on makes
2084 it possible that the waiting_locks_callback() can try to dump
2085 the Lustre kernel debug logs from an interrupt handler. Defer
2086 this log dumping to the expired_lock_main() thread.
2088 Severity : enhancement
2090 Description: Support NFS exporting on 2.6 kernels.
2091 Details : Implement non-rawops metadata methods for NFS server to use without
2092 changing NFS server code.
2095 Frequency : very rare (synthetic metadata workload only)
2097 Description: two racing renames might cause an MDS thread to deadlock
2098 Details : Running the "racer" program may cause one MDS thread to rename
2099 a file from being the source of a rename to being the target of
2100 a rename at exactly the same time that another thread is doing
2101 so, and the second thread has already enqueued these locks after
2102 doing a lookup of the target and is trying to relock them in
2103 order. Ensure that we don't try to re-lock the same resource.
2106 Frequency : only very large systems with liblustre clients
2108 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2109 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2110 all OSTs in parallel. This allows the eviction of stale liblustre
2111 clients to proceed much faster than if they were done in series,
2112 and also offers similar improvements for other set_info RPCs.
2117 Description: excessive CPU usage during initial read phase on client
2118 Details : During the initial read phase on a client, it would agressively
2119 retry readahead on the file, consuming too much CPU and impacting
2120 performance (since 1.4.5.8). Improve the readahead algorithm
2121 to avoid this, and also improve some other common cases (read
2122 of small files in particular, where "small" is files smaller than
2123 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2128 Description: MDS crash when receiving packet with unknown intent.
2129 Details : Do not LBUG in unknown intent case, just return -EFAULT
2131 Severity : enhancement
2132 Bugzilla : 9293, 9385
2133 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2134 Details : Do not serialize getattr (non-intent version) and statfs.
2137 Frequency : occasional, when OST network is overloaded/intermittent
2139 Description: client evicted by OST after bulk IO timeout
2140 Details : If a client sends a bulk IO request (read or write) the OST
2141 may evict the client if it is unresposive to its data GET/PUT
2142 request. This is incorrect if the network is overloaded (takes
2143 too long to transfer the RPC data) or dropped the OST GET/PUT
2144 request. There is no need to evict the client at all, since
2145 the pinger and/or lock callbacks will handle this, and the
2146 client can restart the bulk request.
2149 Frequency : Always when mmapping file with no objects
2151 Description: client crashes when mmapping file with no objects
2152 Details : Check that we actually have objects in a file before doing any
2153 operations on objects in ll_vm_open, ll_vm_close and
2159 Description: Request leak when working with deleted CWD
2160 Details : Introduce advanced request refcount tracking for requests
2161 referenced from lustre intent.
2163 Severity : Enhancement
2165 Description: Cache open file handles on client.
2166 Details : MDS now will return special lock along with openhandle, if
2167 requested and client is allowed to hold openhandle, even if unused,
2168 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2169 closing files for every read/write openration.
2171 Severity : Enhancement
2173 Description: Cache open negative dentries on client when possible.
2174 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2175 negative dentries on lock revocation.
2180 Description: Remounting a client read-only wasn't possible with a zconf mount
2181 Details : It wasn't possible to remount a client read-only with llmount.
2183 Severity : enhancement
2184 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2185 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2186 MPICH is included to add Lustre-specific ADIO interfaces.
2187 This is based closely on the UFS ADIO layer and only differs
2188 in file creation, in order to allow the OST striping to be set.
2189 This is user-contributed code and not supported by CFS.
2194 Description: extended inode attributes (immutable, append-only) work improperly
2195 when 2.4 and 2.6 kernels are used on client/server or vice versa
2196 Details : Introduce kernel-independent values for these flags.
2198 Severity : enhancement
2201 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2202 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2203 granularity. Now, Fractional MB granularity is be supported,
2204 this is very useful for low memory system.
2206 Severity : enhancement
2208 Description: Getattr by fid
2209 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2210 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2211 have locks and avoids getattr rpc requests for it on every stat.
2214 Frequency : Always, for filesystems larger than 2TB
2216 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2217 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2218 filesystems larger than 2TB to be mounted with mballoc enabled.
2221 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2223 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2224 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2225 filesystems larger than 2TB is mounted then the kernel will
2226 silently corrupt the start of the filesystem. CONFIG_LBD is
2227 enabled for all CFS-supported kernels, but the possibility of
2228 this happening with a modified kernel config exists.
2230 Severity : enhancement
2232 Description: add client O_DIRECT support for 2.6 kernels
2233 Details : It is now possible to do O_DIRECT reads and writes to files
2234 in the Lustre client mountpoint on 2.6 kernel clients.
2236 Severity : enhancement
2238 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2239 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2240 parallel, not waiting for response from every OST before sending
2241 a rpc to the next OST.
2246 Description: setattr vs write race when updating file timestamps
2247 Details : Client processes that update a file timestamp into the past
2248 right after writing to the file (e.g. tar) it is possible that
2249 the updated file modification time can be reset to the current
2250 time due to a race between processing the setattr and write RPC.
2252 Severity : enhancement
2254 Description: Bring 'lfs find' closer in line with regular Linux find.
2255 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2256 -print0 options and obtains all the needed info through the lustre
2259 Severity : enhancement
2261 Description: support up to 1024 configured devices on one node
2262 Details : change obd_dev array from statically allocated to dynamically
2263 allocated structs as they are first used to reduce memory usage
2268 Description: Flush dirty partially truncated pages during truncate
2269 Details : Immediatelly flush partially truncated pages in filter_setattr,
2270 this way we completely avoid having any pages in page cache on OST
2271 and can retire ugly workarounds during writes to flush such pages.
2276 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2277 Details : protect inode from truncation within vfs_unlink() context
2278 just take a reference before calling vfs_unlink() and release it
2279 when parent's i_sem is free.
2284 Description: last_id value checked outside lock on OST caused LASSERT failure
2285 Details : If there were multiple MDS->OST object precreate requests in
2286 flight, it was possible that the OST's last object id was checked
2287 outside a lock and incorrectly tripped an assertion. Move checks
2288 inside locks, and discard old precreate requests.
2291 Frequency : always, if extents are used on OSTs
2293 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2294 Details : OSTs using the extents format would not zero the high 16 bits of
2295 the index physical block number. This is not a problem for any
2296 OST filesystems smaller than 16TB, and no kernels support ext3
2297 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2298 new/modified files) and can be fixed for existing filesystems
2299 with e2fsprogs-1.39-cfs1.
2304 Description: import connection selection may be incorrect if timer wraps
2305 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2306 import connections to be ignored if the 32-bit jiffies counter
2307 wraps. Use a 64-bit jiffies counter.
2310 Frequency : very large clusters immediately after boot
2312 Description: LNET request buffers exhausted under heavy short-term load
2313 Details : If a large number of client requests are generated on a service
2314 that has previously never seen so many requests it is possible
2315 that the request buffer growth cannot keep up with the spike in
2316 demand. Instead of dropping incoming requests, they are held in
2317 the LND until the RPC service can accept more requests.
2320 Frequency : Sometimes during replay
2322 Description: Assertion failure in ll_local_open after replay.
2323 Details : If replay happened on an open request reply before we were able
2324 to set replay handler, reply will become not swabbed tripping the
2325 assertion in ll_local_open. Now we set the handler right after
2326 recognising of open request
2329 Frequency : very rare
2331 Description: kernel reports "badness in vsnprintf"
2332 Details : Reading from the "recovery_status" /proc file in small chunks
2333 may cause a negative length in lprocfs_obd_rd_recovery_status()
2334 call to vsnprintf() (which is otherwise harmless). Exit early
2335 if there is no more space in the output buffer.
2337 Severity : enhancement
2339 Description: clear OBD RPC statistics by writing to them
2340 Details : It is now possible to clear the OBD RPC statistics by writing
2341 to the "stats" file.
2346 Description: Client mtime is not the same on different clients after utimes
2347 Details : In some cases, the client was using the utimes() syscall on
2348 a file cached on another node. The clients now validate the
2349 ctime from the MDS + OSTs to determine which one is right.
2354 Description: Inability to activate failout mode
2355 Details : lconf script incorrectly assumed that in pythong string's numeric
2356 value is used in comparisons.
2359 Frequency : always with multiple stripes per file
2361 Description: Inefficient object allocation for mutli-stripe files
2362 Details : When selecting which OSTs to stripe files over, for files with
2363 a stripe count that divides evenly into the number of OSTs,
2364 the MDS is always picking the same starting OST for each file.
2365 Return the OST selection heuristic to the original design.
2370 Description: mount failures may take full timeout to return an error
2371 Details : Under some heavy load conditions it is possible that a
2372 failed mount can wait for the full obd_timeout interval,
2373 possibly several minutes, before reporting an error.
2374 Instead return an error as soon as the status is known.
2376 Frequency : quota enabled and large files being deleted
2378 Description: releasing more than 4GB of quota at once hangs OST
2379 Details : If a user deletes more than 4GB of files on a single OST it
2380 will cause the OST to spin in an infinite loop. Release
2381 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2386 Description: statfs data retrieved from /proc may be stale or zero
2387 Details : When reading per-device statfs data from /proc, in the
2388 {kbytes,files}_{total,free,avail} files, it may appear
2389 as zero or be out of date.
2392 Frequency : systems with MD RAID1 external journal devices
2394 Description: lconf's call to blkid is confused by RAID1 journal devices
2395 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2396 of returning all block devices that match the journal UUID.
2399 Frequency : always, for aggregate stripe size over 4GB
2401 Description: assertion fails when trying to use 4GB stripe size
2402 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2403 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2406 Frequency : always on ppc64
2408 Description: the first write on an ext3 filesystem with mballoc got stuck
2409 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2410 perform endianness conversion.
2412 ------------------------------------------------------------------------------
2414 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2416 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2417 INTEROPERATE with older versions automatically. Please read the
2418 user documentation before upgrading any part of a live system.
2419 * WARNING: Lustre networking configuration changes are required with
2420 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2423 * Support for kernels:
2424 2.6.9-22.0.2.EL (RHEL 4)
2425 2.6.5-7.244 (SLES 9)
2426 2.6.12.6 vanilla (kernel.org)
2429 Severity : enhancement
2430 Bugzilla : 7981/8208
2431 Description: Introduced Lustre Networking (LNET)
2432 Details : LNET is new networking infrastructure for Lustre, it includes
2433 a reorganized network configuration mode (see the user
2434 documentation for full details) as well as support for routing
2435 between different network fabrics. Lustre Networking Devices
2436 (LNDs) for the supported network fabrics have also been
2437 created for this new infrastructure.
2439 Severity : enhancement
2440 Description: Introduced Access control lists
2441 Details : clients can set ACLs on files and directories in order to have
2442 more fine-grained permissions than the standard Unix UGO+RWX.
2443 The MDS must be started with the "-o acl" mount option.
2445 Severity : enhancement
2446 Description: Introduced filesystem quotas
2447 Details : Administrators may now establish per-user quotas on the
2450 Severity : enhancement
2452 Description: Configuration change for the XT3
2453 The PTLLND is now used to run Lustre over Portals on the XT3
2454 The configure option(s) --with-cray-portals are no longer used.
2455 Rather --with-portals=<path-to-portals-includes> is used to
2456 enable building on the XT3. In addition to enable XT3 specific
2457 features the option --enable-cray-xt3 must be used.
2462 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2463 Details : A race between allocating a new llcd and re-getting the llcd_lock
2464 allowed another thread to grab newly-allocated llcd.
2466 Severity : enhancement
2468 Description: 2.6 OST async journal commit and locking fix to improve performance
2469 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2470 2.6 kernels are now async as they already were in 2.4 kernels so
2471 that they can commit concurrently with the network bulk transfer.
2472 For block-allocated files the filter allocation semaphore is held
2473 to avoid filesystem fragmentation during allocation. BKL lock
2474 removed for 2.6 xattr operations where it is no longer needed.
2479 Description: lconf incorrectly determined whether two IP networks could talk
2480 Details : In some more complicated routing and multiple-network
2481 configurations, lconf will avoid trying to make a network
2482 connection to a disjoint part of the IP space. It was doing the
2483 math incorrectly for one set of cases.
2488 Description: Fix for potential infinite loop processing records in an llog.
2489 Details : If an llog record is corrupted/zeroed, it is possible to loop
2490 forever in llog_process(). Validate the llog record length
2491 and skip the remainder of the block on error.
2494 Frequency : occasional (liblustre only)
2496 Description: liblustre could not open files whose last component is a symlink
2497 Details : sysio_path_walk() would incorrectly pass the open intent to
2498 intermediate path components.
2501 Frequency : rare (liblustre only with non-standard tuning)
2502 Bugzilla : 7201 (7350)
2503 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2504 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2505 after it has been released, only do this for referenced locks
2507 Severity : enhancement
2509 Description: specifying an (invalid) directory default stripe_size of -1
2510 would reset the directory default striping
2511 Details : stripe_size -1 was used internally to signal directory stripe
2512 removal, now use "all default" to signal dir stripe removal
2513 as a directory striping of "all default" is not useful
2516 Frequency : common for large clusters running liblustre clients
2518 Description: doing an ls when liblustre clients are running is slow
2519 Details : sending a glimpse AST to a liblustre client waits for every AST
2520 to time out, as liblustre clients will not respond. Since they
2521 cannot cache data we refresh the OST lock LVB from disk instead.
2523 Severity : enhancement
2525 Description: doing an ls at the same time as file IO can be slow
2526 Details : enqueue and other "small" requests can be blocked behind many
2527 large IO requests. Create a new OST IO portal for non-IO
2528 requests so they can be processed faster.
2531 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2533 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2534 created file with mode 000
2535 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2536 allow an ftruncate() as a non-root user to a file with mode 000.
2537 Lustre now respects this flag to disable mode checks when
2538 truncating a file owned by the user
2541 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2544 Description: Revoking locks from clients that went dead or catatonic might take
2546 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2547 cancellation of such locks instant on servers without waiting for
2548 any reply from clients. Clients drops these locks when cancel
2549 notification from server is received without replying.
2552 Frequency : liblustre-only, when liblustre client dies or becomes busy
2554 Description: Doing ls on Linux clients can take a long time with active
2556 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2557 granting such locks to it in the first place if possible. Locks
2558 are taken by proxy on the OST during the read or write and
2559 dropped immediately afterward. Add connect flags handling, do
2560 not grant locks to liblustre clients for glimpse ASTs.
2562 Severity : enhancement
2564 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2565 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2566 sequential reads and behaves badly with applications doing random
2567 reads. Improve it by reading ahead at least read region, and
2568 avoiding excessive large RPC for small reads.
2570 Severity : enhancement
2572 Description: Creating more than 1000 files for a single job may cause a load
2573 imbalance on the OSTs if there are also a large number of OSTs.
2574 Details : qos_prep_create() uses an OST index reseed value that is an
2575 even multiple of the number of available OSTs so that if the
2576 reseed happens in the middle of the object allocation it will
2577 still utilize the OSTs as uniformly as possible.
2582 Description: OST or MDS may oops in ping_evictor_main()
2583 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2584 but doesn't restart at beginning of obd_exports_timed list
2587 Severity : enhancement
2589 Description: improve by-nid export eviction on the MDS and OST
2590 Details : allow multiple exports with the same NID to be evicted at one
2591 time without re-searching the exports list.
2594 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2596 Description: MDS may oops in groups_free()
2597 Details : in rare race conditions a newly allocated group_info struct is
2598 freed again, and this can be NULL. The 2.4 compatibility code
2599 for groups_free() checked for a NULL pointer, but 2.6 did not.
2602 Frequency : common for liblustre clients doing little filesystem IO
2603 Bugzilla : 9352, 7313
2604 Description: server may evict liblustre clients accessing contended locks
2605 Details : if a client is granted a lock or receives a completion AST
2606 with a blocking AST already set it would not reply to the AST
2607 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2610 Frequency : lfs setstripe, only systems with more than 160 OSTs
2612 Description: unable to set striping with a starting offset beyond OST 160
2613 Details : llapi_create_file() incorrectly limited the starting stripe
2614 index to the maximum single-file stripe count.
2617 Frequency : LDAP users only
2619 Description: lconf did not handle in-kernel recovery with LDAP properly
2620 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2622 Severity : enhancement
2624 Description: bind OST threads to NUMA nodes to improve performance
2625 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2626 node and do their allocations there to localize memory access
2628 Severity : enhancement
2630 Description: llmount can determine client NID directly from Myrinet (GM)
2631 Details : the client NID code from gmnalnid was moved directly into
2632 llmount, removing the need to use this or specifying the
2633 client NID explicitly when mounting GM clients with zeroconf
2636 Frequency : if client is started with down MDS
2638 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2639 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2640 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2641 ptlrpc_import_delayed_req() because we didn't check if the
2642 request was interrupted, and we also didn't break out of the
2643 event loop if there was no timeout
2648 Description: data loss during non-page-aligned writes to a single file from
2649 both multiple nodes and multiple threads on one node at same time
2650 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2651 inconsistency led to false short-reads, that were cached and later
2652 used by ->prepare_write() to fill in partially written page,
2653 leading to data loss.
2656 Frequency : always, if lconf --abort_recovery used
2658 Description: lconf --abort_recovery fails with 'Operation not supported'
2659 Details : lconf was attempting to abort recovery on the MDT device and not
2662 Severity : enhancement
2664 Description: remove cleanup logs
2665 Details : replace lconf-generated cleanup logs with lustre internal
2666 cleanup routines. Eliminates the need for client-cleanup and
2669 Severity : enhancement
2671 Description: add support for EAs (user and system) on lustre filesystems
2672 Details : it is now possible to store extended attributes in the Lustre
2673 client filesystem, and with the user_xattr mount option it
2674 is possible to allow users to store EAs on their files also
2676 Severity : enhancement
2678 Description: Add possibility (config option) to show minimal available OST free
2680 Details : When compiled with --enable-mindf configure option, statfs(2)
2681 (and so, df) will return least minimal free space available from
2682 all OSTs as amount of free space on FS, instead of summary of
2683 free spaces of all OSTs.
2685 Severity : enhancement
2687 Description: do not expand extent locks acquired on OST-side
2688 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2689 by server: they are not cached anyway.
2692 Frequency : when mmap is used/binaries executed from Lustre
2694 Description: Unmmap pages before throwing them away from read cache.
2695 Details : llap_shrink cache now attempts to unmap pages before discarding
2696 them (if unmapping failed - do not discard). SLES9 kernel has
2697 extra checks that trigger if this unmapping is not done first.
2702 Description: lconf didn't resolve symlinks before checking to see whether a
2703 given mountpoint was already in use
2706 Frequency : when migrating failover services
2707 Bugzilla : 6395, 9514
2708 Description: When migrating a subset of services from a node (e.g. failback
2709 from a failover service node) the remaining services would
2710 time out and evict clients.
2711 Details : lconf --force (implied by --failover) sets the global obd_timeout
2712 to 5 seconds in order to quickly disconnect, but this caused
2713 other RPCs to time out too quickly. Do not change the global
2714 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2716 Severity : enhancement
2717 Frequency : if MDS is started with down OST
2718 Bugzilla : 9439,5706
2719 Description: Allow startup/shutdown of an MDS without depending on the
2720 availability of the OSTs.
2721 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2722 Add appropriate locking and lov-osc refcounts for safe
2723 cleaning. Add osc abort_inflight calls in case the
2724 synchronize never started.
2727 Frequency : occasional (Cray XT3 only)
2729 Description: root not authorized to access files in CRAY_PORTALS environment
2730 Details : The client process capabilities were not honoured on the MDS in
2731 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2732 been accessed by an authorized user then root was able to access
2733 the file on the local client also. The root user capabilities
2734 are now allowed on the MDS, as this environment has secure UID.
2737 Frequency : occasional
2739 Description: ldiskfs "too long searching" message happens too often
2740 Details : A debugging message (otherwise harmless) prints too often on
2741 the OST console. This has been reduced to only happen when
2742 there are fragmentation problems on the filesystem.
2747 Description: Division by zero in statfs when all OSCs are inactive
2748 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2749 lov_statfs divides by value returned by lov_get_stripecnt().
2753 Bugzilla : 9489, 3273
2754 Description: First write from each client to each OST was only 4kB in size,
2755 to initialize client writeback cache, which caused sub-optimal
2756 RPCs and poor layout on disk for the first writen file.
2757 Details : Clients now request an initial cache grant at (re)connect time
2758 and so that they can start streaming writes to the cache right
2759 away and always do full-sized RPCs if there is enough data.
2760 If the OST is rebooted the client also re-establishes its grant
2761 so that client cached writes will be honoured under the grant.
2766 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2767 Details : Now I/O RPCs go to different portal number and (presumably) fast
2768 lock requests (and glimses) and other RPCs get their own service
2769 threads pool that should be able to service those RPCs
2772 Severity : enhancement
2774 Description: Ability to exchange lustre version between client and servers and
2775 issue warnings at client side if client is too old. Also for
2776 liblustre clients there is ability to refuse connection of too old
2778 Details : New 'version' field is added to connect data structure that is
2779 filled with version info. That info is later checked by server and
2783 Frequency : rare, liblustre only.
2784 Bugzilla : 9296, 9581
2785 Description: Two simultaneous writes from liblustre at offset within same page
2786 might proceed at the same time overwriting eachother with stale
2788 Details : I/O lock withing llu_file_prwv was released too early, before data
2789 actually was hitting the wire. Extended lock-holding time until
2790 server acknowledges receiving data.
2793 Frequency : extremely rare. Never observed in practice.
2795 Description: avoid generating lustre_handle cookie of 0.
2796 Details : class_handle_hash() generates handle cookies by incrementing
2797 global counter, and can hit 0 occasionaly (this is unlikely, but
2798 not impossible, because initial value of cookie counter is
2799 selected randonly). Value of 0 is used as a sentinel meaning
2800 "unassigned handle" --- avoid it. Also coalesce two critical
2801 sections in this function into one.
2803 Severity : enhancement
2805 Description: allow liblustre clients to delegate truncate locking to OST
2806 Details : To avoid overhead of locking, liblustre client instructs OST to
2807 take extent lock in ost_punch() on client's behalf. New connection
2808 flag is added to handle backward compatibility.
2810 Severity : enhancement
2811 Bugzilla : 4928, 7341, 9758
2812 Description: allow number of OST service threads to be specified
2813 Details : a module parameter allows the number of OST service threads
2814 to be specified via "options ost ost_num_threads={N}" in the
2815 OSS's /etc/modules.conf or /etc/modprobe.conf.
2819 Bugzilla : 6146, 9635, 9895
2820 Description: servers crash with bad pointer in target_handle_connect()
2821 Details : In rare cases when a client is reconnecting it was possible that
2822 the connection request was the last reference for that export.
2823 We would temporarily drop the export reference and get a new
2824 one, but this may have been the last reference and the export
2825 was just destroyed. Get new reference before dropping old one.
2827 Severity : enhancement
2828 Frequency : if client is started with failover MDS
2830 Description: Allow multiple MDS hostnames in the mount command
2831 Details : Try to read the configuration from all specified MDS
2832 hostnames during a client mount in case the "primary"
2835 Severity : enhancement
2837 Description: Stop sending data to evicted clients as soon as possible.
2838 Details : Check if the client we are about to send or are sending data to
2839 was evicted already. (Check is done every second of waiting,
2840 for which l_wait_event interface was extended to allow checking
2841 of exit condition at specified intervals).
2844 Frequency : rare, normally only when NFS exporting is done from client
2846 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2848 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2851 Frequency : always, if srand() is called before liblustre initialization
2853 Description: Liblustre uses system PRNG disturbing its usage by user application
2854 Details : Introduce internal to lustre fast and high-quality PRNG for
2855 lustre usage and make liblustre and some other places in generic
2856 lustre code to use it.
2858 Severity : enhancement
2859 Bugzilla : 9477, 9557, 9870
2860 Description: Verify that the MDS configuration logs are updated when xml is
2861 Details : Check if the .xml configuration logs are newer than the config
2862 logs stored on the MDS and report an error if this is the case.
2863 Request --write-conf, or allow starting with --old_conf.
2865 Severity : enhancement
2867 Description: Handle symlinks in the path when checking if Lustre is mounted.
2868 Details : Resolve intermediate symlinks when checking if a client has
2869 mounted a filesystem to avoid duplicate client mounts.
2874 Description: lconf can hit an error exception but still return success.
2875 Details : The lconf command catches the Command error exception at the top
2876 level script context and will exit with the associated exit
2877 status, but doesn't ensure that this exit status is non-zero.
2882 Description: failure of ptlrpc thread startup can cause oops
2883 Details : Starting a ptlrpc service thread can fail if there are a large
2884 number of threads or the server memory is very fragmented.
2885 Handle this without oopsing.
2888 Frequency : always, only if liblustre and non-default acceptor port was used
2890 Description: liblustre cannot connect to servers with non-default acceptor port
2891 Details : tcpnal_set_default_params() was not called and was therefore
2892 ignoring the environment varaible TCPNAL_PORT, as well as other
2893 TCPNAL_ environment variables
2898 Description: two objects could be created on the same OST for a single file
2899 Details : If an OST is down, in some cases it was possible to create two
2900 objects on a single OST for a single file. No problems other
2901 than potential performance impact and spurious error messages.
2905 Bugzilla : 5681, 9562
2906 Description: Client may oops in ll_unhash_aliases
2907 Details : Client dcache may become inconsistent in race condition.
2908 In some cases "getcwd" can fail if the current directory is
2914 Description: Inode refcounting problems in NFS export code
2915 Details : link_raw functions used to call d_instantiate without obtaining
2916 extra inode reference first.
2920 Bugzilla : 9942, 9903
2921 Description: Referencing freed requests leading to crash, memleaks with NFS.
2922 Details : We used to require that call to ll_revalidate_it was always
2923 followed by ll_lookup_it. Also with revalidate_special() it is
2924 possible to call ll_revalidate_it() twice for the same dentry
2925 even if first occurence returned success. This fix changes semantic
2926 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2927 reference on a request referred from the intent.
2928 ll_intent_release() then releases such a request.
2931 Frequency : rare, normally benchmark loads only
2933 Description: unlinked inodes were kept in memory on the client
2934 Details : If a client is repeatedly creating and unlinking files it
2935 can accumulate a lot of stale inodes in the inode slab cache.
2936 If there is no other client load running this can cause the
2937 client node to run out of memory. Instead flush old inodes
2938 from client cache that have the same inode number as a new inode.
2941 Frequency : SLES9 2.6.5 kernel and long filenames only
2942 Bugzilla : 9969, 10379
2943 Description: utime reports stale NFS file handle
2944 Details : SLES9 uses out-of-dentry names in some cases, which confused
2945 the lustre dentry revalidation. Change it to always use the
2949 Frequency : rare, unless heavy write-truncate concurrency is continuous
2950 Bugzilla : 4180, 6984, 7171, 9963, 9331
2951 Description: OST becomes very slow and/or deadlocked during object unlink
2952 Details : filter_destroy() was holding onto the parent directory lock
2953 while truncating+unlinking objects. For very large objects this
2954 may block other threads for a long time and slow overall OST
2955 responsiveness. It may also be possible to get a lock ordering
2956 deadlock in this case, or run out of journal credits because of
2957 the combined truncate+unlink. Solution is to do object truncate
2958 first in one transaction without parent lock, and then do the
2959 final unlink in a new transaction with the parent lock. This
2960 reduces the lock hold time dramatically.
2963 Frequency : rare, 2.4 kernels only
2965 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2966 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2967 before unmounting their filesystem in order to avoid deadlock.
2968 The kernel_locked() function in 2.4 kernels only checks whether
2969 the kernel lock is held, not whether it is this process that is
2970 holding it as 2.6 kernels do.
2975 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2976 Details : The client ptlrpc code may be trying to reconnect to a down
2977 server before a previous connection attempt has timed out.
2978 Increase the reconnect interval to be longer than the connection
2979 timeout interval to avoid sending duplicate connections to
2983 Frequency : echo_client brw_test command
2985 Description: fix echo_client to work with OST preallocated code
2986 Details : OST preallocation code (5137) didn't take echo_client IO path
2987 into account: echo_client calls filter methods outside of any
2988 OST thread and, hence, there is no per-thread preallocated
2989 pages and buffers to use. Solution: hijack pga pages for IO. As
2990 a byproduct, this avoids unnecessary data copying.
2994 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2995 Description: Client can oops in mdc_commit_close() after open replay
2996 Details : It was possible for the MDS to return an open request with no
2997 transaction number in mds_finish_transno() if the client was
2998 evicted, but without actually returning an error. Clients
2999 would later try to replay that open and may trip an assertion
3000 Simplify the client close codepath, and always return an error
3001 from the MDS in case the open is not successful.
3004 Frequency : rare, 2.6 OSTs only
3006 Description: OST may deadlock under high load on fragmented files
3007 Details : If there was a heavy load and highly-fragmented OST filesystems
3008 it was possible to have all the OST threads deadlock waiting on
3009 allocation of biovecs, because the biovecs were not released
3010 until the entire RPC IO was completed. Instead, release biovecs
3011 as soon as they are complete to ensure forward IO progress.
3013 Severity : enhancement
3015 Description: Support for specifying external journal device at mount
3016 Details : If an OST or MDS device is formatted with an external journal
3017 device, this device major/minor is stored in the ext3 superblock
3018 and may not be valid for failover. Allow detecting and
3019 specifying the external journal at mount time.
3024 Description: Mounting an MDS with pending unlinked files may cause oops
3025 Details : target_finish_recovery() calls mds_postrecov() which returned
3026 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3027 considers this an error and immediately begins cleaning up the
3028 lov, just after starting the mds_lov process
3030 Severity : enhancement
3032 Description: Implement 'lfs df' to report actual free space on per-OST basis
3033 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3034 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3035 usage of objects; '-h' to report in human readable format.
3037 ------------------------------------------------------------------------------
3039 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3046 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3047 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3048 at startup during orphan unlinking. Instead, skip buddy update
3049 before setup, as it will be regenerated after recovery is complete.
3054 Description: If an OST is inactive, its locks might reference stale inodes.
3055 Details : lov_change_cbdata() must iterate over all namespaces, even if
3056 they are inactive to clear inode references from the lock.
3058 Severity : enhancement
3059 Frequency : occasional, if non-standard max_dirty_mb used
3061 Description: Client will block write RPCs if not enough grant
3062 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3063 then the client will block writes while waiting for another RPC
3064 to complete instead of consuming its dirty limit. With change
3065 we get improved performance when max_dirty_mb is small.
3067 Severity : enhancement
3068 Bugzilla : 3389, 6253
3069 Description: Add support for supplementary groups on the MDS.
3070 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3071 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3072 lookups for user supplementary groups into a cache.
3076 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3077 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3078 for liblustre O_EXCL is enforced by MDS.
3082 Description: readdir never returns NULL in liblustre.
3083 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3084 dentry in struct dirent.
3088 Frequency : liblustre only, depends on application IO pattern
3089 Description: liblustre clients evicted if not contacting servers
3090 Details : Don't put liblustre clients into the ping_evictor list, so
3091 they will not be evicted by the pinger ever.
3093 Severity : enhancement
3095 Description: Add ability to evict clients by NID from MDS.
3096 Details : By echoing "nid:$NID" string into
3097 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3098 $NID would be instantly evicted from this MDS and from all active
3099 OSTs connected to it.
3103 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3104 Details : lookup_it_finish() used to query file size from OSTs that was not
3109 Description: service threads change working directory to that of init
3110 Details : Starting lustre service threads may pin the working directory
3111 of the parent thread, making that filesystem busy. Threads
3112 now change to the working directory of init to avoid this.
3116 Frequency : during shutdown only
3117 Description: shutdown with a failed MDS or OST can cause unmount to hang
3118 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3123 Frequency : 2.6 only
3124 Description: chmod/chown may include an extra supplementary group
3125 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3126 supplementary group and if none is specified this is used.
3129 Bugzilla : 5479 (6816)
3131 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3132 Details : If lookup is in progress on a file that is unlinked we might try
3133 to revalidate the inode and fail in revalidate after lookup is
3134 complete and ll_file_open() enqueues the open again but
3135 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3138 Frequency : always, if lconf --abort_recovery used
3140 Description: lconf --abort_recovery fails with 'Operation not supported'
3141 Details : lconf was attempting to abort recovery on the MDT device and not
3144 ------------------------------------------------------------------------------
3146 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3151 Frequency : rare (only unsupported configurations with a node running as an
3153 Bugzilla : 6514, 5137
3154 Description: Mounting a Lustre file system on a node running as an OST could
3156 Details : OSTs now preallocates memory needed to write out data at
3157 startup, instead of when needed, to avoid having to
3158 allocate memory in possibly low memory situations.
3159 Specifically, if the file system is mounted on on OST,
3160 memory pressure could force it to try to write out data,
3161 which it needed to allocate memory to do. Due to the low
3162 memory, it would be unable to do so and the node would
3163 become unresponsive.
3165 Severity : enhancement
3167 Description: Addition of lconf --service command line option
3168 Details : lconf now accepts a '--service <arg>' option, which is
3169 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3171 Severity : enhancement
3173 Description: Failover mode is now the default for OSTs.
3174 Details : By default, OSTs will now run in failover mode. To return to
3175 the old behaviour, add '--failout' to the lmc line for OSTs.
3177 Severity : enhancement
3179 Description: Health checks are now provided for MDS and OSTs
3180 Details : Additional detailed health check information on MSD and OSTs
3181 is now provided through the procfs health_check value.
3184 Frequency : occasional, depends on IO load
3186 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3187 after numerous create/delete cycles
3188 Details : The ext3 inode allocation policy would not allocate new inodes
3189 very well on the OSTs because there are no new directories
3190 being created. Instead we look for groups with free space if
3191 the parent directories are nearly full.
3196 Description: Network or server problems during mount may cause partially
3197 mounted clients instead of returning an error.
3198 Details : The config llog parsing code may overwrite the error return
3199 code during mount error handling, returning success instead
3205 Description: MDS can fail to allocate large reply buffers
3206 Details : After long uptimes the MDS can fail to allocate large reply
3207 buffers (e.g. zconf client mount config records) due to memory
3208 fragmentation or consumption by the buffer cache. Preallocate
3209 some large reply buffers so that these replies can be sent even
3210 under memory pressure.
3214 Frequency : rare (liblustre)
3215 Description: fsx running with liblustre complained that using truncate() to
3216 extend the file doesn't work. This patch corrects that issue.
3217 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3218 ATTR_SIZE and lsm use in llu_setattr_raw.
3222 Frequency : rare, only 2.6 kernels
3223 Description: Unusual file access patterns on the MDS may result in inode
3224 data being lost in very rare circumstances.
3225 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3226 code in the 2.6 kernel caused the inode and/or EA data not to
3227 be read from disk, causing single-file corruption.
3231 Frequency : rare, only 2.6 filesystems using extents
3232 Description: Heavy concurrent write and delete load may cause data corruption.
3233 Details : It was possible under high-load situations to have an extent
3234 metadata block in the block device cache from a just-unlinked
3235 file overwrite a newly-allocated data block. We now unmap any
3236 metadata buffers that alias just-allocated data blocks.
3240 Frequency : filesystems with default stripe_count larger than 77
3241 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3242 Details : lconf specifies an inode size of 4096 bytes when the default
3243 stripe_count is larger than 77. This conflicts with the default
3244 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3245 this case to avoid pinning too much memory for large EAs.
3247 ------------------------------------------------------------------------------
3249 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3254 Frequency : rare (extremely heavy IO load with hundreds of clients)
3256 Description: Client is evicted, gets IO error writing to file
3257 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3258 caused clients to be evicted for AST timeouts. The fixes in
3259 bug 5192 mean we no longer need to have such short AST timeouts
3260 so ldlm_timeout has been increased.
3263 Frequency : occasional during --force or --failover shutdown under load
3264 Bugzilla : 5949, 4834
3265 Description: Server oops/LBUG if stopped with --force or --failover under load
3266 Details : a collection of import/export refcount and cleanup ordering
3267 issues fixed for safer force cleanup
3270 Frequency : only filesystems larger than 120 OSTs
3271 Bugzilla : 5990, 6223
3272 Description: lfs getstripe would oops on a very large filesystem
3273 Details : lov_getconfig used kfree on vmalloc'd memory
3276 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3277 Bugzilla : 6242, 6243
3278 Description: reading from files that had been truncated to a non-zero size
3279 but never opened returned no data
3280 Details : ll_file_read() reads zeros from no-object files to EOF
3285 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3286 Details : There's little harm in aborting MDS/OSS recovery and letting it
3287 try again, so I removed the LASSERT and return an error instead.
3289 Severity : enhancement
3291 Description: New debugging infrastructure for tracking down data corruption
3292 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3293 (b) cover more of the client-side code path, and (c) try to narrow
3294 down where problems occurred
3298 Bugzilla : 3819, 4364, 4397, 6313
3299 Description: Racing close and eviction MDS could cause assertion in mds_close
3300 Details : It was possible to get multiple mfd references during close and
3301 client eviction, leading to one thread referencing a freed mfd.
3303 Severity: : enhancement
3304 Bugzilla : 3262, 6359
3305 Description: Attempts to reconnect to servers are now more aggressive.
3306 Details : This builds on the enhanced upcall-less recovery that was added
3307 in 1.4.2. When trying to reconnect to servers, clients will
3308 now try each server in the failover group every 10 seconds. By
3309 default, clients would previously try one server every 25 seconds.
3314 Description: After recovery, certain operations trigger a failed
3315 assertion on a client.
3316 Details : Failing over an mds, using lconf -d --failover, while a
3317 client was doing a readdir() call would cause the client to
3318 LBUG after recovery completed and the readdir() was resent.
3320 Severity : enhancement
3322 Description: Default groups are now added by lconf
3323 Details : You can now run lconf --group <servicename> without having to
3324 manually add groups with lmc.
3327 Frequency : occasional
3329 Description: Nodes with an elan id of 0 trigger a failed assertion
3332 Frequency : always when accessing e.g. tty/console device nodes
3334 Description: tty and some other devices nodes cannot be used on lustre
3335 Details : file's private_data field is used by device data and lustre
3336 values in there got lost. New field was added to struct file to
3337 store fs-specific private data.
3340 Frequency : when exporting Lustre via NFS
3342 Description: NFSD failed occasionally when looking up a path component
3343 Details : NFSD is looking up ".." which was broken in ext3 directories
3344 that had grown large enough to become hashed.
3347 Frequency : Clusters with multiple interfaces not on the same subnet
3349 Description: Nodes will repeatedly try to reconnect to an interface which it
3350 cannot reach and report an error to the log.
3351 Details : Extra peer list entries will be created by lconf with some peers
3352 unreachable. lconf now validates the peer before adding it.
3355 Frequency : Only if a default stripe is set on the filesystem root.
3357 Description: Setting a default stripe on the filesystem root prevented the
3358 filesystem from being remounted.
3359 Details : The client was sending extra request flags in the root getattr
3360 request and did not allocate a reply buffer for the dir EA.
3363 Frequency : occasional, higher if lots of files are accessed by one client
3364 Bugzilla : 6159, 6097
3365 Description: Client trips assertion regarding lsm mismatch/magic
3366 Details : While revalidating inodes the VFS looks up inodes with ifind()
3367 and in rare cases can find an inode that is being freed.
3368 The ll_test_inode() code will free the lsm during ifind()
3369 when it finds an existing inode and then the VFS later attaches
3370 this free lsm to a new inode.
3374 Bugzilla : 6422, 7030
3375 Description: MDS deadlock between mkdir and client eviction
3376 Details : Creating a new file via mkdir or mknod (starting a transaction
3377 and getting the ns lock) can deadlock with client eviction
3378 (gets ns lock and trying to finish a synchronous transaction).
3381 Frequency : occasional
3382 Description: While starting a server, the fsfilt_ext3 module could not be
3384 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3385 kernels. Previously, lconf would still use the ext3 name
3386 when trying to load modules. Now, it will correctly use
3387 ext3 on 2.4 and ldiskfs on 2.6.
3389 Severity : enhancement
3390 Description: The default stripe count has been changed to 1
3391 Details : The interpretation of the default stripe count (0, to lfs
3392 or lmc) has been changed to mean striping across a single
3393 OST, rather than all available. For general usage we have
3394 found a stripe count of 1 or 2 works best.
3396 Severity : enhancement
3397 Description: Add support for compiling against Cray portals.
3398 Details : Conditional compiling for some areas that are different
3402 Frequency : occasional
3403 Bugzilla : 6409, 6834
3404 Description: Creating files with an explicit stripe count may lead to
3405 a failed assertion on the MDS
3406 Details : If some OSTs are full or unavailable, creating files may
3407 trigger a failed assertion on the MDS. Now, Lustre will
3408 try to use other servers or return an error to the
3412 Frequency : occasional
3414 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3415 caused lock timeout during readahead (since 1.4.2).
3416 Details : Processes doing readahead might match a lock that hasn't been
3417 granted yet if there are overlapping and conflicting lock
3418 requests. The readahead process waits on ungranted lock
3419 (original lock is CBPENDING), while OST waits for that process
3420 to cancel CBPENDING read lock and eventually evicts client.
3422 Severity : enhancement
3424 Description: Initial enabling of flock support for clients
3425 Details : Implements fcntl advisory locking and file status functions.
3426 This feature is provided as an optional mount flag (default
3427 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3428 locking are implemented yet, and those that are are not guaranteed
3429 to be completely correct in production environments.
3430 mount -t lustre -o [flock|noflock] ...
3433 Frequency : occasional
3435 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3436 assertion in the ext3 JBD (journaling) layer.
3437 Details : The b_committed_data struct is protected by the big kernel lock
3438 in 2.4 kernels, serializing journal_commit_transaction() and
3439 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3440 access to this struct. In 2.6 kernels there is finer grained
3441 locking to improve SMP performance of the JBD layer.
3445 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3450 Description: merge and cleanup kernel patches.
3453 -----------------------------------------------------------------------------
3455 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3457 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3458 versions of Lustre. Please update all servers and clients to
3459 version 1.4.2 or later at the same time. You must also run
3460 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3461 to update the configuration logs.
3463 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3464 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3465 - swab llog records of type '0' so we get proper header size/idx (5861)
3466 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3467 - fix rename of one directory over another leaking an inode (5953)
3468 - avoid SetPageDirty on 2.6 (5981)
3469 - don't re-add just-being-destroyed locks to the waiting list (5653)
3470 - when creating new directories, inherit the parent's custom
3471 striping settings if present parent (3048)
3472 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3473 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3474 - don't hold client locks on temporary worklist from l_lru (5666)
3475 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3476 - automatically evict dead clients (5921)
3477 - Update file size properly in create+truncate+fstat case (6196)
3478 - Do not unhash mountpoint dentries, do not allow removal of
3480 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3481 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3482 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3483 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3484 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3485 - Correctly handle reads of files with no objects (6243)
3486 - lctl recover will also mark a device active if deactivate used (5933)
3488 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3489 - allow --write-conf on an MDS with different nettype than client (5619)
3490 - don't write config llogs to MDS for mounts not from that MDS (5617)
3491 - lconf should create multiple TCP connections from a client (5201)
3492 - init scripts are now turned off by default; run chkconfig --on
3493 lustre and chkconfig --on lustrefs to use them
3494 - upcalls are no longer needed for clients to recover to failover
3496 - add --abort-recovery option to lconf to abort recovery on device
3498 - add support for an arbitrary number of OSTs (3026)
3499 - Quota support protocol changes.
3500 - forward compatibility changes to wire structs (6007)
3501 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3502 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3503 - improved "lustre status" script
3504 - initialize blocksize for non-regular files (6062)
3505 - added --disable-server and --disable-client configure options (5782)
3506 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3507 - Vanilla 2.4.29 support
3508 - increase maximum number of obd devices to 520 (6242)
3509 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3510 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3511 SLES 9 2.6.5 kernels
3512 - limit stripes per file to 160 (the maximum EA size) (6093)
3514 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3517 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3518 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3519 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3520 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3521 - small ext3 extents cleanups and fixes (5733)
3522 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3523 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3524 - avoid needless client->OST connect, fix handle mismatch (5317)
3525 - fix DLM error path that led to out-of-sync client, long delays (5779)
3526 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3527 - fix several locking issues related to i_size (5492,5624,5654,5672)
3528 - don't move pending lock onto export if it is already evicted (5683)
3529 - fix kernel oops when creating .foo in unlinked directory (5548)
3530 - fix deadlock in obdfilter statistics vs. object create (5811)
3531 - use time_{before,after} to avoid timer jiffies wrap (5882)
3532 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3533 - Do not leak request if server was not able to process it (5154)
3534 - If mds_open unable to find parent dir, make that negative lookup(5154)
3535 - don't create new directories with extent-mapping (5909, 5936)
3537 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3538 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3539 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3540 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3541 - support for RHEL4 kernel on i686 (5773)
3542 - provide error messages when incompatible logs are encountered (5898)
3544 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3545 * version 1.4.0.10 (1.4.1 release candidate 1)
3547 - don't keep a lock reference when lock is not granted (4238)
3548 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3549 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3550 - drop import inflight refcount on signal_completed_replay error (5255)
3551 - unlock page after async write error during send (3677)
3552 - handle missing objects in filter_preprw_read properly (5265)
3553 - no transno return for symlink open, don't save no-trasno open (3440)
3554 - don't try to complete elan receive that already failed (4012)
3555 - free RPC server reply state on error (5406)
3556 - clean up thread from ptlrpc_start_thread() on error (5160)
3557 - readahead could read extra page into cache that wasn't ejected (5388)
3558 - prevent races in class_attach/setup/cleanup/detach (5260)
3559 - don't dereference de->d_inode after l_dput of de (5458)
3560 - use "int" for stripe value returned from lock_to_stripe (5544)
3561 - mballoc allocation and error-checking fixes in 2.6 (5504)
3562 - block device patches to fix I/O request sizes in 2.6 (5482)
3563 - look up hostnames for IB nals (5602)
3564 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3565 - don't start multiple acceptors for the same port (5277)
3566 - fix incorrect LASSERT in mds_getattr_name (5635)
3567 - export a proc file for general "ping" checking (5628)
3568 - fix "lfs check" to not block when the MDS is down (5628)
3570 - service request history (4965)
3571 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3572 - create an "evict_client" /proc entry on OSTs, like the MDS has
3573 - fix mount usage message, return errors per mount(8) (5168)
3574 - change grep [] to grep "[]" in tests so they work in more UMLs
3575 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3576 - remove ancient LOV_MAGIC_V0 EA support (5047)
3577 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3578 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3579 - allow readahead allocations to fail when low on memory (5383)
3580 - mmap locking landed again, after considerable improvement (2828)
3581 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3583 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3586 - send OST transaction number in read/write reply to free req (4966)
3587 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3588 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3589 - the watchdog thread now runs as interruptible (5246)
3590 - flock/lockf fixes (but it's still disabled, pending 5135)
3591 - don't use EXT3 constants in llite code (5094)
3592 - memory shortage at startup could cause assertion (5176)
3594 - reorganization of lov code
3595 - single portals codebase
3597 - add extents/mballoc support (5025)
3598 - direct I/O reads in the obdfilter (4048)
3599 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3601 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3604 - send OST transaction number in read/write reply to free req (4966)
3605 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3606 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3607 - the watchdog thread now runs as interruptible (5246)
3608 - handle missing objects in filter_preprw_read properly (5265)
3609 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3610 - drop import inflight refcount on signal_completed_replay error (5255)
3611 - unlock page after async write error during send (3677)
3612 - return original error code on reconstructed replies (3761)
3613 - no transno return for symlink open, don't save no-trasno open (3440)
3615 - add pid to ldlm debugging output (4922)
3616 - bump the watchdog timeouts -- we can't handle 30sec yet
3617 - extra debugging for orphan dentry/inode bug (5259)
3619 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3622 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3623 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3624 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3625 - fix NULL dereference in /proc/sys/portals/routes (4827)
3626 - allow failed mdc_close() operations to be interrupted (4561)
3627 - stop precreate on OST before MDS would time out on it (4778)
3628 - don't send partial-page writes before EOF from client (4410)
3629 - discard client grant for sub-page writes on large-page clients (4520)
3630 - don't free dentries not owned by NFS code, check generation (4806)
3631 - fix lsm leak if mds_create_objects() fails (4801)
3632 - limit debug_daemon file size, always print CERROR messages (4789)
3633 - use transno after validating reply (3892)
3634 - process timed out requests if import state changes (3754)
3635 - update mtime on OST during writes, return in glimpse (4829)
3636 - add mkfsoptions to LDAP (4679)
3637 - use ->max_readahead method instead of zapping global ra (5039)
3638 - don't interrupt __l_wait_event() during strace
3640 - add software watchdogs to catch hung threads quickly (4941)
3641 - make lustrefs init script start after nfs is mounted
3642 - fix CWARN/ERROR duplication (4930)
3643 - return async write errors to application if possible (2248)
3644 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3645 - print NAL number in %x format (4645)
3646 - update barely-supported suse-2.4.21-171 series (4842)
3647 - support for sles 9 %post scripts
3648 - support for building 2.6 kernel-source packages
3649 - support for sles km_* packages
3651 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3654 - ignore -ENOENT errors in osc_destroy (3639)
3655 - notify osc create thread that OSC is being cleaned up (4600)
3656 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3657 - reconstruct ost_handle() like mds_handle() (4657)
3658 - create a new thread to do import eviction to avoid deadlock (3969)
3659 - let lconf resolve symlinked-to devices (4629)
3660 - don't unlink "objects" from directory with default EA (4554)
3661 - hold socknal file ref over connect in case target is down (4394)
3662 - allow more than 32000 subdirectories in a single directory (3244)
3663 - fix blocks count for O_DIRECT writes (3751)
3664 - OST returns ENOSPC from object create when no space left (4539)
3665 - don't send truncate RPC if file size isn't changing (4410)
3666 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3667 - bind to privileged port in socknal and tcpnal (3689)
3669 - rate limit CERROR/CWARN console message to avoid overload (4519)
3670 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3671 - basic mmap support (3918)
3672 - kernel patch series update from b1_4 (4711)
3674 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3677 - avoid crash during MDS cleanup with OST shut down (2775)
3678 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3679 - don't use bad inodes on the MDS (3744)
3680 - dynamic object preallocation to improve recovery speed (4236)
3681 - don't hold spinlock over lock dumping or change debug flags (4401)
3682 - don't zero obd_dev when it is force cleaned (3651)
3683 - print grants to console if they go negative (4431)
3684 - "lctl deactivate" will stop automatic recovery attempts (3406)
3685 - look for existing locks in ldlm_handle_enqueue() (3764)
3686 - don't resolve lock handle twice in recovery avoiding race (4401)
3687 - revalidate should check working dir is a directory (4134)
3689 - don't always mark "slow" obdfilter messages as errors (4418)
3691 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3694 - don't close LustreDB during write_conf until it is done (3860)
3695 - fix typo in lconf for_each_profile (3821)
3696 - allow dumping logs from multiple threads at one time (3820)
3697 - don't allow multiple threads in OSC recovery (3812)
3698 - fix debug_size parameters (3864)
3699 - fix mds_postrecov to initialize import for llog ctxt (3121)
3700 - replace config semaphore with spinlock (3306)
3701 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3702 - don't allow enqueue to complete on a destroyed export (3822)
3703 - down write_lock before checking llog header bitmap (3825)
3704 - recover from lock replay timeout (3764)
3705 - up llog sem before sending rpc (3652)
3706 - reduce ns lock hold times when setting kms (3267)
3707 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3708 - fix NULL deref and obd_dev leak on setup error (3312)
3709 - replace some LBUG about llog ops with error handling (3841)
3710 - don't match INVALID dentries from d_lookup and spin (3784)
3711 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3712 - fix invalid assertion in ptlrpc_set_wait (3880)
3714 - add libwrap support for the TCP acceptor (3996)
3715 - add /proc/sys/portals/routes for non-root route listing (3994)
3716 - allow setting MDS UUID in .xml (2580)
3717 - print the stack of a process that LBUGs (4228)
3719 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3722 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3723 - make sure to unset replay flag from failed open requests (3440)
3724 - if default stripe count is 0, use OST count for inode size (3636)
3725 - update parent mtime/ctime on client for create/unlink (2611)
3726 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3727 - free recovery state on server during a forced cleanup (3571)
3728 - unregister_reply for resent reqs (3063)
3729 - loop back devices mounting and status check on 2.6 (3563)
3730 - fix resource-creation race that can provoke i_size == 0 (3513)
3731 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3732 - more debugging for page-accounting assertion (3746)
3733 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3734 - avoid deadlock after precreation failure (3758)
3735 - fix race and lock order deadlock in orphan handling (3450, 3750)
3736 - add validity checks when grabbing inodes from l_ast_data (3599)
3738 - add /proc/.../recovery_status to obdfilter (3428)
3739 - lightweight CDEBUG infrastructure, debug daemon (3668)
3740 - change default OSC RPC parameters to be better on small clusters
3741 - turn off OST read cache for files smaller than 32MB
3742 - install man pages and include them in rpms (3100)
3743 - add new init script for (un)mounting lustre filesystems (2593)
3744 - run chkconfig in %post for init scripts (3701)
3745 - drop scimac NAL (unmaintained)
3747 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3750 - clean kiobufs before and after use (3485)
3751 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3752 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3753 - change init script to fail more clearly if not run as root (1528)
3754 - allow clients to reconnect during replay (1742)
3755 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3756 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3757 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3758 - fix cond_resched (3554)
3759 - only evict unfinished clients after recovery (3515)
3760 - allow bulk resend, prevent data loss (3570)
3761 - dynamic ptlrpc request buffer allocation (2102)
3762 - don't allow unlinking open directory if it isn't empty (2904)
3763 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3764 - remove extraneous obd dereference causing LASSERT failure (3334)
3765 - don't use get_cycles() when creating temp. files on the mds (3156)
3766 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3767 - handle EEXIST for set-stripe, set proper directory name (3336)
3769 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3770 - fix ksocknal_fmb_callback() error messages (2918)
3772 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3775 - don't copy lvb into (possibly NULL) reply on error (2983)
3776 - don't deref dentry after dput, don't free lvb on error (2922)
3777 - use the kms to determine writeback rpc length (2947)
3778 - increment oti_logcookies when osc is inactive (2948)
3779 - update client's i_blocks count via lvb messages (2543)
3780 - handle intent open/close of special files properly (1557)
3781 - mount MDS with errors=remount-ro, like obdfilter (2009)
3782 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3783 - don't use cancelling-locks' kms values (2947)
3784 - use highest lock extent for kms, not last one (2925)
3785 - don't dereference ERR_PTR() dentry in error handling path (3107)
3786 - fix thread race in portals_debug_dumplog() (3122)
3787 - create lprocfs device entries at setup instead of at attach (1519)
3788 - common AST error handler, don't evict client on completion race (3145)
3789 - zero nameidata in detach_mnt in 2.6 (3118)
3790 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3791 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3792 - initialize RPC timeout timer earlier for 2.6 (3219)
3793 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3794 - print nal/nid for unknown nid (3258)
3795 - additional checks for oscc recovery before doing precreate (3284)
3796 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3797 - don't crash in mdc_close for bad permissions on open (3285)
3798 - zero i_rdev for non-device files (3147)
3799 - clear page->private before handing to FS, better assertion (3119)
3800 - tune the read pipeline (3236)
3801 - fix incorrect decref of invalidated dentry (2350)
3802 - provide read-ahead stats and refine rpc in flight stats (3328)
3803 - don't hold journal transaction open across create RPC (3313)
3804 - update atime on MDS at close time (3265)
3805 - close LDAP connection when recovering to avoid server load (3315)
3806 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3807 - don't leak open file on MDS after open resend (3325)
3808 - serialize filter_precreate and filter_destroy_precreated (3329)
3809 - loop device shouldn't call sync_dev() for nul device (3092)
3810 - clear page cache after eviction (2766)
3811 - resynchronize MDS->OST in background (2824)
3812 - refuse to mount the same filesystem twice on same mountpoint (3394)
3813 - allow llmount to create routes for mounting behind routers (3320)
3814 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3815 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3816 - fix rare problem with rename on htree directories (3417)
3818 - allow default OST striping configuration per directory (1414)
3819 - fix compilation for qswnal for 2.6 kernels (3125)
3820 - increase maximum number of MDS request buffers for large systems
3821 - change liblustreapi to be useful for external progs like lfsck (3098)
3822 - increase local configuration timeout for slow disks (3353)
3823 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3825 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3828 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3829 - don't overwrite extent policy data in reply if lock was blocked (2901)
3830 - drop filter export grants atomically with removal from device (2663)
3831 - del obd_self_export from work_list in class_disconnect_exports (2908)
3832 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3833 - swab reply message in mdc_close, other PPC fixes (2464)
3834 - fix destroying of named logs (2325)
3835 - overwrite old logs when running lconf --write_conf (2264)
3836 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3837 - fix race in target_handle_connect (2898)
3838 - mds_reint_create() should take same inode create lock (2926)
3839 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3840 - don't close files for self_export to avoid uninitialized obd (2936)
3841 - allow MDS with the same name as client node (2939)
3842 - hold dentry reference for closed log files for unlink (2325)
3843 - reserve space for all logs during transactions (2059)
3844 - don't evict page beyond end of stripe extent (2925)
3845 - don't oops on a deleted current working directory (2399)
3846 - handle hard links to targets without a parent properly (2517)
3847 - don't dereference NULL lock when racing during eviction (2867)
3848 - don't grow lock extents when lots of conflicting locks (2919)
3850 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3853 - account for cache space usage on clients to avoid data loss (974)
3854 - lfsck support in lustre kernel code (2349)
3855 - reduce journal credits needed for BRW writes (2370)
3856 - orphan handling to avoid losing space on client/server crashes
3857 - ptlrpcd can be blocked, stopping ALL progress (2477)
3858 - use lock value blocks to assist in proper KMS, faster stat (1021)
3859 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3860 - recovery for initial connections (2355)
3861 - fixes for mds_cleanup_orphans (1934)
3862 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3863 - block all file creations until orphan recovery completes (1901)
3864 - client remove rq_connection from request struct (2423)
3865 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3866 - recovery timer race (2670)
3867 - mdc_close recovey bug (2532)
3868 - ptlrpc cleanup bug (2710)
3869 - mds timeout on local locks (2588)
3870 - namespace lock held during RPCs (2431)
3871 - handle interrupted sync write properly (2503)
3872 - don't try to handle a message that hasn't been replied to (2699)
3873 - client assert failure during cleanup after abort recovery (2701)
3874 - leak mdc device after failed mount (2712)
3875 - ptlrpc_check_set allows timedout requests to complete (2714)
3876 - wait for inflight reqs when ptlrpcd finishes (2710)
3877 - make sure unregistered services are removed from the srv_list
3878 - reset bulk XID's when resending them (caught by 1138 test)
3879 - unregister_bulk after timeout
3880 - fix lconf error (2694)
3881 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3882 - readahead locks pages, leaves pending causing memory pressure (2673)
3883 - increase OST request buffers to 4096 on large machines (2729)
3884 - fix up permission of existing directories in simple_mkdir (2661)
3885 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3886 - don't assign transno to errored transactions (2742)
3887 - don't delete objects on OST if given a bogus objid from MDS (2751)
3888 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3889 - if rq_no_resend, then timeout request after recovery (2432)
3890 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3891 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3892 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3893 - MDS recovery shouldn't delete objects that it has given out (2730)
3894 - if enqueue arrives after completion, don't clobber LVB (2819)
3895 - don't unlock pages twice when trigger_group_io returns error (2814)
3896 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3897 - don't write pages to disk if there was an error (1450)
3898 - don't ping imports that have recovery disabled (2676)
3899 - take buffered bytes into account when balancing socknal conn (2817)
3900 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3901 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3902 - remove little-endian swabbing of llog records (1987)
3903 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3904 - retry reposting request buffers if they fail (1191)
3905 - grow extent at grant time to avoid granting a revoked lock (2809)
3906 - lock revoke doesn't evict page if covered by a second lock (2765)
3907 - disable VM readahead to avoid reading outside lock extents (2805)
3909 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3910 - updated kernel patches for hp-2.4.20 kernel (2681)
3912 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3915 - fix truncated write corruption (2366)
3916 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3918 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3919 - ensure len(uuid) < 37 in lmc (1171)
3920 - fix ia64 OOPS in llog_test (2255)
3921 - zero end of page at obdfilter for partial page writes (2648)
3922 - don't leave stale dentries around after renames (bug 2428)
3923 - fix timeouts when evicting a client with a single lock held (2642)
3924 - set deadline for the initial HELLO message to drain (2634)
3925 - print out dotted-quad IP addresses in the socknal (2302)
3927 - additional debugging for MDS client eviction problem (2443)
3928 - fix mkfsoptions support for osts (2603, 2604)
3930 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3933 - add series for the vanilla 2.6.0 kernel
3934 - add series for the vanilla 2.4.24 kernel
3935 - add series for a cray x86/64 UL kernel drop
3936 - fix xattr patches for the vanilla 2.4.19 series
3938 - generate true UUIDs in lmc (1171)
3939 - have portals stack dumping break in UML (2466)
3940 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3941 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3942 - wake callback waiting threads on client eviction (2460)
3943 - Add --ptldebug and --subsystem to lmc (1719)
3944 - update assertion to allow safe interrupt allocation
3945 - set rq_no_resend for cancel requests (2432)
3946 - recalculate ptlrpcd timeout after resend (2494)
3947 - call vfs_rmdir when removing pending directories (2368)
3948 - fix renaming a file to itself (2429)
3949 - lmc creates a default one-stripe lov (2454)
3950 - expand procfs space to handle large clusters (2326)
3951 - increase UML stack to avoid overflow
3952 - update lconf's list of debug and subsystem masks
3953 - fix lfs find --obd (2510)
3954 - /proc tunable for disabling filter read caching (2591)
3955 - stop rpm packages from altering slapd.conf (2301)
3956 - disable nagle in the socknal under 0conf (2578)
3957 - choose mds inode size based on stripe count (2572)
3958 - fix kernel-source rpm problems (2516)
3960 - add --disable-doc to avoid pdf generation (2421)
3961 - update documentation, tests, type-os, comments
3962 - avoid format warnings on ia64
3963 - remove the TOE NAL
3964 - tiny code cleanups by removing unused fields
3966 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3969 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3970 - remove the most blatant lies from BUILDING file (2371)
3971 - change default debug level to reasonable production setting
3972 - reduce client side cache size to reduce cache flush time
3973 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3974 - make TCP zerocopy and pinger support enabled by default (2476)
3975 - sync writes completed after process exits caused crashes (2319)
3976 - maintain correct mount count on the MDS (2356)
3977 - backout 1557, because 2316 wasn't really fixed
3978 - better file I/O statistics gathering in /proc
3979 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3980 - another kernel patch to fix zero-copy TCP function export
3981 - don't take duplicate lock when processing re-sent getattr (2420)
3982 - lctl uses obd_self_export instead of creating new conn (2353)
3983 - MDS/OST recovery case which requires object creation asserted (2425)
3984 - move lfs from /usr/sbin to /usr/bin in packages
3985 - fix race between mds_client_add and mds_client_free (2417)
3986 - use kmalloc instead of slabs in portals (2430)
3987 - don't create duplicate records when a failover MDS is present (2442)
3988 - remove unnecessary mount age check (2332)
3989 - don't remove directory inodes from locks prematurely (2451)
3990 - don't break if MDS service name is the same as hostname (2103)
3991 - fix races in client write RPC generation when cache full (2482)
3993 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3996 - remove now-unused request->rq_obd (278)
3997 - if an allocation fails, print out how much memory we've used (1933)
3998 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3999 - add the "configurable stack size" patch to most series files (1256)
4000 - ability to write large log records, for 100+ OST configs (2306)
4001 - fix NULL deref when filter_prep fails (2314)
4002 - fix operator precedence error in filter_sync
4003 - dynamic allocation of socknal TX descriptors (2315)
4004 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4005 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4006 - ensure that CWARN messages in Portals always get to the syslog
4007 - __init/__exit are not for prototype decls (ldlm_init/exit)
4008 - x86-64 compile warning fixes
4009 - fix gateway LMC keyword conflict (2318)
4010 - fix MDS lock inversions in getattr/reint paths (1844)
4011 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4012 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4013 - DLM race condition prevented some lock evictions (2328)
4014 - ENOMEM detection and retry on socknal sends (2230)
4015 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4016 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4017 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4018 - better error messages when a client is rejected during recovery (1505)
4019 - avoid cancelling locks which were never granted, after failure (2330)
4020 - fix i_sem/journal inversion in mds_client_add (2333)
4021 - fix truncate/getattr lock cycle deadlock (2334)
4022 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4023 - fix two rare exit paths which could leak an l_lock() ref (2321)
4024 - fencepost error in MDS/OST orphan recovery (2226)
4025 - make log record alignment 8 bytes (1988)
4026 - lstripe now fails when requested offset > ost_count (2237)
4027 - ensure that all kernel series have a complete list.h (1607)
4028 - fix crashes in special-file operations (2316)
4029 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4031 - allow configurable automake binary, for testing new versions
4032 - small update to the lfs documentation
4034 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4036 * fix negative export reference count in fsfilt_sync (2312)
4038 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4039 * release candidate 0.9.1
4041 - orphans are moved into the PENDING directory for possible recovery
4042 - replayed opens now open by fid for orphan/rename safety (1042)
4043 - last close of an orphan inode generates a transno (683)
4044 - chdir() and mount() now pin the directory entry (1020)
4045 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4046 - discard very old requests without processing them (1502)
4047 - remove some common, well-understood CERRORs (1505)
4048 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4049 - clear "grant" flags in OST replies until OST grant code lands (1644)
4050 - fix read performance by not clobbering i_blksize on client (1598)
4051 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4052 - make LRU size a /proc tunable, clears locks when reduced (707)
4053 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4054 - server locks take references on exports now (1558)
4055 - build fixes for 2.4.20-rh trees (1663)
4056 - return an error from lov_create if all OSCs are inactive (1751)
4057 - fix import levels when a reconnect happens without a timeout (1597)
4058 - exit early from mds_open if we get a lookup error (1749)
4059 - partial page read at EOF wouldn't wait for disk before sending (1642)
4060 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4061 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4062 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4063 - write/truncate lock inversion (1639)
4064 - Don't auto-load obdclass, portals modules during cleanup (1495)
4065 - fix timestamps from jumping to "now" (1763)
4066 - extra journal assertions (1648)
4067 - add an extra multiunlink test (1771)
4068 - fix read_record/write_record API (1776)
4069 - fix leak of offset_extent, possible incorrect i_size later (1772)
4070 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4071 - Debugging for the kqswnal_get_idle_tx problems (1820)
4072 - Allow recovery to be attempted multiple times (1536)
4073 - Write out MDS last_rcvd file after it is first created (1600)
4074 - Fix tx_descriptor leak in failed transmit situations (1827)
4075 - ext3 journaling fixes for assertion failure after IO error (1871)
4076 - class_export_put() on freed export after completion AST error (1896)
4077 - Fix revalidate looping in VFS (1322)
4078 - Don't access a freed export during MDS_REINT timeout (1521)
4079 - Add open-unlink recovery support on the MDS (1673,1764)
4080 - Return an error if no MDS data was read from last_rcvd (1946)
4081 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4082 - Don't setup a disk device that doesn't match exported UUID (317)
4083 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4084 - avoid committing NULL handle in force close
4085 - local.sh is now a one-stripe LOV configuration
4086 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4087 - fix ext3 htree duplicate directory entry corruption (1516)
4088 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4089 - update to new LOV EA format (2097)
4090 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4092 2003-06-15 Phil Schwan <phil@clusterfs.com>
4095 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4096 - per-import/export recovery handling (958, 931, 959)
4097 - multiple last-rcvd slots, for serving multiple FSes (949)
4098 - connections are again shared between multiple imp/exports (963, 964)
4099 - "umount -f" would hang if any requests needed to be sent (393, 978)
4100 - avoid pinning large req buffer by copying for queued messages (989)
4101 - add "uuid" to "lctl device" command to help upcalls (991)
4102 - "open" RPCs with transnos would confuse recovery counters (1037)
4103 - do proper endian conversion of all wire messages (288, 340, 891)
4104 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4105 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4106 - fix ptlrpc_connection leak in target_handle_connect (1174)
4107 - fix import refcounting bug in OST and MDS cleanup (1134)
4108 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4109 - fix very unlikely obd_types race condition (501)
4110 - remove osc_open hack for echo_client (1187)
4111 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4112 - a failure in read_inode2 leads to deadlock (1139)
4113 - cancel ack-locks as soon as transaction is committed (1072)
4114 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4115 - make sure to commitrw after any preprw to avoid deadlock (1162)
4116 - failing to execute a file in a lustre FS would lock inode (1203)
4117 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4118 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4119 - obd_uuid2tgt would walk off the end of the list (1255)
4120 - on IA64 the osc would give portals incorrect bulk size (1258)
4121 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4122 - fix lock inversion caused by new llite matching code (1282)
4123 - limit the number of dirty pages on a client to 10MB (1286)
4124 - timed out locks were not being corrected cancelled (1289)
4125 - fix O_DIRECT above 4GB on IA-32 (1292)
4126 * major user-visible changes
4127 - fail out/fail over policy now controlled by the upcall (993)
4129 - add OBD_PING to check server availability and failure (954)
4130 - lustre messages are now sent in sending host order (288, 340, 891)
4131 - add eadatalen to MDS getattr reply (340)
4132 - OST read replies may contain second buffer, with per-page status (593)
4134 2003-03-11 Phil Schwan <phil@clusterfs.com>
4137 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4138 - failed open()s could cause deadlock; fixed (867, 869)
4139 - stop cancelling OST locks when files are closed (481)
4140 - overlapping XID spaces caused network corruption (851, 853)
4141 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4142 - setattr_raw added, to do single-RPC, server-side setattrs
4143 - lmc/lconf syntax change for OST UUIDs
4144 - fix crashy race condition between ptlrpc_free_req and osc_close
4145 - don't use request in mdc_enqueue if we hit a timeout (889)
4146 - don't set the inode i_size for regular files from the MDS (896)
4147 - handle out of order completion AST (842)
4148 - don't LBUG if a lock request times out after receiving AST (913)
4149 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4150 - if a bad lock AST arrives, send an error instead of dropping entirely
4151 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4152 - fix leak in bulk IO when only partially completed (899, 900, 926)
4153 - fix O_DIRECT for ia64 (55)
4154 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4155 - C-z after timeout could hang a process forever; fixed (977)
4157 - client-side I/O cache (678, 924, 929, 941, 970)
4159 - READPAGE and SETATTRs which don't take server-side locks get
4162 2003-02-11 Phil Schwan <phil@clusterfs.com>
4165 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4166 - Fix fsfilt_extN_readpage() to read a full page of directory
4167 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4168 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4169 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4170 - Most intent operations were removed, in favour of a new RPC mode
4171 that does a single RPC to the server and bypasses most of the VFS
4172 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4173 - Aggressively cancel local locks on DLM servers
4174 - mds_reint_unlink sends EA to the client if it's the last nlink.
4175 client uses that EA to unlink OST objects.
4176 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4177 - recursive symlinks were fixed (439)
4178 - fixed NULL deref in DEBUG_REQ
4179 - filter_update_lastobjid no longer calls sync, which annoyed extN
4180 - fixed multi-client small-writes to a single file problem (445)
4181 - fixed mtime updates during file writes (607)
4182 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4183 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4184 - fixed Linux OST size reporting problem (444, 656)
4185 - OST now updates object mtime with writes or setattr (607, 619)
4186 - client verifies file size before zeroing page past EOF (445)
4187 - OST now writes last allocated objid to disk with allocation (108)
4188 - LOV on echo now works (409)
4190 - mds_reint_unlink sends a new buffer, with the EA included. this
4191 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4192 set if a regular file was being unlinked, and it was the last link
4193 - use PtlGet from the target for bulk writes (315)
4194 - OST now updates object mtime with writes or setattr (607, 619)
4195 - LDLM now has a grant-time callback to revalidate locked items, if
4197 - Many MDS operations were reorganized to combat race conditions
4199 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4200 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4202 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4205 - Fully reactivate OST imports after reconnection (512, others)
4206 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4207 - More graceful error handling for truncating on dead OST (515)
4208 - Don't error out unless we're actually accessing dead stripes (474)
4209 - Fix garbage sizes when stripes are missing (410)
4210 - LRU counters were broken, causing constant lock purge (433, 432)
4211 - garbage on read from stripes with failed OSTs (441)
4212 - mark OSCs as active before reconnecting during recovery (438)
4213 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4214 - lfind did not preserve OST order in output (443)
4215 - symlinks cause hung clients, incorrect data (439)
4216 - stop dereferencing request after dropping refcount (457)
4217 - don't LASSERT(spin_is_locked) on non-SMP (455)
4218 - fixes for many rename() bugs
4219 - fstat didn't correctly synchronize attributes (399)
4220 - server must handle lock cancellation during blocking AST prep (487)
4221 - bulk descriptors were free()d too soon (511)
4222 - fix paths in lconf, which would load incorrect modules (451, 507)
4223 - fix confusing lconf 'host not found' error message (386)
4224 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4225 - fix race condition in mdc_blocking_ast() for inode access (526)
4226 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4227 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4228 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4229 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4230 - fix lstripe in several ways (kernel side) (527)
4231 - fix request leak in ldlm_cli_enqueue (262)
4232 - incorrect OSC was marked inactive after OST failure
4233 - call mds_fs_cleanup before unmounting filesystem (524)
4234 - fix races between taking ns_lock and ldlm_lock_change_resource
4235 - fix races updating LOV export open file list
4236 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4237 - fix recovery NULL deref in ldlm_cli_cancel_unused
4238 - fix some DLM races by using new hash table for lock handles (419)
4239 - permit the client to specify desired inodes, at replay
4240 - duplicate requests when we queue them for replay reintegration
4241 - fix last_rcvd offset calculation
4242 - sync after each recovered transaction, so we always make progress
4243 - never, not always, ERESTART requests without transnos
4244 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4246 - skip replay if the MDS says that the client is already connected
4247 - don't check for a recovery-enabled export to match lctl's UUID
4248 - don't INC_USE_COUNT for phantom exports
4249 - don't crash when cleaning up phantom exports (567)
4250 - don't double-finish or set replay data for errored mdc_open requests
4251 - abort requests when they time out, so we don't get old replies
4252 - send/receive replies for AST messages again
4253 - if the client says that it doesn't have the lock, cancel it on the
4255 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4256 mark it as destroyed, it will all work out when decref is called
4257 - fix module use counts (22, 581)
4259 - ASTs now expect a reply (server cancels lock on error reply)
4261 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4264 - fix many simultaneous client startup (392)
4265 - fix dentry->d_it clobbering
4266 - credentials weren't being shipped for readdir/getattr operations
4267 - remove invalid assertions triggered during some concurrent MD
4269 - proper Lustre versions added (336, 389)
4270 - fix memory leak for create error case (398)
4271 - fix LOV locking bug that would get cli/srv out of sync
4272 - fix echo client over LOV (409)
4273 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4274 - fix double-O_EXCL intent crash (424)
4275 - avoid sending multiple lock CANCELs (352)
4277 - MDS can do multi-client recovery (modulo bugs in new code)
4279 - many updates, edits, cleanups
4281 2002-11-18 Phil Schwan <phil@clusterfs.com>
4284 - fix null d_it dereference (346)
4285 - fix full OST/dbench hang (333)
4286 - fix permission problem with file removal (286)
4287 - fix removal of OSCs from LOV when they fail
4288 - fix NULL deref during bulk timeout (214)
4289 - fix problems related to multiple filesystems on one MDS (241)
4290 - fixed serious subtle metadata locking bugs
4291 - free locks on clients when inodes are removed due to memory
4293 - fix inode pointer in lock data (285)
4294 - partial support for multiple MDS on a single host (241)
4295 - data locks weren't cancelled at clear_inode time (290, 311)
4296 - intent locks could lead to unbounded lock growth (205)
4297 - added a maximum lock count, an LRU list, and a flusher
4298 - fix multiple rename (365)
4299 - properly abstracted the echo client
4300 - OSC locked 1 byte too many; fixed
4301 - rewrote brw callback code:
4302 - fixed recovery bugs related to LOVs (306)
4303 - fixed too-many-pages-in-one-write crash (191)
4304 - fixed (again) crash in sync_io_timeout (214)
4305 - probably fixed callback-related race (385)
4307 - Add capability to MDS protocol
4308 - LDLM cancellations and callbacks on different portals
4310 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4313 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4314 - defer open object destruction to close time (601981, 138)
4315 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4316 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4317 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4318 - don't call obd_setattr in ll_file_release for destroyed objects
4319 * protocol change to lustre_msg: move |version| and add |flags|
4320 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4321 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4322 * added replay of create, unlink, link and rename operations during
4323 MDS failover; recovery should be much more robust now
4324 * remove failed OSCs from LOVs (only lov_create uses this so far)
4325 * the lustre-HOWTO was brought (more) up to date (582544)
4327 2002-10-23 Phil Schwan <phil@clusterfs.com>
4330 - in-use dentries weren't being reused properly (617851)
4331 - prevent multiple LDLM setup (599178)
4332 - fix LOV size calculations for truncate (617853)
4333 - fix client handling of MDS intent errors (POSIX)
4334 - fix permission bug in lovstripe.c test (624321)
4335 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4336 - truncate past end of file could corrupt data
4337 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4338 - a race in recovery could return ETIMEDOUT to apps (623947)
4339 - building outside the source directory was fixed
4340 * the lustre-HOWTO was brought (more) up to date (582544)
4341 * major progress was made on recovery functionality
4343 2002-10-10 Phil Schwan <phil@clusterfs.com>
4346 - recovery deadlock fix
4347 - rm -rf causes LBUG fix (617817)
4348 - file open by multiple tasks fix (618962)
4349 - directory permissions bugs (602707 and 620007)
4350 - journal_stop fixed with locking (611313)
4351 - O_APPEND failures resolved (618273, perhaps 614459)
4352 - lconf PATH fix (619770)
4353 - IA64 build fix (621450)
4354 - RPC buffer sizes scale with amount of memory
4356 2002-10-01 Phil Schwan <phil@clusterfs.com>
4359 - locks would be cancelled without throwing away data pages,
4360 resulting in inconsistent data (605627)
4361 - inode attributes were not always being refreshed (605627, 612449)
4362 - lconf now continues to cleanup after lctl reports an error
4363 - MDS now enforces user permissions (602707)
4364 - lprocfs cleanup fixed, but not yet enabled (614157)
4365 - fixed infinite server hang, should a client not respond to an AST
4366 - avoid going into recovery if user calls readlink() with a buffer
4367 that's too small (613941)
4368 - AST RPCs no longer require replies (614867) -- this may be changed
4369 - don't crash server if client sends an IOV that's too big (611336)
4370 - fixed lock conversion deadlock (611892)
4371 - fixed the following of symlinks (614622)
4372 * recovery: the server can remove locks from a client that dies, other
4373 clients can make progress
4374 * more extN patch fixes
4375 * compile-time configurable ptlrpc buffer allocations
4377 - collaborative read cache document
4378 - Lustre Lite Performance CDR document-in-progress
4380 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4383 - fix typo in patch-2.4.18
4385 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4388 - clear ptlrpc request each time in handle_incoming_request()
4389 - unlink of files now destroys the object on the OST
4391 2002-09-19 Peter Braam <braam@clusterfs.com>
4393 * add hard link support
4394 * change obdfile creation method
4395 * kernel patch changed
4397 2002-09-19 Peter Braam <braam@clusterfs.com>
4400 - stack overflow bug in extN fixed
4402 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4404 * documentation updates
4405 - add man pages for config tools
4406 - update tests/README to describe testing with new config tools
4407 - finish metadata API descriptions
4408 * bug fixes and cleanups
4409 - statfs workaround for 16TB limit
4410 - LOV stripe allocation improved, can stripe on subset of OSTs
4411 - LOV file size/IO offset was wrong for files > 4GB in size
4412 - object EA data was being dropped, caused files to be unreadable
4413 - memory overflow with non-LOV OST caused memory corruption
4414 - fixed regression tests to work with new config tools, obdfilter
4415 - fixed bug when directory size became larger than 1 block
4416 - fixed bug (for single client case) when PWD was deleted
4417 - invalidate local directory pages when doing intent-based ops
4418 - avoid LDLM oops when lock callback contained bad data
4420 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4422 * documentation updates
4423 * bug fixes and cleanups
4424 - configuration tools
4427 - 64-bit compile warnings
4428 - 64-bit internal statfs data
4430 * test_brw on persistent OST devices
4432 * lprocfs (disabled)
4434 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4436 * documentation updates
4437 * bug fixes and cleanups
4438 * configuration tools
4440 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4443 * many small fixes to 0_5_4
4444 * io/network handling
4445 * thinkos in MDS operations
4447 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4450 * crucial basic fixes to 0.5.3
4451 * IOR, Iozone work over Elan
4454 2002-08-07 Phil Schwan <phil@clusterfs.com>
4455 * version 0_5_3, our first alpha
4456 * we use the new Portals iovs
4457 * documentation updates
4458 * bug fixes and cleanups
4459 * small changes in the DLM wire protocol
4461 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4462 * version 0_5_1 with some initial stability,
4463 * locking on MD and file I/O.
4464 * documentation updates
4465 * several bug fixes since 0.5.0
4466 * small changes in wire protocol
4468 2002-07-18 Phil Schwan <phil@clusterfs.com>
4470 * delivered as Lustre Light Alpha
4471 * fixed a crash after handling invalid MDS requests
4472 * fixed directory pages for architectures with non-4k pages sizes
4474 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4475 * release version v0_4_4
4476 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4478 * Fixes a number of interruption problems with OST operations.
4479 * Update documentation for portals header changes
4480 * Move all wire protocol structs/defines to lustre_idl.h
4481 * Fixes symlink length bug.
4482 * Add tcpdump to repository.
4484 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4485 * release version v0_4_3
4486 * Fixes statfs for inodes on extN.
4487 * Fixes bug in runtests which would delete /etc/hosts.
4488 * Use 64-bit object IDs wherever possible (not into VFS though)
4489 Remove ost_get_info, which is unused by lustre, and out of date.
4491 2002-07-03 Peter Braam <braam@clusterfs.com>
4492 * release version v0_4_2 Fixes a lookup error (type not passed)
4493 * move forward to head of Portals
4494 * move forward to latest Lustre kernel
4496 2002-06-25 Peter Braam <braam@clusterfs.com>
4497 * release version v0_4_1. Hopefully stable on single node use.