1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.13.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).
15 Severity : enhancement
17 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
22 Description: Fix replay if there is an un-replied request and open
23 Details : In some cases, older replay request will revert the
24 mcd->mcd_last_xid on MDS which is used to record the client's
27 Severity : enhancement
29 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
31 Severity : enhancement
33 Description: Add rhel5 support to HEAD.
35 Serverity : enhancement
37 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
42 Description: Don't allow skipping OSTs if index has been specified.
43 Details : Don't allow skipping OSTs if index has been specified, make locking
44 in internal create lots better.
48 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
49 Details : access to bitfield in structure is always rounded to long
50 and this produce problem with not atomic change any bit.
54 Description: Lustre make rpms failed.
55 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
56 builds Lustre rpms from tarball.
59 Frequency : rare on shutdown ost
61 Description: If llog cancel was not send before clean_exports phase, this can
62 produce deadlock in llog code.
63 Details : If llog thread has last reference to obd and call class_import_put
64 this produce deadlock because llog_cleanup_commit_master wait when
65 last llog_commit_thread exited, but this never success because was
66 called from llog_commit_thread.
70 Description: allow userland application know is lost one of stripes.
71 Details : fill lvb_blocks with error code on ost and return it to
72 application if error flag found.
76 Description: NULL lov_tgts causing MDS oops
77 Details : more safe checks for NULL lov_tgts for avoid oops.
79 Severity : enhancement
81 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
85 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
86 Details : If an OST is down the MDS will hang indefinitely in
87 obd_statfs_rqset() waiting for the statfs data. While for
88 MDS QOS usage of statfs, it should not stuck in waiting.
90 Severity : enhancement
92 Description: remote_acl support
93 Details : Support ACL-based permission check for remote user.
94 Support setfacl/getfacl for remote user with the utils
95 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
96 the system "{s,g}etfacl" utils.
98 Severity : enhancement
100 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
102 Severity : enhancement
104 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
108 Description: make mgs_setparam() handle fsname containing dash
109 Details : fsname containing a dash does not work with lctl conf_param
111 Severity : enhancement
113 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
115 Severity : enhancement
117 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
121 Frequency : rare, at shutdown
122 Description: access already free / zero obd_namespace.
123 Details : if client_disconnect_export was called without force flag set,
124 and exist connect request in flight, this can produce access to
125 NULL pointer (or already free pointer) when connect_interpret
126 store ocd flags in obd_namespace.
130 Frequency : only at startup
131 Description: not alloc memory with spinlock held.
132 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
133 if any spinlock held.
135 Severity : enhancement
137 Description: make lustre randomly fail allocating memory
138 Details : Make lustre randomly failed allocating memory for testing purpose.
140 Severity : enhancement
142 Description: lost problems with lov objid file
143 Details : Fixes some scability and access to not inited memory problems
144 in work with lov objdid file.
149 Description: lfs find does not continue on file error
150 Details : Continue other files processing when a file/dir is absent.
154 Description: Inconsistent usage of lustre_pack_reply()
155 Details : Standardize the usage of lustre_pack_reply() such that it
156 always generate a CERROR on failure.
159 Frequency : occasional
161 Description: MDS hang or stay in waiting lock
162 Details : If client receive lock with CBPENDING flag ldlm need send lock
163 cancel as separate rpc, to avoid situation when cancel request
164 can't processed due all i/o threads stay in wait lock.
168 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
172 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
176 Description: Update to SLES9 kernel-2.6.5-7.287.3.
180 Description: Update to RHEL4 latest kernel.
182 Severity : enhancement
184 Description: Build SLES10 patchless client fails
185 Details : The configure was broken by run ./configure with
186 --with-linux-obj=.... argument for patchless client. When the
187 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
188 can't search header adequately. Use absolute path such as
189 -I($LINUX)/include instead.
193 Description: interrupt oig_wait produce painc on resend.
194 Details : brw_redo_request can be used for resend requests from ptlrpcd and
195 private set, and this produce situation when rq_ptlrpcd_data not
196 copyed to new allocated request and triggered LBUG on assert
197 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
198 ptlrpcd set if request is changed and can be safety changed to use
203 Description: LASSERT_{REQ,REP}SWAB macros are buggy
204 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
205 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
206 these macros and replace them with their difinitions instead.
210 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
211 Details : Update which_patch & target file for SLES10 latest kernel.
214 Description: Remove client patches from SLES 10 kernel.
215 Details : This causes SLES 10 clients to behave as patchless clients
216 even on a Lustre-patched (server) kernel.
218 Severity : enhancement
220 Description: self-adjustable client's lru lists
221 Details : use adaptive algorithm for managing client cached locks lru
222 lists according to current server load, other client's work
223 pattern, memory activities, etc. Both, server and client
224 side namespaces provide number of proc tunables for controlling
227 Severity : enhancement
229 Description: light-weight GSS support
230 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
231 and reduce performance overhead.
233 Severity : enhancement
235 Description: Linux keyring support
236 Details : Support using service of Linux keyring for Lustre GSS internal
237 context refresh/cache mechanism.
241 Description: Fix errors in lfs documentation
242 Details : Fixes man pages
246 Description: don't use GFP_* in generic Lustre code.
247 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
251 Description: obdclass is limited by single OBD_ALLOC(idarray)
252 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
256 Description: warnings with build patchless client with vanila 2.6.19 and up
257 Details : change old ctl_table style and replace ctl_table/ctl_table_header
258 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
262 Frequency : Only for SLES
263 Description: sanity_quota fail test_1
264 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
265 file, which makes TSTID[2] inunique.
269 Frequency : Only for SLES9
270 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
271 Details : Update target/ChangeLog/which_patch .
275 Frequency : Only for test_10 in sanity-quota.sh
276 Description: a bug in quota test code
277 Details : checking a test flag in a wrong place
281 Frequency : Only for test_13 in sanity-quota.sh
282 Description: a bug in quota test script
283 Details : shouldn't check overfull allocated quota
287 Description: jbd statistics
288 Details : Port older jbd statistics patch for sles10
290 Severity : enhancement
292 Description: Add printing inode info into message about error in writepage.
297 Description: reply_lock_interpret crash due to race with it and lock cancel.
298 Details : Do not replay locks that are being cancelled. Do not reference
299 locks by their address during replay, just by their handle.
303 Frequency : When flocks are used.
304 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
305 Details : Flock locks might destroy just granted lock if it could be merged
306 with another existing flock, this is done in completion handler,
307 so teach ldlm_cli_enquque_fini that this is a valid case for
313 Description: Oops in read and write path when failing to allocate lock.
314 Details : Check if lock allocation failed and return error back.
318 Description: lstripe command fails for valid OST index
319 Details : The stripe offset is compared to lov->desc.ld_tgt_count
320 instead of lov->desc.ld_active_tgt_count.
324 Description: sanity.sh failed test 103
325 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
330 Description: df doesn't work properly if diskfs blocksize != 4K
331 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
335 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
336 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
340 Description: obd_health_check_timeout too short
341 Details : set obd_health_check_timeout as 1.5x of obd_timeout
345 Description: llapi_file_create() does not allow some changes
346 Details : add llapi_file_open() that allows specifying the mode and
347 open flags, and also returns an open file handle.
351 Description: merge and cleanup kernel patches.
352 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
356 Description: Add journal checksum support.(Kernel part)
357 Details : The journal checksum feature adds two new flags i.e
358 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
359 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
360 indicates that the commit block contains the checksum for
361 the blocks described by the descriptor blocks. Now commit
362 record can be sent to disk without waiting for descriptor
363 blocks to be written to disk. This behavior is controlled
364 using JBD2_FEATURE_ASYNC_COMMIT flag.
368 Description: OSS needs mutliple precreate threads
369 Details : Add ability to start more than one create thread per OSS.
373 Description: Sanity.sh test_65k failed: setstripe should have succeeded
374 Details : execute lfs setstripe on client
378 Description: mds_obd_create error creating tmp object
379 Details : When the user sets quota on root, llog will be affected and can't
380 create files and write files.
383 Frequency : Always on ia64 patchless client, and possibly others.
385 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
386 Details : This allows the patchless client to be loaded on architectures
391 Description: RedHat Update kernel for RHEL5
392 Details : Add the kernel config file for RHEL5.
396 Description: RedHat Update kernel for RHEL5
397 Details : Modify the kernel config file more closer RHEL5.
401 Description: Build failure against Centos5 (RHEL5)
402 Details : Define PAGE_SIZE when it isn't present.
406 Description: test_7 on sanity-quota.sh can't be used on separate machines
407 Details : add facet to handle it
411 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
412 Details : with NFS, the anon dentry's parent was set to itself in
413 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
414 corresponding dentry other than use rec->ur_name.
416 Severity : enhancement
418 Description: lfs setstripe enhancement
419 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
423 Description: enable data checksumming by default
424 Details : enable checksum by default, allow --disable-checksum
425 configure option and "-o nochecksum" mount option. Checksums
426 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
430 Description: lustre support for RHEL5
431 Details : Add support for RHEL5.
435 Description: Client eviction due to failover config
436 Details : after a connection loss, the lustre client should attempt to
437 reconnect to the last active server first before trying the
438 other potential connections.
442 Description: when mds and osts use different quota unit(32bit and 64bit),
443 quota will be released repeatly.
444 Details : void sending multiple quota reqs to mds, which will keep the
445 status between the reqs.
449 Description: osts not allocated evenly to files
450 Details : change the condition to increase offset_idx
453 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
455 Description: Data corruption for OSTs that are formatted larger than 2TB
457 Details : When generating the bio request for lustre file writes the
458 sector number would overflow a temporary variable before being
459 used for the IO. The data reads correctly from Lustre (which
460 will overflow in a similar manner) but other file data or
461 filesystem metadata may be corrupted in some cases.
465 Description: Tune the kernel for good SCSI performance.
466 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
467 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
472 Description: rewrite ext2-derived code in obdclass/uuid.c
473 Details : rewrite inherited code (uuid parsing code from ext2 utils)
474 from scratch preserving functionality.
478 Description: "lfs find -obd UUID" prints directories
479 Details : "lfs find -obd UUID" will return all directory names instead
480 of just file names. It is incorrect because the directories
481 do not reside on the OSTs.
485 Description: use i_size_read and i_size_write in 2.6 port
486 Details : replace inode->i_size access with i_size_read/write()
489 Frequency : when removing large files
491 Description: scheduling issue during removal of large Lustre files
492 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
493 It causes scheduling issues when removing large files (17TB in the
497 Frequency : only with liblustre clients on XT3
499 Description: evictions taking too long
500 Details : allow llrd to evict clients directly on OSTs
503 Frequency : only on ppc
505 Description: /proc/fs/lustre/devices broken on ppc
506 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
507 We should make sure the type of 'index' is loff_t and then cast
508 explicitly as needed below. Do not assign an explicitly cast
512 Frequency : only for rhel5
514 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
515 Details : Modify the target file & which_kernel.
517 Severity : enhancement
519 Description: omit set fsid for export NFS
520 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
521 as FS_REQUIRES_DEV for avoid problems with generate fsid.
525 Frequency : Always, for kernels after 2.6.16
526 Description: Fix warning idr_remove called for id=.. which is not allocated.
527 Details : Last kernels save old s_dev before kill super and not allow
528 to restore from callback - restore it before call kill_anon_super.
532 Description: buffer overruns could theoretically occur
533 Details : llapi_semantic_traverse() modifies the "path" argument by
534 appending values to the end of the origin string, and a
535 overrun may occur. Adding buffer overrun check in liblustreapi.
539 Description: Fix error on 'ls .' at the top of the Lustre mount.
540 Details : Don't revalidate dentry if it is a root dentry.
544 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
545 Details : Modify vm-tunables-rhel4.patch.
549 Description: Kernel config for 2.6.18-vanilla.
550 Details : Modify targets/2.6-vanilla.target.in.
551 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
552 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
553 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
554 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
558 Description: improve handling recoverable errors
559 Details : if request processig with error which can be recoverable on server
560 request should be resend, otherwise page released from cache and
565 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
566 Details : Modify target file & which_patch.
567 A flaw was found in the IA32 system call emulation provided
568 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
569 value could be stored in the %RAX register, which could trigger an
570 out-of-bounds system call table access. An untrusted local user
571 could exploit this flaw to run code in the kernel
572 (ie a root privilege escalation). (CVE-2007-4573).
576 Description: O_DIRECT bypasses client statistics.
577 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
578 read_ahead_stats were updated. copied the stats section from
579 osc_send_oap_rpc() into async_internal().
583 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
587 Description: change order of libsysio includes
588 Details : '#include sysio.h' should always come before '#include xtio.h'
592 Description: handle "serious error: objid * already exists" more gracefully
593 Details : If LAST_ID value on disk is smaller than the objects existing in
594 the O/0/d* directories, it indicates disk corruption and causes an
595 LBUG(). If the object is 0-length, then we should use the existing
596 object. This will help to avoid a full fsck in most cases.
598 Severity : enhancement
600 Description: adapt the lustre_config script to support the upgrade case
601 Details : Add "-u" option for lustre_config script to support upgrading 1.4
602 server targets to 1.6 in parallel.
606 Description: To avoid grant space > avaible space when the disk is almost
607 full. Without this patch you might see the error "grant XXXX >
608 available" or some LBUG about grant, when the disk is almost
610 Details : In filter_check_grant, for non_grant cache write, we should
611 check the left space by if (*left > ungranted + bytes), instead
612 of (*left > ungranted), because only we are sure the left space
613 is enough for another "bytes", then the ungrant space should be
614 increase. In client, we should update cl_avail_grant only there
615 is OBD_MD_FLGRANT in the reply.
619 Description: Update RHEL 4 kernel to fix local root privilege escalation.
620 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
621 described in CVE-2007-4573. This problem could allow untrusted
622 local users to gain root access.
625 Frequency : when using O_DIRECT and quotas
627 Description: Incorrect file ownership on O_DIRECT output files
628 Details : block usage reported by 'lfs quota' does not take into account
629 files that have been written with O_DIRECT.
634 Description: touch file failed when fs is not full
635 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
636 otherwise we can get ENOSP while fs is not full.
640 Description: parallel lock callbacks
641 Details : Instead of sending blocking and completion callbacks as separated
642 requests, adding them to a set and sending in parallel.
645 Frequency : only for Cray XT3
646 Bugzilla : 12829/13455
647 Description: Changing primary group doesn't change the group lustre assigns to
649 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
650 overridden with the primary group provided by the group upcall,
651 whereas the supplied fsgid can be trusted if it is in the list of
652 supplementary groups returned by the group upcall.
654 Severity : enhancement
656 Description: Allow masking D_WARNING, D_ERROR messages from console
657 Details : Console messages can now be disabled via lnet.printk.
661 Description: User code with malformed file open parameter crashes client node
662 Details : Before packing join_file req, all the related reference should be
663 checked carefully in case some malformed flags cause fake
664 join_file req on client.
668 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
669 Details : ldlm_completion_ast() assumes that a lock is granted when the req
670 mode is equal to the granted mode. However, it should also check
671 that LDLM_FL_CP_REQD is not set.
675 Description: Heavy nfs access might result in deadlocks
676 Details : After ELC code landed, it is now improper to enqueue any mds
677 locks under och_sem, because enqueue might want to decide to
678 cancel open locks for same inode we are holding och_sem for.
682 Description: Client eviction while running blogbench
683 Details : A lot of unlink operations with concurrent I/O can lead to a
684 deadlock causing evictions. To address the problem, the number of
685 oustanding OST_DESTROY requests is now throttled to
686 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
687 callbacks are processed in priority.
691 Description: enable ACLs on MDS by default
692 Details : ACLs must be enabled on MDS by default.
695 Frequency : PPC/PPC64 only
697 Description: conflicts between asm-ppc64/types.h and lustre_types.h
698 Details : fix duplicated definitions between asm-ppc64/types.h and
699 lustre_types.h on PPC.
701 --------------------------------------------------------------------------------
703 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
705 * Support for kernels:
706 2.6.5-7.283 (SLES 9),
707 2.6.9-55.EL (RHEL 4),
708 2.6.16.46-0.14 (SLES 10),
709 2.6.18.8 vanilla (kernel.org)
710 * Client support for unpatched kernels:
711 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
712 2.6.16 - 2.6.22 vanilla (kernel.org)
713 * Due to recently discovered recovery problems, we do not recommend
714 using patchless RHEL 4 clients with this or any earlier release.
715 * Recommended e2fsprogs version: 1.39.cfs8
716 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
717 * Starting with this release, the ldiskfs backing filesystem required
718 by Lustre is now in its own package, lustre-ldiskfs. This package
719 should be installed. It is versioned separately from Lustre and
720 may be released separately in future.
724 Description: block reactivating mgc import until all deactivates complete
725 Details : Fix race when failing back MDT/MGS to itself (testing)
728 Frequency : at statup only
730 Description: mds_lov_synchronize race leads to various problems
731 Details : simultaneous MDT->OST connections at startup can cause the
732 sync to abort, leaving the OSC in a bad state.
734 Severity : enhancement
736 Description: add optional extra BUILD_VERSION info
737 Details : add a new environment variable (namely LUSTRE_VERS) which allows
738 to override the lustre version.
741 Frequency : 2.6.18 servers only
743 Description: ll_kern_mount() doesn't release the module reference
744 Details : The ldiskfs module reference count never drops down to 0
745 because ll_kern_mount() doesn't release the module reference.
750 Description: server LBUG when using old ost_num_threads parameter
751 Details : Accept the old ost_num_threads parameter but warn that it
752 is deprecated, and fix an off-by-one error that caused an LBUG.
757 Description: Transient SCSI error results in persistent IO issue
758 Details : iobuf->dr_error is not reinitialized to 0 between two
762 Frequency : sometimes when underlying device returns I/O errors
764 Description: OSTs not going read-only during write failures
765 Details : OSTs are not remounted read-only when the journal commit threads
766 get I/O errors because fsfilt_ext3 calls journal_start/stop()
767 instead of the ext3 wrappers.
771 Description: poor connect scaling with increasing client count
772 Details : Don't run filter_grant_sanity_check for more than 100 exports
773 to improve scaling for large numbers of clients.
776 Frequency : SLES10 only
778 Description: sanity-quota.sh quotacheck failed: rc = -22
779 Details : Quotas cannot be enabled on SLES10.
782 Frequency : liblustre clients only
784 Description: getdirentries does not give error when run on compute nodes
785 Details : getdirentries does not fail when the size specified as an argument
786 is too small to contain at least one entry
788 Severity : enhancement
790 Description: Add LNET router traceability for debug purposes
791 Details : If a checksum failure occurs with a router as part of the
792 IO path, the NID of the last router that forwarded the bulk data
793 is printed so it can be identified.
798 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
799 Details : Due to a race condition, liblustre clients were occasionally
802 Severity : enhancement
804 Description: lfs setstripe use optional parameters instead of postional
807 Severity : enhancement
809 Description: Nanosecond timestamp support for ldiskfs
810 Details : The on-disk ldiskfs filesystem has added support for nanosecond
811 resolution timestamps. There is not yet support for this at
812 the Lustre filesystem level.
815 Frequency : during server recovery
817 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
818 Details : request with rq_no_resend flag not awake l_wait_event if they get a
822 Frequency : nfs export on patchless client
824 Description: connectathon hang when test nfs export over patchless client
825 Details : Disconnected dentry cannot be found with lookup, so we do not need
826 to unhash it or make it invalid
829 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
830 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
831 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
836 Description: open req refcounting wrong on reconnect
837 Details : If reconnect happened between getting open reply from server and
838 call to mdc_set_replay_data in ll_file_open, we will schedule
839 replay for unreferenced request that we are about to free.
840 Subsequent close will crash in variety of ways.
841 Check that request is still eligible for replay in
842 mdc_set_replay_data().
847 Description: disable writes to filesystem when reading health_check file
848 Details : the default for reading the health_check proc file has changed
849 to NOT do a journal transaction and write to disk, because this
850 can cause reads of the /proc file to hang and block HA state
851 checking on a healthy but otherwise heavily loaded system. It
852 is possible to return to the previous behaviour during configure
853 with --enable-health-write.
855 Severity : enhancement
857 Description: 64-bit inode version
858 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
859 to the inode. This will be required for version-based recovery.
864 Description: MDS fails to start if a duplicate client export is detected
865 Details : in some rare cases it was possible for a client to connect to
866 an MDS multiple times. Upon recovery the MDS would detect this
867 and fail during startup. Handle this more gracefully.
869 Severity : enhancement
871 Description: Add -o localflock option to simulate old noflock
873 Details : This will achieve local-only flock/fcntl locks
879 Description: log_commit_thread vs filter_destroy race leads to crash
880 Details : Take import reference before releasing llog record semaphore
885 Description: Wrong request locking in request set processing
886 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
887 imp_delayed_list, in this place should be used imp_lock.
890 Frequency : when reconnection
892 Description: Grant Leak when osc reconnect to OST
893 Details : When osc reconnect ost, OST(filter) should check whether it
894 should grant more space to client by comparing fed_grant and
895 cl_avail_grant, and return the granted space to client instead
896 of "new granted" space, because client will call osc_init_grant
897 to update the client grant space info.
900 Frequency : when client reconnect to OST
902 Description: Grant Leak when osc do resend and replay bulk write
903 Details : When osc reconnect to OST, OST(filter)should clear grant info of
904 bulk write request, because the grant info while be sync between
905 OSC and OST when reconnect, and we should ignore the grant info
906 these of resend/replay write req.
911 Description: Grant space more than avaiable left space sometimes.
912 Details : When then OST is about to be full, if two bulk writing from
913 different clients came to OST. Accord the avaliable space of the
914 OST, the first req should be permitted, and the second one
915 should be denied by ENOSPC. But if the seconde arrived before
916 the first one is commited. The OST might wrongly permit second
917 writing, which will cause grant space > avaiable space.
920 Frequency : when client is evicted
922 Description: Grant might be wrongly erased when osc is evicted by OST
923 Details : when the import is evicted by server, it will fork another
924 thread ptlrpc_invalidate_import_thread to invalidate the
925 import, where the grant will be set to 0. While the original
926 thread will update the grant it got when connecting. So if
927 the former happened latter, the grant will be wrongly errased
928 because of this race.
933 Description: Checking Stale with correct fid
934 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
935 to check whether it is stale, because sometimes, we want the
936 enqueue happened anyway, and op_data.fid2 will not be initialized.
938 Severity : enhancement
940 Description: update patchless client
941 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
944 Frequency : only with 2.4 kernel
946 Description: random memory corruption
947 Details : size of struct ll_inode_info is to big for union inode.u and this
948 can be cause of random memory corruption.
953 Description: Memory leak in recovery
954 Details : Lov_mds_md was not free in an error handler in mds_create_object.
955 It should also check obd_fail before fsfilt_start, otherwise if
956 fsfilt_start return -EROFS,(failover mds during mds recovery).
957 then the req will return with repmsg->transno = 0 and rc = EROFS.
958 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
959 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
960 be freed no matter whether fsfilt_commit success or not.
963 Frequency : only with huge count clients
965 Description: Prevents from taking the superblock lock in llap_from_page for
967 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
968 for llap_from_page call in ll_removepage prevents from taking the
969 superblock lock for a soon died page.
974 Description: Not check open intent error before release open handle
975 Details : in some rare cases, the open intent error is not checked before
976 release open handle, which may cause
977 ASSERTION(open_req->rq_transno != 0), because it tries to release
978 the failed open handle.
983 Description: Set cat log bitmap only after create log success.
984 Details : in some rare cases, the cat log bitmap is set too early. and it
985 should be set only after create log success.
989 Description: Accessing a block bevice can re-enable I/O when Lustre is
990 tearing down a device.
991 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
996 Description: service threads may hog cpus when there are a lot of requests
997 Details : Insert cond_resched to give other threads a chance to use some CPU
1002 Description: the cat log was not initialized in recovery
1003 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1004 unlink log on mds will not be initialized until mds post
1005 recovery. And also in mds post recovery, the unlink log will
1006 initialization will be done asynchronausly, so there will be race
1007 between add unlink log and unlink log initialization.
1011 Description: brw_stats were being printed incorrectly
1012 Details : brw_stats were being printed as log2 but all of them were not
1013 recorded as log2. Also remove some code duplication arising from
1014 filter_tally_{read,write}.
1018 Frequency : rare, only in recovery.
1019 Description: ASSERTION(req->rq_type != LI_POISON) failed
1020 Details : imp_lock should be held while iterating over imp_sending_list for
1021 prevent destroy request after get timeout in ptlrpc_queue_wait.
1025 Description: replay-single.sh test 52 fails
1026 Details : A lock's skiplist need to be cleanup when it being unlinked
1027 from its resource list.
1031 Description: Short directio read returns full requested size rather than
1033 Details : Direct I/O operations should return actual amount of bytes
1034 transferred rather than requested size.
1036 Severity : enhancement
1038 Description: metadata RPC reduction (e.g. for rm performance)
1039 Details : decrease the amount of synchronous RPC between clients and servers
1040 by canceling conflicing lock before the operation on the client
1041 and packing thier handles into the main operation RPC to server.
1043 Severity : enhancement
1045 Description: Async OSC create to avoid the blocking unnecessarily.
1046 Details : If a OST has no remain object, system will block on the creating
1047 when need to create a new object on this OST. Now, ways use
1048 pre-created objects when available, instead of blocking on an
1049 empty osc while others are not empty. If we must block, we block
1050 for the shortest possible period of time.
1052 Severity : enhancement
1054 Description: refine locking for avoid write wrong info into lov_objid file
1055 Details : fix possible races with add new target and write/update data in
1058 --------------------------------------------------------------------------------
1060 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1065 Frequency : on some architectures
1067 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1068 Details : Uninitialized flags sometimes cause configuration commands to
1072 Frequency : patchless clients only
1074 Description: missing __iget() symbol export
1075 Details : The __iget() symbol export is missing. To avoid the need for
1076 this on patchless clients the deathrow inode reaper is turned
1077 off, and we depend on the VM to clean up old inodes. This
1078 dependency was during via the fix for bug 12181.
1080 --------------------------------------------------------------------------------
1082 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1084 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1085 INTEROPERATE with older versions automatically. In many cases a
1086 special upgrade step is needed. Please read the
1087 user documentation before upgrading any part of a live system.
1088 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1089 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1090 * WARNING: Lustre configuration and startup changes are required with
1091 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1097 Severity : enhancement
1099 Description: MountConf
1100 Details : Lustre configuration is now managed via mkfs and mount
1101 commands instead of lmc and lconf. New obd types (MGS, MGC)
1102 are added for dynamic configuration management. See
1103 https://mail.clusterfs.com/wikis/lustre/MountConf for
1106 Severity : enhancement
1108 Description: dynamic OST addition
1109 Details : OSTs can now be added to a live filesystem
1111 Severity : enhancement
1113 Description: startup order invariance
1114 Details : MDTs and OSTs can be started in any order. Clients only
1115 require the MDT to complete startup.
1117 Severity : enhancement
1119 Description: parallel, asynchronous orphan cleanup
1120 Details : orphan cleanup is now performed in separate threads for each
1121 OST, allowing parallel non-blocking operation.
1123 Severity : enhancement
1125 Description: optimized stripe assignment
1126 Details : stripe assignments are now made based on ost space available,
1127 ost previous usage, and OSS previous usage, in order to try
1128 to optimize storage space and networking resources.
1130 Severity : enhancement
1132 Description: Permanently set tunables
1133 Details : All writable /proc/fs/lustre tunables can now be permanently
1134 set on a per-server basis, at mkfs time or on a live system.
1136 Severity : enhancement
1138 Description: Lustre message v2
1139 Details : Add lustre message format v2.
1141 Severity : enhancement
1143 Description: client OST exclusion list
1144 Details : Clients can be started with a list of OSTs that should be
1145 declared "inactive" for known non-responsive OSTs.
1149 Description: SPEC SFS validation failure on NFS v2 over lustre.
1150 Details : Changes the blocksize for regular files to be 2x RPC size,
1151 and not depend on stripe size.
1153 Severity : enhancement
1155 Description: Multiple MD RPCs in flight.
1156 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1157 To avoid overly-overloading MDS, introduce a limit on number of
1158 MDS RPCs in flight for a single client and add /proc controls
1159 to adjust this limit.
1161 Severity : enhancement
1163 Description: client read/write statistics
1164 Details : Add client read/write call usage stats for performance
1165 analysis of user processes.
1166 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1167 file access. extents_stats shows chunk size distribution.
1168 extents_stats_per_process show chunk size distribution per
1171 Severity : enhancement
1173 Description: mds statistics
1174 Details : Add detailed mds operations statistics in
1175 /proc/fs/lustre/mds/*/stats.
1179 Description: Failure of copying files with lustre special EAs.
1180 Details : Client side always return success for setxattr call for lustre
1181 special xattr (currently only "trusted.lov").
1186 Description: Refcount LNET uuids
1187 Details : The global LNET uuid list grew linearly with every startup;
1188 refcount repeated list entries instead of always adding to
1191 Severity : enhancement
1193 Description: Dynamic service threads
1194 Details : Within a small range, start extra service threads
1195 automatically when the request queue builds up.
1198 Frequency : mixed-endian client/server environments
1200 Description: mixed-endian crashes
1201 Details : The new msg_v2 system had some failures in mixed-endian
1204 Severity : enhancement
1206 Description: Easy OST removal
1207 Details : OSTs can be permanently deactivated with e.g. 'lctl
1208 conf_param lustre-OST0001.osc.active=0'
1210 Severity : enhancement
1212 Description: MGS proc entries
1213 Details : Added basic proc entries for the MGS showing what filesystems
1216 Severity : enhancement
1218 Description: provide MGS failover
1219 Details : Added config lock reacquisition after MGS server failover.
1221 Severity : enhancement
1223 Description: add Linux 2.4 support
1224 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1228 Description: a large application tries to do I/O to the same resource and dies
1229 in the middle of it.
1230 Details : Check the req->rq_arrival time after the call to
1231 ost_brw_lock_get(), but before we do anything about
1232 processing it & sending the BULK transfer request. This
1233 should help move old stale pending locks off the queue as
1234 quickly as obd_timeout.
1237 Frequency : when an incorrect nid is specified during startup
1239 Description: ptlrpc connect to non-existant node causes kernel crash
1240 Details : LNET can't be re-entered from an event callback, which
1241 happened when we expire a message after the export has been
1242 cleaned up. Instead, hand the zombie cleanup off to another
1245 Severity : enhancement
1247 Description: plain/inodebits lock performance improvement
1248 Details : Grouping plain/inodebits in granted list by their request modes
1249 and bits policy, thus improving the performance of search through
1253 Frequency : only if OST filesystem is corrupted
1255 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1256 Details : for a short time RPCs with bulk IO are in the replay list,
1257 but replay of bulk IOs is unimplemented. If the OST filesystem
1258 is corrupted due to disk cache incoherency and then replay is
1259 started it is possible to trip an assertion. Avoid putting
1260 committed RPCs into the replay list at all to avoid this issue.
1263 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1265 Description: System hang on startup
1266 Details : This bug allowed the liblustre (e.g. catamount) client to
1267 return to the app before handling all startup RPCs. This
1268 could leave the node unresponsive to lustre network traffic
1269 and manifested as a server ptllnd timeout.
1271 Severity : enhancement
1273 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1274 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1275 boolean to control whether to print peer debug info when a
1276 client's RPC times out.
1279 Frequency : only for kernels with patches from Lustre below 1.4.3
1281 Description: Remove old rdonly API
1282 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1285 Frequency : only for devices with external journals
1287 Description: Set external device read-only also
1288 Details : During a commanded failover stop, we set the disk device
1289 read-only while the server shuts down. We now also set any
1290 external journal device read-only at the same time.
1293 Frequency : when upgrading from 1.4 while trying to change parameters
1295 Description: The wrong (new) MDC name was used when setting parameters for
1296 upgraded MDT's. Also allows changing of OSC (and MDC)
1297 parameters if --writeconf is specified at tunefs upgrade time.
1300 Frequency : when setting specific ost indicies
1302 Description: QOS code breaks on skipped indicies
1303 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1304 created with --index need not be sequential.
1306 Severity : enhancement
1308 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1309 Details : The uninit_groups feature works in conjunction with the kernel
1310 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1311 up the pass1 processing of e2fsck. This is a read-only feature
1312 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1313 mount filesystems that have had this feature enabled.
1315 Severity : enhancement
1317 Description: Improve multi-block allocation algorithm to avoid fragmentation
1318 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1319 allocation locality and avoid filesystem fragmentation.
1321 ------------------------------------------------------------------------------
1323 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1325 * Support for kernels:
1326 2.4.21-47.0.1.EL (RHEL 3)
1327 2.6.5-7.283 (SLES 9)
1328 2.6.9-42.0.10.EL (RHEL 4)
1329 2.6.12.6 vanilla (kernel.org)
1330 2.6.16.27-0.9 (SLES 10)
1331 * Recommended e2fsprogs version: 1.39.cfs5
1333 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1337 Frequency : occasional, depends on client load and configuration
1338 Bugzilla : 12181, 12203
1339 Description: data loss for recently-modified files
1341 Details : In some cases it is possible that recently written or created
1342 files may not be written to disk in a timely manner (this should
1343 normally be within 30s unless client IO load is very high).
1344 The problem appears as zero-length files or files that are a
1345 multiple of 1MB in size after a client crash or client eviction
1346 that are missing data at the end of the file.
1348 This problem is more likely to be hit on clients where files are
1349 repeatedly created and unlinked in the same directory, clients
1350 have a large amount of RAM, have many CPUs, the filesystem has
1351 many OSTs, the clients are rebooted frequently, and/or the files
1352 are not accessed by other nodes after being written.
1354 The presence of the problem can be detected by looking at
1355 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1356 smaller than the second (nr_unused) then dirty files will not
1357 be flushed automatically to disk. "sync; sleep 10" should be
1358 run several times on the node before unmounting it to update
1359 Lustre (this is also safe to run on nodes without this problem).
1361 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1362 that can cause this same problem, so customers using that kernel
1363 also need to update the kernel in addition to Lustre. In order
1364 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1366 It is normal that files written just before a client crash (less
1367 than 30s) may not yet have been flushed to disk, even for local
1371 Frequency : frequent on thin XT3 nodes
1373 Description: UUID collision on thin XT3 Linux nodes
1374 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1375 randomly, since we relied on an insufficiently-seeded PRNG.
1380 Description: OSS hangs after "All ost request buffers busy"
1381 Details : A deadlock between quota and journal operations caused OSS
1382 hangs after printing "All ost request buffers busy."
1385 Frequency : always on liblustre builds
1387 Description: Cleanup compiler warnings on liblustre
1390 Frequency : always on liblustre builds on XT3
1392 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1396 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1397 > 32000 subdirectories
1398 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1399 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1400 subdirectory count crosses 32000. This will aid e2fsck to
1401 correctly handle more than 32000 subdirectories.
1404 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1406 Description: System hang on startup
1407 Details : This bug allowed the liblustre (e.g. catamount) client to
1408 return to the app before handling all startup RPCs. This
1409 could leave the node unresponsive to lustre network traffic
1410 and manifested as a server ptllnd timeout.
1412 Severity : enhancement
1414 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1415 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1416 boolean to control whether to print peer debug info when a
1417 client's RPC times out.
1422 Description: make O_SYNC working on 2.6 kernels
1423 Details : 2.6 kernels use different method for mark pages for write,
1424 so need add a code to lustre for O_SYNC work.
1429 Description: Failure to close file and release space on NFS
1430 Details : Put inode details into lock acquired in ll_intent_file_open.
1431 Use mdc_intent_lock in ll_intent_open to properly
1432 detect all kind of errors unhandled by mdc_enqueue.
1437 Description: proc file read during shutdown sometimes raced obd removal,
1439 Details : Add lock to prevent obd access after proc file removal.
1442 Frequency : Only for files larger than 4GB on 32-bit clients.
1444 Description: improperly doing page alignment of locks
1445 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1446 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1451 Description: RPCs being resent when they shouldn't be
1452 Details : Some RPCs that should not be resent are being resent. This
1453 can cause inconsistencies in the RPC state machine. Do not
1454 resend such requests.
1457 Frequency : rare, only with NFS export
1459 Description: Crash on NFS re-export node
1460 Details : under very unusual load conditions an assertion is hit in
1461 ll_intent_file_open()
1464 Frequency : only if OST filesystem is corrupted
1466 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1467 Details : for a short time RPCs with bulk IO are in the replay list,
1468 but replay of bulk IOs is unimplemented. If the OST filesystem
1469 is corrupted due to disk cache incoherency and then replay is
1470 started it is possible to trip an assertion. Avoid putting
1471 committed RPCs into the replay list at all to avoid this issue.
1476 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1477 Details : Large single O_DIRECT read and write calls can fail to allocate
1478 a sufficiently large buffer to process the request. In case of
1479 allocation failure the allocation is retried with a smaller
1480 buffer and broken into smaller requests.
1482 Severity : enhancement
1484 Description: Add -o localflock option to simulate old noflock behaviour.
1485 Details : This will achieve local-only flock/fcntl locks coherentness.
1490 Description: versioning check is incomplete
1491 Details : Checking the version difference of client vs. server, report
1492 error if the gap is too big.
1497 Description: add support PG_writeback bit
1498 Details : add support for PG_writeback bit for Lustre, for more carefull
1499 work with page cache in 2.6 kernel. This also fix some deadlocks
1500 and remove hack for work O_SYNC with 2.6 kernel.
1502 Severity : enhancement
1504 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1505 Details : The uninit_groups feature works in conjunction with the kernel
1506 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1507 up the pass1 processing of e2fsck. This is a read-only feature
1508 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1509 mount filesystems that have had this feature enabled.
1511 Severity : enhancement
1513 Description: Improve multi-block allocation algorithm to avoid fragmentation
1514 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1515 allocation locality and avoid filesystem fragmentation.
1517 ------------------------------------------------------------------------------
1519 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1521 * Support for kernels:
1522 2.6.9-42.0.3.EL (RHEL 4)
1523 2.6.5-7.276 (SLES 9)
1524 2.4.21-47.0.1.EL (RHEL 3)
1525 2.6.12.6 vanilla (kernel.org)
1526 2.6.16.21-0.8 (SLES10)
1527 * Recommended e2fsprogs version: 1.39.cfs2-0
1529 * The backwards-compatible /proc/sys/portals symlink has been removed
1530 in this release. Before upgrading, please ensure that you change
1531 any configuration scripts or /etc/sysctl.conf files that access
1532 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1533 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1534 in advance of the upgrade on any system running Lustre 1.4.6 or
1535 newer, since /proc/sys/lnet was added in that version.
1536 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1540 Frequency : only when quota is used
1542 Description: avoid scanning export list for quota master
1543 Details : Change the algorithms to avoid scanning export list in order
1544 to improve the efficiency.
1547 Frequency : MDS failover only, very rarely
1549 Description: "went back in time" messages on mds failover
1550 Details : The greatest transno may be lost when the current operation
1551 finishes with an error (transno==0) and the client's last_rcvd
1552 record is over-written. Save the greatest transno in the
1553 mds_last_transno for this case.
1556 Frequency : always for specific kernels and striping counts
1558 Description: client may get "Matching packet too big" without ACL support
1559 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1560 when trying to access files in certain configurations. The
1561 clients should in fact be denied when mounting because they do
1562 not understand ACLs.
1565 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1567 Description: many clients connecting with IO in progress causes connect timeouts
1568 Details : Avoid synchronous journal commits to avoid delays caused by many
1569 clients connecting/disconnecting when bulk IO is in progress.
1570 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1571 OST_IO_PORTAL to avoid delays behind potentially many pending
1575 Frequency : occasionally with multiple writers to a single file
1577 Description: shared writes to file may result in wrong size reported by stat()
1578 Details : Allow growing of kms when extent lock is cancelled
1581 Frequency : always with random mmap IO to multi-striped file
1583 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1584 Details : Only if the hole is at the end of OST object so that kms is too
1585 small. Fix is to increase kms accordingly in ll_nopage.
1588 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1590 Description: writes to a missing object would leak memory on the OST
1591 Details : If there is an inconsistency between the MDS and OST filesystems,
1592 such that the MDS references an object that doesn't exist, writes
1593 to that object will leak memory due to incorrect cleanup in the
1594 error handling path, eventually running out of memory on the OST.
1599 Description: Creating too long symlink causes lustre errors
1600 Details : Check symlink and name lengths before sending requests to MDS.
1603 Frequency : only if flock is enabled (not on by default)
1605 Description: posix locks not released on fd closure on 2.6.9+
1606 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1607 kernels, this caused such locks not to be released on fd close and
1608 then assertions on fs unmount about still used locks.
1611 Frequency : MDS failover only, very rarely
1613 Description: clients may get ASSERTION(granted_lock != NULL)
1614 Details : When request was taking a long time, and a client was resending
1615 a getattr by name lock request. The were multiple lock requests
1616 with the same client lock handle and
1617 mds_getattr_name->fixup_handle_for_resent_request found one of the
1618 lock handles but later failed with ASSERTION(granted_lock != NULL).
1623 Description: handle->h_buffer_credits > 0, assertion failure
1624 Details : h_buffer_credits is zero after truncate, causing assertion
1625 failure. This patch extends the transaction or creates a new
1629 Frequency : NFS re-export or patchless client
1630 Bugzilla : 11179, 10796
1631 Description: Crash on NFS re-export node (__d_move)
1632 Details : We do not want to hash the dentry if we don't have a lock.
1633 But if this dentry is later used in d_move, we'd hit uninitialised
1634 list head d_hash, so we just do this to init d_hash field but
1635 leave dentry unhashed.
1638 Frequency : NFS re-export or patchless client
1640 Description: NFS exports has problem with symbolic link
1641 Details : lustre client didn't properly install dentry when re-exported
1642 to NFS or running patchless client.
1645 Frequency : NFS re-export or patchless client
1647 Description: Various nfs/patchless fixes.
1648 Details : fixes reuse disconected alias for lookup process - this fixes
1649 warning "find_exported_dentry: npd != pd",
1650 fix permission error with open files at nfs.
1651 fix apply umask when do revalidate.
1654 Frequency : occasional
1656 Description: Crash on NFS re-export node
1657 Details : calling clear_page() on the wrong pointer triggered oops in
1658 generic_mapping_read().
1661 Frequency : rarely, using O_DIRECT IO
1663 Description: unaligned directio crashes client with LASSERT
1664 Details : check for unaligned buffers before trying any requests.
1667 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1669 Description: stale data returned from RAID cache
1670 Details : If only a small amount of IO is done to the RAID device before
1671 reading it again it is possible to get stale data from the RAID
1672 cache instead of reading it from disk.
1675 Frequency : always for sles10 kernel
1677 Description: sles10 support
1678 Details : ll_follow_link: compile fixes and using of nd_set_link
1679 under newer kernels.
1682 Frequency : depends on arch, kernel and compiler version, always on sles10
1685 Description: recursive or deep enough symlinks cause stack overflow
1686 Details : getting rid of large stack-allocated variable in
1690 Frequency : depends on hardware
1692 Description: lustre write performance loss in the SLES10 kernel
1693 Details : the performance loss is caused by using of write barriers in the
1694 ext3 code. The SLES10 kernel turns barrier support on by
1695 default. The fix is to undo that change for ldiskfs.
1697 ------------------------------------------------------------------------------
1699 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1701 * Support for kernels:
1702 2.6.9-42.0.3EL (RHEL 4)
1703 2.6.5-7.276 (SLES 9)
1704 2.4.21-47.0.1.EL (RHEL 3)
1705 2.6.12.6 vanilla (kernel.org)
1709 Frequency : quota enabled and large files being deleted
1711 Description: releasing more than 4GB of quota at once hangs OST
1712 Details : If a user deletes more than 4GB of files on a single OST it
1713 will cause the OST to spin in an infinite loop. Release
1714 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1719 Description: statfs data retrieved from /proc may be stale or zero
1720 Details : When reading per-device statfs data from /proc, in the
1721 {kbytes,files}_{total,free,avail} files, it may appear
1722 as zero or be out of date.
1725 Frequency : systems with MD RAID1 external journal devices
1727 Description: lconf's call to blkid is confused by RAID1 journal devices
1728 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1729 of returning all block devices that match the journal UUID.
1732 Frequency : always, for aggregate stripe size over 4GB
1734 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1735 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1736 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1739 Frequency : always if "lfs find" used on a local file/directory
1741 Description: "lfs find" segfaults if used on a local file/directory
1742 Details : The case where a directory component was not specified wasn't
1743 handled correctly. Handle this properly.
1746 Frequency : always on ppc64
1748 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1749 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1750 perform endianness conversion.
1753 Frequency : rarely (truncate to non-zero file size after write under load)
1754 Bugzilla : 10730, 10687
1755 Description: Files padded with zeros to next 4K multiple
1756 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1757 it is possible that files that are truncated to a non-zero size
1758 immediately after being written are filled with zero bytes beyond
1759 the truncated size. No file data is lost.
1761 Severity : enhancement
1763 Description: Allow recovery/failover for liblustre clients.
1764 Details : liblustre clients were unaware of failover configurations until
1767 Severity : enhancement
1769 Description: user file locks should fail when not mounting with flock option
1770 Details : Set up an error-returning stub in ll_file_operations.lock field
1771 to prevent incorrect behaviour when client is mounted without
1772 flock option. Also, set up properly f_op->flock field for
1776 Frequency : always on ia64
1778 Description: "lfs df" loops on printing out MDS statfs information
1779 Details : The obd_ioctl_data was not initialized and in some systems
1780 this caused a failure during the ioctl that did not return
1781 an error. Initialize the struct and return an error on failure.
1784 Frequency : SLES 9 only
1786 Description: Error of copying files with lustre special EAs as root
1787 Details : Client side always return success for setxattr call for lustre
1788 special xattr (currently only "trusted.lov").
1791 Frequency : rarely on clusters with both ia64+i386 clients
1793 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1794 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1795 clients are doing concurrent writes to the same file it is
1796 possible that the ia64 clients may LASSERT because the OST
1797 extent locks are not PAGE_SIZE aligned. Ensure that grown
1798 locks are always aligned on the request boundary.
1801 Frequency : specific use, occasional
1803 Description: Overwriting in use executable truncates on-disk binary image
1804 Details : If one node attempts to overwrite an executable in use by
1805 another node, we now correctly return ETXTBSY instead of
1806 truncating the file.
1808 Severity : enhancement
1810 Description: Async OSC create to avoid the blocking unnecessarily.
1811 Details : If a OST has no remain object, system will block on the creating
1812 when need to create a new object on this OST. Now, ways use
1813 pre-created objects when available, instead of blocking on an
1814 empty osc while others are not empty. If we must block, we block
1815 for the shortest possible period of time.
1820 Description: chmod on Lustre root is propagated to other clients
1821 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1822 invalid by the follow_mount time.
1825 Frequency : liblustre clients only
1827 Description: Race in 'instant cancel' lock handling could lead to such locks
1828 never to be granted in case of SMP MDS
1829 Details : Do not destroy not yet granted but cbpending locks in
1833 Frequency : replay/resend of open
1835 Description: non null lock assetion failure in mds_intent_policy
1836 Details : Trying to replay/resend lockless open requests resulted in
1837 mds_open() returning 0 with no lock. Now it sets a flag if
1838 a lock is going to be returned.
1840 Severity : enhancement
1842 Description: Checksum enhancements
1843 Details : New checksum enhancements allow for resending RPCs that failed
1846 Severity : enhancement
1848 Description: Tunables on number of dirty pages in cacche
1849 Details : Allow to set limit on number of dirty pages cached.
1854 Description: client crash on unmount - lock still has references
1855 Details : In some error handling cases it was possible to leak a lock
1856 reference on a client while accessing a file. This was not
1857 harmful to the client during operation, but would cause the
1858 client to crash when the filesystem is unmounted.
1861 Frequency : specific case, rare
1863 Description: ETXTBSY on mds though file not in use by client
1864 Details : ETXTBSY is no longer incorrectly returned when attempting to
1865 chmod or chown a directory that the user previously tried to
1866 execute or a currently-executing binary.
1869 Frequency : extremely rare except on liblustre-based clients
1871 Description: Lustre space not freed when files are deleted
1872 Details : Clean up open-unlinked files after client eviction. Previously
1873 the unlink was skipped and the files remained as orphans.
1878 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1879 Details : In some cases it was possible to send a blocking callback to
1880 a client doing a glimpse, even though that client didn't get
1881 a lock granted. When the glimpse lock is cancelled on the OST
1882 the freed lock is left on the waiting list and corrupted the list.
1885 Frequency : all core dumps
1887 Description: Broke core dumps to lustre
1888 Details : Negative dentry may be unhashed if parent does not have UPDATE
1889 lock, but some callers, e.g. do_coredump, expect dentry to be
1890 hashed after successful create, hash it in ll_create_it.
1892 ------------------------------------------------------------------------------
1894 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1896 * Support for kernels:
1897 2.6.9-42.0.2.EL (RHEL 4)
1898 2.6.5-7.276 (SLES 9)
1899 2.4.21-40.EL (RHEL 3)
1900 2.6.12.6 vanilla (kernel.org)
1904 Frequency : always on RHEL 3
1906 Description: Number of open files grows over time
1907 Details : The number of open files grows over time, whether or not
1908 Lustre is started. This was due to a filp leak introduced
1909 by one of our kernel patches.
1911 ------------------------------------------------------------------------------
1913 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1915 * Support for kernels:
1916 2.6.9-42.EL (RHEL 4)
1917 2.6.5-7.276 (SLES 9)
1918 2.4.21-40.EL (RHEL 3)
1919 2.6.12.6 vanilla (kernel.org)
1924 Bugzilla : 5719, 9635, 9792, 9684
1925 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1926 Details : If a server is under heavy load and cannot reply to new
1927 connection requests before the client resends the (re)connect,
1928 the connection handling code can behave badly if two service
1929 threads are concurrently handing separate (re)connections from
1930 the same client. Add better locking to the connection handling
1931 code, and ensure that only a single connection will be processed
1932 for a given client UUID, even if the lock is dropped.
1934 Severity : enhancement
1936 Description: add TCP zero-copy support to kernel
1937 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1938 sends if the hardware supports scatter-gather and checksumming.
1939 This allows socklnd to do client-write and server-read more
1940 efficiently and reduce CPU utilization from skbuf copying.
1943 Frequency : only if NFS exporting from client
1945 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1946 Details : If setting attributes on a file created under NFS that had
1947 never been opened it would be possible to oops the client
1948 if the file had no objects.
1951 Frequency : always for liblustre
1953 Description: liblustre client does MDS+OSTs setattr RPC for each write
1954 Details : When doing a write from a liblustre client, the client
1955 incorrectly issued an RPC to the MDS and each OST the file was
1956 striped over in order to update the timestamps. When writing
1957 with small chunks and many clients this could overwhelm the MDS
1958 with RPCs. In all cases it would slow down the write because
1959 these RPCs are unnecessary.
1961 Severity : enhancement
1963 Description: allow number of MDS service threads to be changed at module load
1964 Details : It is now possible to change the number of MDS service threads
1965 running. Adding "options mds mds_num_threads={N}" to the MDS's
1966 /etc/modprobe.conf will set the number of threads for the next
1967 time Lustre is restarted (assuming the "mds" module is also
1968 reloaded at that time). The default number of threads will
1969 stay the same, 32 for most systems.
1974 Description: OST crash if filesystem is unformatted or corrupt
1975 Details : If an OST is started on a device that has never been formatted
1976 or if the filesystem is corrupt and cannot even mount then the
1977 error handling cleanup routines would dereference a NULL pointer.
1982 Description: NULL pointer deref in llap_from_page.
1983 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1984 incorrect) mapping if the page was truncated/reclaimed while it was
1985 searched for. Check for this condition and skip such pages when
1986 doing readahead. Introduce extra check to llap_from_page() to
1987 verify page->mapping->host is non-NULL (so page is not anonymous).
1990 Frequency : Sometimes when using sys_sendfile
1992 Description: "page not covered by a lock" warnings from ll_readpage
1993 Details : sendfile called ll_readpage without right page locks present.
1994 Now we introduced ll_file_sendfile that does necessary locking
1995 around call to generic_file_sendfile() much like we do in
1999 Frequency : with certain MDS communication failures at client mount time
2001 Description: NULL pointer deref after failed client mount
2002 Details : a client connection request may delayed by the network layer
2003 and not be sent until after the PTLRPC layer has timed out the
2004 request. If the client fails the mount immediately it will try
2005 to clean up before the network times out the request. Add a
2006 reference from the request import to the obd device and delay
2007 the cleanup until the network drops the request.
2010 Frequency : occasionally during client (re)connect
2012 Description: assertion failure during client (re)connect
2013 Details : processing a client connection request may be delayed by the
2014 client or server longer than the client connect timeout. This
2015 causes the client to resend the connection request. If the
2016 original connection request is replied in this interval, the
2017 client may trip an assertion failure in ptlrpc_connect_interpret()
2018 which thought it would be the only running connect process.
2021 Frequency : only with obd_echo servers and clients that are rebooted
2023 Description: kernel BUG accessing uninitialized data structure
2024 Details : When running an obd_echo server it did not start the ping_evictor
2025 thread, and when a client was evicted an uninitialized data
2026 structure was accessed. Start the ping_evictor in the RPC
2027 service startup instead of the OBD startup.
2029 Severity : enhancement
2030 Bugzilla : 10193 (patchless)
2031 Description: Remove dependency on various unexported kernel interfaces.
2032 Details : No longer need reparent_to_init, exit_mm, exit_files,
2033 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2036 Frequency : rare (only users of deprecated and unsupported LDAP config)
2038 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2039 Details : LDAP apparently contains 'lustreName' attributes instead of
2040 'name'. A simple remapping of the name is sufficient.
2043 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2045 Description: waiting_locks_callback trips kernel BUG if client is evicted
2046 Details : Running with the dump_on_timeout debug flag turned on makes
2047 it possible that the waiting_locks_callback() can try to dump
2048 the Lustre kernel debug logs from an interrupt handler. Defer
2049 this log dumping to the expired_lock_main() thread.
2051 Severity : enhancement
2053 Description: Support NFS exporting on 2.6 kernels.
2054 Details : Implement non-rawops metadata methods for NFS server to use without
2055 changing NFS server code.
2058 Frequency : very rare (synthetic metadata workload only)
2060 Description: two racing renames might cause an MDS thread to deadlock
2061 Details : Running the "racer" program may cause one MDS thread to rename
2062 a file from being the source of a rename to being the target of
2063 a rename at exactly the same time that another thread is doing
2064 so, and the second thread has already enqueued these locks after
2065 doing a lookup of the target and is trying to relock them in
2066 order. Ensure that we don't try to re-lock the same resource.
2069 Frequency : only very large systems with liblustre clients
2071 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2072 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2073 all OSTs in parallel. This allows the eviction of stale liblustre
2074 clients to proceed much faster than if they were done in series,
2075 and also offers similar improvements for other set_info RPCs.
2080 Description: excessive CPU usage during initial read phase on client
2081 Details : During the initial read phase on a client, it would agressively
2082 retry readahead on the file, consuming too much CPU and impacting
2083 performance (since 1.4.5.8). Improve the readahead algorithm
2084 to avoid this, and also improve some other common cases (read
2085 of small files in particular, where "small" is files smaller than
2086 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2091 Description: MDS crash when receiving packet with unknown intent.
2092 Details : Do not LBUG in unknown intent case, just return -EFAULT
2094 Severity : enhancement
2095 Bugzilla : 9293, 9385
2096 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2097 Details : Do not serialize getattr (non-intent version) and statfs.
2100 Frequency : occasional, when OST network is overloaded/intermittent
2102 Description: client evicted by OST after bulk IO timeout
2103 Details : If a client sends a bulk IO request (read or write) the OST
2104 may evict the client if it is unresposive to its data GET/PUT
2105 request. This is incorrect if the network is overloaded (takes
2106 too long to transfer the RPC data) or dropped the OST GET/PUT
2107 request. There is no need to evict the client at all, since
2108 the pinger and/or lock callbacks will handle this, and the
2109 client can restart the bulk request.
2112 Frequency : Always when mmapping file with no objects
2114 Description: client crashes when mmapping file with no objects
2115 Details : Check that we actually have objects in a file before doing any
2116 operations on objects in ll_vm_open, ll_vm_close and
2122 Description: Request leak when working with deleted CWD
2123 Details : Introduce advanced request refcount tracking for requests
2124 referenced from lustre intent.
2126 Severity : Enhancement
2128 Description: Cache open file handles on client.
2129 Details : MDS now will return special lock along with openhandle, if
2130 requested and client is allowed to hold openhandle, even if unused,
2131 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2132 closing files for every read/write openration.
2134 Severity : Enhancement
2136 Description: Cache open negative dentries on client when possible.
2137 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2138 negative dentries on lock revocation.
2143 Description: Remounting a client read-only wasn't possible with a zconf mount
2144 Details : It wasn't possible to remount a client read-only with llmount.
2146 Severity : enhancement
2147 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2148 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2149 MPICH is included to add Lustre-specific ADIO interfaces.
2150 This is based closely on the UFS ADIO layer and only differs
2151 in file creation, in order to allow the OST striping to be set.
2152 This is user-contributed code and not supported by CFS.
2157 Description: extended inode attributes (immutable, append-only) work improperly
2158 when 2.4 and 2.6 kernels are used on client/server or vice versa
2159 Details : Introduce kernel-independent values for these flags.
2161 Severity : enhancement
2164 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2165 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2166 granularity. Now, Fractional MB granularity is be supported,
2167 this is very useful for low memory system.
2169 Severity : enhancement
2171 Description: Getattr by fid
2172 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2173 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2174 have locks and avoids getattr rpc requests for it on every stat.
2177 Frequency : Always, for filesystems larger than 2TB
2179 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2180 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2181 filesystems larger than 2TB to be mounted with mballoc enabled.
2184 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2186 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2187 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2188 filesystems larger than 2TB is mounted then the kernel will
2189 silently corrupt the start of the filesystem. CONFIG_LBD is
2190 enabled for all CFS-supported kernels, but the possibility of
2191 this happening with a modified kernel config exists.
2193 Severity : enhancement
2195 Description: add client O_DIRECT support for 2.6 kernels
2196 Details : It is now possible to do O_DIRECT reads and writes to files
2197 in the Lustre client mountpoint on 2.6 kernel clients.
2199 Severity : enhancement
2201 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2202 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2203 parallel, not waiting for response from every OST before sending
2204 a rpc to the next OST.
2209 Description: setattr vs write race when updating file timestamps
2210 Details : Client processes that update a file timestamp into the past
2211 right after writing to the file (e.g. tar) it is possible that
2212 the updated file modification time can be reset to the current
2213 time due to a race between processing the setattr and write RPC.
2215 Severity : enhancement
2217 Description: Bring 'lfs find' closer in line with regular Linux find.
2218 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2219 -print0 options and obtains all the needed info through the lustre
2222 Severity : enhancement
2224 Description: support up to 1024 configured devices on one node
2225 Details : change obd_dev array from statically allocated to dynamically
2226 allocated structs as they are first used to reduce memory usage
2231 Description: Flush dirty partially truncated pages during truncate
2232 Details : Immediatelly flush partially truncated pages in filter_setattr,
2233 this way we completely avoid having any pages in page cache on OST
2234 and can retire ugly workarounds during writes to flush such pages.
2239 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2240 Details : protect inode from truncation within vfs_unlink() context
2241 just take a reference before calling vfs_unlink() and release it
2242 when parent's i_sem is free.
2247 Description: last_id value checked outside lock on OST caused LASSERT failure
2248 Details : If there were multiple MDS->OST object precreate requests in
2249 flight, it was possible that the OST's last object id was checked
2250 outside a lock and incorrectly tripped an assertion. Move checks
2251 inside locks, and discard old precreate requests.
2254 Frequency : always, if extents are used on OSTs
2256 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2257 Details : OSTs using the extents format would not zero the high 16 bits of
2258 the index physical block number. This is not a problem for any
2259 OST filesystems smaller than 16TB, and no kernels support ext3
2260 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2261 new/modified files) and can be fixed for existing filesystems
2262 with e2fsprogs-1.39-cfs1.
2267 Description: import connection selection may be incorrect if timer wraps
2268 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2269 import connections to be ignored if the 32-bit jiffies counter
2270 wraps. Use a 64-bit jiffies counter.
2273 Frequency : very large clusters immediately after boot
2275 Description: LNET request buffers exhausted under heavy short-term load
2276 Details : If a large number of client requests are generated on a service
2277 that has previously never seen so many requests it is possible
2278 that the request buffer growth cannot keep up with the spike in
2279 demand. Instead of dropping incoming requests, they are held in
2280 the LND until the RPC service can accept more requests.
2283 Frequency : Sometimes during replay
2285 Description: Assertion failure in ll_local_open after replay.
2286 Details : If replay happened on an open request reply before we were able
2287 to set replay handler, reply will become not swabbed tripping the
2288 assertion in ll_local_open. Now we set the handler right after
2289 recognising of open request
2292 Frequency : very rare
2294 Description: kernel reports "badness in vsnprintf"
2295 Details : Reading from the "recovery_status" /proc file in small chunks
2296 may cause a negative length in lprocfs_obd_rd_recovery_status()
2297 call to vsnprintf() (which is otherwise harmless). Exit early
2298 if there is no more space in the output buffer.
2300 Severity : enhancement
2302 Description: clear OBD RPC statistics by writing to them
2303 Details : It is now possible to clear the OBD RPC statistics by writing
2304 to the "stats" file.
2309 Description: Client mtime is not the same on different clients after utimes
2310 Details : In some cases, the client was using the utimes() syscall on
2311 a file cached on another node. The clients now validate the
2312 ctime from the MDS + OSTs to determine which one is right.
2317 Description: Inability to activate failout mode
2318 Details : lconf script incorrectly assumed that in pythong string's numeric
2319 value is used in comparisons.
2322 Frequency : always with multiple stripes per file
2324 Description: Inefficient object allocation for mutli-stripe files
2325 Details : When selecting which OSTs to stripe files over, for files with
2326 a stripe count that divides evenly into the number of OSTs,
2327 the MDS is always picking the same starting OST for each file.
2328 Return the OST selection heuristic to the original design.
2333 Description: mount failures may take full timeout to return an error
2334 Details : Under some heavy load conditions it is possible that a
2335 failed mount can wait for the full obd_timeout interval,
2336 possibly several minutes, before reporting an error.
2337 Instead return an error as soon as the status is known.
2339 Frequency : quota enabled and large files being deleted
2341 Description: releasing more than 4GB of quota at once hangs OST
2342 Details : If a user deletes more than 4GB of files on a single OST it
2343 will cause the OST to spin in an infinite loop. Release
2344 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2349 Description: statfs data retrieved from /proc may be stale or zero
2350 Details : When reading per-device statfs data from /proc, in the
2351 {kbytes,files}_{total,free,avail} files, it may appear
2352 as zero or be out of date.
2355 Frequency : systems with MD RAID1 external journal devices
2357 Description: lconf's call to blkid is confused by RAID1 journal devices
2358 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2359 of returning all block devices that match the journal UUID.
2362 Frequency : always, for aggregate stripe size over 4GB
2364 Description: assertion fails when trying to use 4GB stripe size
2365 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2366 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2369 Frequency : always on ppc64
2371 Description: the first write on an ext3 filesystem with mballoc got stuck
2372 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2373 perform endianness conversion.
2375 ------------------------------------------------------------------------------
2377 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2379 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2380 INTEROPERATE with older versions automatically. Please read the
2381 user documentation before upgrading any part of a live system.
2382 * WARNING: Lustre networking configuration changes are required with
2383 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2386 * Support for kernels:
2387 2.6.9-22.0.2.EL (RHEL 4)
2388 2.6.5-7.244 (SLES 9)
2389 2.6.12.6 vanilla (kernel.org)
2392 Severity : enhancement
2393 Bugzilla : 7981/8208
2394 Description: Introduced Lustre Networking (LNET)
2395 Details : LNET is new networking infrastructure for Lustre, it includes
2396 a reorganized network configuration mode (see the user
2397 documentation for full details) as well as support for routing
2398 between different network fabrics. Lustre Networking Devices
2399 (LNDs) for the supported network fabrics have also been
2400 created for this new infrastructure.
2402 Severity : enhancement
2403 Description: Introduced Access control lists
2404 Details : clients can set ACLs on files and directories in order to have
2405 more fine-grained permissions than the standard Unix UGO+RWX.
2406 The MDS must be started with the "-o acl" mount option.
2408 Severity : enhancement
2409 Description: Introduced filesystem quotas
2410 Details : Administrators may now establish per-user quotas on the
2413 Severity : enhancement
2415 Description: Configuration change for the XT3
2416 The PTLLND is now used to run Lustre over Portals on the XT3
2417 The configure option(s) --with-cray-portals are no longer used.
2418 Rather --with-portals=<path-to-portals-includes> is used to
2419 enable building on the XT3. In addition to enable XT3 specific
2420 features the option --enable-cray-xt3 must be used.
2425 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2426 Details : A race between allocating a new llcd and re-getting the llcd_lock
2427 allowed another thread to grab newly-allocated llcd.
2429 Severity : enhancement
2431 Description: 2.6 OST async journal commit and locking fix to improve performance
2432 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2433 2.6 kernels are now async as they already were in 2.4 kernels so
2434 that they can commit concurrently with the network bulk transfer.
2435 For block-allocated files the filter allocation semaphore is held
2436 to avoid filesystem fragmentation during allocation. BKL lock
2437 removed for 2.6 xattr operations where it is no longer needed.
2442 Description: lconf incorrectly determined whether two IP networks could talk
2443 Details : In some more complicated routing and multiple-network
2444 configurations, lconf will avoid trying to make a network
2445 connection to a disjoint part of the IP space. It was doing the
2446 math incorrectly for one set of cases.
2451 Description: Fix for potential infinite loop processing records in an llog.
2452 Details : If an llog record is corrupted/zeroed, it is possible to loop
2453 forever in llog_process(). Validate the llog record length
2454 and skip the remainder of the block on error.
2457 Frequency : occasional (liblustre only)
2459 Description: liblustre could not open files whose last component is a symlink
2460 Details : sysio_path_walk() would incorrectly pass the open intent to
2461 intermediate path components.
2464 Frequency : rare (liblustre only with non-standard tuning)
2465 Bugzilla : 7201 (7350)
2466 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2467 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2468 after it has been released, only do this for referenced locks
2470 Severity : enhancement
2472 Description: specifying an (invalid) directory default stripe_size of -1
2473 would reset the directory default striping
2474 Details : stripe_size -1 was used internally to signal directory stripe
2475 removal, now use "all default" to signal dir stripe removal
2476 as a directory striping of "all default" is not useful
2479 Frequency : common for large clusters running liblustre clients
2481 Description: doing an ls when liblustre clients are running is slow
2482 Details : sending a glimpse AST to a liblustre client waits for every AST
2483 to time out, as liblustre clients will not respond. Since they
2484 cannot cache data we refresh the OST lock LVB from disk instead.
2486 Severity : enhancement
2488 Description: doing an ls at the same time as file IO can be slow
2489 Details : enqueue and other "small" requests can be blocked behind many
2490 large IO requests. Create a new OST IO portal for non-IO
2491 requests so they can be processed faster.
2494 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2496 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2497 created file with mode 000
2498 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2499 allow an ftruncate() as a non-root user to a file with mode 000.
2500 Lustre now respects this flag to disable mode checks when
2501 truncating a file owned by the user
2504 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2507 Description: Revoking locks from clients that went dead or catatonic might take
2509 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2510 cancellation of such locks instant on servers without waiting for
2511 any reply from clients. Clients drops these locks when cancel
2512 notification from server is received without replying.
2515 Frequency : liblustre-only, when liblustre client dies or becomes busy
2517 Description: Doing ls on Linux clients can take a long time with active
2519 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2520 granting such locks to it in the first place if possible. Locks
2521 are taken by proxy on the OST during the read or write and
2522 dropped immediately afterward. Add connect flags handling, do
2523 not grant locks to liblustre clients for glimpse ASTs.
2525 Severity : enhancement
2527 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2528 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2529 sequential reads and behaves badly with applications doing random
2530 reads. Improve it by reading ahead at least read region, and
2531 avoiding excessive large RPC for small reads.
2533 Severity : enhancement
2535 Description: Creating more than 1000 files for a single job may cause a load
2536 imbalance on the OSTs if there are also a large number of OSTs.
2537 Details : qos_prep_create() uses an OST index reseed value that is an
2538 even multiple of the number of available OSTs so that if the
2539 reseed happens in the middle of the object allocation it will
2540 still utilize the OSTs as uniformly as possible.
2545 Description: OST or MDS may oops in ping_evictor_main()
2546 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2547 but doesn't restart at beginning of obd_exports_timed list
2550 Severity : enhancement
2552 Description: improve by-nid export eviction on the MDS and OST
2553 Details : allow multiple exports with the same NID to be evicted at one
2554 time without re-searching the exports list.
2557 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2559 Description: MDS may oops in groups_free()
2560 Details : in rare race conditions a newly allocated group_info struct is
2561 freed again, and this can be NULL. The 2.4 compatibility code
2562 for groups_free() checked for a NULL pointer, but 2.6 did not.
2565 Frequency : common for liblustre clients doing little filesystem IO
2566 Bugzilla : 9352, 7313
2567 Description: server may evict liblustre clients accessing contended locks
2568 Details : if a client is granted a lock or receives a completion AST
2569 with a blocking AST already set it would not reply to the AST
2570 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2573 Frequency : lfs setstripe, only systems with more than 160 OSTs
2575 Description: unable to set striping with a starting offset beyond OST 160
2576 Details : llapi_create_file() incorrectly limited the starting stripe
2577 index to the maximum single-file stripe count.
2580 Frequency : LDAP users only
2582 Description: lconf did not handle in-kernel recovery with LDAP properly
2583 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2585 Severity : enhancement
2587 Description: bind OST threads to NUMA nodes to improve performance
2588 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2589 node and do their allocations there to localize memory access
2591 Severity : enhancement
2593 Description: llmount can determine client NID directly from Myrinet (GM)
2594 Details : the client NID code from gmnalnid was moved directly into
2595 llmount, removing the need to use this or specifying the
2596 client NID explicitly when mounting GM clients with zeroconf
2599 Frequency : if client is started with down MDS
2601 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2602 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2603 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2604 ptlrpc_import_delayed_req() because we didn't check if the
2605 request was interrupted, and we also didn't break out of the
2606 event loop if there was no timeout
2611 Description: data loss during non-page-aligned writes to a single file from
2612 both multiple nodes and multiple threads on one node at same time
2613 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2614 inconsistency led to false short-reads, that were cached and later
2615 used by ->prepare_write() to fill in partially written page,
2616 leading to data loss.
2619 Frequency : always, if lconf --abort_recovery used
2621 Description: lconf --abort_recovery fails with 'Operation not supported'
2622 Details : lconf was attempting to abort recovery on the MDT device and not
2625 Severity : enhancement
2627 Description: remove cleanup logs
2628 Details : replace lconf-generated cleanup logs with lustre internal
2629 cleanup routines. Eliminates the need for client-cleanup and
2632 Severity : enhancement
2634 Description: add support for EAs (user and system) on lustre filesystems
2635 Details : it is now possible to store extended attributes in the Lustre
2636 client filesystem, and with the user_xattr mount option it
2637 is possible to allow users to store EAs on their files also
2639 Severity : enhancement
2641 Description: Add possibility (config option) to show minimal available OST free
2643 Details : When compiled with --enable-mindf configure option, statfs(2)
2644 (and so, df) will return least minimal free space available from
2645 all OSTs as amount of free space on FS, instead of summary of
2646 free spaces of all OSTs.
2648 Severity : enhancement
2650 Description: do not expand extent locks acquired on OST-side
2651 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2652 by server: they are not cached anyway.
2655 Frequency : when mmap is used/binaries executed from Lustre
2657 Description: Unmmap pages before throwing them away from read cache.
2658 Details : llap_shrink cache now attempts to unmap pages before discarding
2659 them (if unmapping failed - do not discard). SLES9 kernel has
2660 extra checks that trigger if this unmapping is not done first.
2665 Description: lconf didn't resolve symlinks before checking to see whether a
2666 given mountpoint was already in use
2669 Frequency : when migrating failover services
2670 Bugzilla : 6395, 9514
2671 Description: When migrating a subset of services from a node (e.g. failback
2672 from a failover service node) the remaining services would
2673 time out and evict clients.
2674 Details : lconf --force (implied by --failover) sets the global obd_timeout
2675 to 5 seconds in order to quickly disconnect, but this caused
2676 other RPCs to time out too quickly. Do not change the global
2677 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2679 Severity : enhancement
2680 Frequency : if MDS is started with down OST
2681 Bugzilla : 9439,5706
2682 Description: Allow startup/shutdown of an MDS without depending on the
2683 availability of the OSTs.
2684 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2685 Add appropriate locking and lov-osc refcounts for safe
2686 cleaning. Add osc abort_inflight calls in case the
2687 synchronize never started.
2690 Frequency : occasional (Cray XT3 only)
2692 Description: root not authorized to access files in CRAY_PORTALS environment
2693 Details : The client process capabilities were not honoured on the MDS in
2694 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2695 been accessed by an authorized user then root was able to access
2696 the file on the local client also. The root user capabilities
2697 are now allowed on the MDS, as this environment has secure UID.
2700 Frequency : occasional
2702 Description: ldiskfs "too long searching" message happens too often
2703 Details : A debugging message (otherwise harmless) prints too often on
2704 the OST console. This has been reduced to only happen when
2705 there are fragmentation problems on the filesystem.
2710 Description: Division by zero in statfs when all OSCs are inactive
2711 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2712 lov_statfs divides by value returned by lov_get_stripecnt().
2716 Bugzilla : 9489, 3273
2717 Description: First write from each client to each OST was only 4kB in size,
2718 to initialize client writeback cache, which caused sub-optimal
2719 RPCs and poor layout on disk for the first writen file.
2720 Details : Clients now request an initial cache grant at (re)connect time
2721 and so that they can start streaming writes to the cache right
2722 away and always do full-sized RPCs if there is enough data.
2723 If the OST is rebooted the client also re-establishes its grant
2724 so that client cached writes will be honoured under the grant.
2729 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2730 Details : Now I/O RPCs go to different portal number and (presumably) fast
2731 lock requests (and glimses) and other RPCs get their own service
2732 threads pool that should be able to service those RPCs
2735 Severity : enhancement
2737 Description: Ability to exchange lustre version between client and servers and
2738 issue warnings at client side if client is too old. Also for
2739 liblustre clients there is ability to refuse connection of too old
2741 Details : New 'version' field is added to connect data structure that is
2742 filled with version info. That info is later checked by server and
2746 Frequency : rare, liblustre only.
2747 Bugzilla : 9296, 9581
2748 Description: Two simultaneous writes from liblustre at offset within same page
2749 might proceed at the same time overwriting eachother with stale
2751 Details : I/O lock withing llu_file_prwv was released too early, before data
2752 actually was hitting the wire. Extended lock-holding time until
2753 server acknowledges receiving data.
2756 Frequency : extremely rare. Never observed in practice.
2758 Description: avoid generating lustre_handle cookie of 0.
2759 Details : class_handle_hash() generates handle cookies by incrementing
2760 global counter, and can hit 0 occasionaly (this is unlikely, but
2761 not impossible, because initial value of cookie counter is
2762 selected randonly). Value of 0 is used as a sentinel meaning
2763 "unassigned handle" --- avoid it. Also coalesce two critical
2764 sections in this function into one.
2766 Severity : enhancement
2768 Description: allow liblustre clients to delegate truncate locking to OST
2769 Details : To avoid overhead of locking, liblustre client instructs OST to
2770 take extent lock in ost_punch() on client's behalf. New connection
2771 flag is added to handle backward compatibility.
2773 Severity : enhancement
2774 Bugzilla : 4928, 7341, 9758
2775 Description: allow number of OST service threads to be specified
2776 Details : a module parameter allows the number of OST service threads
2777 to be specified via "options ost ost_num_threads={N}" in the
2778 OSS's /etc/modules.conf or /etc/modprobe.conf.
2782 Bugzilla : 6146, 9635, 9895
2783 Description: servers crash with bad pointer in target_handle_connect()
2784 Details : In rare cases when a client is reconnecting it was possible that
2785 the connection request was the last reference for that export.
2786 We would temporarily drop the export reference and get a new
2787 one, but this may have been the last reference and the export
2788 was just destroyed. Get new reference before dropping old one.
2790 Severity : enhancement
2791 Frequency : if client is started with failover MDS
2793 Description: Allow multiple MDS hostnames in the mount command
2794 Details : Try to read the configuration from all specified MDS
2795 hostnames during a client mount in case the "primary"
2798 Severity : enhancement
2800 Description: Stop sending data to evicted clients as soon as possible.
2801 Details : Check if the client we are about to send or are sending data to
2802 was evicted already. (Check is done every second of waiting,
2803 for which l_wait_event interface was extended to allow checking
2804 of exit condition at specified intervals).
2807 Frequency : rare, normally only when NFS exporting is done from client
2809 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2811 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2814 Frequency : always, if srand() is called before liblustre initialization
2816 Description: Liblustre uses system PRNG disturbing its usage by user application
2817 Details : Introduce internal to lustre fast and high-quality PRNG for
2818 lustre usage and make liblustre and some other places in generic
2819 lustre code to use it.
2821 Severity : enhancement
2822 Bugzilla : 9477, 9557, 9870
2823 Description: Verify that the MDS configuration logs are updated when xml is
2824 Details : Check if the .xml configuration logs are newer than the config
2825 logs stored on the MDS and report an error if this is the case.
2826 Request --write-conf, or allow starting with --old_conf.
2828 Severity : enhancement
2830 Description: Handle symlinks in the path when checking if Lustre is mounted.
2831 Details : Resolve intermediate symlinks when checking if a client has
2832 mounted a filesystem to avoid duplicate client mounts.
2837 Description: lconf can hit an error exception but still return success.
2838 Details : The lconf command catches the Command error exception at the top
2839 level script context and will exit with the associated exit
2840 status, but doesn't ensure that this exit status is non-zero.
2845 Description: failure of ptlrpc thread startup can cause oops
2846 Details : Starting a ptlrpc service thread can fail if there are a large
2847 number of threads or the server memory is very fragmented.
2848 Handle this without oopsing.
2851 Frequency : always, only if liblustre and non-default acceptor port was used
2853 Description: liblustre cannot connect to servers with non-default acceptor port
2854 Details : tcpnal_set_default_params() was not called and was therefore
2855 ignoring the environment varaible TCPNAL_PORT, as well as other
2856 TCPNAL_ environment variables
2861 Description: two objects could be created on the same OST for a single file
2862 Details : If an OST is down, in some cases it was possible to create two
2863 objects on a single OST for a single file. No problems other
2864 than potential performance impact and spurious error messages.
2868 Bugzilla : 5681, 9562
2869 Description: Client may oops in ll_unhash_aliases
2870 Details : Client dcache may become inconsistent in race condition.
2871 In some cases "getcwd" can fail if the current directory is
2877 Description: Inode refcounting problems in NFS export code
2878 Details : link_raw functions used to call d_instantiate without obtaining
2879 extra inode reference first.
2883 Bugzilla : 9942, 9903
2884 Description: Referencing freed requests leading to crash, memleaks with NFS.
2885 Details : We used to require that call to ll_revalidate_it was always
2886 followed by ll_lookup_it. Also with revalidate_special() it is
2887 possible to call ll_revalidate_it() twice for the same dentry
2888 even if first occurence returned success. This fix changes semantic
2889 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2890 reference on a request referred from the intent.
2891 ll_intent_release() then releases such a request.
2894 Frequency : rare, normally benchmark loads only
2896 Description: unlinked inodes were kept in memory on the client
2897 Details : If a client is repeatedly creating and unlinking files it
2898 can accumulate a lot of stale inodes in the inode slab cache.
2899 If there is no other client load running this can cause the
2900 client node to run out of memory. Instead flush old inodes
2901 from client cache that have the same inode number as a new inode.
2904 Frequency : SLES9 2.6.5 kernel and long filenames only
2905 Bugzilla : 9969, 10379
2906 Description: utime reports stale NFS file handle
2907 Details : SLES9 uses out-of-dentry names in some cases, which confused
2908 the lustre dentry revalidation. Change it to always use the
2912 Frequency : rare, unless heavy write-truncate concurrency is continuous
2913 Bugzilla : 4180, 6984, 7171, 9963, 9331
2914 Description: OST becomes very slow and/or deadlocked during object unlink
2915 Details : filter_destroy() was holding onto the parent directory lock
2916 while truncating+unlinking objects. For very large objects this
2917 may block other threads for a long time and slow overall OST
2918 responsiveness. It may also be possible to get a lock ordering
2919 deadlock in this case, or run out of journal credits because of
2920 the combined truncate+unlink. Solution is to do object truncate
2921 first in one transaction without parent lock, and then do the
2922 final unlink in a new transaction with the parent lock. This
2923 reduces the lock hold time dramatically.
2926 Frequency : rare, 2.4 kernels only
2928 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2929 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2930 before unmounting their filesystem in order to avoid deadlock.
2931 The kernel_locked() function in 2.4 kernels only checks whether
2932 the kernel lock is held, not whether it is this process that is
2933 holding it as 2.6 kernels do.
2938 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2939 Details : The client ptlrpc code may be trying to reconnect to a down
2940 server before a previous connection attempt has timed out.
2941 Increase the reconnect interval to be longer than the connection
2942 timeout interval to avoid sending duplicate connections to
2946 Frequency : echo_client brw_test command
2948 Description: fix echo_client to work with OST preallocated code
2949 Details : OST preallocation code (5137) didn't take echo_client IO path
2950 into account: echo_client calls filter methods outside of any
2951 OST thread and, hence, there is no per-thread preallocated
2952 pages and buffers to use. Solution: hijack pga pages for IO. As
2953 a byproduct, this avoids unnecessary data copying.
2957 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2958 Description: Client can oops in mdc_commit_close() after open replay
2959 Details : It was possible for the MDS to return an open request with no
2960 transaction number in mds_finish_transno() if the client was
2961 evicted, but without actually returning an error. Clients
2962 would later try to replay that open and may trip an assertion
2963 Simplify the client close codepath, and always return an error
2964 from the MDS in case the open is not successful.
2967 Frequency : rare, 2.6 OSTs only
2969 Description: OST may deadlock under high load on fragmented files
2970 Details : If there was a heavy load and highly-fragmented OST filesystems
2971 it was possible to have all the OST threads deadlock waiting on
2972 allocation of biovecs, because the biovecs were not released
2973 until the entire RPC IO was completed. Instead, release biovecs
2974 as soon as they are complete to ensure forward IO progress.
2976 Severity : enhancement
2978 Description: Support for specifying external journal device at mount
2979 Details : If an OST or MDS device is formatted with an external journal
2980 device, this device major/minor is stored in the ext3 superblock
2981 and may not be valid for failover. Allow detecting and
2982 specifying the external journal at mount time.
2987 Description: Mounting an MDS with pending unlinked files may cause oops
2988 Details : target_finish_recovery() calls mds_postrecov() which returned
2989 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2990 considers this an error and immediately begins cleaning up the
2991 lov, just after starting the mds_lov process
2993 Severity : enhancement
2995 Description: Implement 'lfs df' to report actual free space on per-OST basis
2996 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2997 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2998 usage of objects; '-h' to report in human readable format.
3000 ------------------------------------------------------------------------------
3002 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3009 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3010 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3011 at startup during orphan unlinking. Instead, skip buddy update
3012 before setup, as it will be regenerated after recovery is complete.
3017 Description: If an OST is inactive, its locks might reference stale inodes.
3018 Details : lov_change_cbdata() must iterate over all namespaces, even if
3019 they are inactive to clear inode references from the lock.
3021 Severity : enhancement
3022 Frequency : occasional, if non-standard max_dirty_mb used
3024 Description: Client will block write RPCs if not enough grant
3025 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3026 then the client will block writes while waiting for another RPC
3027 to complete instead of consuming its dirty limit. With change
3028 we get improved performance when max_dirty_mb is small.
3030 Severity : enhancement
3031 Bugzilla : 3389, 6253
3032 Description: Add support for supplementary groups on the MDS.
3033 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3034 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3035 lookups for user supplementary groups into a cache.
3039 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3040 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3041 for liblustre O_EXCL is enforced by MDS.
3045 Description: readdir never returns NULL in liblustre.
3046 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3047 dentry in struct dirent.
3051 Frequency : liblustre only, depends on application IO pattern
3052 Description: liblustre clients evicted if not contacting servers
3053 Details : Don't put liblustre clients into the ping_evictor list, so
3054 they will not be evicted by the pinger ever.
3056 Severity : enhancement
3058 Description: Add ability to evict clients by NID from MDS.
3059 Details : By echoing "nid:$NID" string into
3060 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3061 $NID would be instantly evicted from this MDS and from all active
3062 OSTs connected to it.
3066 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3067 Details : lookup_it_finish() used to query file size from OSTs that was not
3072 Description: service threads change working directory to that of init
3073 Details : Starting lustre service threads may pin the working directory
3074 of the parent thread, making that filesystem busy. Threads
3075 now change to the working directory of init to avoid this.
3079 Frequency : during shutdown only
3080 Description: shutdown with a failed MDS or OST can cause unmount to hang
3081 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3086 Frequency : 2.6 only
3087 Description: chmod/chown may include an extra supplementary group
3088 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3089 supplementary group and if none is specified this is used.
3092 Bugzilla : 5479 (6816)
3094 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3095 Details : If lookup is in progress on a file that is unlinked we might try
3096 to revalidate the inode and fail in revalidate after lookup is
3097 complete and ll_file_open() enqueues the open again but
3098 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3101 Frequency : always, if lconf --abort_recovery used
3103 Description: lconf --abort_recovery fails with 'Operation not supported'
3104 Details : lconf was attempting to abort recovery on the MDT device and not
3107 ------------------------------------------------------------------------------
3109 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3114 Frequency : rare (only unsupported configurations with a node running as an
3116 Bugzilla : 6514, 5137
3117 Description: Mounting a Lustre file system on a node running as an OST could
3119 Details : OSTs now preallocates memory needed to write out data at
3120 startup, instead of when needed, to avoid having to
3121 allocate memory in possibly low memory situations.
3122 Specifically, if the file system is mounted on on OST,
3123 memory pressure could force it to try to write out data,
3124 which it needed to allocate memory to do. Due to the low
3125 memory, it would be unable to do so and the node would
3126 become unresponsive.
3128 Severity : enhancement
3130 Description: Addition of lconf --service command line option
3131 Details : lconf now accepts a '--service <arg>' option, which is
3132 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3134 Severity : enhancement
3136 Description: Failover mode is now the default for OSTs.
3137 Details : By default, OSTs will now run in failover mode. To return to
3138 the old behaviour, add '--failout' to the lmc line for OSTs.
3140 Severity : enhancement
3142 Description: Health checks are now provided for MDS and OSTs
3143 Details : Additional detailed health check information on MSD and OSTs
3144 is now provided through the procfs health_check value.
3147 Frequency : occasional, depends on IO load
3149 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3150 after numerous create/delete cycles
3151 Details : The ext3 inode allocation policy would not allocate new inodes
3152 very well on the OSTs because there are no new directories
3153 being created. Instead we look for groups with free space if
3154 the parent directories are nearly full.
3159 Description: Network or server problems during mount may cause partially
3160 mounted clients instead of returning an error.
3161 Details : The config llog parsing code may overwrite the error return
3162 code during mount error handling, returning success instead
3168 Description: MDS can fail to allocate large reply buffers
3169 Details : After long uptimes the MDS can fail to allocate large reply
3170 buffers (e.g. zconf client mount config records) due to memory
3171 fragmentation or consumption by the buffer cache. Preallocate
3172 some large reply buffers so that these replies can be sent even
3173 under memory pressure.
3177 Frequency : rare (liblustre)
3178 Description: fsx running with liblustre complained that using truncate() to
3179 extend the file doesn't work. This patch corrects that issue.
3180 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3181 ATTR_SIZE and lsm use in llu_setattr_raw.
3185 Frequency : rare, only 2.6 kernels
3186 Description: Unusual file access patterns on the MDS may result in inode
3187 data being lost in very rare circumstances.
3188 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3189 code in the 2.6 kernel caused the inode and/or EA data not to
3190 be read from disk, causing single-file corruption.
3194 Frequency : rare, only 2.6 filesystems using extents
3195 Description: Heavy concurrent write and delete load may cause data corruption.
3196 Details : It was possible under high-load situations to have an extent
3197 metadata block in the block device cache from a just-unlinked
3198 file overwrite a newly-allocated data block. We now unmap any
3199 metadata buffers that alias just-allocated data blocks.
3203 Frequency : filesystems with default stripe_count larger than 77
3204 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3205 Details : lconf specifies an inode size of 4096 bytes when the default
3206 stripe_count is larger than 77. This conflicts with the default
3207 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3208 this case to avoid pinning too much memory for large EAs.
3210 ------------------------------------------------------------------------------
3212 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3217 Frequency : rare (extremely heavy IO load with hundreds of clients)
3219 Description: Client is evicted, gets IO error writing to file
3220 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3221 caused clients to be evicted for AST timeouts. The fixes in
3222 bug 5192 mean we no longer need to have such short AST timeouts
3223 so ldlm_timeout has been increased.
3226 Frequency : occasional during --force or --failover shutdown under load
3227 Bugzilla : 5949, 4834
3228 Description: Server oops/LBUG if stopped with --force or --failover under load
3229 Details : a collection of import/export refcount and cleanup ordering
3230 issues fixed for safer force cleanup
3233 Frequency : only filesystems larger than 120 OSTs
3234 Bugzilla : 5990, 6223
3235 Description: lfs getstripe would oops on a very large filesystem
3236 Details : lov_getconfig used kfree on vmalloc'd memory
3239 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3240 Bugzilla : 6242, 6243
3241 Description: reading from files that had been truncated to a non-zero size
3242 but never opened returned no data
3243 Details : ll_file_read() reads zeros from no-object files to EOF
3248 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3249 Details : There's little harm in aborting MDS/OSS recovery and letting it
3250 try again, so I removed the LASSERT and return an error instead.
3252 Severity : enhancement
3254 Description: New debugging infrastructure for tracking down data corruption
3255 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3256 (b) cover more of the client-side code path, and (c) try to narrow
3257 down where problems occurred
3261 Bugzilla : 3819, 4364, 4397, 6313
3262 Description: Racing close and eviction MDS could cause assertion in mds_close
3263 Details : It was possible to get multiple mfd references during close and
3264 client eviction, leading to one thread referencing a freed mfd.
3266 Severity: : enhancement
3267 Bugzilla : 3262, 6359
3268 Description: Attempts to reconnect to servers are now more aggressive.
3269 Details : This builds on the enhanced upcall-less recovery that was added
3270 in 1.4.2. When trying to reconnect to servers, clients will
3271 now try each server in the failover group every 10 seconds. By
3272 default, clients would previously try one server every 25 seconds.
3277 Description: After recovery, certain operations trigger a failed
3278 assertion on a client.
3279 Details : Failing over an mds, using lconf -d --failover, while a
3280 client was doing a readdir() call would cause the client to
3281 LBUG after recovery completed and the readdir() was resent.
3283 Severity : enhancement
3285 Description: Default groups are now added by lconf
3286 Details : You can now run lconf --group <servicename> without having to
3287 manually add groups with lmc.
3290 Frequency : occasional
3292 Description: Nodes with an elan id of 0 trigger a failed assertion
3295 Frequency : always when accessing e.g. tty/console device nodes
3297 Description: tty and some other devices nodes cannot be used on lustre
3298 Details : file's private_data field is used by device data and lustre
3299 values in there got lost. New field was added to struct file to
3300 store fs-specific private data.
3303 Frequency : when exporting Lustre via NFS
3305 Description: NFSD failed occasionally when looking up a path component
3306 Details : NFSD is looking up ".." which was broken in ext3 directories
3307 that had grown large enough to become hashed.
3310 Frequency : Clusters with multiple interfaces not on the same subnet
3312 Description: Nodes will repeatedly try to reconnect to an interface which it
3313 cannot reach and report an error to the log.
3314 Details : Extra peer list entries will be created by lconf with some peers
3315 unreachable. lconf now validates the peer before adding it.
3318 Frequency : Only if a default stripe is set on the filesystem root.
3320 Description: Setting a default stripe on the filesystem root prevented the
3321 filesystem from being remounted.
3322 Details : The client was sending extra request flags in the root getattr
3323 request and did not allocate a reply buffer for the dir EA.
3326 Frequency : occasional, higher if lots of files are accessed by one client
3327 Bugzilla : 6159, 6097
3328 Description: Client trips assertion regarding lsm mismatch/magic
3329 Details : While revalidating inodes the VFS looks up inodes with ifind()
3330 and in rare cases can find an inode that is being freed.
3331 The ll_test_inode() code will free the lsm during ifind()
3332 when it finds an existing inode and then the VFS later attaches
3333 this free lsm to a new inode.
3337 Bugzilla : 6422, 7030
3338 Description: MDS deadlock between mkdir and client eviction
3339 Details : Creating a new file via mkdir or mknod (starting a transaction
3340 and getting the ns lock) can deadlock with client eviction
3341 (gets ns lock and trying to finish a synchronous transaction).
3344 Frequency : occasional
3345 Description: While starting a server, the fsfilt_ext3 module could not be
3347 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3348 kernels. Previously, lconf would still use the ext3 name
3349 when trying to load modules. Now, it will correctly use
3350 ext3 on 2.4 and ldiskfs on 2.6.
3352 Severity : enhancement
3353 Description: The default stripe count has been changed to 1
3354 Details : The interpretation of the default stripe count (0, to lfs
3355 or lmc) has been changed to mean striping across a single
3356 OST, rather than all available. For general usage we have
3357 found a stripe count of 1 or 2 works best.
3359 Severity : enhancement
3360 Description: Add support for compiling against Cray portals.
3361 Details : Conditional compiling for some areas that are different
3365 Frequency : occasional
3366 Bugzilla : 6409, 6834
3367 Description: Creating files with an explicit stripe count may lead to
3368 a failed assertion on the MDS
3369 Details : If some OSTs are full or unavailable, creating files may
3370 trigger a failed assertion on the MDS. Now, Lustre will
3371 try to use other servers or return an error to the
3375 Frequency : occasional
3377 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3378 caused lock timeout during readahead (since 1.4.2).
3379 Details : Processes doing readahead might match a lock that hasn't been
3380 granted yet if there are overlapping and conflicting lock
3381 requests. The readahead process waits on ungranted lock
3382 (original lock is CBPENDING), while OST waits for that process
3383 to cancel CBPENDING read lock and eventually evicts client.
3385 Severity : enhancement
3387 Description: Initial enabling of flock support for clients
3388 Details : Implements fcntl advisory locking and file status functions.
3389 This feature is provided as an optional mount flag (default
3390 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3391 locking are implemented yet, and those that are are not guaranteed
3392 to be completely correct in production environments.
3393 mount -t lustre -o [flock|noflock] ...
3396 Frequency : occasional
3398 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3399 assertion in the ext3 JBD (journaling) layer.
3400 Details : The b_committed_data struct is protected by the big kernel lock
3401 in 2.4 kernels, serializing journal_commit_transaction() and
3402 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3403 access to this struct. In 2.6 kernels there is finer grained
3404 locking to improve SMP performance of the JBD layer.
3408 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3413 Description: merge and cleanup kernel patches.
3416 -----------------------------------------------------------------------------
3418 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3420 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3421 versions of Lustre. Please update all servers and clients to
3422 version 1.4.2 or later at the same time. You must also run
3423 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3424 to update the configuration logs.
3426 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3427 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3428 - swab llog records of type '0' so we get proper header size/idx (5861)
3429 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3430 - fix rename of one directory over another leaking an inode (5953)
3431 - avoid SetPageDirty on 2.6 (5981)
3432 - don't re-add just-being-destroyed locks to the waiting list (5653)
3433 - when creating new directories, inherit the parent's custom
3434 striping settings if present parent (3048)
3435 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3436 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3437 - don't hold client locks on temporary worklist from l_lru (5666)
3438 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3439 - automatically evict dead clients (5921)
3440 - Update file size properly in create+truncate+fstat case (6196)
3441 - Do not unhash mountpoint dentries, do not allow removal of
3443 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3444 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3445 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3446 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3447 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3448 - Correctly handle reads of files with no objects (6243)
3449 - lctl recover will also mark a device active if deactivate used (5933)
3451 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3452 - allow --write-conf on an MDS with different nettype than client (5619)
3453 - don't write config llogs to MDS for mounts not from that MDS (5617)
3454 - lconf should create multiple TCP connections from a client (5201)
3455 - init scripts are now turned off by default; run chkconfig --on
3456 lustre and chkconfig --on lustrefs to use them
3457 - upcalls are no longer needed for clients to recover to failover
3459 - add --abort-recovery option to lconf to abort recovery on device
3461 - add support for an arbitrary number of OSTs (3026)
3462 - Quota support protocol changes.
3463 - forward compatibility changes to wire structs (6007)
3464 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3465 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3466 - improved "lustre status" script
3467 - initialize blocksize for non-regular files (6062)
3468 - added --disable-server and --disable-client configure options (5782)
3469 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3470 - Vanilla 2.4.29 support
3471 - increase maximum number of obd devices to 520 (6242)
3472 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3473 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3474 SLES 9 2.6.5 kernels
3475 - limit stripes per file to 160 (the maximum EA size) (6093)
3477 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3480 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3481 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3482 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3483 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3484 - small ext3 extents cleanups and fixes (5733)
3485 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3486 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3487 - avoid needless client->OST connect, fix handle mismatch (5317)
3488 - fix DLM error path that led to out-of-sync client, long delays (5779)
3489 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3490 - fix several locking issues related to i_size (5492,5624,5654,5672)
3491 - don't move pending lock onto export if it is already evicted (5683)
3492 - fix kernel oops when creating .foo in unlinked directory (5548)
3493 - fix deadlock in obdfilter statistics vs. object create (5811)
3494 - use time_{before,after} to avoid timer jiffies wrap (5882)
3495 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3496 - Do not leak request if server was not able to process it (5154)
3497 - If mds_open unable to find parent dir, make that negative lookup(5154)
3498 - don't create new directories with extent-mapping (5909, 5936)
3500 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3501 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3502 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3503 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3504 - support for RHEL4 kernel on i686 (5773)
3505 - provide error messages when incompatible logs are encountered (5898)
3507 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3508 * version 1.4.0.10 (1.4.1 release candidate 1)
3510 - don't keep a lock reference when lock is not granted (4238)
3511 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3512 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3513 - drop import inflight refcount on signal_completed_replay error (5255)
3514 - unlock page after async write error during send (3677)
3515 - handle missing objects in filter_preprw_read properly (5265)
3516 - no transno return for symlink open, don't save no-trasno open (3440)
3517 - don't try to complete elan receive that already failed (4012)
3518 - free RPC server reply state on error (5406)
3519 - clean up thread from ptlrpc_start_thread() on error (5160)
3520 - readahead could read extra page into cache that wasn't ejected (5388)
3521 - prevent races in class_attach/setup/cleanup/detach (5260)
3522 - don't dereference de->d_inode after l_dput of de (5458)
3523 - use "int" for stripe value returned from lock_to_stripe (5544)
3524 - mballoc allocation and error-checking fixes in 2.6 (5504)
3525 - block device patches to fix I/O request sizes in 2.6 (5482)
3526 - look up hostnames for IB nals (5602)
3527 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3528 - don't start multiple acceptors for the same port (5277)
3529 - fix incorrect LASSERT in mds_getattr_name (5635)
3530 - export a proc file for general "ping" checking (5628)
3531 - fix "lfs check" to not block when the MDS is down (5628)
3533 - service request history (4965)
3534 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3535 - create an "evict_client" /proc entry on OSTs, like the MDS has
3536 - fix mount usage message, return errors per mount(8) (5168)
3537 - change grep [] to grep "[]" in tests so they work in more UMLs
3538 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3539 - remove ancient LOV_MAGIC_V0 EA support (5047)
3540 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3541 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3542 - allow readahead allocations to fail when low on memory (5383)
3543 - mmap locking landed again, after considerable improvement (2828)
3544 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3546 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3549 - send OST transaction number in read/write reply to free req (4966)
3550 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3551 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3552 - the watchdog thread now runs as interruptible (5246)
3553 - flock/lockf fixes (but it's still disabled, pending 5135)
3554 - don't use EXT3 constants in llite code (5094)
3555 - memory shortage at startup could cause assertion (5176)
3557 - reorganization of lov code
3558 - single portals codebase
3560 - add extents/mballoc support (5025)
3561 - direct I/O reads in the obdfilter (4048)
3562 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3564 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3567 - send OST transaction number in read/write reply to free req (4966)
3568 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3569 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3570 - the watchdog thread now runs as interruptible (5246)
3571 - handle missing objects in filter_preprw_read properly (5265)
3572 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3573 - drop import inflight refcount on signal_completed_replay error (5255)
3574 - unlock page after async write error during send (3677)
3575 - return original error code on reconstructed replies (3761)
3576 - no transno return for symlink open, don't save no-trasno open (3440)
3578 - add pid to ldlm debugging output (4922)
3579 - bump the watchdog timeouts -- we can't handle 30sec yet
3580 - extra debugging for orphan dentry/inode bug (5259)
3582 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3585 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3586 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3587 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3588 - fix NULL dereference in /proc/sys/portals/routes (4827)
3589 - allow failed mdc_close() operations to be interrupted (4561)
3590 - stop precreate on OST before MDS would time out on it (4778)
3591 - don't send partial-page writes before EOF from client (4410)
3592 - discard client grant for sub-page writes on large-page clients (4520)
3593 - don't free dentries not owned by NFS code, check generation (4806)
3594 - fix lsm leak if mds_create_objects() fails (4801)
3595 - limit debug_daemon file size, always print CERROR messages (4789)
3596 - use transno after validating reply (3892)
3597 - process timed out requests if import state changes (3754)
3598 - update mtime on OST during writes, return in glimpse (4829)
3599 - add mkfsoptions to LDAP (4679)
3600 - use ->max_readahead method instead of zapping global ra (5039)
3601 - don't interrupt __l_wait_event() during strace
3603 - add software watchdogs to catch hung threads quickly (4941)
3604 - make lustrefs init script start after nfs is mounted
3605 - fix CWARN/ERROR duplication (4930)
3606 - return async write errors to application if possible (2248)
3607 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3608 - print NAL number in %x format (4645)
3609 - update barely-supported suse-2.4.21-171 series (4842)
3610 - support for sles 9 %post scripts
3611 - support for building 2.6 kernel-source packages
3612 - support for sles km_* packages
3614 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3617 - ignore -ENOENT errors in osc_destroy (3639)
3618 - notify osc create thread that OSC is being cleaned up (4600)
3619 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3620 - reconstruct ost_handle() like mds_handle() (4657)
3621 - create a new thread to do import eviction to avoid deadlock (3969)
3622 - let lconf resolve symlinked-to devices (4629)
3623 - don't unlink "objects" from directory with default EA (4554)
3624 - hold socknal file ref over connect in case target is down (4394)
3625 - allow more than 32000 subdirectories in a single directory (3244)
3626 - fix blocks count for O_DIRECT writes (3751)
3627 - OST returns ENOSPC from object create when no space left (4539)
3628 - don't send truncate RPC if file size isn't changing (4410)
3629 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3630 - bind to privileged port in socknal and tcpnal (3689)
3632 - rate limit CERROR/CWARN console message to avoid overload (4519)
3633 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3634 - basic mmap support (3918)
3635 - kernel patch series update from b1_4 (4711)
3637 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3640 - avoid crash during MDS cleanup with OST shut down (2775)
3641 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3642 - don't use bad inodes on the MDS (3744)
3643 - dynamic object preallocation to improve recovery speed (4236)
3644 - don't hold spinlock over lock dumping or change debug flags (4401)
3645 - don't zero obd_dev when it is force cleaned (3651)
3646 - print grants to console if they go negative (4431)
3647 - "lctl deactivate" will stop automatic recovery attempts (3406)
3648 - look for existing locks in ldlm_handle_enqueue() (3764)
3649 - don't resolve lock handle twice in recovery avoiding race (4401)
3650 - revalidate should check working dir is a directory (4134)
3652 - don't always mark "slow" obdfilter messages as errors (4418)
3654 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3657 - don't close LustreDB during write_conf until it is done (3860)
3658 - fix typo in lconf for_each_profile (3821)
3659 - allow dumping logs from multiple threads at one time (3820)
3660 - don't allow multiple threads in OSC recovery (3812)
3661 - fix debug_size parameters (3864)
3662 - fix mds_postrecov to initialize import for llog ctxt (3121)
3663 - replace config semaphore with spinlock (3306)
3664 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3665 - don't allow enqueue to complete on a destroyed export (3822)
3666 - down write_lock before checking llog header bitmap (3825)
3667 - recover from lock replay timeout (3764)
3668 - up llog sem before sending rpc (3652)
3669 - reduce ns lock hold times when setting kms (3267)
3670 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3671 - fix NULL deref and obd_dev leak on setup error (3312)
3672 - replace some LBUG about llog ops with error handling (3841)
3673 - don't match INVALID dentries from d_lookup and spin (3784)
3674 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3675 - fix invalid assertion in ptlrpc_set_wait (3880)
3677 - add libwrap support for the TCP acceptor (3996)
3678 - add /proc/sys/portals/routes for non-root route listing (3994)
3679 - allow setting MDS UUID in .xml (2580)
3680 - print the stack of a process that LBUGs (4228)
3682 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3685 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3686 - make sure to unset replay flag from failed open requests (3440)
3687 - if default stripe count is 0, use OST count for inode size (3636)
3688 - update parent mtime/ctime on client for create/unlink (2611)
3689 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3690 - free recovery state on server during a forced cleanup (3571)
3691 - unregister_reply for resent reqs (3063)
3692 - loop back devices mounting and status check on 2.6 (3563)
3693 - fix resource-creation race that can provoke i_size == 0 (3513)
3694 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3695 - more debugging for page-accounting assertion (3746)
3696 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3697 - avoid deadlock after precreation failure (3758)
3698 - fix race and lock order deadlock in orphan handling (3450, 3750)
3699 - add validity checks when grabbing inodes from l_ast_data (3599)
3701 - add /proc/.../recovery_status to obdfilter (3428)
3702 - lightweight CDEBUG infrastructure, debug daemon (3668)
3703 - change default OSC RPC parameters to be better on small clusters
3704 - turn off OST read cache for files smaller than 32MB
3705 - install man pages and include them in rpms (3100)
3706 - add new init script for (un)mounting lustre filesystems (2593)
3707 - run chkconfig in %post for init scripts (3701)
3708 - drop scimac NAL (unmaintained)
3710 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3713 - clean kiobufs before and after use (3485)
3714 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3715 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3716 - change init script to fail more clearly if not run as root (1528)
3717 - allow clients to reconnect during replay (1742)
3718 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3719 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3720 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3721 - fix cond_resched (3554)
3722 - only evict unfinished clients after recovery (3515)
3723 - allow bulk resend, prevent data loss (3570)
3724 - dynamic ptlrpc request buffer allocation (2102)
3725 - don't allow unlinking open directory if it isn't empty (2904)
3726 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3727 - remove extraneous obd dereference causing LASSERT failure (3334)
3728 - don't use get_cycles() when creating temp. files on the mds (3156)
3729 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3730 - handle EEXIST for set-stripe, set proper directory name (3336)
3732 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3733 - fix ksocknal_fmb_callback() error messages (2918)
3735 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3738 - don't copy lvb into (possibly NULL) reply on error (2983)
3739 - don't deref dentry after dput, don't free lvb on error (2922)
3740 - use the kms to determine writeback rpc length (2947)
3741 - increment oti_logcookies when osc is inactive (2948)
3742 - update client's i_blocks count via lvb messages (2543)
3743 - handle intent open/close of special files properly (1557)
3744 - mount MDS with errors=remount-ro, like obdfilter (2009)
3745 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3746 - don't use cancelling-locks' kms values (2947)
3747 - use highest lock extent for kms, not last one (2925)
3748 - don't dereference ERR_PTR() dentry in error handling path (3107)
3749 - fix thread race in portals_debug_dumplog() (3122)
3750 - create lprocfs device entries at setup instead of at attach (1519)
3751 - common AST error handler, don't evict client on completion race (3145)
3752 - zero nameidata in detach_mnt in 2.6 (3118)
3753 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3754 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3755 - initialize RPC timeout timer earlier for 2.6 (3219)
3756 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3757 - print nal/nid for unknown nid (3258)
3758 - additional checks for oscc recovery before doing precreate (3284)
3759 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3760 - don't crash in mdc_close for bad permissions on open (3285)
3761 - zero i_rdev for non-device files (3147)
3762 - clear page->private before handing to FS, better assertion (3119)
3763 - tune the read pipeline (3236)
3764 - fix incorrect decref of invalidated dentry (2350)
3765 - provide read-ahead stats and refine rpc in flight stats (3328)
3766 - don't hold journal transaction open across create RPC (3313)
3767 - update atime on MDS at close time (3265)
3768 - close LDAP connection when recovering to avoid server load (3315)
3769 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3770 - don't leak open file on MDS after open resend (3325)
3771 - serialize filter_precreate and filter_destroy_precreated (3329)
3772 - loop device shouldn't call sync_dev() for nul device (3092)
3773 - clear page cache after eviction (2766)
3774 - resynchronize MDS->OST in background (2824)
3775 - refuse to mount the same filesystem twice on same mountpoint (3394)
3776 - allow llmount to create routes for mounting behind routers (3320)
3777 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3778 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3779 - fix rare problem with rename on htree directories (3417)
3781 - allow default OST striping configuration per directory (1414)
3782 - fix compilation for qswnal for 2.6 kernels (3125)
3783 - increase maximum number of MDS request buffers for large systems
3784 - change liblustreapi to be useful for external progs like lfsck (3098)
3785 - increase local configuration timeout for slow disks (3353)
3786 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3788 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3791 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3792 - don't overwrite extent policy data in reply if lock was blocked (2901)
3793 - drop filter export grants atomically with removal from device (2663)
3794 - del obd_self_export from work_list in class_disconnect_exports (2908)
3795 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3796 - swab reply message in mdc_close, other PPC fixes (2464)
3797 - fix destroying of named logs (2325)
3798 - overwrite old logs when running lconf --write_conf (2264)
3799 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3800 - fix race in target_handle_connect (2898)
3801 - mds_reint_create() should take same inode create lock (2926)
3802 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3803 - don't close files for self_export to avoid uninitialized obd (2936)
3804 - allow MDS with the same name as client node (2939)
3805 - hold dentry reference for closed log files for unlink (2325)
3806 - reserve space for all logs during transactions (2059)
3807 - don't evict page beyond end of stripe extent (2925)
3808 - don't oops on a deleted current working directory (2399)
3809 - handle hard links to targets without a parent properly (2517)
3810 - don't dereference NULL lock when racing during eviction (2867)
3811 - don't grow lock extents when lots of conflicting locks (2919)
3813 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3816 - account for cache space usage on clients to avoid data loss (974)
3817 - lfsck support in lustre kernel code (2349)
3818 - reduce journal credits needed for BRW writes (2370)
3819 - orphan handling to avoid losing space on client/server crashes
3820 - ptlrpcd can be blocked, stopping ALL progress (2477)
3821 - use lock value blocks to assist in proper KMS, faster stat (1021)
3822 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3823 - recovery for initial connections (2355)
3824 - fixes for mds_cleanup_orphans (1934)
3825 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3826 - block all file creations until orphan recovery completes (1901)
3827 - client remove rq_connection from request struct (2423)
3828 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3829 - recovery timer race (2670)
3830 - mdc_close recovey bug (2532)
3831 - ptlrpc cleanup bug (2710)
3832 - mds timeout on local locks (2588)
3833 - namespace lock held during RPCs (2431)
3834 - handle interrupted sync write properly (2503)
3835 - don't try to handle a message that hasn't been replied to (2699)
3836 - client assert failure during cleanup after abort recovery (2701)
3837 - leak mdc device after failed mount (2712)
3838 - ptlrpc_check_set allows timedout requests to complete (2714)
3839 - wait for inflight reqs when ptlrpcd finishes (2710)
3840 - make sure unregistered services are removed from the srv_list
3841 - reset bulk XID's when resending them (caught by 1138 test)
3842 - unregister_bulk after timeout
3843 - fix lconf error (2694)
3844 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3845 - readahead locks pages, leaves pending causing memory pressure (2673)
3846 - increase OST request buffers to 4096 on large machines (2729)
3847 - fix up permission of existing directories in simple_mkdir (2661)
3848 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3849 - don't assign transno to errored transactions (2742)
3850 - don't delete objects on OST if given a bogus objid from MDS (2751)
3851 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3852 - if rq_no_resend, then timeout request after recovery (2432)
3853 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3854 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3855 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3856 - MDS recovery shouldn't delete objects that it has given out (2730)
3857 - if enqueue arrives after completion, don't clobber LVB (2819)
3858 - don't unlock pages twice when trigger_group_io returns error (2814)
3859 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3860 - don't write pages to disk if there was an error (1450)
3861 - don't ping imports that have recovery disabled (2676)
3862 - take buffered bytes into account when balancing socknal conn (2817)
3863 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3864 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3865 - remove little-endian swabbing of llog records (1987)
3866 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3867 - retry reposting request buffers if they fail (1191)
3868 - grow extent at grant time to avoid granting a revoked lock (2809)
3869 - lock revoke doesn't evict page if covered by a second lock (2765)
3870 - disable VM readahead to avoid reading outside lock extents (2805)
3872 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3873 - updated kernel patches for hp-2.4.20 kernel (2681)
3875 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3878 - fix truncated write corruption (2366)
3879 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3881 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3882 - ensure len(uuid) < 37 in lmc (1171)
3883 - fix ia64 OOPS in llog_test (2255)
3884 - zero end of page at obdfilter for partial page writes (2648)
3885 - don't leave stale dentries around after renames (bug 2428)
3886 - fix timeouts when evicting a client with a single lock held (2642)
3887 - set deadline for the initial HELLO message to drain (2634)
3888 - print out dotted-quad IP addresses in the socknal (2302)
3890 - additional debugging for MDS client eviction problem (2443)
3891 - fix mkfsoptions support for osts (2603, 2604)
3893 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3896 - add series for the vanilla 2.6.0 kernel
3897 - add series for the vanilla 2.4.24 kernel
3898 - add series for a cray x86/64 UL kernel drop
3899 - fix xattr patches for the vanilla 2.4.19 series
3901 - generate true UUIDs in lmc (1171)
3902 - have portals stack dumping break in UML (2466)
3903 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3904 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3905 - wake callback waiting threads on client eviction (2460)
3906 - Add --ptldebug and --subsystem to lmc (1719)
3907 - update assertion to allow safe interrupt allocation
3908 - set rq_no_resend for cancel requests (2432)
3909 - recalculate ptlrpcd timeout after resend (2494)
3910 - call vfs_rmdir when removing pending directories (2368)
3911 - fix renaming a file to itself (2429)
3912 - lmc creates a default one-stripe lov (2454)
3913 - expand procfs space to handle large clusters (2326)
3914 - increase UML stack to avoid overflow
3915 - update lconf's list of debug and subsystem masks
3916 - fix lfs find --obd (2510)
3917 - /proc tunable for disabling filter read caching (2591)
3918 - stop rpm packages from altering slapd.conf (2301)
3919 - disable nagle in the socknal under 0conf (2578)
3920 - choose mds inode size based on stripe count (2572)
3921 - fix kernel-source rpm problems (2516)
3923 - add --disable-doc to avoid pdf generation (2421)
3924 - update documentation, tests, type-os, comments
3925 - avoid format warnings on ia64
3926 - remove the TOE NAL
3927 - tiny code cleanups by removing unused fields
3929 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3932 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3933 - remove the most blatant lies from BUILDING file (2371)
3934 - change default debug level to reasonable production setting
3935 - reduce client side cache size to reduce cache flush time
3936 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3937 - make TCP zerocopy and pinger support enabled by default (2476)
3938 - sync writes completed after process exits caused crashes (2319)
3939 - maintain correct mount count on the MDS (2356)
3940 - backout 1557, because 2316 wasn't really fixed
3941 - better file I/O statistics gathering in /proc
3942 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3943 - another kernel patch to fix zero-copy TCP function export
3944 - don't take duplicate lock when processing re-sent getattr (2420)
3945 - lctl uses obd_self_export instead of creating new conn (2353)
3946 - MDS/OST recovery case which requires object creation asserted (2425)
3947 - move lfs from /usr/sbin to /usr/bin in packages
3948 - fix race between mds_client_add and mds_client_free (2417)
3949 - use kmalloc instead of slabs in portals (2430)
3950 - don't create duplicate records when a failover MDS is present (2442)
3951 - remove unnecessary mount age check (2332)
3952 - don't remove directory inodes from locks prematurely (2451)
3953 - don't break if MDS service name is the same as hostname (2103)
3954 - fix races in client write RPC generation when cache full (2482)
3956 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3959 - remove now-unused request->rq_obd (278)
3960 - if an allocation fails, print out how much memory we've used (1933)
3961 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3962 - add the "configurable stack size" patch to most series files (1256)
3963 - ability to write large log records, for 100+ OST configs (2306)
3964 - fix NULL deref when filter_prep fails (2314)
3965 - fix operator precedence error in filter_sync
3966 - dynamic allocation of socknal TX descriptors (2315)
3967 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3968 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3969 - ensure that CWARN messages in Portals always get to the syslog
3970 - __init/__exit are not for prototype decls (ldlm_init/exit)
3971 - x86-64 compile warning fixes
3972 - fix gateway LMC keyword conflict (2318)
3973 - fix MDS lock inversions in getattr/reint paths (1844)
3974 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3975 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3976 - DLM race condition prevented some lock evictions (2328)
3977 - ENOMEM detection and retry on socknal sends (2230)
3978 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3979 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3980 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3981 - better error messages when a client is rejected during recovery (1505)
3982 - avoid cancelling locks which were never granted, after failure (2330)
3983 - fix i_sem/journal inversion in mds_client_add (2333)
3984 - fix truncate/getattr lock cycle deadlock (2334)
3985 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3986 - fix two rare exit paths which could leak an l_lock() ref (2321)
3987 - fencepost error in MDS/OST orphan recovery (2226)
3988 - make log record alignment 8 bytes (1988)
3989 - lstripe now fails when requested offset > ost_count (2237)
3990 - ensure that all kernel series have a complete list.h (1607)
3991 - fix crashes in special-file operations (2316)
3992 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3994 - allow configurable automake binary, for testing new versions
3995 - small update to the lfs documentation
3997 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3999 * fix negative export reference count in fsfilt_sync (2312)
4001 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4002 * release candidate 0.9.1
4004 - orphans are moved into the PENDING directory for possible recovery
4005 - replayed opens now open by fid for orphan/rename safety (1042)
4006 - last close of an orphan inode generates a transno (683)
4007 - chdir() and mount() now pin the directory entry (1020)
4008 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4009 - discard very old requests without processing them (1502)
4010 - remove some common, well-understood CERRORs (1505)
4011 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4012 - clear "grant" flags in OST replies until OST grant code lands (1644)
4013 - fix read performance by not clobbering i_blksize on client (1598)
4014 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4015 - make LRU size a /proc tunable, clears locks when reduced (707)
4016 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4017 - server locks take references on exports now (1558)
4018 - build fixes for 2.4.20-rh trees (1663)
4019 - return an error from lov_create if all OSCs are inactive (1751)
4020 - fix import levels when a reconnect happens without a timeout (1597)
4021 - exit early from mds_open if we get a lookup error (1749)
4022 - partial page read at EOF wouldn't wait for disk before sending (1642)
4023 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4024 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4025 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4026 - write/truncate lock inversion (1639)
4027 - Don't auto-load obdclass, portals modules during cleanup (1495)
4028 - fix timestamps from jumping to "now" (1763)
4029 - extra journal assertions (1648)
4030 - add an extra multiunlink test (1771)
4031 - fix read_record/write_record API (1776)
4032 - fix leak of offset_extent, possible incorrect i_size later (1772)
4033 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4034 - Debugging for the kqswnal_get_idle_tx problems (1820)
4035 - Allow recovery to be attempted multiple times (1536)
4036 - Write out MDS last_rcvd file after it is first created (1600)
4037 - Fix tx_descriptor leak in failed transmit situations (1827)
4038 - ext3 journaling fixes for assertion failure after IO error (1871)
4039 - class_export_put() on freed export after completion AST error (1896)
4040 - Fix revalidate looping in VFS (1322)
4041 - Don't access a freed export during MDS_REINT timeout (1521)
4042 - Add open-unlink recovery support on the MDS (1673,1764)
4043 - Return an error if no MDS data was read from last_rcvd (1946)
4044 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4045 - Don't setup a disk device that doesn't match exported UUID (317)
4046 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4047 - avoid committing NULL handle in force close
4048 - local.sh is now a one-stripe LOV configuration
4049 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4050 - fix ext3 htree duplicate directory entry corruption (1516)
4051 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4052 - update to new LOV EA format (2097)
4053 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4055 2003-06-15 Phil Schwan <phil@clusterfs.com>
4058 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4059 - per-import/export recovery handling (958, 931, 959)
4060 - multiple last-rcvd slots, for serving multiple FSes (949)
4061 - connections are again shared between multiple imp/exports (963, 964)
4062 - "umount -f" would hang if any requests needed to be sent (393, 978)
4063 - avoid pinning large req buffer by copying for queued messages (989)
4064 - add "uuid" to "lctl device" command to help upcalls (991)
4065 - "open" RPCs with transnos would confuse recovery counters (1037)
4066 - do proper endian conversion of all wire messages (288, 340, 891)
4067 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4068 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4069 - fix ptlrpc_connection leak in target_handle_connect (1174)
4070 - fix import refcounting bug in OST and MDS cleanup (1134)
4071 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4072 - fix very unlikely obd_types race condition (501)
4073 - remove osc_open hack for echo_client (1187)
4074 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4075 - a failure in read_inode2 leads to deadlock (1139)
4076 - cancel ack-locks as soon as transaction is committed (1072)
4077 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4078 - make sure to commitrw after any preprw to avoid deadlock (1162)
4079 - failing to execute a file in a lustre FS would lock inode (1203)
4080 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4081 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4082 - obd_uuid2tgt would walk off the end of the list (1255)
4083 - on IA64 the osc would give portals incorrect bulk size (1258)
4084 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4085 - fix lock inversion caused by new llite matching code (1282)
4086 - limit the number of dirty pages on a client to 10MB (1286)
4087 - timed out locks were not being corrected cancelled (1289)
4088 - fix O_DIRECT above 4GB on IA-32 (1292)
4089 * major user-visible changes
4090 - fail out/fail over policy now controlled by the upcall (993)
4092 - add OBD_PING to check server availability and failure (954)
4093 - lustre messages are now sent in sending host order (288, 340, 891)
4094 - add eadatalen to MDS getattr reply (340)
4095 - OST read replies may contain second buffer, with per-page status (593)
4097 2003-03-11 Phil Schwan <phil@clusterfs.com>
4100 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4101 - failed open()s could cause deadlock; fixed (867, 869)
4102 - stop cancelling OST locks when files are closed (481)
4103 - overlapping XID spaces caused network corruption (851, 853)
4104 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4105 - setattr_raw added, to do single-RPC, server-side setattrs
4106 - lmc/lconf syntax change for OST UUIDs
4107 - fix crashy race condition between ptlrpc_free_req and osc_close
4108 - don't use request in mdc_enqueue if we hit a timeout (889)
4109 - don't set the inode i_size for regular files from the MDS (896)
4110 - handle out of order completion AST (842)
4111 - don't LBUG if a lock request times out after receiving AST (913)
4112 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4113 - if a bad lock AST arrives, send an error instead of dropping entirely
4114 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4115 - fix leak in bulk IO when only partially completed (899, 900, 926)
4116 - fix O_DIRECT for ia64 (55)
4117 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4118 - C-z after timeout could hang a process forever; fixed (977)
4120 - client-side I/O cache (678, 924, 929, 941, 970)
4122 - READPAGE and SETATTRs which don't take server-side locks get
4125 2003-02-11 Phil Schwan <phil@clusterfs.com>
4128 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4129 - Fix fsfilt_extN_readpage() to read a full page of directory
4130 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4131 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4132 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4133 - Most intent operations were removed, in favour of a new RPC mode
4134 that does a single RPC to the server and bypasses most of the VFS
4135 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4136 - Aggressively cancel local locks on DLM servers
4137 - mds_reint_unlink sends EA to the client if it's the last nlink.
4138 client uses that EA to unlink OST objects.
4139 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4140 - recursive symlinks were fixed (439)
4141 - fixed NULL deref in DEBUG_REQ
4142 - filter_update_lastobjid no longer calls sync, which annoyed extN
4143 - fixed multi-client small-writes to a single file problem (445)
4144 - fixed mtime updates during file writes (607)
4145 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4146 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4147 - fixed Linux OST size reporting problem (444, 656)
4148 - OST now updates object mtime with writes or setattr (607, 619)
4149 - client verifies file size before zeroing page past EOF (445)
4150 - OST now writes last allocated objid to disk with allocation (108)
4151 - LOV on echo now works (409)
4153 - mds_reint_unlink sends a new buffer, with the EA included. this
4154 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4155 set if a regular file was being unlinked, and it was the last link
4156 - use PtlGet from the target for bulk writes (315)
4157 - OST now updates object mtime with writes or setattr (607, 619)
4158 - LDLM now has a grant-time callback to revalidate locked items, if
4160 - Many MDS operations were reorganized to combat race conditions
4162 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4163 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4165 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4168 - Fully reactivate OST imports after reconnection (512, others)
4169 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4170 - More graceful error handling for truncating on dead OST (515)
4171 - Don't error out unless we're actually accessing dead stripes (474)
4172 - Fix garbage sizes when stripes are missing (410)
4173 - LRU counters were broken, causing constant lock purge (433, 432)
4174 - garbage on read from stripes with failed OSTs (441)
4175 - mark OSCs as active before reconnecting during recovery (438)
4176 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4177 - lfind did not preserve OST order in output (443)
4178 - symlinks cause hung clients, incorrect data (439)
4179 - stop dereferencing request after dropping refcount (457)
4180 - don't LASSERT(spin_is_locked) on non-SMP (455)
4181 - fixes for many rename() bugs
4182 - fstat didn't correctly synchronize attributes (399)
4183 - server must handle lock cancellation during blocking AST prep (487)
4184 - bulk descriptors were free()d too soon (511)
4185 - fix paths in lconf, which would load incorrect modules (451, 507)
4186 - fix confusing lconf 'host not found' error message (386)
4187 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4188 - fix race condition in mdc_blocking_ast() for inode access (526)
4189 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4190 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4191 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4192 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4193 - fix lstripe in several ways (kernel side) (527)
4194 - fix request leak in ldlm_cli_enqueue (262)
4195 - incorrect OSC was marked inactive after OST failure
4196 - call mds_fs_cleanup before unmounting filesystem (524)
4197 - fix races between taking ns_lock and ldlm_lock_change_resource
4198 - fix races updating LOV export open file list
4199 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4200 - fix recovery NULL deref in ldlm_cli_cancel_unused
4201 - fix some DLM races by using new hash table for lock handles (419)
4202 - permit the client to specify desired inodes, at replay
4203 - duplicate requests when we queue them for replay reintegration
4204 - fix last_rcvd offset calculation
4205 - sync after each recovered transaction, so we always make progress
4206 - never, not always, ERESTART requests without transnos
4207 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4209 - skip replay if the MDS says that the client is already connected
4210 - don't check for a recovery-enabled export to match lctl's UUID
4211 - don't INC_USE_COUNT for phantom exports
4212 - don't crash when cleaning up phantom exports (567)
4213 - don't double-finish or set replay data for errored mdc_open requests
4214 - abort requests when they time out, so we don't get old replies
4215 - send/receive replies for AST messages again
4216 - if the client says that it doesn't have the lock, cancel it on the
4218 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4219 mark it as destroyed, it will all work out when decref is called
4220 - fix module use counts (22, 581)
4222 - ASTs now expect a reply (server cancels lock on error reply)
4224 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4227 - fix many simultaneous client startup (392)
4228 - fix dentry->d_it clobbering
4229 - credentials weren't being shipped for readdir/getattr operations
4230 - remove invalid assertions triggered during some concurrent MD
4232 - proper Lustre versions added (336, 389)
4233 - fix memory leak for create error case (398)
4234 - fix LOV locking bug that would get cli/srv out of sync
4235 - fix echo client over LOV (409)
4236 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4237 - fix double-O_EXCL intent crash (424)
4238 - avoid sending multiple lock CANCELs (352)
4240 - MDS can do multi-client recovery (modulo bugs in new code)
4242 - many updates, edits, cleanups
4244 2002-11-18 Phil Schwan <phil@clusterfs.com>
4247 - fix null d_it dereference (346)
4248 - fix full OST/dbench hang (333)
4249 - fix permission problem with file removal (286)
4250 - fix removal of OSCs from LOV when they fail
4251 - fix NULL deref during bulk timeout (214)
4252 - fix problems related to multiple filesystems on one MDS (241)
4253 - fixed serious subtle metadata locking bugs
4254 - free locks on clients when inodes are removed due to memory
4256 - fix inode pointer in lock data (285)
4257 - partial support for multiple MDS on a single host (241)
4258 - data locks weren't cancelled at clear_inode time (290, 311)
4259 - intent locks could lead to unbounded lock growth (205)
4260 - added a maximum lock count, an LRU list, and a flusher
4261 - fix multiple rename (365)
4262 - properly abstracted the echo client
4263 - OSC locked 1 byte too many; fixed
4264 - rewrote brw callback code:
4265 - fixed recovery bugs related to LOVs (306)
4266 - fixed too-many-pages-in-one-write crash (191)
4267 - fixed (again) crash in sync_io_timeout (214)
4268 - probably fixed callback-related race (385)
4270 - Add capability to MDS protocol
4271 - LDLM cancellations and callbacks on different portals
4273 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4276 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4277 - defer open object destruction to close time (601981, 138)
4278 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4279 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4280 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4281 - don't call obd_setattr in ll_file_release for destroyed objects
4282 * protocol change to lustre_msg: move |version| and add |flags|
4283 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4284 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4285 * added replay of create, unlink, link and rename operations during
4286 MDS failover; recovery should be much more robust now
4287 * remove failed OSCs from LOVs (only lov_create uses this so far)
4288 * the lustre-HOWTO was brought (more) up to date (582544)
4290 2002-10-23 Phil Schwan <phil@clusterfs.com>
4293 - in-use dentries weren't being reused properly (617851)
4294 - prevent multiple LDLM setup (599178)
4295 - fix LOV size calculations for truncate (617853)
4296 - fix client handling of MDS intent errors (POSIX)
4297 - fix permission bug in lovstripe.c test (624321)
4298 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4299 - truncate past end of file could corrupt data
4300 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4301 - a race in recovery could return ETIMEDOUT to apps (623947)
4302 - building outside the source directory was fixed
4303 * the lustre-HOWTO was brought (more) up to date (582544)
4304 * major progress was made on recovery functionality
4306 2002-10-10 Phil Schwan <phil@clusterfs.com>
4309 - recovery deadlock fix
4310 - rm -rf causes LBUG fix (617817)
4311 - file open by multiple tasks fix (618962)
4312 - directory permissions bugs (602707 and 620007)
4313 - journal_stop fixed with locking (611313)
4314 - O_APPEND failures resolved (618273, perhaps 614459)
4315 - lconf PATH fix (619770)
4316 - IA64 build fix (621450)
4317 - RPC buffer sizes scale with amount of memory
4319 2002-10-01 Phil Schwan <phil@clusterfs.com>
4322 - locks would be cancelled without throwing away data pages,
4323 resulting in inconsistent data (605627)
4324 - inode attributes were not always being refreshed (605627, 612449)
4325 - lconf now continues to cleanup after lctl reports an error
4326 - MDS now enforces user permissions (602707)
4327 - lprocfs cleanup fixed, but not yet enabled (614157)
4328 - fixed infinite server hang, should a client not respond to an AST
4329 - avoid going into recovery if user calls readlink() with a buffer
4330 that's too small (613941)
4331 - AST RPCs no longer require replies (614867) -- this may be changed
4332 - don't crash server if client sends an IOV that's too big (611336)
4333 - fixed lock conversion deadlock (611892)
4334 - fixed the following of symlinks (614622)
4335 * recovery: the server can remove locks from a client that dies, other
4336 clients can make progress
4337 * more extN patch fixes
4338 * compile-time configurable ptlrpc buffer allocations
4340 - collaborative read cache document
4341 - Lustre Lite Performance CDR document-in-progress
4343 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4346 - fix typo in patch-2.4.18
4348 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4351 - clear ptlrpc request each time in handle_incoming_request()
4352 - unlink of files now destroys the object on the OST
4354 2002-09-19 Peter Braam <braam@clusterfs.com>
4356 * add hard link support
4357 * change obdfile creation method
4358 * kernel patch changed
4360 2002-09-19 Peter Braam <braam@clusterfs.com>
4363 - stack overflow bug in extN fixed
4365 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4367 * documentation updates
4368 - add man pages for config tools
4369 - update tests/README to describe testing with new config tools
4370 - finish metadata API descriptions
4371 * bug fixes and cleanups
4372 - statfs workaround for 16TB limit
4373 - LOV stripe allocation improved, can stripe on subset of OSTs
4374 - LOV file size/IO offset was wrong for files > 4GB in size
4375 - object EA data was being dropped, caused files to be unreadable
4376 - memory overflow with non-LOV OST caused memory corruption
4377 - fixed regression tests to work with new config tools, obdfilter
4378 - fixed bug when directory size became larger than 1 block
4379 - fixed bug (for single client case) when PWD was deleted
4380 - invalidate local directory pages when doing intent-based ops
4381 - avoid LDLM oops when lock callback contained bad data
4383 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4385 * documentation updates
4386 * bug fixes and cleanups
4387 - configuration tools
4390 - 64-bit compile warnings
4391 - 64-bit internal statfs data
4393 * test_brw on persistent OST devices
4395 * lprocfs (disabled)
4397 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4399 * documentation updates
4400 * bug fixes and cleanups
4401 * configuration tools
4403 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4406 * many small fixes to 0_5_4
4407 * io/network handling
4408 * thinkos in MDS operations
4410 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4413 * crucial basic fixes to 0.5.3
4414 * IOR, Iozone work over Elan
4417 2002-08-07 Phil Schwan <phil@clusterfs.com>
4418 * version 0_5_3, our first alpha
4419 * we use the new Portals iovs
4420 * documentation updates
4421 * bug fixes and cleanups
4422 * small changes in the DLM wire protocol
4424 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4425 * version 0_5_1 with some initial stability,
4426 * locking on MD and file I/O.
4427 * documentation updates
4428 * several bug fixes since 0.5.0
4429 * small changes in wire protocol
4431 2002-07-18 Phil Schwan <phil@clusterfs.com>
4433 * delivered as Lustre Light Alpha
4434 * fixed a crash after handling invalid MDS requests
4435 * fixed directory pages for architectures with non-4k pages sizes
4437 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4438 * release version v0_4_4
4439 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4441 * Fixes a number of interruption problems with OST operations.
4442 * Update documentation for portals header changes
4443 * Move all wire protocol structs/defines to lustre_idl.h
4444 * Fixes symlink length bug.
4445 * Add tcpdump to repository.
4447 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4448 * release version v0_4_3
4449 * Fixes statfs for inodes on extN.
4450 * Fixes bug in runtests which would delete /etc/hosts.
4451 * Use 64-bit object IDs wherever possible (not into VFS though)
4452 Remove ost_get_info, which is unused by lustre, and out of date.
4454 2002-07-03 Peter Braam <braam@clusterfs.com>
4455 * release version v0_4_2 Fixes a lookup error (type not passed)
4456 * move forward to head of Portals
4457 * move forward to latest Lustre kernel
4459 2002-06-25 Peter Braam <braam@clusterfs.com>
4460 * release version v0_4_1. Hopefully stable on single node use.