1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.5 (SLES 10),
6 2.6.18-53.1.13.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
15 Severity : enhancement
17 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
19 Severity : enhancement
21 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
26 Description: Fix replay if there is an un-replied request and open
27 Details : In some cases, older replay request will revert the
28 mcd->mcd_last_xid on MDS which is used to record the client's
31 Severity : enhancement
33 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
35 Severity : enhancement
37 Description: Add rhel5 support to HEAD.
39 Serverity : enhancement
41 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
46 Description: Don't allow skipping OSTs if index has been specified.
47 Details : Don't allow skipping OSTs if index has been specified, make locking
48 in internal create lots better.
52 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
53 Details : access to bitfield in structure is always rounded to long
54 and this produce problem with not atomic change any bit.
58 Description: Lustre make rpms failed.
59 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
60 builds Lustre rpms from tarball.
63 Frequency : rare on shutdown ost
65 Description: If llog cancel was not send before clean_exports phase, this can
66 produce deadlock in llog code.
67 Details : If llog thread has last reference to obd and call class_import_put
68 this produce deadlock because llog_cleanup_commit_master wait when
69 last llog_commit_thread exited, but this never success because was
70 called from llog_commit_thread.
74 Description: allow userland application know is lost one of stripes.
75 Details : fill lvb_blocks with error code on ost and return it to
76 application if error flag found.
80 Description: NULL lov_tgts causing MDS oops
81 Details : more safe checks for NULL lov_tgts for avoid oops.
83 Severity : enhancement
85 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
89 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
90 Details : If an OST is down the MDS will hang indefinitely in
91 obd_statfs_rqset() waiting for the statfs data. While for
92 MDS QOS usage of statfs, it should not stuck in waiting.
94 Severity : enhancement
96 Description: remote_acl support
97 Details : Support ACL-based permission check for remote user.
98 Support setfacl/getfacl for remote user with the utils
99 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
100 the system "{s,g}etfacl" utils.
102 Severity : enhancement
104 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
106 Severity : enhancement
108 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
112 Description: make mgs_setparam() handle fsname containing dash
113 Details : fsname containing a dash does not work with lctl conf_param
115 Severity : enhancement
117 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
119 Severity : enhancement
121 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
125 Frequency : rare, at shutdown
126 Description: access already free / zero obd_namespace.
127 Details : if client_disconnect_export was called without force flag set,
128 and exist connect request in flight, this can produce access to
129 NULL pointer (or already free pointer) when connect_interpret
130 store ocd flags in obd_namespace.
134 Frequency : only at startup
135 Description: not alloc memory with spinlock held.
136 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
137 if any spinlock held.
139 Severity : enhancement
141 Description: make lustre randomly fail allocating memory
142 Details : Make lustre randomly failed allocating memory for testing purpose.
144 Severity : enhancement
146 Description: lost problems with lov objid file
147 Details : Fixes some scability and access to not inited memory problems
148 in work with lov objdid file.
153 Description: lfs find does not continue on file error
154 Details : Continue other files processing when a file/dir is absent.
158 Description: Inconsistent usage of lustre_pack_reply()
159 Details : Standardize the usage of lustre_pack_reply() such that it
160 always generate a CERROR on failure.
163 Frequency : occasional
165 Description: MDS hang or stay in waiting lock
166 Details : If client receive lock with CBPENDING flag ldlm need send lock
167 cancel as separate rpc, to avoid situation when cancel request
168 can't processed due all i/o threads stay in wait lock.
172 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
176 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
180 Description: Update to SLES9 kernel-2.6.5-7.287.3.
184 Description: Update to RHEL4 latest kernel.
186 Severity : enhancement
188 Description: Build SLES10 patchless client fails
189 Details : The configure was broken by run ./configure with
190 --with-linux-obj=.... argument for patchless client. When the
191 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
192 can't search header adequately. Use absolute path such as
193 -I($LINUX)/include instead.
197 Description: interrupt oig_wait produce painc on resend.
198 Details : brw_redo_request can be used for resend requests from ptlrpcd and
199 private set, and this produce situation when rq_ptlrpcd_data not
200 copyed to new allocated request and triggered LBUG on assert
201 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
202 ptlrpcd set if request is changed and can be safety changed to use
207 Description: LASSERT_{REQ,REP}SWAB macros are buggy
208 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
209 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
210 these macros and replace them with their difinitions instead.
214 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
215 Details : Update which_patch & target file for SLES10 latest kernel.
218 Description: Remove client patches from SLES 10 kernel.
219 Details : This causes SLES 10 clients to behave as patchless clients
220 even on a Lustre-patched (server) kernel.
222 Severity : enhancement
224 Description: self-adjustable client's lru lists
225 Details : use adaptive algorithm for managing client cached locks lru
226 lists according to current server load, other client's work
227 pattern, memory activities, etc. Both, server and client
228 side namespaces provide number of proc tunables for controlling
231 Severity : enhancement
233 Description: light-weight GSS support
234 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
235 and reduce performance overhead.
237 Severity : enhancement
239 Description: Linux keyring support
240 Details : Support using service of Linux keyring for Lustre GSS internal
241 context refresh/cache mechanism.
245 Description: Fix errors in lfs documentation
246 Details : Fixes man pages
250 Description: don't use GFP_* in generic Lustre code.
251 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
255 Description: obdclass is limited by single OBD_ALLOC(idarray)
256 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
260 Description: warnings with build patchless client with vanila 2.6.19 and up
261 Details : change old ctl_table style and replace ctl_table/ctl_table_header
262 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
266 Frequency : Only for SLES
267 Description: sanity_quota fail test_1
268 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
269 file, which makes TSTID[2] inunique.
273 Frequency : Only for SLES9
274 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
275 Details : Update target/ChangeLog/which_patch .
279 Frequency : Only for test_10 in sanity-quota.sh
280 Description: a bug in quota test code
281 Details : checking a test flag in a wrong place
285 Frequency : Only for test_13 in sanity-quota.sh
286 Description: a bug in quota test script
287 Details : shouldn't check overfull allocated quota
291 Description: jbd statistics
292 Details : Port older jbd statistics patch for sles10
294 Severity : enhancement
296 Description: Add printing inode info into message about error in writepage.
301 Description: reply_lock_interpret crash due to race with it and lock cancel.
302 Details : Do not replay locks that are being cancelled. Do not reference
303 locks by their address during replay, just by their handle.
307 Frequency : When flocks are used.
308 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
309 Details : Flock locks might destroy just granted lock if it could be merged
310 with another existing flock, this is done in completion handler,
311 so teach ldlm_cli_enquque_fini that this is a valid case for
317 Description: Oops in read and write path when failing to allocate lock.
318 Details : Check if lock allocation failed and return error back.
322 Description: lstripe command fails for valid OST index
323 Details : The stripe offset is compared to lov->desc.ld_tgt_count
324 instead of lov->desc.ld_active_tgt_count.
328 Description: sanity.sh failed test 103
329 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
334 Description: df doesn't work properly if diskfs blocksize != 4K
335 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
339 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
340 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
344 Description: obd_health_check_timeout too short
345 Details : set obd_health_check_timeout as 1.5x of obd_timeout
349 Description: llapi_file_create() does not allow some changes
350 Details : add llapi_file_open() that allows specifying the mode and
351 open flags, and also returns an open file handle.
355 Description: merge and cleanup kernel patches.
356 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
360 Description: Add journal checksum support.(Kernel part)
361 Details : The journal checksum feature adds two new flags i.e
362 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
363 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
364 indicates that the commit block contains the checksum for
365 the blocks described by the descriptor blocks. Now commit
366 record can be sent to disk without waiting for descriptor
367 blocks to be written to disk. This behavior is controlled
368 using JBD2_FEATURE_ASYNC_COMMIT flag.
372 Description: OSS needs mutliple precreate threads
373 Details : Add ability to start more than one create thread per OSS.
377 Description: Sanity.sh test_65k failed: setstripe should have succeeded
378 Details : execute lfs setstripe on client
382 Description: mds_obd_create error creating tmp object
383 Details : When the user sets quota on root, llog will be affected and can't
384 create files and write files.
387 Frequency : Always on ia64 patchless client, and possibly others.
389 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
390 Details : This allows the patchless client to be loaded on architectures
395 Description: RedHat Update kernel for RHEL5
396 Details : Add the kernel config file for RHEL5.
400 Description: RedHat Update kernel for RHEL5
401 Details : Modify the kernel config file more closer RHEL5.
405 Description: Build failure against Centos5 (RHEL5)
406 Details : Define PAGE_SIZE when it isn't present.
410 Description: test_7 on sanity-quota.sh can't be used on separate machines
411 Details : add facet to handle it
415 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
416 Details : with NFS, the anon dentry's parent was set to itself in
417 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
418 corresponding dentry other than use rec->ur_name.
420 Severity : enhancement
422 Description: lfs setstripe enhancement
423 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
427 Description: enable data checksumming by default
428 Details : enable checksum by default, allow --disable-checksum
429 configure option and "-o nochecksum" mount option. Checksums
430 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
434 Description: lustre support for RHEL5
435 Details : Add support for RHEL5.
439 Description: Client eviction due to failover config
440 Details : after a connection loss, the lustre client should attempt to
441 reconnect to the last active server first before trying the
442 other potential connections.
446 Description: when mds and osts use different quota unit(32bit and 64bit),
447 quota will be released repeatly.
448 Details : void sending multiple quota reqs to mds, which will keep the
449 status between the reqs.
453 Description: osts not allocated evenly to files
454 Details : change the condition to increase offset_idx
457 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
459 Description: Data corruption for OSTs that are formatted larger than 2TB
461 Details : When generating the bio request for lustre file writes the
462 sector number would overflow a temporary variable before being
463 used for the IO. The data reads correctly from Lustre (which
464 will overflow in a similar manner) but other file data or
465 filesystem metadata may be corrupted in some cases.
469 Description: Tune the kernel for good SCSI performance.
470 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
471 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
476 Description: rewrite ext2-derived code in obdclass/uuid.c
477 Details : rewrite inherited code (uuid parsing code from ext2 utils)
478 from scratch preserving functionality.
482 Description: "lfs find -obd UUID" prints directories
483 Details : "lfs find -obd UUID" will return all directory names instead
484 of just file names. It is incorrect because the directories
485 do not reside on the OSTs.
489 Description: use i_size_read and i_size_write in 2.6 port
490 Details : replace inode->i_size access with i_size_read/write()
493 Frequency : when removing large files
495 Description: scheduling issue during removal of large Lustre files
496 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
497 It causes scheduling issues when removing large files (17TB in the
501 Frequency : only with liblustre clients on XT3
503 Description: evictions taking too long
504 Details : allow llrd to evict clients directly on OSTs
507 Frequency : only on ppc
509 Description: /proc/fs/lustre/devices broken on ppc
510 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
511 We should make sure the type of 'index' is loff_t and then cast
512 explicitly as needed below. Do not assign an explicitly cast
516 Frequency : only for rhel5
518 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
519 Details : Modify the target file & which_kernel.
521 Severity : enhancement
523 Description: omit set fsid for export NFS
524 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
525 as FS_REQUIRES_DEV for avoid problems with generate fsid.
529 Frequency : Always, for kernels after 2.6.16
530 Description: Fix warning idr_remove called for id=.. which is not allocated.
531 Details : Last kernels save old s_dev before kill super and not allow
532 to restore from callback - restore it before call kill_anon_super.
536 Description: buffer overruns could theoretically occur
537 Details : llapi_semantic_traverse() modifies the "path" argument by
538 appending values to the end of the origin string, and a
539 overrun may occur. Adding buffer overrun check in liblustreapi.
543 Description: Fix error on 'ls .' at the top of the Lustre mount.
544 Details : Don't revalidate dentry if it is a root dentry.
548 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
549 Details : Modify vm-tunables-rhel4.patch.
553 Description: Kernel config for 2.6.18-vanilla.
554 Details : Modify targets/2.6-vanilla.target.in.
555 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
556 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
557 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
558 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
562 Description: improve handling recoverable errors
563 Details : if request processig with error which can be recoverable on server
564 request should be resend, otherwise page released from cache and
569 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
570 Details : Modify target file & which_patch.
571 A flaw was found in the IA32 system call emulation provided
572 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
573 value could be stored in the %RAX register, which could trigger an
574 out-of-bounds system call table access. An untrusted local user
575 could exploit this flaw to run code in the kernel
576 (ie a root privilege escalation). (CVE-2007-4573).
580 Description: O_DIRECT bypasses client statistics.
581 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
582 read_ahead_stats were updated. copied the stats section from
583 osc_send_oap_rpc() into async_internal().
587 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
591 Description: change order of libsysio includes
592 Details : '#include sysio.h' should always come before '#include xtio.h'
596 Description: handle "serious error: objid * already exists" more gracefully
597 Details : If LAST_ID value on disk is smaller than the objects existing in
598 the O/0/d* directories, it indicates disk corruption and causes an
599 LBUG(). If the object is 0-length, then we should use the existing
600 object. This will help to avoid a full fsck in most cases.
602 Severity : enhancement
604 Description: adapt the lustre_config script to support the upgrade case
605 Details : Add "-u" option for lustre_config script to support upgrading 1.4
606 server targets to 1.6 in parallel.
610 Description: To avoid grant space > avaible space when the disk is almost
611 full. Without this patch you might see the error "grant XXXX >
612 available" or some LBUG about grant, when the disk is almost
614 Details : In filter_check_grant, for non_grant cache write, we should
615 check the left space by if (*left > ungranted + bytes), instead
616 of (*left > ungranted), because only we are sure the left space
617 is enough for another "bytes", then the ungrant space should be
618 increase. In client, we should update cl_avail_grant only there
619 is OBD_MD_FLGRANT in the reply.
623 Description: Update RHEL 4 kernel to fix local root privilege escalation.
624 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
625 described in CVE-2007-4573. This problem could allow untrusted
626 local users to gain root access.
629 Frequency : when using O_DIRECT and quotas
631 Description: Incorrect file ownership on O_DIRECT output files
632 Details : block usage reported by 'lfs quota' does not take into account
633 files that have been written with O_DIRECT.
638 Description: touch file failed when fs is not full
639 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
640 otherwise we can get ENOSP while fs is not full.
644 Description: parallel lock callbacks
645 Details : Instead of sending blocking and completion callbacks as separated
646 requests, adding them to a set and sending in parallel.
649 Frequency : only for Cray XT3
650 Bugzilla : 12829/13455
651 Description: Changing primary group doesn't change the group lustre assigns to
653 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
654 overridden with the primary group provided by the group upcall,
655 whereas the supplied fsgid can be trusted if it is in the list of
656 supplementary groups returned by the group upcall.
658 Severity : enhancement
660 Description: Allow masking D_WARNING, D_ERROR messages from console
661 Details : Console messages can now be disabled via lnet.printk.
665 Description: User code with malformed file open parameter crashes client node
666 Details : Before packing join_file req, all the related reference should be
667 checked carefully in case some malformed flags cause fake
668 join_file req on client.
672 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
673 Details : ldlm_completion_ast() assumes that a lock is granted when the req
674 mode is equal to the granted mode. However, it should also check
675 that LDLM_FL_CP_REQD is not set.
679 Description: Heavy nfs access might result in deadlocks
680 Details : After ELC code landed, it is now improper to enqueue any mds
681 locks under och_sem, because enqueue might want to decide to
682 cancel open locks for same inode we are holding och_sem for.
686 Description: Client eviction while running blogbench
687 Details : A lot of unlink operations with concurrent I/O can lead to a
688 deadlock causing evictions. To address the problem, the number of
689 oustanding OST_DESTROY requests is now throttled to
690 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
691 callbacks are processed in priority.
695 Description: enable ACLs on MDS by default
696 Details : ACLs must be enabled on MDS by default.
699 Frequency : PPC/PPC64 only
701 Description: conflicts between asm-ppc64/types.h and lustre_types.h
702 Details : fix duplicated definitions between asm-ppc64/types.h and
703 lustre_types.h on PPC.
705 --------------------------------------------------------------------------------
707 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
709 * Support for kernels:
710 2.6.5-7.283 (SLES 9),
711 2.6.9-55.EL (RHEL 4),
712 2.6.16.46-0.14 (SLES 10),
713 2.6.18.8 vanilla (kernel.org)
714 * Client support for unpatched kernels:
715 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
716 2.6.16 - 2.6.22 vanilla (kernel.org)
717 * Due to recently discovered recovery problems, we do not recommend
718 using patchless RHEL 4 clients with this or any earlier release.
719 * Recommended e2fsprogs version: 1.39.cfs8
720 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
721 * Starting with this release, the ldiskfs backing filesystem required
722 by Lustre is now in its own package, lustre-ldiskfs. This package
723 should be installed. It is versioned separately from Lustre and
724 may be released separately in future.
728 Description: block reactivating mgc import until all deactivates complete
729 Details : Fix race when failing back MDT/MGS to itself (testing)
732 Frequency : at statup only
734 Description: mds_lov_synchronize race leads to various problems
735 Details : simultaneous MDT->OST connections at startup can cause the
736 sync to abort, leaving the OSC in a bad state.
738 Severity : enhancement
740 Description: add optional extra BUILD_VERSION info
741 Details : add a new environment variable (namely LUSTRE_VERS) which allows
742 to override the lustre version.
745 Frequency : 2.6.18 servers only
747 Description: ll_kern_mount() doesn't release the module reference
748 Details : The ldiskfs module reference count never drops down to 0
749 because ll_kern_mount() doesn't release the module reference.
754 Description: server LBUG when using old ost_num_threads parameter
755 Details : Accept the old ost_num_threads parameter but warn that it
756 is deprecated, and fix an off-by-one error that caused an LBUG.
761 Description: Transient SCSI error results in persistent IO issue
762 Details : iobuf->dr_error is not reinitialized to 0 between two
766 Frequency : sometimes when underlying device returns I/O errors
768 Description: OSTs not going read-only during write failures
769 Details : OSTs are not remounted read-only when the journal commit threads
770 get I/O errors because fsfilt_ext3 calls journal_start/stop()
771 instead of the ext3 wrappers.
775 Description: poor connect scaling with increasing client count
776 Details : Don't run filter_grant_sanity_check for more than 100 exports
777 to improve scaling for large numbers of clients.
780 Frequency : SLES10 only
782 Description: sanity-quota.sh quotacheck failed: rc = -22
783 Details : Quotas cannot be enabled on SLES10.
786 Frequency : liblustre clients only
788 Description: getdirentries does not give error when run on compute nodes
789 Details : getdirentries does not fail when the size specified as an argument
790 is too small to contain at least one entry
792 Severity : enhancement
794 Description: Add LNET router traceability for debug purposes
795 Details : If a checksum failure occurs with a router as part of the
796 IO path, the NID of the last router that forwarded the bulk data
797 is printed so it can be identified.
802 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
803 Details : Due to a race condition, liblustre clients were occasionally
806 Severity : enhancement
808 Description: lfs setstripe use optional parameters instead of postional
811 Severity : enhancement
813 Description: Nanosecond timestamp support for ldiskfs
814 Details : The on-disk ldiskfs filesystem has added support for nanosecond
815 resolution timestamps. There is not yet support for this at
816 the Lustre filesystem level.
819 Frequency : during server recovery
821 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
822 Details : request with rq_no_resend flag not awake l_wait_event if they get a
826 Frequency : nfs export on patchless client
828 Description: connectathon hang when test nfs export over patchless client
829 Details : Disconnected dentry cannot be found with lookup, so we do not need
830 to unhash it or make it invalid
833 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
834 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
835 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
840 Description: open req refcounting wrong on reconnect
841 Details : If reconnect happened between getting open reply from server and
842 call to mdc_set_replay_data in ll_file_open, we will schedule
843 replay for unreferenced request that we are about to free.
844 Subsequent close will crash in variety of ways.
845 Check that request is still eligible for replay in
846 mdc_set_replay_data().
851 Description: disable writes to filesystem when reading health_check file
852 Details : the default for reading the health_check proc file has changed
853 to NOT do a journal transaction and write to disk, because this
854 can cause reads of the /proc file to hang and block HA state
855 checking on a healthy but otherwise heavily loaded system. It
856 is possible to return to the previous behaviour during configure
857 with --enable-health-write.
859 Severity : enhancement
861 Description: 64-bit inode version
862 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
863 to the inode. This will be required for version-based recovery.
868 Description: MDS fails to start if a duplicate client export is detected
869 Details : in some rare cases it was possible for a client to connect to
870 an MDS multiple times. Upon recovery the MDS would detect this
871 and fail during startup. Handle this more gracefully.
873 Severity : enhancement
875 Description: Add -o localflock option to simulate old noflock
877 Details : This will achieve local-only flock/fcntl locks
883 Description: log_commit_thread vs filter_destroy race leads to crash
884 Details : Take import reference before releasing llog record semaphore
889 Description: Wrong request locking in request set processing
890 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
891 imp_delayed_list, in this place should be used imp_lock.
894 Frequency : when reconnection
896 Description: Grant Leak when osc reconnect to OST
897 Details : When osc reconnect ost, OST(filter) should check whether it
898 should grant more space to client by comparing fed_grant and
899 cl_avail_grant, and return the granted space to client instead
900 of "new granted" space, because client will call osc_init_grant
901 to update the client grant space info.
904 Frequency : when client reconnect to OST
906 Description: Grant Leak when osc do resend and replay bulk write
907 Details : When osc reconnect to OST, OST(filter)should clear grant info of
908 bulk write request, because the grant info while be sync between
909 OSC and OST when reconnect, and we should ignore the grant info
910 these of resend/replay write req.
915 Description: Grant space more than avaiable left space sometimes.
916 Details : When then OST is about to be full, if two bulk writing from
917 different clients came to OST. Accord the avaliable space of the
918 OST, the first req should be permitted, and the second one
919 should be denied by ENOSPC. But if the seconde arrived before
920 the first one is commited. The OST might wrongly permit second
921 writing, which will cause grant space > avaiable space.
924 Frequency : when client is evicted
926 Description: Grant might be wrongly erased when osc is evicted by OST
927 Details : when the import is evicted by server, it will fork another
928 thread ptlrpc_invalidate_import_thread to invalidate the
929 import, where the grant will be set to 0. While the original
930 thread will update the grant it got when connecting. So if
931 the former happened latter, the grant will be wrongly errased
932 because of this race.
937 Description: Checking Stale with correct fid
938 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
939 to check whether it is stale, because sometimes, we want the
940 enqueue happened anyway, and op_data.fid2 will not be initialized.
942 Severity : enhancement
944 Description: update patchless client
945 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
948 Frequency : only with 2.4 kernel
950 Description: random memory corruption
951 Details : size of struct ll_inode_info is to big for union inode.u and this
952 can be cause of random memory corruption.
957 Description: Memory leak in recovery
958 Details : Lov_mds_md was not free in an error handler in mds_create_object.
959 It should also check obd_fail before fsfilt_start, otherwise if
960 fsfilt_start return -EROFS,(failover mds during mds recovery).
961 then the req will return with repmsg->transno = 0 and rc = EROFS.
962 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
963 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
964 be freed no matter whether fsfilt_commit success or not.
967 Frequency : only with huge count clients
969 Description: Prevents from taking the superblock lock in llap_from_page for
971 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
972 for llap_from_page call in ll_removepage prevents from taking the
973 superblock lock for a soon died page.
978 Description: Not check open intent error before release open handle
979 Details : in some rare cases, the open intent error is not checked before
980 release open handle, which may cause
981 ASSERTION(open_req->rq_transno != 0), because it tries to release
982 the failed open handle.
987 Description: Set cat log bitmap only after create log success.
988 Details : in some rare cases, the cat log bitmap is set too early. and it
989 should be set only after create log success.
993 Description: Accessing a block bevice can re-enable I/O when Lustre is
994 tearing down a device.
995 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1000 Description: service threads may hog cpus when there are a lot of requests
1001 Details : Insert cond_resched to give other threads a chance to use some CPU
1006 Description: the cat log was not initialized in recovery
1007 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1008 unlink log on mds will not be initialized until mds post
1009 recovery. And also in mds post recovery, the unlink log will
1010 initialization will be done asynchronausly, so there will be race
1011 between add unlink log and unlink log initialization.
1015 Description: brw_stats were being printed incorrectly
1016 Details : brw_stats were being printed as log2 but all of them were not
1017 recorded as log2. Also remove some code duplication arising from
1018 filter_tally_{read,write}.
1022 Frequency : rare, only in recovery.
1023 Description: ASSERTION(req->rq_type != LI_POISON) failed
1024 Details : imp_lock should be held while iterating over imp_sending_list for
1025 prevent destroy request after get timeout in ptlrpc_queue_wait.
1029 Description: replay-single.sh test 52 fails
1030 Details : A lock's skiplist need to be cleanup when it being unlinked
1031 from its resource list.
1035 Description: Short directio read returns full requested size rather than
1037 Details : Direct I/O operations should return actual amount of bytes
1038 transferred rather than requested size.
1040 Severity : enhancement
1042 Description: metadata RPC reduction (e.g. for rm performance)
1043 Details : decrease the amount of synchronous RPC between clients and servers
1044 by canceling conflicing lock before the operation on the client
1045 and packing thier handles into the main operation RPC to server.
1047 Severity : enhancement
1049 Description: Async OSC create to avoid the blocking unnecessarily.
1050 Details : If a OST has no remain object, system will block on the creating
1051 when need to create a new object on this OST. Now, ways use
1052 pre-created objects when available, instead of blocking on an
1053 empty osc while others are not empty. If we must block, we block
1054 for the shortest possible period of time.
1056 Severity : enhancement
1058 Description: refine locking for avoid write wrong info into lov_objid file
1059 Details : fix possible races with add new target and write/update data in
1062 --------------------------------------------------------------------------------
1064 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1069 Frequency : on some architectures
1071 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1072 Details : Uninitialized flags sometimes cause configuration commands to
1076 Frequency : patchless clients only
1078 Description: missing __iget() symbol export
1079 Details : The __iget() symbol export is missing. To avoid the need for
1080 this on patchless clients the deathrow inode reaper is turned
1081 off, and we depend on the VM to clean up old inodes. This
1082 dependency was during via the fix for bug 12181.
1084 --------------------------------------------------------------------------------
1086 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1088 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1089 INTEROPERATE with older versions automatically. In many cases a
1090 special upgrade step is needed. Please read the
1091 user documentation before upgrading any part of a live system.
1092 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1093 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1094 * WARNING: Lustre configuration and startup changes are required with
1095 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1101 Severity : enhancement
1103 Description: MountConf
1104 Details : Lustre configuration is now managed via mkfs and mount
1105 commands instead of lmc and lconf. New obd types (MGS, MGC)
1106 are added for dynamic configuration management. See
1107 https://mail.clusterfs.com/wikis/lustre/MountConf for
1110 Severity : enhancement
1112 Description: dynamic OST addition
1113 Details : OSTs can now be added to a live filesystem
1115 Severity : enhancement
1117 Description: startup order invariance
1118 Details : MDTs and OSTs can be started in any order. Clients only
1119 require the MDT to complete startup.
1121 Severity : enhancement
1123 Description: parallel, asynchronous orphan cleanup
1124 Details : orphan cleanup is now performed in separate threads for each
1125 OST, allowing parallel non-blocking operation.
1127 Severity : enhancement
1129 Description: optimized stripe assignment
1130 Details : stripe assignments are now made based on ost space available,
1131 ost previous usage, and OSS previous usage, in order to try
1132 to optimize storage space and networking resources.
1134 Severity : enhancement
1136 Description: Permanently set tunables
1137 Details : All writable /proc/fs/lustre tunables can now be permanently
1138 set on a per-server basis, at mkfs time or on a live system.
1140 Severity : enhancement
1142 Description: Lustre message v2
1143 Details : Add lustre message format v2.
1145 Severity : enhancement
1147 Description: client OST exclusion list
1148 Details : Clients can be started with a list of OSTs that should be
1149 declared "inactive" for known non-responsive OSTs.
1153 Description: SPEC SFS validation failure on NFS v2 over lustre.
1154 Details : Changes the blocksize for regular files to be 2x RPC size,
1155 and not depend on stripe size.
1157 Severity : enhancement
1159 Description: Multiple MD RPCs in flight.
1160 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1161 To avoid overly-overloading MDS, introduce a limit on number of
1162 MDS RPCs in flight for a single client and add /proc controls
1163 to adjust this limit.
1165 Severity : enhancement
1167 Description: client read/write statistics
1168 Details : Add client read/write call usage stats for performance
1169 analysis of user processes.
1170 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1171 file access. extents_stats shows chunk size distribution.
1172 extents_stats_per_process show chunk size distribution per
1175 Severity : enhancement
1177 Description: mds statistics
1178 Details : Add detailed mds operations statistics in
1179 /proc/fs/lustre/mds/*/stats.
1183 Description: Failure of copying files with lustre special EAs.
1184 Details : Client side always return success for setxattr call for lustre
1185 special xattr (currently only "trusted.lov").
1190 Description: Refcount LNET uuids
1191 Details : The global LNET uuid list grew linearly with every startup;
1192 refcount repeated list entries instead of always adding to
1195 Severity : enhancement
1197 Description: Dynamic service threads
1198 Details : Within a small range, start extra service threads
1199 automatically when the request queue builds up.
1202 Frequency : mixed-endian client/server environments
1204 Description: mixed-endian crashes
1205 Details : The new msg_v2 system had some failures in mixed-endian
1208 Severity : enhancement
1210 Description: Easy OST removal
1211 Details : OSTs can be permanently deactivated with e.g. 'lctl
1212 conf_param lustre-OST0001.osc.active=0'
1214 Severity : enhancement
1216 Description: MGS proc entries
1217 Details : Added basic proc entries for the MGS showing what filesystems
1220 Severity : enhancement
1222 Description: provide MGS failover
1223 Details : Added config lock reacquisition after MGS server failover.
1225 Severity : enhancement
1227 Description: add Linux 2.4 support
1228 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1232 Description: a large application tries to do I/O to the same resource and dies
1233 in the middle of it.
1234 Details : Check the req->rq_arrival time after the call to
1235 ost_brw_lock_get(), but before we do anything about
1236 processing it & sending the BULK transfer request. This
1237 should help move old stale pending locks off the queue as
1238 quickly as obd_timeout.
1241 Frequency : when an incorrect nid is specified during startup
1243 Description: ptlrpc connect to non-existant node causes kernel crash
1244 Details : LNET can't be re-entered from an event callback, which
1245 happened when we expire a message after the export has been
1246 cleaned up. Instead, hand the zombie cleanup off to another
1249 Severity : enhancement
1251 Description: plain/inodebits lock performance improvement
1252 Details : Grouping plain/inodebits in granted list by their request modes
1253 and bits policy, thus improving the performance of search through
1257 Frequency : only if OST filesystem is corrupted
1259 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1260 Details : for a short time RPCs with bulk IO are in the replay list,
1261 but replay of bulk IOs is unimplemented. If the OST filesystem
1262 is corrupted due to disk cache incoherency and then replay is
1263 started it is possible to trip an assertion. Avoid putting
1264 committed RPCs into the replay list at all to avoid this issue.
1267 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1269 Description: System hang on startup
1270 Details : This bug allowed the liblustre (e.g. catamount) client to
1271 return to the app before handling all startup RPCs. This
1272 could leave the node unresponsive to lustre network traffic
1273 and manifested as a server ptllnd timeout.
1275 Severity : enhancement
1277 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1278 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1279 boolean to control whether to print peer debug info when a
1280 client's RPC times out.
1283 Frequency : only for kernels with patches from Lustre below 1.4.3
1285 Description: Remove old rdonly API
1286 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1289 Frequency : only for devices with external journals
1291 Description: Set external device read-only also
1292 Details : During a commanded failover stop, we set the disk device
1293 read-only while the server shuts down. We now also set any
1294 external journal device read-only at the same time.
1297 Frequency : when upgrading from 1.4 while trying to change parameters
1299 Description: The wrong (new) MDC name was used when setting parameters for
1300 upgraded MDT's. Also allows changing of OSC (and MDC)
1301 parameters if --writeconf is specified at tunefs upgrade time.
1304 Frequency : when setting specific ost indicies
1306 Description: QOS code breaks on skipped indicies
1307 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1308 created with --index need not be sequential.
1310 Severity : enhancement
1312 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1313 Details : The uninit_groups feature works in conjunction with the kernel
1314 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1315 up the pass1 processing of e2fsck. This is a read-only feature
1316 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1317 mount filesystems that have had this feature enabled.
1319 Severity : enhancement
1321 Description: Improve multi-block allocation algorithm to avoid fragmentation
1322 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1323 allocation locality and avoid filesystem fragmentation.
1325 ------------------------------------------------------------------------------
1327 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1329 * Support for kernels:
1330 2.4.21-47.0.1.EL (RHEL 3)
1331 2.6.5-7.283 (SLES 9)
1332 2.6.9-42.0.10.EL (RHEL 4)
1333 2.6.12.6 vanilla (kernel.org)
1334 2.6.16.27-0.9 (SLES 10)
1335 * Recommended e2fsprogs version: 1.39.cfs5
1337 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1341 Frequency : occasional, depends on client load and configuration
1342 Bugzilla : 12181, 12203
1343 Description: data loss for recently-modified files
1345 Details : In some cases it is possible that recently written or created
1346 files may not be written to disk in a timely manner (this should
1347 normally be within 30s unless client IO load is very high).
1348 The problem appears as zero-length files or files that are a
1349 multiple of 1MB in size after a client crash or client eviction
1350 that are missing data at the end of the file.
1352 This problem is more likely to be hit on clients where files are
1353 repeatedly created and unlinked in the same directory, clients
1354 have a large amount of RAM, have many CPUs, the filesystem has
1355 many OSTs, the clients are rebooted frequently, and/or the files
1356 are not accessed by other nodes after being written.
1358 The presence of the problem can be detected by looking at
1359 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1360 smaller than the second (nr_unused) then dirty files will not
1361 be flushed automatically to disk. "sync; sleep 10" should be
1362 run several times on the node before unmounting it to update
1363 Lustre (this is also safe to run on nodes without this problem).
1365 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1366 that can cause this same problem, so customers using that kernel
1367 also need to update the kernel in addition to Lustre. In order
1368 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1370 It is normal that files written just before a client crash (less
1371 than 30s) may not yet have been flushed to disk, even for local
1375 Frequency : frequent on thin XT3 nodes
1377 Description: UUID collision on thin XT3 Linux nodes
1378 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1379 randomly, since we relied on an insufficiently-seeded PRNG.
1384 Description: OSS hangs after "All ost request buffers busy"
1385 Details : A deadlock between quota and journal operations caused OSS
1386 hangs after printing "All ost request buffers busy."
1389 Frequency : always on liblustre builds
1391 Description: Cleanup compiler warnings on liblustre
1394 Frequency : always on liblustre builds on XT3
1396 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1400 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1401 > 32000 subdirectories
1402 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1403 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1404 subdirectory count crosses 32000. This will aid e2fsck to
1405 correctly handle more than 32000 subdirectories.
1408 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1410 Description: System hang on startup
1411 Details : This bug allowed the liblustre (e.g. catamount) client to
1412 return to the app before handling all startup RPCs. This
1413 could leave the node unresponsive to lustre network traffic
1414 and manifested as a server ptllnd timeout.
1416 Severity : enhancement
1418 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1419 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1420 boolean to control whether to print peer debug info when a
1421 client's RPC times out.
1426 Description: make O_SYNC working on 2.6 kernels
1427 Details : 2.6 kernels use different method for mark pages for write,
1428 so need add a code to lustre for O_SYNC work.
1433 Description: Failure to close file and release space on NFS
1434 Details : Put inode details into lock acquired in ll_intent_file_open.
1435 Use mdc_intent_lock in ll_intent_open to properly
1436 detect all kind of errors unhandled by mdc_enqueue.
1441 Description: proc file read during shutdown sometimes raced obd removal,
1443 Details : Add lock to prevent obd access after proc file removal.
1446 Frequency : Only for files larger than 4GB on 32-bit clients.
1448 Description: improperly doing page alignment of locks
1449 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1450 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1455 Description: RPCs being resent when they shouldn't be
1456 Details : Some RPCs that should not be resent are being resent. This
1457 can cause inconsistencies in the RPC state machine. Do not
1458 resend such requests.
1461 Frequency : rare, only with NFS export
1463 Description: Crash on NFS re-export node
1464 Details : under very unusual load conditions an assertion is hit in
1465 ll_intent_file_open()
1468 Frequency : only if OST filesystem is corrupted
1470 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1471 Details : for a short time RPCs with bulk IO are in the replay list,
1472 but replay of bulk IOs is unimplemented. If the OST filesystem
1473 is corrupted due to disk cache incoherency and then replay is
1474 started it is possible to trip an assertion. Avoid putting
1475 committed RPCs into the replay list at all to avoid this issue.
1480 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1481 Details : Large single O_DIRECT read and write calls can fail to allocate
1482 a sufficiently large buffer to process the request. In case of
1483 allocation failure the allocation is retried with a smaller
1484 buffer and broken into smaller requests.
1486 Severity : enhancement
1488 Description: Add -o localflock option to simulate old noflock behaviour.
1489 Details : This will achieve local-only flock/fcntl locks coherentness.
1494 Description: versioning check is incomplete
1495 Details : Checking the version difference of client vs. server, report
1496 error if the gap is too big.
1501 Description: add support PG_writeback bit
1502 Details : add support for PG_writeback bit for Lustre, for more carefull
1503 work with page cache in 2.6 kernel. This also fix some deadlocks
1504 and remove hack for work O_SYNC with 2.6 kernel.
1506 Severity : enhancement
1508 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1509 Details : The uninit_groups feature works in conjunction with the kernel
1510 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1511 up the pass1 processing of e2fsck. This is a read-only feature
1512 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1513 mount filesystems that have had this feature enabled.
1515 Severity : enhancement
1517 Description: Improve multi-block allocation algorithm to avoid fragmentation
1518 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1519 allocation locality and avoid filesystem fragmentation.
1521 ------------------------------------------------------------------------------
1523 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1525 * Support for kernels:
1526 2.6.9-42.0.3.EL (RHEL 4)
1527 2.6.5-7.276 (SLES 9)
1528 2.4.21-47.0.1.EL (RHEL 3)
1529 2.6.12.6 vanilla (kernel.org)
1530 2.6.16.21-0.8 (SLES10)
1531 * Recommended e2fsprogs version: 1.39.cfs2-0
1533 * The backwards-compatible /proc/sys/portals symlink has been removed
1534 in this release. Before upgrading, please ensure that you change
1535 any configuration scripts or /etc/sysctl.conf files that access
1536 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1537 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1538 in advance of the upgrade on any system running Lustre 1.4.6 or
1539 newer, since /proc/sys/lnet was added in that version.
1540 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1544 Frequency : only when quota is used
1546 Description: avoid scanning export list for quota master
1547 Details : Change the algorithms to avoid scanning export list in order
1548 to improve the efficiency.
1551 Frequency : MDS failover only, very rarely
1553 Description: "went back in time" messages on mds failover
1554 Details : The greatest transno may be lost when the current operation
1555 finishes with an error (transno==0) and the client's last_rcvd
1556 record is over-written. Save the greatest transno in the
1557 mds_last_transno for this case.
1560 Frequency : always for specific kernels and striping counts
1562 Description: client may get "Matching packet too big" without ACL support
1563 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1564 when trying to access files in certain configurations. The
1565 clients should in fact be denied when mounting because they do
1566 not understand ACLs.
1569 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1571 Description: many clients connecting with IO in progress causes connect timeouts
1572 Details : Avoid synchronous journal commits to avoid delays caused by many
1573 clients connecting/disconnecting when bulk IO is in progress.
1574 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1575 OST_IO_PORTAL to avoid delays behind potentially many pending
1579 Frequency : occasionally with multiple writers to a single file
1581 Description: shared writes to file may result in wrong size reported by stat()
1582 Details : Allow growing of kms when extent lock is cancelled
1585 Frequency : always with random mmap IO to multi-striped file
1587 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1588 Details : Only if the hole is at the end of OST object so that kms is too
1589 small. Fix is to increase kms accordingly in ll_nopage.
1592 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1594 Description: writes to a missing object would leak memory on the OST
1595 Details : If there is an inconsistency between the MDS and OST filesystems,
1596 such that the MDS references an object that doesn't exist, writes
1597 to that object will leak memory due to incorrect cleanup in the
1598 error handling path, eventually running out of memory on the OST.
1603 Description: Creating too long symlink causes lustre errors
1604 Details : Check symlink and name lengths before sending requests to MDS.
1607 Frequency : only if flock is enabled (not on by default)
1609 Description: posix locks not released on fd closure on 2.6.9+
1610 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1611 kernels, this caused such locks not to be released on fd close and
1612 then assertions on fs unmount about still used locks.
1615 Frequency : MDS failover only, very rarely
1617 Description: clients may get ASSERTION(granted_lock != NULL)
1618 Details : When request was taking a long time, and a client was resending
1619 a getattr by name lock request. The were multiple lock requests
1620 with the same client lock handle and
1621 mds_getattr_name->fixup_handle_for_resent_request found one of the
1622 lock handles but later failed with ASSERTION(granted_lock != NULL).
1627 Description: handle->h_buffer_credits > 0, assertion failure
1628 Details : h_buffer_credits is zero after truncate, causing assertion
1629 failure. This patch extends the transaction or creates a new
1633 Frequency : NFS re-export or patchless client
1634 Bugzilla : 11179, 10796
1635 Description: Crash on NFS re-export node (__d_move)
1636 Details : We do not want to hash the dentry if we don't have a lock.
1637 But if this dentry is later used in d_move, we'd hit uninitialised
1638 list head d_hash, so we just do this to init d_hash field but
1639 leave dentry unhashed.
1642 Frequency : NFS re-export or patchless client
1644 Description: NFS exports has problem with symbolic link
1645 Details : lustre client didn't properly install dentry when re-exported
1646 to NFS or running patchless client.
1649 Frequency : NFS re-export or patchless client
1651 Description: Various nfs/patchless fixes.
1652 Details : fixes reuse disconected alias for lookup process - this fixes
1653 warning "find_exported_dentry: npd != pd",
1654 fix permission error with open files at nfs.
1655 fix apply umask when do revalidate.
1658 Frequency : occasional
1660 Description: Crash on NFS re-export node
1661 Details : calling clear_page() on the wrong pointer triggered oops in
1662 generic_mapping_read().
1665 Frequency : rarely, using O_DIRECT IO
1667 Description: unaligned directio crashes client with LASSERT
1668 Details : check for unaligned buffers before trying any requests.
1671 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1673 Description: stale data returned from RAID cache
1674 Details : If only a small amount of IO is done to the RAID device before
1675 reading it again it is possible to get stale data from the RAID
1676 cache instead of reading it from disk.
1679 Frequency : always for sles10 kernel
1681 Description: sles10 support
1682 Details : ll_follow_link: compile fixes and using of nd_set_link
1683 under newer kernels.
1686 Frequency : depends on arch, kernel and compiler version, always on sles10
1689 Description: recursive or deep enough symlinks cause stack overflow
1690 Details : getting rid of large stack-allocated variable in
1694 Frequency : depends on hardware
1696 Description: lustre write performance loss in the SLES10 kernel
1697 Details : the performance loss is caused by using of write barriers in the
1698 ext3 code. The SLES10 kernel turns barrier support on by
1699 default. The fix is to undo that change for ldiskfs.
1701 ------------------------------------------------------------------------------
1703 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1705 * Support for kernels:
1706 2.6.9-42.0.3EL (RHEL 4)
1707 2.6.5-7.276 (SLES 9)
1708 2.4.21-47.0.1.EL (RHEL 3)
1709 2.6.12.6 vanilla (kernel.org)
1713 Frequency : quota enabled and large files being deleted
1715 Description: releasing more than 4GB of quota at once hangs OST
1716 Details : If a user deletes more than 4GB of files on a single OST it
1717 will cause the OST to spin in an infinite loop. Release
1718 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1723 Description: statfs data retrieved from /proc may be stale or zero
1724 Details : When reading per-device statfs data from /proc, in the
1725 {kbytes,files}_{total,free,avail} files, it may appear
1726 as zero or be out of date.
1729 Frequency : systems with MD RAID1 external journal devices
1731 Description: lconf's call to blkid is confused by RAID1 journal devices
1732 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1733 of returning all block devices that match the journal UUID.
1736 Frequency : always, for aggregate stripe size over 4GB
1738 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1739 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1740 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1743 Frequency : always if "lfs find" used on a local file/directory
1745 Description: "lfs find" segfaults if used on a local file/directory
1746 Details : The case where a directory component was not specified wasn't
1747 handled correctly. Handle this properly.
1750 Frequency : always on ppc64
1752 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1753 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1754 perform endianness conversion.
1757 Frequency : rarely (truncate to non-zero file size after write under load)
1758 Bugzilla : 10730, 10687
1759 Description: Files padded with zeros to next 4K multiple
1760 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1761 it is possible that files that are truncated to a non-zero size
1762 immediately after being written are filled with zero bytes beyond
1763 the truncated size. No file data is lost.
1765 Severity : enhancement
1767 Description: Allow recovery/failover for liblustre clients.
1768 Details : liblustre clients were unaware of failover configurations until
1771 Severity : enhancement
1773 Description: user file locks should fail when not mounting with flock option
1774 Details : Set up an error-returning stub in ll_file_operations.lock field
1775 to prevent incorrect behaviour when client is mounted without
1776 flock option. Also, set up properly f_op->flock field for
1780 Frequency : always on ia64
1782 Description: "lfs df" loops on printing out MDS statfs information
1783 Details : The obd_ioctl_data was not initialized and in some systems
1784 this caused a failure during the ioctl that did not return
1785 an error. Initialize the struct and return an error on failure.
1788 Frequency : SLES 9 only
1790 Description: Error of copying files with lustre special EAs as root
1791 Details : Client side always return success for setxattr call for lustre
1792 special xattr (currently only "trusted.lov").
1795 Frequency : rarely on clusters with both ia64+i386 clients
1797 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1798 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1799 clients are doing concurrent writes to the same file it is
1800 possible that the ia64 clients may LASSERT because the OST
1801 extent locks are not PAGE_SIZE aligned. Ensure that grown
1802 locks are always aligned on the request boundary.
1805 Frequency : specific use, occasional
1807 Description: Overwriting in use executable truncates on-disk binary image
1808 Details : If one node attempts to overwrite an executable in use by
1809 another node, we now correctly return ETXTBSY instead of
1810 truncating the file.
1812 Severity : enhancement
1814 Description: Async OSC create to avoid the blocking unnecessarily.
1815 Details : If a OST has no remain object, system will block on the creating
1816 when need to create a new object on this OST. Now, ways use
1817 pre-created objects when available, instead of blocking on an
1818 empty osc while others are not empty. If we must block, we block
1819 for the shortest possible period of time.
1824 Description: chmod on Lustre root is propagated to other clients
1825 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1826 invalid by the follow_mount time.
1829 Frequency : liblustre clients only
1831 Description: Race in 'instant cancel' lock handling could lead to such locks
1832 never to be granted in case of SMP MDS
1833 Details : Do not destroy not yet granted but cbpending locks in
1837 Frequency : replay/resend of open
1839 Description: non null lock assetion failure in mds_intent_policy
1840 Details : Trying to replay/resend lockless open requests resulted in
1841 mds_open() returning 0 with no lock. Now it sets a flag if
1842 a lock is going to be returned.
1844 Severity : enhancement
1846 Description: Checksum enhancements
1847 Details : New checksum enhancements allow for resending RPCs that failed
1850 Severity : enhancement
1852 Description: Tunables on number of dirty pages in cacche
1853 Details : Allow to set limit on number of dirty pages cached.
1858 Description: client crash on unmount - lock still has references
1859 Details : In some error handling cases it was possible to leak a lock
1860 reference on a client while accessing a file. This was not
1861 harmful to the client during operation, but would cause the
1862 client to crash when the filesystem is unmounted.
1865 Frequency : specific case, rare
1867 Description: ETXTBSY on mds though file not in use by client
1868 Details : ETXTBSY is no longer incorrectly returned when attempting to
1869 chmod or chown a directory that the user previously tried to
1870 execute or a currently-executing binary.
1873 Frequency : extremely rare except on liblustre-based clients
1875 Description: Lustre space not freed when files are deleted
1876 Details : Clean up open-unlinked files after client eviction. Previously
1877 the unlink was skipped and the files remained as orphans.
1882 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1883 Details : In some cases it was possible to send a blocking callback to
1884 a client doing a glimpse, even though that client didn't get
1885 a lock granted. When the glimpse lock is cancelled on the OST
1886 the freed lock is left on the waiting list and corrupted the list.
1889 Frequency : all core dumps
1891 Description: Broke core dumps to lustre
1892 Details : Negative dentry may be unhashed if parent does not have UPDATE
1893 lock, but some callers, e.g. do_coredump, expect dentry to be
1894 hashed after successful create, hash it in ll_create_it.
1896 ------------------------------------------------------------------------------
1898 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1900 * Support for kernels:
1901 2.6.9-42.0.2.EL (RHEL 4)
1902 2.6.5-7.276 (SLES 9)
1903 2.4.21-40.EL (RHEL 3)
1904 2.6.12.6 vanilla (kernel.org)
1908 Frequency : always on RHEL 3
1910 Description: Number of open files grows over time
1911 Details : The number of open files grows over time, whether or not
1912 Lustre is started. This was due to a filp leak introduced
1913 by one of our kernel patches.
1915 ------------------------------------------------------------------------------
1917 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1919 * Support for kernels:
1920 2.6.9-42.EL (RHEL 4)
1921 2.6.5-7.276 (SLES 9)
1922 2.4.21-40.EL (RHEL 3)
1923 2.6.12.6 vanilla (kernel.org)
1928 Bugzilla : 5719, 9635, 9792, 9684
1929 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1930 Details : If a server is under heavy load and cannot reply to new
1931 connection requests before the client resends the (re)connect,
1932 the connection handling code can behave badly if two service
1933 threads are concurrently handing separate (re)connections from
1934 the same client. Add better locking to the connection handling
1935 code, and ensure that only a single connection will be processed
1936 for a given client UUID, even if the lock is dropped.
1938 Severity : enhancement
1940 Description: add TCP zero-copy support to kernel
1941 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1942 sends if the hardware supports scatter-gather and checksumming.
1943 This allows socklnd to do client-write and server-read more
1944 efficiently and reduce CPU utilization from skbuf copying.
1947 Frequency : only if NFS exporting from client
1949 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1950 Details : If setting attributes on a file created under NFS that had
1951 never been opened it would be possible to oops the client
1952 if the file had no objects.
1955 Frequency : always for liblustre
1957 Description: liblustre client does MDS+OSTs setattr RPC for each write
1958 Details : When doing a write from a liblustre client, the client
1959 incorrectly issued an RPC to the MDS and each OST the file was
1960 striped over in order to update the timestamps. When writing
1961 with small chunks and many clients this could overwhelm the MDS
1962 with RPCs. In all cases it would slow down the write because
1963 these RPCs are unnecessary.
1965 Severity : enhancement
1967 Description: allow number of MDS service threads to be changed at module load
1968 Details : It is now possible to change the number of MDS service threads
1969 running. Adding "options mds mds_num_threads={N}" to the MDS's
1970 /etc/modprobe.conf will set the number of threads for the next
1971 time Lustre is restarted (assuming the "mds" module is also
1972 reloaded at that time). The default number of threads will
1973 stay the same, 32 for most systems.
1978 Description: OST crash if filesystem is unformatted or corrupt
1979 Details : If an OST is started on a device that has never been formatted
1980 or if the filesystem is corrupt and cannot even mount then the
1981 error handling cleanup routines would dereference a NULL pointer.
1986 Description: NULL pointer deref in llap_from_page.
1987 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1988 incorrect) mapping if the page was truncated/reclaimed while it was
1989 searched for. Check for this condition and skip such pages when
1990 doing readahead. Introduce extra check to llap_from_page() to
1991 verify page->mapping->host is non-NULL (so page is not anonymous).
1994 Frequency : Sometimes when using sys_sendfile
1996 Description: "page not covered by a lock" warnings from ll_readpage
1997 Details : sendfile called ll_readpage without right page locks present.
1998 Now we introduced ll_file_sendfile that does necessary locking
1999 around call to generic_file_sendfile() much like we do in
2003 Frequency : with certain MDS communication failures at client mount time
2005 Description: NULL pointer deref after failed client mount
2006 Details : a client connection request may delayed by the network layer
2007 and not be sent until after the PTLRPC layer has timed out the
2008 request. If the client fails the mount immediately it will try
2009 to clean up before the network times out the request. Add a
2010 reference from the request import to the obd device and delay
2011 the cleanup until the network drops the request.
2014 Frequency : occasionally during client (re)connect
2016 Description: assertion failure during client (re)connect
2017 Details : processing a client connection request may be delayed by the
2018 client or server longer than the client connect timeout. This
2019 causes the client to resend the connection request. If the
2020 original connection request is replied in this interval, the
2021 client may trip an assertion failure in ptlrpc_connect_interpret()
2022 which thought it would be the only running connect process.
2025 Frequency : only with obd_echo servers and clients that are rebooted
2027 Description: kernel BUG accessing uninitialized data structure
2028 Details : When running an obd_echo server it did not start the ping_evictor
2029 thread, and when a client was evicted an uninitialized data
2030 structure was accessed. Start the ping_evictor in the RPC
2031 service startup instead of the OBD startup.
2033 Severity : enhancement
2034 Bugzilla : 10193 (patchless)
2035 Description: Remove dependency on various unexported kernel interfaces.
2036 Details : No longer need reparent_to_init, exit_mm, exit_files,
2037 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2040 Frequency : rare (only users of deprecated and unsupported LDAP config)
2042 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2043 Details : LDAP apparently contains 'lustreName' attributes instead of
2044 'name'. A simple remapping of the name is sufficient.
2047 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2049 Description: waiting_locks_callback trips kernel BUG if client is evicted
2050 Details : Running with the dump_on_timeout debug flag turned on makes
2051 it possible that the waiting_locks_callback() can try to dump
2052 the Lustre kernel debug logs from an interrupt handler. Defer
2053 this log dumping to the expired_lock_main() thread.
2055 Severity : enhancement
2057 Description: Support NFS exporting on 2.6 kernels.
2058 Details : Implement non-rawops metadata methods for NFS server to use without
2059 changing NFS server code.
2062 Frequency : very rare (synthetic metadata workload only)
2064 Description: two racing renames might cause an MDS thread to deadlock
2065 Details : Running the "racer" program may cause one MDS thread to rename
2066 a file from being the source of a rename to being the target of
2067 a rename at exactly the same time that another thread is doing
2068 so, and the second thread has already enqueued these locks after
2069 doing a lookup of the target and is trying to relock them in
2070 order. Ensure that we don't try to re-lock the same resource.
2073 Frequency : only very large systems with liblustre clients
2075 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2076 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2077 all OSTs in parallel. This allows the eviction of stale liblustre
2078 clients to proceed much faster than if they were done in series,
2079 and also offers similar improvements for other set_info RPCs.
2084 Description: excessive CPU usage during initial read phase on client
2085 Details : During the initial read phase on a client, it would agressively
2086 retry readahead on the file, consuming too much CPU and impacting
2087 performance (since 1.4.5.8). Improve the readahead algorithm
2088 to avoid this, and also improve some other common cases (read
2089 of small files in particular, where "small" is files smaller than
2090 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2095 Description: MDS crash when receiving packet with unknown intent.
2096 Details : Do not LBUG in unknown intent case, just return -EFAULT
2098 Severity : enhancement
2099 Bugzilla : 9293, 9385
2100 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2101 Details : Do not serialize getattr (non-intent version) and statfs.
2104 Frequency : occasional, when OST network is overloaded/intermittent
2106 Description: client evicted by OST after bulk IO timeout
2107 Details : If a client sends a bulk IO request (read or write) the OST
2108 may evict the client if it is unresposive to its data GET/PUT
2109 request. This is incorrect if the network is overloaded (takes
2110 too long to transfer the RPC data) or dropped the OST GET/PUT
2111 request. There is no need to evict the client at all, since
2112 the pinger and/or lock callbacks will handle this, and the
2113 client can restart the bulk request.
2116 Frequency : Always when mmapping file with no objects
2118 Description: client crashes when mmapping file with no objects
2119 Details : Check that we actually have objects in a file before doing any
2120 operations on objects in ll_vm_open, ll_vm_close and
2126 Description: Request leak when working with deleted CWD
2127 Details : Introduce advanced request refcount tracking for requests
2128 referenced from lustre intent.
2130 Severity : Enhancement
2132 Description: Cache open file handles on client.
2133 Details : MDS now will return special lock along with openhandle, if
2134 requested and client is allowed to hold openhandle, even if unused,
2135 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2136 closing files for every read/write openration.
2138 Severity : Enhancement
2140 Description: Cache open negative dentries on client when possible.
2141 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2142 negative dentries on lock revocation.
2147 Description: Remounting a client read-only wasn't possible with a zconf mount
2148 Details : It wasn't possible to remount a client read-only with llmount.
2150 Severity : enhancement
2151 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2152 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2153 MPICH is included to add Lustre-specific ADIO interfaces.
2154 This is based closely on the UFS ADIO layer and only differs
2155 in file creation, in order to allow the OST striping to be set.
2156 This is user-contributed code and not supported by CFS.
2161 Description: extended inode attributes (immutable, append-only) work improperly
2162 when 2.4 and 2.6 kernels are used on client/server or vice versa
2163 Details : Introduce kernel-independent values for these flags.
2165 Severity : enhancement
2168 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2169 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2170 granularity. Now, Fractional MB granularity is be supported,
2171 this is very useful for low memory system.
2173 Severity : enhancement
2175 Description: Getattr by fid
2176 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2177 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2178 have locks and avoids getattr rpc requests for it on every stat.
2181 Frequency : Always, for filesystems larger than 2TB
2183 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2184 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2185 filesystems larger than 2TB to be mounted with mballoc enabled.
2188 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2190 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2191 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2192 filesystems larger than 2TB is mounted then the kernel will
2193 silently corrupt the start of the filesystem. CONFIG_LBD is
2194 enabled for all CFS-supported kernels, but the possibility of
2195 this happening with a modified kernel config exists.
2197 Severity : enhancement
2199 Description: add client O_DIRECT support for 2.6 kernels
2200 Details : It is now possible to do O_DIRECT reads and writes to files
2201 in the Lustre client mountpoint on 2.6 kernel clients.
2203 Severity : enhancement
2205 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2206 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2207 parallel, not waiting for response from every OST before sending
2208 a rpc to the next OST.
2213 Description: setattr vs write race when updating file timestamps
2214 Details : Client processes that update a file timestamp into the past
2215 right after writing to the file (e.g. tar) it is possible that
2216 the updated file modification time can be reset to the current
2217 time due to a race between processing the setattr and write RPC.
2219 Severity : enhancement
2221 Description: Bring 'lfs find' closer in line with regular Linux find.
2222 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2223 -print0 options and obtains all the needed info through the lustre
2226 Severity : enhancement
2228 Description: support up to 1024 configured devices on one node
2229 Details : change obd_dev array from statically allocated to dynamically
2230 allocated structs as they are first used to reduce memory usage
2235 Description: Flush dirty partially truncated pages during truncate
2236 Details : Immediatelly flush partially truncated pages in filter_setattr,
2237 this way we completely avoid having any pages in page cache on OST
2238 and can retire ugly workarounds during writes to flush such pages.
2243 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2244 Details : protect inode from truncation within vfs_unlink() context
2245 just take a reference before calling vfs_unlink() and release it
2246 when parent's i_sem is free.
2251 Description: last_id value checked outside lock on OST caused LASSERT failure
2252 Details : If there were multiple MDS->OST object precreate requests in
2253 flight, it was possible that the OST's last object id was checked
2254 outside a lock and incorrectly tripped an assertion. Move checks
2255 inside locks, and discard old precreate requests.
2258 Frequency : always, if extents are used on OSTs
2260 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2261 Details : OSTs using the extents format would not zero the high 16 bits of
2262 the index physical block number. This is not a problem for any
2263 OST filesystems smaller than 16TB, and no kernels support ext3
2264 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2265 new/modified files) and can be fixed for existing filesystems
2266 with e2fsprogs-1.39-cfs1.
2271 Description: import connection selection may be incorrect if timer wraps
2272 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2273 import connections to be ignored if the 32-bit jiffies counter
2274 wraps. Use a 64-bit jiffies counter.
2277 Frequency : very large clusters immediately after boot
2279 Description: LNET request buffers exhausted under heavy short-term load
2280 Details : If a large number of client requests are generated on a service
2281 that has previously never seen so many requests it is possible
2282 that the request buffer growth cannot keep up with the spike in
2283 demand. Instead of dropping incoming requests, they are held in
2284 the LND until the RPC service can accept more requests.
2287 Frequency : Sometimes during replay
2289 Description: Assertion failure in ll_local_open after replay.
2290 Details : If replay happened on an open request reply before we were able
2291 to set replay handler, reply will become not swabbed tripping the
2292 assertion in ll_local_open. Now we set the handler right after
2293 recognising of open request
2296 Frequency : very rare
2298 Description: kernel reports "badness in vsnprintf"
2299 Details : Reading from the "recovery_status" /proc file in small chunks
2300 may cause a negative length in lprocfs_obd_rd_recovery_status()
2301 call to vsnprintf() (which is otherwise harmless). Exit early
2302 if there is no more space in the output buffer.
2304 Severity : enhancement
2306 Description: clear OBD RPC statistics by writing to them
2307 Details : It is now possible to clear the OBD RPC statistics by writing
2308 to the "stats" file.
2313 Description: Client mtime is not the same on different clients after utimes
2314 Details : In some cases, the client was using the utimes() syscall on
2315 a file cached on another node. The clients now validate the
2316 ctime from the MDS + OSTs to determine which one is right.
2321 Description: Inability to activate failout mode
2322 Details : lconf script incorrectly assumed that in pythong string's numeric
2323 value is used in comparisons.
2326 Frequency : always with multiple stripes per file
2328 Description: Inefficient object allocation for mutli-stripe files
2329 Details : When selecting which OSTs to stripe files over, for files with
2330 a stripe count that divides evenly into the number of OSTs,
2331 the MDS is always picking the same starting OST for each file.
2332 Return the OST selection heuristic to the original design.
2337 Description: mount failures may take full timeout to return an error
2338 Details : Under some heavy load conditions it is possible that a
2339 failed mount can wait for the full obd_timeout interval,
2340 possibly several minutes, before reporting an error.
2341 Instead return an error as soon as the status is known.
2343 Frequency : quota enabled and large files being deleted
2345 Description: releasing more than 4GB of quota at once hangs OST
2346 Details : If a user deletes more than 4GB of files on a single OST it
2347 will cause the OST to spin in an infinite loop. Release
2348 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2353 Description: statfs data retrieved from /proc may be stale or zero
2354 Details : When reading per-device statfs data from /proc, in the
2355 {kbytes,files}_{total,free,avail} files, it may appear
2356 as zero or be out of date.
2359 Frequency : systems with MD RAID1 external journal devices
2361 Description: lconf's call to blkid is confused by RAID1 journal devices
2362 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2363 of returning all block devices that match the journal UUID.
2366 Frequency : always, for aggregate stripe size over 4GB
2368 Description: assertion fails when trying to use 4GB stripe size
2369 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2370 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2373 Frequency : always on ppc64
2375 Description: the first write on an ext3 filesystem with mballoc got stuck
2376 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2377 perform endianness conversion.
2379 ------------------------------------------------------------------------------
2381 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2383 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2384 INTEROPERATE with older versions automatically. Please read the
2385 user documentation before upgrading any part of a live system.
2386 * WARNING: Lustre networking configuration changes are required with
2387 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2390 * Support for kernels:
2391 2.6.9-22.0.2.EL (RHEL 4)
2392 2.6.5-7.244 (SLES 9)
2393 2.6.12.6 vanilla (kernel.org)
2396 Severity : enhancement
2397 Bugzilla : 7981/8208
2398 Description: Introduced Lustre Networking (LNET)
2399 Details : LNET is new networking infrastructure for Lustre, it includes
2400 a reorganized network configuration mode (see the user
2401 documentation for full details) as well as support for routing
2402 between different network fabrics. Lustre Networking Devices
2403 (LNDs) for the supported network fabrics have also been
2404 created for this new infrastructure.
2406 Severity : enhancement
2407 Description: Introduced Access control lists
2408 Details : clients can set ACLs on files and directories in order to have
2409 more fine-grained permissions than the standard Unix UGO+RWX.
2410 The MDS must be started with the "-o acl" mount option.
2412 Severity : enhancement
2413 Description: Introduced filesystem quotas
2414 Details : Administrators may now establish per-user quotas on the
2417 Severity : enhancement
2419 Description: Configuration change for the XT3
2420 The PTLLND is now used to run Lustre over Portals on the XT3
2421 The configure option(s) --with-cray-portals are no longer used.
2422 Rather --with-portals=<path-to-portals-includes> is used to
2423 enable building on the XT3. In addition to enable XT3 specific
2424 features the option --enable-cray-xt3 must be used.
2429 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2430 Details : A race between allocating a new llcd and re-getting the llcd_lock
2431 allowed another thread to grab newly-allocated llcd.
2433 Severity : enhancement
2435 Description: 2.6 OST async journal commit and locking fix to improve performance
2436 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2437 2.6 kernels are now async as they already were in 2.4 kernels so
2438 that they can commit concurrently with the network bulk transfer.
2439 For block-allocated files the filter allocation semaphore is held
2440 to avoid filesystem fragmentation during allocation. BKL lock
2441 removed for 2.6 xattr operations where it is no longer needed.
2446 Description: lconf incorrectly determined whether two IP networks could talk
2447 Details : In some more complicated routing and multiple-network
2448 configurations, lconf will avoid trying to make a network
2449 connection to a disjoint part of the IP space. It was doing the
2450 math incorrectly for one set of cases.
2455 Description: Fix for potential infinite loop processing records in an llog.
2456 Details : If an llog record is corrupted/zeroed, it is possible to loop
2457 forever in llog_process(). Validate the llog record length
2458 and skip the remainder of the block on error.
2461 Frequency : occasional (liblustre only)
2463 Description: liblustre could not open files whose last component is a symlink
2464 Details : sysio_path_walk() would incorrectly pass the open intent to
2465 intermediate path components.
2468 Frequency : rare (liblustre only with non-standard tuning)
2469 Bugzilla : 7201 (7350)
2470 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2471 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2472 after it has been released, only do this for referenced locks
2474 Severity : enhancement
2476 Description: specifying an (invalid) directory default stripe_size of -1
2477 would reset the directory default striping
2478 Details : stripe_size -1 was used internally to signal directory stripe
2479 removal, now use "all default" to signal dir stripe removal
2480 as a directory striping of "all default" is not useful
2483 Frequency : common for large clusters running liblustre clients
2485 Description: doing an ls when liblustre clients are running is slow
2486 Details : sending a glimpse AST to a liblustre client waits for every AST
2487 to time out, as liblustre clients will not respond. Since they
2488 cannot cache data we refresh the OST lock LVB from disk instead.
2490 Severity : enhancement
2492 Description: doing an ls at the same time as file IO can be slow
2493 Details : enqueue and other "small" requests can be blocked behind many
2494 large IO requests. Create a new OST IO portal for non-IO
2495 requests so they can be processed faster.
2498 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2500 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2501 created file with mode 000
2502 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2503 allow an ftruncate() as a non-root user to a file with mode 000.
2504 Lustre now respects this flag to disable mode checks when
2505 truncating a file owned by the user
2508 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2511 Description: Revoking locks from clients that went dead or catatonic might take
2513 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2514 cancellation of such locks instant on servers without waiting for
2515 any reply from clients. Clients drops these locks when cancel
2516 notification from server is received without replying.
2519 Frequency : liblustre-only, when liblustre client dies or becomes busy
2521 Description: Doing ls on Linux clients can take a long time with active
2523 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2524 granting such locks to it in the first place if possible. Locks
2525 are taken by proxy on the OST during the read or write and
2526 dropped immediately afterward. Add connect flags handling, do
2527 not grant locks to liblustre clients for glimpse ASTs.
2529 Severity : enhancement
2531 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2532 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2533 sequential reads and behaves badly with applications doing random
2534 reads. Improve it by reading ahead at least read region, and
2535 avoiding excessive large RPC for small reads.
2537 Severity : enhancement
2539 Description: Creating more than 1000 files for a single job may cause a load
2540 imbalance on the OSTs if there are also a large number of OSTs.
2541 Details : qos_prep_create() uses an OST index reseed value that is an
2542 even multiple of the number of available OSTs so that if the
2543 reseed happens in the middle of the object allocation it will
2544 still utilize the OSTs as uniformly as possible.
2549 Description: OST or MDS may oops in ping_evictor_main()
2550 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2551 but doesn't restart at beginning of obd_exports_timed list
2554 Severity : enhancement
2556 Description: improve by-nid export eviction on the MDS and OST
2557 Details : allow multiple exports with the same NID to be evicted at one
2558 time without re-searching the exports list.
2561 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2563 Description: MDS may oops in groups_free()
2564 Details : in rare race conditions a newly allocated group_info struct is
2565 freed again, and this can be NULL. The 2.4 compatibility code
2566 for groups_free() checked for a NULL pointer, but 2.6 did not.
2569 Frequency : common for liblustre clients doing little filesystem IO
2570 Bugzilla : 9352, 7313
2571 Description: server may evict liblustre clients accessing contended locks
2572 Details : if a client is granted a lock or receives a completion AST
2573 with a blocking AST already set it would not reply to the AST
2574 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2577 Frequency : lfs setstripe, only systems with more than 160 OSTs
2579 Description: unable to set striping with a starting offset beyond OST 160
2580 Details : llapi_create_file() incorrectly limited the starting stripe
2581 index to the maximum single-file stripe count.
2584 Frequency : LDAP users only
2586 Description: lconf did not handle in-kernel recovery with LDAP properly
2587 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2589 Severity : enhancement
2591 Description: bind OST threads to NUMA nodes to improve performance
2592 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2593 node and do their allocations there to localize memory access
2595 Severity : enhancement
2597 Description: llmount can determine client NID directly from Myrinet (GM)
2598 Details : the client NID code from gmnalnid was moved directly into
2599 llmount, removing the need to use this or specifying the
2600 client NID explicitly when mounting GM clients with zeroconf
2603 Frequency : if client is started with down MDS
2605 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2606 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2607 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2608 ptlrpc_import_delayed_req() because we didn't check if the
2609 request was interrupted, and we also didn't break out of the
2610 event loop if there was no timeout
2615 Description: data loss during non-page-aligned writes to a single file from
2616 both multiple nodes and multiple threads on one node at same time
2617 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2618 inconsistency led to false short-reads, that were cached and later
2619 used by ->prepare_write() to fill in partially written page,
2620 leading to data loss.
2623 Frequency : always, if lconf --abort_recovery used
2625 Description: lconf --abort_recovery fails with 'Operation not supported'
2626 Details : lconf was attempting to abort recovery on the MDT device and not
2629 Severity : enhancement
2631 Description: remove cleanup logs
2632 Details : replace lconf-generated cleanup logs with lustre internal
2633 cleanup routines. Eliminates the need for client-cleanup and
2636 Severity : enhancement
2638 Description: add support for EAs (user and system) on lustre filesystems
2639 Details : it is now possible to store extended attributes in the Lustre
2640 client filesystem, and with the user_xattr mount option it
2641 is possible to allow users to store EAs on their files also
2643 Severity : enhancement
2645 Description: Add possibility (config option) to show minimal available OST free
2647 Details : When compiled with --enable-mindf configure option, statfs(2)
2648 (and so, df) will return least minimal free space available from
2649 all OSTs as amount of free space on FS, instead of summary of
2650 free spaces of all OSTs.
2652 Severity : enhancement
2654 Description: do not expand extent locks acquired on OST-side
2655 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2656 by server: they are not cached anyway.
2659 Frequency : when mmap is used/binaries executed from Lustre
2661 Description: Unmmap pages before throwing them away from read cache.
2662 Details : llap_shrink cache now attempts to unmap pages before discarding
2663 them (if unmapping failed - do not discard). SLES9 kernel has
2664 extra checks that trigger if this unmapping is not done first.
2669 Description: lconf didn't resolve symlinks before checking to see whether a
2670 given mountpoint was already in use
2673 Frequency : when migrating failover services
2674 Bugzilla : 6395, 9514
2675 Description: When migrating a subset of services from a node (e.g. failback
2676 from a failover service node) the remaining services would
2677 time out and evict clients.
2678 Details : lconf --force (implied by --failover) sets the global obd_timeout
2679 to 5 seconds in order to quickly disconnect, but this caused
2680 other RPCs to time out too quickly. Do not change the global
2681 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2683 Severity : enhancement
2684 Frequency : if MDS is started with down OST
2685 Bugzilla : 9439,5706
2686 Description: Allow startup/shutdown of an MDS without depending on the
2687 availability of the OSTs.
2688 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2689 Add appropriate locking and lov-osc refcounts for safe
2690 cleaning. Add osc abort_inflight calls in case the
2691 synchronize never started.
2694 Frequency : occasional (Cray XT3 only)
2696 Description: root not authorized to access files in CRAY_PORTALS environment
2697 Details : The client process capabilities were not honoured on the MDS in
2698 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2699 been accessed by an authorized user then root was able to access
2700 the file on the local client also. The root user capabilities
2701 are now allowed on the MDS, as this environment has secure UID.
2704 Frequency : occasional
2706 Description: ldiskfs "too long searching" message happens too often
2707 Details : A debugging message (otherwise harmless) prints too often on
2708 the OST console. This has been reduced to only happen when
2709 there are fragmentation problems on the filesystem.
2714 Description: Division by zero in statfs when all OSCs are inactive
2715 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2716 lov_statfs divides by value returned by lov_get_stripecnt().
2720 Bugzilla : 9489, 3273
2721 Description: First write from each client to each OST was only 4kB in size,
2722 to initialize client writeback cache, which caused sub-optimal
2723 RPCs and poor layout on disk for the first writen file.
2724 Details : Clients now request an initial cache grant at (re)connect time
2725 and so that they can start streaming writes to the cache right
2726 away and always do full-sized RPCs if there is enough data.
2727 If the OST is rebooted the client also re-establishes its grant
2728 so that client cached writes will be honoured under the grant.
2733 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2734 Details : Now I/O RPCs go to different portal number and (presumably) fast
2735 lock requests (and glimses) and other RPCs get their own service
2736 threads pool that should be able to service those RPCs
2739 Severity : enhancement
2741 Description: Ability to exchange lustre version between client and servers and
2742 issue warnings at client side if client is too old. Also for
2743 liblustre clients there is ability to refuse connection of too old
2745 Details : New 'version' field is added to connect data structure that is
2746 filled with version info. That info is later checked by server and
2750 Frequency : rare, liblustre only.
2751 Bugzilla : 9296, 9581
2752 Description: Two simultaneous writes from liblustre at offset within same page
2753 might proceed at the same time overwriting eachother with stale
2755 Details : I/O lock withing llu_file_prwv was released too early, before data
2756 actually was hitting the wire. Extended lock-holding time until
2757 server acknowledges receiving data.
2760 Frequency : extremely rare. Never observed in practice.
2762 Description: avoid generating lustre_handle cookie of 0.
2763 Details : class_handle_hash() generates handle cookies by incrementing
2764 global counter, and can hit 0 occasionaly (this is unlikely, but
2765 not impossible, because initial value of cookie counter is
2766 selected randonly). Value of 0 is used as a sentinel meaning
2767 "unassigned handle" --- avoid it. Also coalesce two critical
2768 sections in this function into one.
2770 Severity : enhancement
2772 Description: allow liblustre clients to delegate truncate locking to OST
2773 Details : To avoid overhead of locking, liblustre client instructs OST to
2774 take extent lock in ost_punch() on client's behalf. New connection
2775 flag is added to handle backward compatibility.
2777 Severity : enhancement
2778 Bugzilla : 4928, 7341, 9758
2779 Description: allow number of OST service threads to be specified
2780 Details : a module parameter allows the number of OST service threads
2781 to be specified via "options ost ost_num_threads={N}" in the
2782 OSS's /etc/modules.conf or /etc/modprobe.conf.
2786 Bugzilla : 6146, 9635, 9895
2787 Description: servers crash with bad pointer in target_handle_connect()
2788 Details : In rare cases when a client is reconnecting it was possible that
2789 the connection request was the last reference for that export.
2790 We would temporarily drop the export reference and get a new
2791 one, but this may have been the last reference and the export
2792 was just destroyed. Get new reference before dropping old one.
2794 Severity : enhancement
2795 Frequency : if client is started with failover MDS
2797 Description: Allow multiple MDS hostnames in the mount command
2798 Details : Try to read the configuration from all specified MDS
2799 hostnames during a client mount in case the "primary"
2802 Severity : enhancement
2804 Description: Stop sending data to evicted clients as soon as possible.
2805 Details : Check if the client we are about to send or are sending data to
2806 was evicted already. (Check is done every second of waiting,
2807 for which l_wait_event interface was extended to allow checking
2808 of exit condition at specified intervals).
2811 Frequency : rare, normally only when NFS exporting is done from client
2813 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2815 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2818 Frequency : always, if srand() is called before liblustre initialization
2820 Description: Liblustre uses system PRNG disturbing its usage by user application
2821 Details : Introduce internal to lustre fast and high-quality PRNG for
2822 lustre usage and make liblustre and some other places in generic
2823 lustre code to use it.
2825 Severity : enhancement
2826 Bugzilla : 9477, 9557, 9870
2827 Description: Verify that the MDS configuration logs are updated when xml is
2828 Details : Check if the .xml configuration logs are newer than the config
2829 logs stored on the MDS and report an error if this is the case.
2830 Request --write-conf, or allow starting with --old_conf.
2832 Severity : enhancement
2834 Description: Handle symlinks in the path when checking if Lustre is mounted.
2835 Details : Resolve intermediate symlinks when checking if a client has
2836 mounted a filesystem to avoid duplicate client mounts.
2841 Description: lconf can hit an error exception but still return success.
2842 Details : The lconf command catches the Command error exception at the top
2843 level script context and will exit with the associated exit
2844 status, but doesn't ensure that this exit status is non-zero.
2849 Description: failure of ptlrpc thread startup can cause oops
2850 Details : Starting a ptlrpc service thread can fail if there are a large
2851 number of threads or the server memory is very fragmented.
2852 Handle this without oopsing.
2855 Frequency : always, only if liblustre and non-default acceptor port was used
2857 Description: liblustre cannot connect to servers with non-default acceptor port
2858 Details : tcpnal_set_default_params() was not called and was therefore
2859 ignoring the environment varaible TCPNAL_PORT, as well as other
2860 TCPNAL_ environment variables
2865 Description: two objects could be created on the same OST for a single file
2866 Details : If an OST is down, in some cases it was possible to create two
2867 objects on a single OST for a single file. No problems other
2868 than potential performance impact and spurious error messages.
2872 Bugzilla : 5681, 9562
2873 Description: Client may oops in ll_unhash_aliases
2874 Details : Client dcache may become inconsistent in race condition.
2875 In some cases "getcwd" can fail if the current directory is
2881 Description: Inode refcounting problems in NFS export code
2882 Details : link_raw functions used to call d_instantiate without obtaining
2883 extra inode reference first.
2887 Bugzilla : 9942, 9903
2888 Description: Referencing freed requests leading to crash, memleaks with NFS.
2889 Details : We used to require that call to ll_revalidate_it was always
2890 followed by ll_lookup_it. Also with revalidate_special() it is
2891 possible to call ll_revalidate_it() twice for the same dentry
2892 even if first occurence returned success. This fix changes semantic
2893 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2894 reference on a request referred from the intent.
2895 ll_intent_release() then releases such a request.
2898 Frequency : rare, normally benchmark loads only
2900 Description: unlinked inodes were kept in memory on the client
2901 Details : If a client is repeatedly creating and unlinking files it
2902 can accumulate a lot of stale inodes in the inode slab cache.
2903 If there is no other client load running this can cause the
2904 client node to run out of memory. Instead flush old inodes
2905 from client cache that have the same inode number as a new inode.
2908 Frequency : SLES9 2.6.5 kernel and long filenames only
2909 Bugzilla : 9969, 10379
2910 Description: utime reports stale NFS file handle
2911 Details : SLES9 uses out-of-dentry names in some cases, which confused
2912 the lustre dentry revalidation. Change it to always use the
2916 Frequency : rare, unless heavy write-truncate concurrency is continuous
2917 Bugzilla : 4180, 6984, 7171, 9963, 9331
2918 Description: OST becomes very slow and/or deadlocked during object unlink
2919 Details : filter_destroy() was holding onto the parent directory lock
2920 while truncating+unlinking objects. For very large objects this
2921 may block other threads for a long time and slow overall OST
2922 responsiveness. It may also be possible to get a lock ordering
2923 deadlock in this case, or run out of journal credits because of
2924 the combined truncate+unlink. Solution is to do object truncate
2925 first in one transaction without parent lock, and then do the
2926 final unlink in a new transaction with the parent lock. This
2927 reduces the lock hold time dramatically.
2930 Frequency : rare, 2.4 kernels only
2932 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2933 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2934 before unmounting their filesystem in order to avoid deadlock.
2935 The kernel_locked() function in 2.4 kernels only checks whether
2936 the kernel lock is held, not whether it is this process that is
2937 holding it as 2.6 kernels do.
2942 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2943 Details : The client ptlrpc code may be trying to reconnect to a down
2944 server before a previous connection attempt has timed out.
2945 Increase the reconnect interval to be longer than the connection
2946 timeout interval to avoid sending duplicate connections to
2950 Frequency : echo_client brw_test command
2952 Description: fix echo_client to work with OST preallocated code
2953 Details : OST preallocation code (5137) didn't take echo_client IO path
2954 into account: echo_client calls filter methods outside of any
2955 OST thread and, hence, there is no per-thread preallocated
2956 pages and buffers to use. Solution: hijack pga pages for IO. As
2957 a byproduct, this avoids unnecessary data copying.
2961 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2962 Description: Client can oops in mdc_commit_close() after open replay
2963 Details : It was possible for the MDS to return an open request with no
2964 transaction number in mds_finish_transno() if the client was
2965 evicted, but without actually returning an error. Clients
2966 would later try to replay that open and may trip an assertion
2967 Simplify the client close codepath, and always return an error
2968 from the MDS in case the open is not successful.
2971 Frequency : rare, 2.6 OSTs only
2973 Description: OST may deadlock under high load on fragmented files
2974 Details : If there was a heavy load and highly-fragmented OST filesystems
2975 it was possible to have all the OST threads deadlock waiting on
2976 allocation of biovecs, because the biovecs were not released
2977 until the entire RPC IO was completed. Instead, release biovecs
2978 as soon as they are complete to ensure forward IO progress.
2980 Severity : enhancement
2982 Description: Support for specifying external journal device at mount
2983 Details : If an OST or MDS device is formatted with an external journal
2984 device, this device major/minor is stored in the ext3 superblock
2985 and may not be valid for failover. Allow detecting and
2986 specifying the external journal at mount time.
2991 Description: Mounting an MDS with pending unlinked files may cause oops
2992 Details : target_finish_recovery() calls mds_postrecov() which returned
2993 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2994 considers this an error and immediately begins cleaning up the
2995 lov, just after starting the mds_lov process
2997 Severity : enhancement
2999 Description: Implement 'lfs df' to report actual free space on per-OST basis
3000 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3001 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3002 usage of objects; '-h' to report in human readable format.
3004 ------------------------------------------------------------------------------
3006 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
3013 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3014 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3015 at startup during orphan unlinking. Instead, skip buddy update
3016 before setup, as it will be regenerated after recovery is complete.
3021 Description: If an OST is inactive, its locks might reference stale inodes.
3022 Details : lov_change_cbdata() must iterate over all namespaces, even if
3023 they are inactive to clear inode references from the lock.
3025 Severity : enhancement
3026 Frequency : occasional, if non-standard max_dirty_mb used
3028 Description: Client will block write RPCs if not enough grant
3029 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3030 then the client will block writes while waiting for another RPC
3031 to complete instead of consuming its dirty limit. With change
3032 we get improved performance when max_dirty_mb is small.
3034 Severity : enhancement
3035 Bugzilla : 3389, 6253
3036 Description: Add support for supplementary groups on the MDS.
3037 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3038 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3039 lookups for user supplementary groups into a cache.
3043 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3044 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3045 for liblustre O_EXCL is enforced by MDS.
3049 Description: readdir never returns NULL in liblustre.
3050 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3051 dentry in struct dirent.
3055 Frequency : liblustre only, depends on application IO pattern
3056 Description: liblustre clients evicted if not contacting servers
3057 Details : Don't put liblustre clients into the ping_evictor list, so
3058 they will not be evicted by the pinger ever.
3060 Severity : enhancement
3062 Description: Add ability to evict clients by NID from MDS.
3063 Details : By echoing "nid:$NID" string into
3064 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3065 $NID would be instantly evicted from this MDS and from all active
3066 OSTs connected to it.
3070 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3071 Details : lookup_it_finish() used to query file size from OSTs that was not
3076 Description: service threads change working directory to that of init
3077 Details : Starting lustre service threads may pin the working directory
3078 of the parent thread, making that filesystem busy. Threads
3079 now change to the working directory of init to avoid this.
3083 Frequency : during shutdown only
3084 Description: shutdown with a failed MDS or OST can cause unmount to hang
3085 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3090 Frequency : 2.6 only
3091 Description: chmod/chown may include an extra supplementary group
3092 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3093 supplementary group and if none is specified this is used.
3096 Bugzilla : 5479 (6816)
3098 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3099 Details : If lookup is in progress on a file that is unlinked we might try
3100 to revalidate the inode and fail in revalidate after lookup is
3101 complete and ll_file_open() enqueues the open again but
3102 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3105 Frequency : always, if lconf --abort_recovery used
3107 Description: lconf --abort_recovery fails with 'Operation not supported'
3108 Details : lconf was attempting to abort recovery on the MDT device and not
3111 ------------------------------------------------------------------------------
3113 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3118 Frequency : rare (only unsupported configurations with a node running as an
3120 Bugzilla : 6514, 5137
3121 Description: Mounting a Lustre file system on a node running as an OST could
3123 Details : OSTs now preallocates memory needed to write out data at
3124 startup, instead of when needed, to avoid having to
3125 allocate memory in possibly low memory situations.
3126 Specifically, if the file system is mounted on on OST,
3127 memory pressure could force it to try to write out data,
3128 which it needed to allocate memory to do. Due to the low
3129 memory, it would be unable to do so and the node would
3130 become unresponsive.
3132 Severity : enhancement
3134 Description: Addition of lconf --service command line option
3135 Details : lconf now accepts a '--service <arg>' option, which is
3136 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3138 Severity : enhancement
3140 Description: Failover mode is now the default for OSTs.
3141 Details : By default, OSTs will now run in failover mode. To return to
3142 the old behaviour, add '--failout' to the lmc line for OSTs.
3144 Severity : enhancement
3146 Description: Health checks are now provided for MDS and OSTs
3147 Details : Additional detailed health check information on MSD and OSTs
3148 is now provided through the procfs health_check value.
3151 Frequency : occasional, depends on IO load
3153 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3154 after numerous create/delete cycles
3155 Details : The ext3 inode allocation policy would not allocate new inodes
3156 very well on the OSTs because there are no new directories
3157 being created. Instead we look for groups with free space if
3158 the parent directories are nearly full.
3163 Description: Network or server problems during mount may cause partially
3164 mounted clients instead of returning an error.
3165 Details : The config llog parsing code may overwrite the error return
3166 code during mount error handling, returning success instead
3172 Description: MDS can fail to allocate large reply buffers
3173 Details : After long uptimes the MDS can fail to allocate large reply
3174 buffers (e.g. zconf client mount config records) due to memory
3175 fragmentation or consumption by the buffer cache. Preallocate
3176 some large reply buffers so that these replies can be sent even
3177 under memory pressure.
3181 Frequency : rare (liblustre)
3182 Description: fsx running with liblustre complained that using truncate() to
3183 extend the file doesn't work. This patch corrects that issue.
3184 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3185 ATTR_SIZE and lsm use in llu_setattr_raw.
3189 Frequency : rare, only 2.6 kernels
3190 Description: Unusual file access patterns on the MDS may result in inode
3191 data being lost in very rare circumstances.
3192 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3193 code in the 2.6 kernel caused the inode and/or EA data not to
3194 be read from disk, causing single-file corruption.
3198 Frequency : rare, only 2.6 filesystems using extents
3199 Description: Heavy concurrent write and delete load may cause data corruption.
3200 Details : It was possible under high-load situations to have an extent
3201 metadata block in the block device cache from a just-unlinked
3202 file overwrite a newly-allocated data block. We now unmap any
3203 metadata buffers that alias just-allocated data blocks.
3207 Frequency : filesystems with default stripe_count larger than 77
3208 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3209 Details : lconf specifies an inode size of 4096 bytes when the default
3210 stripe_count is larger than 77. This conflicts with the default
3211 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3212 this case to avoid pinning too much memory for large EAs.
3214 ------------------------------------------------------------------------------
3216 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3221 Frequency : rare (extremely heavy IO load with hundreds of clients)
3223 Description: Client is evicted, gets IO error writing to file
3224 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3225 caused clients to be evicted for AST timeouts. The fixes in
3226 bug 5192 mean we no longer need to have such short AST timeouts
3227 so ldlm_timeout has been increased.
3230 Frequency : occasional during --force or --failover shutdown under load
3231 Bugzilla : 5949, 4834
3232 Description: Server oops/LBUG if stopped with --force or --failover under load
3233 Details : a collection of import/export refcount and cleanup ordering
3234 issues fixed for safer force cleanup
3237 Frequency : only filesystems larger than 120 OSTs
3238 Bugzilla : 5990, 6223
3239 Description: lfs getstripe would oops on a very large filesystem
3240 Details : lov_getconfig used kfree on vmalloc'd memory
3243 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3244 Bugzilla : 6242, 6243
3245 Description: reading from files that had been truncated to a non-zero size
3246 but never opened returned no data
3247 Details : ll_file_read() reads zeros from no-object files to EOF
3252 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3253 Details : There's little harm in aborting MDS/OSS recovery and letting it
3254 try again, so I removed the LASSERT and return an error instead.
3256 Severity : enhancement
3258 Description: New debugging infrastructure for tracking down data corruption
3259 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3260 (b) cover more of the client-side code path, and (c) try to narrow
3261 down where problems occurred
3265 Bugzilla : 3819, 4364, 4397, 6313
3266 Description: Racing close and eviction MDS could cause assertion in mds_close
3267 Details : It was possible to get multiple mfd references during close and
3268 client eviction, leading to one thread referencing a freed mfd.
3270 Severity: : enhancement
3271 Bugzilla : 3262, 6359
3272 Description: Attempts to reconnect to servers are now more aggressive.
3273 Details : This builds on the enhanced upcall-less recovery that was added
3274 in 1.4.2. When trying to reconnect to servers, clients will
3275 now try each server in the failover group every 10 seconds. By
3276 default, clients would previously try one server every 25 seconds.
3281 Description: After recovery, certain operations trigger a failed
3282 assertion on a client.
3283 Details : Failing over an mds, using lconf -d --failover, while a
3284 client was doing a readdir() call would cause the client to
3285 LBUG after recovery completed and the readdir() was resent.
3287 Severity : enhancement
3289 Description: Default groups are now added by lconf
3290 Details : You can now run lconf --group <servicename> without having to
3291 manually add groups with lmc.
3294 Frequency : occasional
3296 Description: Nodes with an elan id of 0 trigger a failed assertion
3299 Frequency : always when accessing e.g. tty/console device nodes
3301 Description: tty and some other devices nodes cannot be used on lustre
3302 Details : file's private_data field is used by device data and lustre
3303 values in there got lost. New field was added to struct file to
3304 store fs-specific private data.
3307 Frequency : when exporting Lustre via NFS
3309 Description: NFSD failed occasionally when looking up a path component
3310 Details : NFSD is looking up ".." which was broken in ext3 directories
3311 that had grown large enough to become hashed.
3314 Frequency : Clusters with multiple interfaces not on the same subnet
3316 Description: Nodes will repeatedly try to reconnect to an interface which it
3317 cannot reach and report an error to the log.
3318 Details : Extra peer list entries will be created by lconf with some peers
3319 unreachable. lconf now validates the peer before adding it.
3322 Frequency : Only if a default stripe is set on the filesystem root.
3324 Description: Setting a default stripe on the filesystem root prevented the
3325 filesystem from being remounted.
3326 Details : The client was sending extra request flags in the root getattr
3327 request and did not allocate a reply buffer for the dir EA.
3330 Frequency : occasional, higher if lots of files are accessed by one client
3331 Bugzilla : 6159, 6097
3332 Description: Client trips assertion regarding lsm mismatch/magic
3333 Details : While revalidating inodes the VFS looks up inodes with ifind()
3334 and in rare cases can find an inode that is being freed.
3335 The ll_test_inode() code will free the lsm during ifind()
3336 when it finds an existing inode and then the VFS later attaches
3337 this free lsm to a new inode.
3341 Bugzilla : 6422, 7030
3342 Description: MDS deadlock between mkdir and client eviction
3343 Details : Creating a new file via mkdir or mknod (starting a transaction
3344 and getting the ns lock) can deadlock with client eviction
3345 (gets ns lock and trying to finish a synchronous transaction).
3348 Frequency : occasional
3349 Description: While starting a server, the fsfilt_ext3 module could not be
3351 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3352 kernels. Previously, lconf would still use the ext3 name
3353 when trying to load modules. Now, it will correctly use
3354 ext3 on 2.4 and ldiskfs on 2.6.
3356 Severity : enhancement
3357 Description: The default stripe count has been changed to 1
3358 Details : The interpretation of the default stripe count (0, to lfs
3359 or lmc) has been changed to mean striping across a single
3360 OST, rather than all available. For general usage we have
3361 found a stripe count of 1 or 2 works best.
3363 Severity : enhancement
3364 Description: Add support for compiling against Cray portals.
3365 Details : Conditional compiling for some areas that are different
3369 Frequency : occasional
3370 Bugzilla : 6409, 6834
3371 Description: Creating files with an explicit stripe count may lead to
3372 a failed assertion on the MDS
3373 Details : If some OSTs are full or unavailable, creating files may
3374 trigger a failed assertion on the MDS. Now, Lustre will
3375 try to use other servers or return an error to the
3379 Frequency : occasional
3381 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3382 caused lock timeout during readahead (since 1.4.2).
3383 Details : Processes doing readahead might match a lock that hasn't been
3384 granted yet if there are overlapping and conflicting lock
3385 requests. The readahead process waits on ungranted lock
3386 (original lock is CBPENDING), while OST waits for that process
3387 to cancel CBPENDING read lock and eventually evicts client.
3389 Severity : enhancement
3391 Description: Initial enabling of flock support for clients
3392 Details : Implements fcntl advisory locking and file status functions.
3393 This feature is provided as an optional mount flag (default
3394 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3395 locking are implemented yet, and those that are are not guaranteed
3396 to be completely correct in production environments.
3397 mount -t lustre -o [flock|noflock] ...
3400 Frequency : occasional
3402 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3403 assertion in the ext3 JBD (journaling) layer.
3404 Details : The b_committed_data struct is protected by the big kernel lock
3405 in 2.4 kernels, serializing journal_commit_transaction() and
3406 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3407 access to this struct. In 2.6 kernels there is finer grained
3408 locking to improve SMP performance of the JBD layer.
3412 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3417 Description: merge and cleanup kernel patches.
3420 -----------------------------------------------------------------------------
3422 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3424 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3425 versions of Lustre. Please update all servers and clients to
3426 version 1.4.2 or later at the same time. You must also run
3427 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3428 to update the configuration logs.
3430 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3431 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3432 - swab llog records of type '0' so we get proper header size/idx (5861)
3433 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3434 - fix rename of one directory over another leaking an inode (5953)
3435 - avoid SetPageDirty on 2.6 (5981)
3436 - don't re-add just-being-destroyed locks to the waiting list (5653)
3437 - when creating new directories, inherit the parent's custom
3438 striping settings if present parent (3048)
3439 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3440 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3441 - don't hold client locks on temporary worklist from l_lru (5666)
3442 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3443 - automatically evict dead clients (5921)
3444 - Update file size properly in create+truncate+fstat case (6196)
3445 - Do not unhash mountpoint dentries, do not allow removal of
3447 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3448 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3449 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3450 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3451 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3452 - Correctly handle reads of files with no objects (6243)
3453 - lctl recover will also mark a device active if deactivate used (5933)
3455 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3456 - allow --write-conf on an MDS with different nettype than client (5619)
3457 - don't write config llogs to MDS for mounts not from that MDS (5617)
3458 - lconf should create multiple TCP connections from a client (5201)
3459 - init scripts are now turned off by default; run chkconfig --on
3460 lustre and chkconfig --on lustrefs to use them
3461 - upcalls are no longer needed for clients to recover to failover
3463 - add --abort-recovery option to lconf to abort recovery on device
3465 - add support for an arbitrary number of OSTs (3026)
3466 - Quota support protocol changes.
3467 - forward compatibility changes to wire structs (6007)
3468 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3469 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3470 - improved "lustre status" script
3471 - initialize blocksize for non-regular files (6062)
3472 - added --disable-server and --disable-client configure options (5782)
3473 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3474 - Vanilla 2.4.29 support
3475 - increase maximum number of obd devices to 520 (6242)
3476 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3477 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3478 SLES 9 2.6.5 kernels
3479 - limit stripes per file to 160 (the maximum EA size) (6093)
3481 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3484 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3485 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3486 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3487 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3488 - small ext3 extents cleanups and fixes (5733)
3489 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3490 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3491 - avoid needless client->OST connect, fix handle mismatch (5317)
3492 - fix DLM error path that led to out-of-sync client, long delays (5779)
3493 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3494 - fix several locking issues related to i_size (5492,5624,5654,5672)
3495 - don't move pending lock onto export if it is already evicted (5683)
3496 - fix kernel oops when creating .foo in unlinked directory (5548)
3497 - fix deadlock in obdfilter statistics vs. object create (5811)
3498 - use time_{before,after} to avoid timer jiffies wrap (5882)
3499 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3500 - Do not leak request if server was not able to process it (5154)
3501 - If mds_open unable to find parent dir, make that negative lookup(5154)
3502 - don't create new directories with extent-mapping (5909, 5936)
3504 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3505 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3506 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3507 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3508 - support for RHEL4 kernel on i686 (5773)
3509 - provide error messages when incompatible logs are encountered (5898)
3511 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3512 * version 1.4.0.10 (1.4.1 release candidate 1)
3514 - don't keep a lock reference when lock is not granted (4238)
3515 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3516 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3517 - drop import inflight refcount on signal_completed_replay error (5255)
3518 - unlock page after async write error during send (3677)
3519 - handle missing objects in filter_preprw_read properly (5265)
3520 - no transno return for symlink open, don't save no-trasno open (3440)
3521 - don't try to complete elan receive that already failed (4012)
3522 - free RPC server reply state on error (5406)
3523 - clean up thread from ptlrpc_start_thread() on error (5160)
3524 - readahead could read extra page into cache that wasn't ejected (5388)
3525 - prevent races in class_attach/setup/cleanup/detach (5260)
3526 - don't dereference de->d_inode after l_dput of de (5458)
3527 - use "int" for stripe value returned from lock_to_stripe (5544)
3528 - mballoc allocation and error-checking fixes in 2.6 (5504)
3529 - block device patches to fix I/O request sizes in 2.6 (5482)
3530 - look up hostnames for IB nals (5602)
3531 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3532 - don't start multiple acceptors for the same port (5277)
3533 - fix incorrect LASSERT in mds_getattr_name (5635)
3534 - export a proc file for general "ping" checking (5628)
3535 - fix "lfs check" to not block when the MDS is down (5628)
3537 - service request history (4965)
3538 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3539 - create an "evict_client" /proc entry on OSTs, like the MDS has
3540 - fix mount usage message, return errors per mount(8) (5168)
3541 - change grep [] to grep "[]" in tests so they work in more UMLs
3542 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3543 - remove ancient LOV_MAGIC_V0 EA support (5047)
3544 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3545 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3546 - allow readahead allocations to fail when low on memory (5383)
3547 - mmap locking landed again, after considerable improvement (2828)
3548 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3550 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3553 - send OST transaction number in read/write reply to free req (4966)
3554 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3555 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3556 - the watchdog thread now runs as interruptible (5246)
3557 - flock/lockf fixes (but it's still disabled, pending 5135)
3558 - don't use EXT3 constants in llite code (5094)
3559 - memory shortage at startup could cause assertion (5176)
3561 - reorganization of lov code
3562 - single portals codebase
3564 - add extents/mballoc support (5025)
3565 - direct I/O reads in the obdfilter (4048)
3566 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3568 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3571 - send OST transaction number in read/write reply to free req (4966)
3572 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3573 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3574 - the watchdog thread now runs as interruptible (5246)
3575 - handle missing objects in filter_preprw_read properly (5265)
3576 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3577 - drop import inflight refcount on signal_completed_replay error (5255)
3578 - unlock page after async write error during send (3677)
3579 - return original error code on reconstructed replies (3761)
3580 - no transno return for symlink open, don't save no-trasno open (3440)
3582 - add pid to ldlm debugging output (4922)
3583 - bump the watchdog timeouts -- we can't handle 30sec yet
3584 - extra debugging for orphan dentry/inode bug (5259)
3586 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3589 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3590 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3591 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3592 - fix NULL dereference in /proc/sys/portals/routes (4827)
3593 - allow failed mdc_close() operations to be interrupted (4561)
3594 - stop precreate on OST before MDS would time out on it (4778)
3595 - don't send partial-page writes before EOF from client (4410)
3596 - discard client grant for sub-page writes on large-page clients (4520)
3597 - don't free dentries not owned by NFS code, check generation (4806)
3598 - fix lsm leak if mds_create_objects() fails (4801)
3599 - limit debug_daemon file size, always print CERROR messages (4789)
3600 - use transno after validating reply (3892)
3601 - process timed out requests if import state changes (3754)
3602 - update mtime on OST during writes, return in glimpse (4829)
3603 - add mkfsoptions to LDAP (4679)
3604 - use ->max_readahead method instead of zapping global ra (5039)
3605 - don't interrupt __l_wait_event() during strace
3607 - add software watchdogs to catch hung threads quickly (4941)
3608 - make lustrefs init script start after nfs is mounted
3609 - fix CWARN/ERROR duplication (4930)
3610 - return async write errors to application if possible (2248)
3611 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3612 - print NAL number in %x format (4645)
3613 - update barely-supported suse-2.4.21-171 series (4842)
3614 - support for sles 9 %post scripts
3615 - support for building 2.6 kernel-source packages
3616 - support for sles km_* packages
3618 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3621 - ignore -ENOENT errors in osc_destroy (3639)
3622 - notify osc create thread that OSC is being cleaned up (4600)
3623 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3624 - reconstruct ost_handle() like mds_handle() (4657)
3625 - create a new thread to do import eviction to avoid deadlock (3969)
3626 - let lconf resolve symlinked-to devices (4629)
3627 - don't unlink "objects" from directory with default EA (4554)
3628 - hold socknal file ref over connect in case target is down (4394)
3629 - allow more than 32000 subdirectories in a single directory (3244)
3630 - fix blocks count for O_DIRECT writes (3751)
3631 - OST returns ENOSPC from object create when no space left (4539)
3632 - don't send truncate RPC if file size isn't changing (4410)
3633 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3634 - bind to privileged port in socknal and tcpnal (3689)
3636 - rate limit CERROR/CWARN console message to avoid overload (4519)
3637 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3638 - basic mmap support (3918)
3639 - kernel patch series update from b1_4 (4711)
3641 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3644 - avoid crash during MDS cleanup with OST shut down (2775)
3645 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3646 - don't use bad inodes on the MDS (3744)
3647 - dynamic object preallocation to improve recovery speed (4236)
3648 - don't hold spinlock over lock dumping or change debug flags (4401)
3649 - don't zero obd_dev when it is force cleaned (3651)
3650 - print grants to console if they go negative (4431)
3651 - "lctl deactivate" will stop automatic recovery attempts (3406)
3652 - look for existing locks in ldlm_handle_enqueue() (3764)
3653 - don't resolve lock handle twice in recovery avoiding race (4401)
3654 - revalidate should check working dir is a directory (4134)
3656 - don't always mark "slow" obdfilter messages as errors (4418)
3658 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3661 - don't close LustreDB during write_conf until it is done (3860)
3662 - fix typo in lconf for_each_profile (3821)
3663 - allow dumping logs from multiple threads at one time (3820)
3664 - don't allow multiple threads in OSC recovery (3812)
3665 - fix debug_size parameters (3864)
3666 - fix mds_postrecov to initialize import for llog ctxt (3121)
3667 - replace config semaphore with spinlock (3306)
3668 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3669 - don't allow enqueue to complete on a destroyed export (3822)
3670 - down write_lock before checking llog header bitmap (3825)
3671 - recover from lock replay timeout (3764)
3672 - up llog sem before sending rpc (3652)
3673 - reduce ns lock hold times when setting kms (3267)
3674 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3675 - fix NULL deref and obd_dev leak on setup error (3312)
3676 - replace some LBUG about llog ops with error handling (3841)
3677 - don't match INVALID dentries from d_lookup and spin (3784)
3678 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3679 - fix invalid assertion in ptlrpc_set_wait (3880)
3681 - add libwrap support for the TCP acceptor (3996)
3682 - add /proc/sys/portals/routes for non-root route listing (3994)
3683 - allow setting MDS UUID in .xml (2580)
3684 - print the stack of a process that LBUGs (4228)
3686 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3689 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3690 - make sure to unset replay flag from failed open requests (3440)
3691 - if default stripe count is 0, use OST count for inode size (3636)
3692 - update parent mtime/ctime on client for create/unlink (2611)
3693 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3694 - free recovery state on server during a forced cleanup (3571)
3695 - unregister_reply for resent reqs (3063)
3696 - loop back devices mounting and status check on 2.6 (3563)
3697 - fix resource-creation race that can provoke i_size == 0 (3513)
3698 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3699 - more debugging for page-accounting assertion (3746)
3700 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3701 - avoid deadlock after precreation failure (3758)
3702 - fix race and lock order deadlock in orphan handling (3450, 3750)
3703 - add validity checks when grabbing inodes from l_ast_data (3599)
3705 - add /proc/.../recovery_status to obdfilter (3428)
3706 - lightweight CDEBUG infrastructure, debug daemon (3668)
3707 - change default OSC RPC parameters to be better on small clusters
3708 - turn off OST read cache for files smaller than 32MB
3709 - install man pages and include them in rpms (3100)
3710 - add new init script for (un)mounting lustre filesystems (2593)
3711 - run chkconfig in %post for init scripts (3701)
3712 - drop scimac NAL (unmaintained)
3714 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3717 - clean kiobufs before and after use (3485)
3718 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3719 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3720 - change init script to fail more clearly if not run as root (1528)
3721 - allow clients to reconnect during replay (1742)
3722 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3723 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3724 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3725 - fix cond_resched (3554)
3726 - only evict unfinished clients after recovery (3515)
3727 - allow bulk resend, prevent data loss (3570)
3728 - dynamic ptlrpc request buffer allocation (2102)
3729 - don't allow unlinking open directory if it isn't empty (2904)
3730 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3731 - remove extraneous obd dereference causing LASSERT failure (3334)
3732 - don't use get_cycles() when creating temp. files on the mds (3156)
3733 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3734 - handle EEXIST for set-stripe, set proper directory name (3336)
3736 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3737 - fix ksocknal_fmb_callback() error messages (2918)
3739 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3742 - don't copy lvb into (possibly NULL) reply on error (2983)
3743 - don't deref dentry after dput, don't free lvb on error (2922)
3744 - use the kms to determine writeback rpc length (2947)
3745 - increment oti_logcookies when osc is inactive (2948)
3746 - update client's i_blocks count via lvb messages (2543)
3747 - handle intent open/close of special files properly (1557)
3748 - mount MDS with errors=remount-ro, like obdfilter (2009)
3749 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3750 - don't use cancelling-locks' kms values (2947)
3751 - use highest lock extent for kms, not last one (2925)
3752 - don't dereference ERR_PTR() dentry in error handling path (3107)
3753 - fix thread race in portals_debug_dumplog() (3122)
3754 - create lprocfs device entries at setup instead of at attach (1519)
3755 - common AST error handler, don't evict client on completion race (3145)
3756 - zero nameidata in detach_mnt in 2.6 (3118)
3757 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3758 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3759 - initialize RPC timeout timer earlier for 2.6 (3219)
3760 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3761 - print nal/nid for unknown nid (3258)
3762 - additional checks for oscc recovery before doing precreate (3284)
3763 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3764 - don't crash in mdc_close for bad permissions on open (3285)
3765 - zero i_rdev for non-device files (3147)
3766 - clear page->private before handing to FS, better assertion (3119)
3767 - tune the read pipeline (3236)
3768 - fix incorrect decref of invalidated dentry (2350)
3769 - provide read-ahead stats and refine rpc in flight stats (3328)
3770 - don't hold journal transaction open across create RPC (3313)
3771 - update atime on MDS at close time (3265)
3772 - close LDAP connection when recovering to avoid server load (3315)
3773 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3774 - don't leak open file on MDS after open resend (3325)
3775 - serialize filter_precreate and filter_destroy_precreated (3329)
3776 - loop device shouldn't call sync_dev() for nul device (3092)
3777 - clear page cache after eviction (2766)
3778 - resynchronize MDS->OST in background (2824)
3779 - refuse to mount the same filesystem twice on same mountpoint (3394)
3780 - allow llmount to create routes for mounting behind routers (3320)
3781 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3782 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3783 - fix rare problem with rename on htree directories (3417)
3785 - allow default OST striping configuration per directory (1414)
3786 - fix compilation for qswnal for 2.6 kernels (3125)
3787 - increase maximum number of MDS request buffers for large systems
3788 - change liblustreapi to be useful for external progs like lfsck (3098)
3789 - increase local configuration timeout for slow disks (3353)
3790 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3792 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3795 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3796 - don't overwrite extent policy data in reply if lock was blocked (2901)
3797 - drop filter export grants atomically with removal from device (2663)
3798 - del obd_self_export from work_list in class_disconnect_exports (2908)
3799 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3800 - swab reply message in mdc_close, other PPC fixes (2464)
3801 - fix destroying of named logs (2325)
3802 - overwrite old logs when running lconf --write_conf (2264)
3803 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3804 - fix race in target_handle_connect (2898)
3805 - mds_reint_create() should take same inode create lock (2926)
3806 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3807 - don't close files for self_export to avoid uninitialized obd (2936)
3808 - allow MDS with the same name as client node (2939)
3809 - hold dentry reference for closed log files for unlink (2325)
3810 - reserve space for all logs during transactions (2059)
3811 - don't evict page beyond end of stripe extent (2925)
3812 - don't oops on a deleted current working directory (2399)
3813 - handle hard links to targets without a parent properly (2517)
3814 - don't dereference NULL lock when racing during eviction (2867)
3815 - don't grow lock extents when lots of conflicting locks (2919)
3817 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3820 - account for cache space usage on clients to avoid data loss (974)
3821 - lfsck support in lustre kernel code (2349)
3822 - reduce journal credits needed for BRW writes (2370)
3823 - orphan handling to avoid losing space on client/server crashes
3824 - ptlrpcd can be blocked, stopping ALL progress (2477)
3825 - use lock value blocks to assist in proper KMS, faster stat (1021)
3826 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3827 - recovery for initial connections (2355)
3828 - fixes for mds_cleanup_orphans (1934)
3829 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3830 - block all file creations until orphan recovery completes (1901)
3831 - client remove rq_connection from request struct (2423)
3832 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3833 - recovery timer race (2670)
3834 - mdc_close recovey bug (2532)
3835 - ptlrpc cleanup bug (2710)
3836 - mds timeout on local locks (2588)
3837 - namespace lock held during RPCs (2431)
3838 - handle interrupted sync write properly (2503)
3839 - don't try to handle a message that hasn't been replied to (2699)
3840 - client assert failure during cleanup after abort recovery (2701)
3841 - leak mdc device after failed mount (2712)
3842 - ptlrpc_check_set allows timedout requests to complete (2714)
3843 - wait for inflight reqs when ptlrpcd finishes (2710)
3844 - make sure unregistered services are removed from the srv_list
3845 - reset bulk XID's when resending them (caught by 1138 test)
3846 - unregister_bulk after timeout
3847 - fix lconf error (2694)
3848 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3849 - readahead locks pages, leaves pending causing memory pressure (2673)
3850 - increase OST request buffers to 4096 on large machines (2729)
3851 - fix up permission of existing directories in simple_mkdir (2661)
3852 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3853 - don't assign transno to errored transactions (2742)
3854 - don't delete objects on OST if given a bogus objid from MDS (2751)
3855 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3856 - if rq_no_resend, then timeout request after recovery (2432)
3857 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3858 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3859 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3860 - MDS recovery shouldn't delete objects that it has given out (2730)
3861 - if enqueue arrives after completion, don't clobber LVB (2819)
3862 - don't unlock pages twice when trigger_group_io returns error (2814)
3863 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3864 - don't write pages to disk if there was an error (1450)
3865 - don't ping imports that have recovery disabled (2676)
3866 - take buffered bytes into account when balancing socknal conn (2817)
3867 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3868 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3869 - remove little-endian swabbing of llog records (1987)
3870 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3871 - retry reposting request buffers if they fail (1191)
3872 - grow extent at grant time to avoid granting a revoked lock (2809)
3873 - lock revoke doesn't evict page if covered by a second lock (2765)
3874 - disable VM readahead to avoid reading outside lock extents (2805)
3876 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3877 - updated kernel patches for hp-2.4.20 kernel (2681)
3879 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3882 - fix truncated write corruption (2366)
3883 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3885 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3886 - ensure len(uuid) < 37 in lmc (1171)
3887 - fix ia64 OOPS in llog_test (2255)
3888 - zero end of page at obdfilter for partial page writes (2648)
3889 - don't leave stale dentries around after renames (bug 2428)
3890 - fix timeouts when evicting a client with a single lock held (2642)
3891 - set deadline for the initial HELLO message to drain (2634)
3892 - print out dotted-quad IP addresses in the socknal (2302)
3894 - additional debugging for MDS client eviction problem (2443)
3895 - fix mkfsoptions support for osts (2603, 2604)
3897 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3900 - add series for the vanilla 2.6.0 kernel
3901 - add series for the vanilla 2.4.24 kernel
3902 - add series for a cray x86/64 UL kernel drop
3903 - fix xattr patches for the vanilla 2.4.19 series
3905 - generate true UUIDs in lmc (1171)
3906 - have portals stack dumping break in UML (2466)
3907 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3908 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3909 - wake callback waiting threads on client eviction (2460)
3910 - Add --ptldebug and --subsystem to lmc (1719)
3911 - update assertion to allow safe interrupt allocation
3912 - set rq_no_resend for cancel requests (2432)
3913 - recalculate ptlrpcd timeout after resend (2494)
3914 - call vfs_rmdir when removing pending directories (2368)
3915 - fix renaming a file to itself (2429)
3916 - lmc creates a default one-stripe lov (2454)
3917 - expand procfs space to handle large clusters (2326)
3918 - increase UML stack to avoid overflow
3919 - update lconf's list of debug and subsystem masks
3920 - fix lfs find --obd (2510)
3921 - /proc tunable for disabling filter read caching (2591)
3922 - stop rpm packages from altering slapd.conf (2301)
3923 - disable nagle in the socknal under 0conf (2578)
3924 - choose mds inode size based on stripe count (2572)
3925 - fix kernel-source rpm problems (2516)
3927 - add --disable-doc to avoid pdf generation (2421)
3928 - update documentation, tests, type-os, comments
3929 - avoid format warnings on ia64
3930 - remove the TOE NAL
3931 - tiny code cleanups by removing unused fields
3933 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3936 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3937 - remove the most blatant lies from BUILDING file (2371)
3938 - change default debug level to reasonable production setting
3939 - reduce client side cache size to reduce cache flush time
3940 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3941 - make TCP zerocopy and pinger support enabled by default (2476)
3942 - sync writes completed after process exits caused crashes (2319)
3943 - maintain correct mount count on the MDS (2356)
3944 - backout 1557, because 2316 wasn't really fixed
3945 - better file I/O statistics gathering in /proc
3946 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3947 - another kernel patch to fix zero-copy TCP function export
3948 - don't take duplicate lock when processing re-sent getattr (2420)
3949 - lctl uses obd_self_export instead of creating new conn (2353)
3950 - MDS/OST recovery case which requires object creation asserted (2425)
3951 - move lfs from /usr/sbin to /usr/bin in packages
3952 - fix race between mds_client_add and mds_client_free (2417)
3953 - use kmalloc instead of slabs in portals (2430)
3954 - don't create duplicate records when a failover MDS is present (2442)
3955 - remove unnecessary mount age check (2332)
3956 - don't remove directory inodes from locks prematurely (2451)
3957 - don't break if MDS service name is the same as hostname (2103)
3958 - fix races in client write RPC generation when cache full (2482)
3960 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3963 - remove now-unused request->rq_obd (278)
3964 - if an allocation fails, print out how much memory we've used (1933)
3965 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3966 - add the "configurable stack size" patch to most series files (1256)
3967 - ability to write large log records, for 100+ OST configs (2306)
3968 - fix NULL deref when filter_prep fails (2314)
3969 - fix operator precedence error in filter_sync
3970 - dynamic allocation of socknal TX descriptors (2315)
3971 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3972 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3973 - ensure that CWARN messages in Portals always get to the syslog
3974 - __init/__exit are not for prototype decls (ldlm_init/exit)
3975 - x86-64 compile warning fixes
3976 - fix gateway LMC keyword conflict (2318)
3977 - fix MDS lock inversions in getattr/reint paths (1844)
3978 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3979 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3980 - DLM race condition prevented some lock evictions (2328)
3981 - ENOMEM detection and retry on socknal sends (2230)
3982 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3983 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3984 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3985 - better error messages when a client is rejected during recovery (1505)
3986 - avoid cancelling locks which were never granted, after failure (2330)
3987 - fix i_sem/journal inversion in mds_client_add (2333)
3988 - fix truncate/getattr lock cycle deadlock (2334)
3989 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3990 - fix two rare exit paths which could leak an l_lock() ref (2321)
3991 - fencepost error in MDS/OST orphan recovery (2226)
3992 - make log record alignment 8 bytes (1988)
3993 - lstripe now fails when requested offset > ost_count (2237)
3994 - ensure that all kernel series have a complete list.h (1607)
3995 - fix crashes in special-file operations (2316)
3996 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3998 - allow configurable automake binary, for testing new versions
3999 - small update to the lfs documentation
4001 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4003 * fix negative export reference count in fsfilt_sync (2312)
4005 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4006 * release candidate 0.9.1
4008 - orphans are moved into the PENDING directory for possible recovery
4009 - replayed opens now open by fid for orphan/rename safety (1042)
4010 - last close of an orphan inode generates a transno (683)
4011 - chdir() and mount() now pin the directory entry (1020)
4012 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4013 - discard very old requests without processing them (1502)
4014 - remove some common, well-understood CERRORs (1505)
4015 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4016 - clear "grant" flags in OST replies until OST grant code lands (1644)
4017 - fix read performance by not clobbering i_blksize on client (1598)
4018 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4019 - make LRU size a /proc tunable, clears locks when reduced (707)
4020 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4021 - server locks take references on exports now (1558)
4022 - build fixes for 2.4.20-rh trees (1663)
4023 - return an error from lov_create if all OSCs are inactive (1751)
4024 - fix import levels when a reconnect happens without a timeout (1597)
4025 - exit early from mds_open if we get a lookup error (1749)
4026 - partial page read at EOF wouldn't wait for disk before sending (1642)
4027 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4028 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4029 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4030 - write/truncate lock inversion (1639)
4031 - Don't auto-load obdclass, portals modules during cleanup (1495)
4032 - fix timestamps from jumping to "now" (1763)
4033 - extra journal assertions (1648)
4034 - add an extra multiunlink test (1771)
4035 - fix read_record/write_record API (1776)
4036 - fix leak of offset_extent, possible incorrect i_size later (1772)
4037 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4038 - Debugging for the kqswnal_get_idle_tx problems (1820)
4039 - Allow recovery to be attempted multiple times (1536)
4040 - Write out MDS last_rcvd file after it is first created (1600)
4041 - Fix tx_descriptor leak in failed transmit situations (1827)
4042 - ext3 journaling fixes for assertion failure after IO error (1871)
4043 - class_export_put() on freed export after completion AST error (1896)
4044 - Fix revalidate looping in VFS (1322)
4045 - Don't access a freed export during MDS_REINT timeout (1521)
4046 - Add open-unlink recovery support on the MDS (1673,1764)
4047 - Return an error if no MDS data was read from last_rcvd (1946)
4048 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4049 - Don't setup a disk device that doesn't match exported UUID (317)
4050 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4051 - avoid committing NULL handle in force close
4052 - local.sh is now a one-stripe LOV configuration
4053 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4054 - fix ext3 htree duplicate directory entry corruption (1516)
4055 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4056 - update to new LOV EA format (2097)
4057 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4059 2003-06-15 Phil Schwan <phil@clusterfs.com>
4062 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4063 - per-import/export recovery handling (958, 931, 959)
4064 - multiple last-rcvd slots, for serving multiple FSes (949)
4065 - connections are again shared between multiple imp/exports (963, 964)
4066 - "umount -f" would hang if any requests needed to be sent (393, 978)
4067 - avoid pinning large req buffer by copying for queued messages (989)
4068 - add "uuid" to "lctl device" command to help upcalls (991)
4069 - "open" RPCs with transnos would confuse recovery counters (1037)
4070 - do proper endian conversion of all wire messages (288, 340, 891)
4071 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4072 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4073 - fix ptlrpc_connection leak in target_handle_connect (1174)
4074 - fix import refcounting bug in OST and MDS cleanup (1134)
4075 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4076 - fix very unlikely obd_types race condition (501)
4077 - remove osc_open hack for echo_client (1187)
4078 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4079 - a failure in read_inode2 leads to deadlock (1139)
4080 - cancel ack-locks as soon as transaction is committed (1072)
4081 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4082 - make sure to commitrw after any preprw to avoid deadlock (1162)
4083 - failing to execute a file in a lustre FS would lock inode (1203)
4084 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4085 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4086 - obd_uuid2tgt would walk off the end of the list (1255)
4087 - on IA64 the osc would give portals incorrect bulk size (1258)
4088 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4089 - fix lock inversion caused by new llite matching code (1282)
4090 - limit the number of dirty pages on a client to 10MB (1286)
4091 - timed out locks were not being corrected cancelled (1289)
4092 - fix O_DIRECT above 4GB on IA-32 (1292)
4093 * major user-visible changes
4094 - fail out/fail over policy now controlled by the upcall (993)
4096 - add OBD_PING to check server availability and failure (954)
4097 - lustre messages are now sent in sending host order (288, 340, 891)
4098 - add eadatalen to MDS getattr reply (340)
4099 - OST read replies may contain second buffer, with per-page status (593)
4101 2003-03-11 Phil Schwan <phil@clusterfs.com>
4104 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4105 - failed open()s could cause deadlock; fixed (867, 869)
4106 - stop cancelling OST locks when files are closed (481)
4107 - overlapping XID spaces caused network corruption (851, 853)
4108 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4109 - setattr_raw added, to do single-RPC, server-side setattrs
4110 - lmc/lconf syntax change for OST UUIDs
4111 - fix crashy race condition between ptlrpc_free_req and osc_close
4112 - don't use request in mdc_enqueue if we hit a timeout (889)
4113 - don't set the inode i_size for regular files from the MDS (896)
4114 - handle out of order completion AST (842)
4115 - don't LBUG if a lock request times out after receiving AST (913)
4116 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4117 - if a bad lock AST arrives, send an error instead of dropping entirely
4118 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4119 - fix leak in bulk IO when only partially completed (899, 900, 926)
4120 - fix O_DIRECT for ia64 (55)
4121 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4122 - C-z after timeout could hang a process forever; fixed (977)
4124 - client-side I/O cache (678, 924, 929, 941, 970)
4126 - READPAGE and SETATTRs which don't take server-side locks get
4129 2003-02-11 Phil Schwan <phil@clusterfs.com>
4132 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4133 - Fix fsfilt_extN_readpage() to read a full page of directory
4134 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4135 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4136 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4137 - Most intent operations were removed, in favour of a new RPC mode
4138 that does a single RPC to the server and bypasses most of the VFS
4139 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4140 - Aggressively cancel local locks on DLM servers
4141 - mds_reint_unlink sends EA to the client if it's the last nlink.
4142 client uses that EA to unlink OST objects.
4143 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4144 - recursive symlinks were fixed (439)
4145 - fixed NULL deref in DEBUG_REQ
4146 - filter_update_lastobjid no longer calls sync, which annoyed extN
4147 - fixed multi-client small-writes to a single file problem (445)
4148 - fixed mtime updates during file writes (607)
4149 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4150 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4151 - fixed Linux OST size reporting problem (444, 656)
4152 - OST now updates object mtime with writes or setattr (607, 619)
4153 - client verifies file size before zeroing page past EOF (445)
4154 - OST now writes last allocated objid to disk with allocation (108)
4155 - LOV on echo now works (409)
4157 - mds_reint_unlink sends a new buffer, with the EA included. this
4158 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4159 set if a regular file was being unlinked, and it was the last link
4160 - use PtlGet from the target for bulk writes (315)
4161 - OST now updates object mtime with writes or setattr (607, 619)
4162 - LDLM now has a grant-time callback to revalidate locked items, if
4164 - Many MDS operations were reorganized to combat race conditions
4166 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4167 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4169 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4172 - Fully reactivate OST imports after reconnection (512, others)
4173 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4174 - More graceful error handling for truncating on dead OST (515)
4175 - Don't error out unless we're actually accessing dead stripes (474)
4176 - Fix garbage sizes when stripes are missing (410)
4177 - LRU counters were broken, causing constant lock purge (433, 432)
4178 - garbage on read from stripes with failed OSTs (441)
4179 - mark OSCs as active before reconnecting during recovery (438)
4180 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4181 - lfind did not preserve OST order in output (443)
4182 - symlinks cause hung clients, incorrect data (439)
4183 - stop dereferencing request after dropping refcount (457)
4184 - don't LASSERT(spin_is_locked) on non-SMP (455)
4185 - fixes for many rename() bugs
4186 - fstat didn't correctly synchronize attributes (399)
4187 - server must handle lock cancellation during blocking AST prep (487)
4188 - bulk descriptors were free()d too soon (511)
4189 - fix paths in lconf, which would load incorrect modules (451, 507)
4190 - fix confusing lconf 'host not found' error message (386)
4191 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4192 - fix race condition in mdc_blocking_ast() for inode access (526)
4193 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4194 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4195 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4196 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4197 - fix lstripe in several ways (kernel side) (527)
4198 - fix request leak in ldlm_cli_enqueue (262)
4199 - incorrect OSC was marked inactive after OST failure
4200 - call mds_fs_cleanup before unmounting filesystem (524)
4201 - fix races between taking ns_lock and ldlm_lock_change_resource
4202 - fix races updating LOV export open file list
4203 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4204 - fix recovery NULL deref in ldlm_cli_cancel_unused
4205 - fix some DLM races by using new hash table for lock handles (419)
4206 - permit the client to specify desired inodes, at replay
4207 - duplicate requests when we queue them for replay reintegration
4208 - fix last_rcvd offset calculation
4209 - sync after each recovered transaction, so we always make progress
4210 - never, not always, ERESTART requests without transnos
4211 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4213 - skip replay if the MDS says that the client is already connected
4214 - don't check for a recovery-enabled export to match lctl's UUID
4215 - don't INC_USE_COUNT for phantom exports
4216 - don't crash when cleaning up phantom exports (567)
4217 - don't double-finish or set replay data for errored mdc_open requests
4218 - abort requests when they time out, so we don't get old replies
4219 - send/receive replies for AST messages again
4220 - if the client says that it doesn't have the lock, cancel it on the
4222 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4223 mark it as destroyed, it will all work out when decref is called
4224 - fix module use counts (22, 581)
4226 - ASTs now expect a reply (server cancels lock on error reply)
4228 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4231 - fix many simultaneous client startup (392)
4232 - fix dentry->d_it clobbering
4233 - credentials weren't being shipped for readdir/getattr operations
4234 - remove invalid assertions triggered during some concurrent MD
4236 - proper Lustre versions added (336, 389)
4237 - fix memory leak for create error case (398)
4238 - fix LOV locking bug that would get cli/srv out of sync
4239 - fix echo client over LOV (409)
4240 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4241 - fix double-O_EXCL intent crash (424)
4242 - avoid sending multiple lock CANCELs (352)
4244 - MDS can do multi-client recovery (modulo bugs in new code)
4246 - many updates, edits, cleanups
4248 2002-11-18 Phil Schwan <phil@clusterfs.com>
4251 - fix null d_it dereference (346)
4252 - fix full OST/dbench hang (333)
4253 - fix permission problem with file removal (286)
4254 - fix removal of OSCs from LOV when they fail
4255 - fix NULL deref during bulk timeout (214)
4256 - fix problems related to multiple filesystems on one MDS (241)
4257 - fixed serious subtle metadata locking bugs
4258 - free locks on clients when inodes are removed due to memory
4260 - fix inode pointer in lock data (285)
4261 - partial support for multiple MDS on a single host (241)
4262 - data locks weren't cancelled at clear_inode time (290, 311)
4263 - intent locks could lead to unbounded lock growth (205)
4264 - added a maximum lock count, an LRU list, and a flusher
4265 - fix multiple rename (365)
4266 - properly abstracted the echo client
4267 - OSC locked 1 byte too many; fixed
4268 - rewrote brw callback code:
4269 - fixed recovery bugs related to LOVs (306)
4270 - fixed too-many-pages-in-one-write crash (191)
4271 - fixed (again) crash in sync_io_timeout (214)
4272 - probably fixed callback-related race (385)
4274 - Add capability to MDS protocol
4275 - LDLM cancellations and callbacks on different portals
4277 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4280 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4281 - defer open object destruction to close time (601981, 138)
4282 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4283 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4284 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4285 - don't call obd_setattr in ll_file_release for destroyed objects
4286 * protocol change to lustre_msg: move |version| and add |flags|
4287 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4288 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4289 * added replay of create, unlink, link and rename operations during
4290 MDS failover; recovery should be much more robust now
4291 * remove failed OSCs from LOVs (only lov_create uses this so far)
4292 * the lustre-HOWTO was brought (more) up to date (582544)
4294 2002-10-23 Phil Schwan <phil@clusterfs.com>
4297 - in-use dentries weren't being reused properly (617851)
4298 - prevent multiple LDLM setup (599178)
4299 - fix LOV size calculations for truncate (617853)
4300 - fix client handling of MDS intent errors (POSIX)
4301 - fix permission bug in lovstripe.c test (624321)
4302 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4303 - truncate past end of file could corrupt data
4304 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4305 - a race in recovery could return ETIMEDOUT to apps (623947)
4306 - building outside the source directory was fixed
4307 * the lustre-HOWTO was brought (more) up to date (582544)
4308 * major progress was made on recovery functionality
4310 2002-10-10 Phil Schwan <phil@clusterfs.com>
4313 - recovery deadlock fix
4314 - rm -rf causes LBUG fix (617817)
4315 - file open by multiple tasks fix (618962)
4316 - directory permissions bugs (602707 and 620007)
4317 - journal_stop fixed with locking (611313)
4318 - O_APPEND failures resolved (618273, perhaps 614459)
4319 - lconf PATH fix (619770)
4320 - IA64 build fix (621450)
4321 - RPC buffer sizes scale with amount of memory
4323 2002-10-01 Phil Schwan <phil@clusterfs.com>
4326 - locks would be cancelled without throwing away data pages,
4327 resulting in inconsistent data (605627)
4328 - inode attributes were not always being refreshed (605627, 612449)
4329 - lconf now continues to cleanup after lctl reports an error
4330 - MDS now enforces user permissions (602707)
4331 - lprocfs cleanup fixed, but not yet enabled (614157)
4332 - fixed infinite server hang, should a client not respond to an AST
4333 - avoid going into recovery if user calls readlink() with a buffer
4334 that's too small (613941)
4335 - AST RPCs no longer require replies (614867) -- this may be changed
4336 - don't crash server if client sends an IOV that's too big (611336)
4337 - fixed lock conversion deadlock (611892)
4338 - fixed the following of symlinks (614622)
4339 * recovery: the server can remove locks from a client that dies, other
4340 clients can make progress
4341 * more extN patch fixes
4342 * compile-time configurable ptlrpc buffer allocations
4344 - collaborative read cache document
4345 - Lustre Lite Performance CDR document-in-progress
4347 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4350 - fix typo in patch-2.4.18
4352 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4355 - clear ptlrpc request each time in handle_incoming_request()
4356 - unlink of files now destroys the object on the OST
4358 2002-09-19 Peter Braam <braam@clusterfs.com>
4360 * add hard link support
4361 * change obdfile creation method
4362 * kernel patch changed
4364 2002-09-19 Peter Braam <braam@clusterfs.com>
4367 - stack overflow bug in extN fixed
4369 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4371 * documentation updates
4372 - add man pages for config tools
4373 - update tests/README to describe testing with new config tools
4374 - finish metadata API descriptions
4375 * bug fixes and cleanups
4376 - statfs workaround for 16TB limit
4377 - LOV stripe allocation improved, can stripe on subset of OSTs
4378 - LOV file size/IO offset was wrong for files > 4GB in size
4379 - object EA data was being dropped, caused files to be unreadable
4380 - memory overflow with non-LOV OST caused memory corruption
4381 - fixed regression tests to work with new config tools, obdfilter
4382 - fixed bug when directory size became larger than 1 block
4383 - fixed bug (for single client case) when PWD was deleted
4384 - invalidate local directory pages when doing intent-based ops
4385 - avoid LDLM oops when lock callback contained bad data
4387 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4389 * documentation updates
4390 * bug fixes and cleanups
4391 - configuration tools
4394 - 64-bit compile warnings
4395 - 64-bit internal statfs data
4397 * test_brw on persistent OST devices
4399 * lprocfs (disabled)
4401 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4403 * documentation updates
4404 * bug fixes and cleanups
4405 * configuration tools
4407 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4410 * many small fixes to 0_5_4
4411 * io/network handling
4412 * thinkos in MDS operations
4414 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4417 * crucial basic fixes to 0.5.3
4418 * IOR, Iozone work over Elan
4421 2002-08-07 Phil Schwan <phil@clusterfs.com>
4422 * version 0_5_3, our first alpha
4423 * we use the new Portals iovs
4424 * documentation updates
4425 * bug fixes and cleanups
4426 * small changes in the DLM wire protocol
4428 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4429 * version 0_5_1 with some initial stability,
4430 * locking on MD and file I/O.
4431 * documentation updates
4432 * several bug fixes since 0.5.0
4433 * small changes in wire protocol
4435 2002-07-18 Phil Schwan <phil@clusterfs.com>
4437 * delivered as Lustre Light Alpha
4438 * fixed a crash after handling invalid MDS requests
4439 * fixed directory pages for architectures with non-4k pages sizes
4441 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4442 * release version v0_4_4
4443 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4445 * Fixes a number of interruption problems with OST operations.
4446 * Update documentation for portals header changes
4447 * Move all wire protocol structs/defines to lustre_idl.h
4448 * Fixes symlink length bug.
4449 * Add tcpdump to repository.
4451 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4452 * release version v0_4_3
4453 * Fixes statfs for inodes on extN.
4454 * Fixes bug in runtests which would delete /etc/hosts.
4455 * Use 64-bit object IDs wherever possible (not into VFS though)
4456 Remove ost_get_info, which is unused by lustre, and out of date.
4458 2002-07-03 Peter Braam <braam@clusterfs.com>
4459 * release version v0_4_2 Fixes a lookup error (type not passed)
4460 * move forward to head of Portals
4461 * move forward to latest Lustre kernel
4463 2002-06-25 Peter Braam <braam@clusterfs.com>
4464 * release version v0_4_1. Hopefully stable on single node use.