1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.4.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
15 Severity : enhancement
17 Description: Add rhel5 support to HEAD.
19 Serverity : enhancement
21 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
26 Description: Don't allow skipping OSTs if index has been specified.
27 Details : Don't allow skipping OSTs if index has been specified, make locking
28 in internal create lots better.
32 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
33 Details : access to bitfield in structure is always rounded to long
34 and this produce problem with not atomic change any bit.
38 Description: Lustre make rpms failed.
39 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
40 builds Lustre rpms from tarball.
43 Frequency : rare on shutdown ost
45 Description: If llog cancel was not send before clean_exports phase, this can
46 produce deadlock in llog code.
47 Details : If llog thread has last reference to obd and call class_import_put
48 this produce deadlock because llog_cleanup_commit_master wait when
49 last llog_commit_thread exited, but this never success because was
50 called from llog_commit_thread.
54 Description: allow userland application know is lost one of stripes.
55 Details : fill lvb_blocks with error code on ost and return it to
56 application if error flag found.
60 Description: NULL lov_tgts causing MDS oops
61 Details : more safe checks for NULL lov_tgts for avoid oops.
63 Severity : enhancement
65 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
69 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
70 Details : If an OST is down the MDS will hang indefinitely in
71 obd_statfs_rqset() waiting for the statfs data. While for
72 MDS QOS usage of statfs, it should not stuck in waiting.
74 Severity : enhancement
76 Description: remote_acl support
77 Details : Support ACL-based permission check for remote user.
78 Support setfacl/getfacl for remote user with the utils
79 "lfs {l,r}{s,g}etfacl" which follow the same parameter format as
80 the system "{s,g}etfacl" utils.
82 Severity : enhancement
84 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
86 Severity : enhancement
88 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
92 Description: make mgs_setparam() handle fsname containing dash
93 Details : fsname containing a dash does not work with lctl conf_param
95 Severity : enhancement
97 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
99 Severity : enhancement
101 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
105 Frequency : rare, at shutdown
106 Description: access already free / zero obd_namespace.
107 Details : if client_disconnect_export was called without force flag set,
108 and exist connect request in flight, this can produce access to
109 NULL pointer (or already free pointer) when connect_interpret
110 store ocd flags in obd_namespace.
114 Frequency : only at startup
115 Description: not alloc memory with spinlock held.
116 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
117 if any spinlock held.
119 Severity : enhancement
121 Description: make lustre randomly fail allocating memory
122 Details : Make lustre randomly failed allocating memory for testing purpose.
124 Severity : enhancement
126 Description: lost problems with lov objid file
127 Details : Fixes some scability and access to not inited memory problems
128 in work with lov objdid file.
133 Description: lfs find does not continue on file error
134 Details : Continue other files processing when a file/dir is absent.
138 Description: Inconsistent usage of lustre_pack_reply()
139 Details : Standardize the usage of lustre_pack_reply() such that it
140 always generate a CERROR on failure.
143 Frequency : occasional
145 Description: MDS hang or stay in waiting lock
146 Details : If client receive lock with CBPENDING flag ldlm need send lock
147 cancel as separate rpc, to avoid situation when cancel request
148 can't processed due all i/o threads stay in wait lock.
152 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
156 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
160 Description: Update to SLES9 kernel-2.6.5-7.287.3.
164 Description: Update to RHEL4 latest kernel.
166 Severity : enhancement
168 Description: Build SLES10 patchless client fails
169 Details : The configure was broken by run ./configure with
170 --with-linux-obj=.... argument for patchless client. When the
171 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
172 can't search header adequately. Use absolute path such as
173 -I($LINUX)/include instead.
177 Description: interrupt oig_wait produce painc on resend.
178 Details : brw_redo_request can be used for resend requests from ptlrpcd and
179 private set, and this produce situation when rq_ptlrpcd_data not
180 copyed to new allocated request and triggered LBUG on assert
181 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
182 ptlrpcd set if request is changed and can be safety changed to use
187 Description: LASSERT_{REQ,REP}SWAB macros are buggy
188 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
189 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
190 these macros and replace them with their difinitions instead.
194 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
195 Details : Update which_patch & target file for SLES10 latest kernel.
198 Description: Remove client patches from SLES 10 kernel.
199 Details : This causes SLES 10 clients to behave as patchless clients
200 even on a Lustre-patched (server) kernel.
202 Severity : enhancement
204 Description: self-adjustable client's lru lists
205 Details : use adaptive algorithm for managing client cached locks lru
206 lists according to current server load, other client's work
207 pattern, memory activities, etc. Both, server and client
208 side namespaces provide number of proc tunables for controlling
211 Severity : enhancement
213 Description: light-weight GSS support
214 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
215 and reduce performance overhead.
217 Severity : enhancement
219 Description: Linux keyring support
220 Details : Support using service of Linux keyring for Lustre GSS internal
221 context refresh/cache mechanism.
225 Description: Fix errors in lfs documentation
226 Details : Fixes man pages
230 Description: don't use GFP_* in generic Lustre code.
231 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
235 Description: obdclass is limited by single OBD_ALLOC(idarray)
236 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
240 Description: warnings with build patchless client with vanila 2.6.19 and up
241 Details : change old ctl_table style and replace ctl_table/ctl_table_header
242 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
246 Frequency : Only for SLES
247 Description: sanity_quota fail test_1
248 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
249 file, which makes TSTID[2] inunique.
253 Frequency : Only for SLES9
254 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
255 Details : Update target/ChangeLog/which_patch .
259 Frequency : Only for test_10 in sanity-quota.sh
260 Description: a bug in quota test code
261 Details : checking a test flag in a wrong place
265 Frequency : Only for test_13 in sanity-quota.sh
266 Description: a bug in quota test script
267 Details : shouldn't check overfull allocated quota
271 Description: jbd statistics
272 Details : Port older jbd statistics patch for sles10
274 Severity : enhancement
276 Description: Add printing inode info into message about error in writepage.
281 Description: reply_lock_interpret crash due to race with it and lock cancel.
282 Details : Do not replay locks that are being cancelled. Do not reference
283 locks by their address during replay, just by their handle.
287 Frequency : When flocks are used.
288 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
289 Details : Flock locks might destroy just granted lock if it could be merged
290 with another existing flock, this is done in completion handler,
291 so teach ldlm_cli_enquque_fini that this is a valid case for
297 Description: Oops in read and write path when failing to allocate lock.
298 Details : Check if lock allocation failed and return error back.
302 Description: lstripe command fails for valid OST index
303 Details : The stripe offset is compared to lov->desc.ld_tgt_count
304 instead of lov->desc.ld_active_tgt_count.
308 Description: sanity.sh failed test 103
309 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
314 Description: df doesn't work properly if diskfs blocksize != 4K
315 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
319 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
320 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
324 Description: obd_health_check_timeout too short
325 Details : set obd_health_check_timeout as 1.5x of obd_timeout
329 Description: llapi_file_create() does not allow some changes
330 Details : add llapi_file_open() that allows specifying the mode and
331 open flags, and also returns an open file handle.
335 Description: merge and cleanup kernel patches.
336 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
340 Description: Add journal checksum support.(Kernel part)
341 Details : The journal checksum feature adds two new flags i.e
342 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
343 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
344 indicates that the commit block contains the checksum for
345 the blocks described by the descriptor blocks. Now commit
346 record can be sent to disk without waiting for descriptor
347 blocks to be written to disk. This behavior is controlled
348 using JBD2_FEATURE_ASYNC_COMMIT flag.
352 Description: OSS needs mutliple precreate threads
353 Details : Add ability to start more than one create thread per OSS.
357 Description: Sanity.sh test_65k failed: setstripe should have succeeded
358 Details : execute lfs setstripe on client
362 Description: mds_obd_create error creating tmp object
363 Details : When the user sets quota on root, llog will be affected and can't
364 create files and write files.
367 Frequency : Always on ia64 patchless client, and possibly others.
369 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
370 Details : This allows the patchless client to be loaded on architectures
375 Description: RedHat Update kernel for RHEL5
376 Details : Add the kernel config file for RHEL5.
380 Description: RedHat Update kernel for RHEL5
381 Details : Modify the kernel config file more closer RHEL5.
385 Description: Build failure against Centos5 (RHEL5)
386 Details : Define PAGE_SIZE when it isn't present.
390 Description: test_7 on sanity-quota.sh can't be used on separate machines
391 Details : add facet to handle it
395 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
396 Details : with NFS, the anon dentry's parent was set to itself in
397 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
398 corresponding dentry other than use rec->ur_name.
400 Severity : enhancement
402 Description: lfs setstripe enhancement
403 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
407 Description: enable data checksumming by default
408 Details : enable checksum by default, allow --disable-checksum
409 configure option and "-o nochecksum" mount option. Checksums
410 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
414 Description: lustre support for RHEL5
415 Details : Add support for RHEL5.
419 Description: Client eviction due to failover config
420 Details : after a connection loss, the lustre client should attempt to
421 reconnect to the last active server first before trying the
422 other potential connections.
426 Description: when mds and osts use different quota unit(32bit and 64bit),
427 quota will be released repeatly.
428 Details : void sending multiple quota reqs to mds, which will keep the
429 status between the reqs.
433 Description: osts not allocated evenly to files
434 Details : change the condition to increase offset_idx
437 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
439 Description: Data corruption for OSTs that are formatted larger than 2TB
441 Details : When generating the bio request for lustre file writes the
442 sector number would overflow a temporary variable before being
443 used for the IO. The data reads correctly from Lustre (which
444 will overflow in a similar manner) but other file data or
445 filesystem metadata may be corrupted in some cases.
449 Description: Tune the kernel for good SCSI performance.
450 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
451 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
456 Description: rewrite ext2-derived code in obdclass/uuid.c
457 Details : rewrite inherited code (uuid parsing code from ext2 utils)
458 from scratch preserving functionality.
462 Description: "lfs find -obd UUID" prints directories
463 Details : "lfs find -obd UUID" will return all directory names instead
464 of just file names. It is incorrect because the directories
465 do not reside on the OSTs.
469 Description: use i_size_read and i_size_write in 2.6 port
470 Details : replace inode->i_size access with i_size_read/write()
473 Frequency : when removing large files
475 Description: scheduling issue during removal of large Lustre files
476 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
477 It causes scheduling issues when removing large files (17TB in the
481 Frequency : only with liblustre clients on XT3
483 Description: evictions taking too long
484 Details : allow llrd to evict clients directly on OSTs
487 Frequency : only on ppc
489 Description: /proc/fs/lustre/devices broken on ppc
490 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
491 We should make sure the type of 'index' is loff_t and then cast
492 explicitly as needed below. Do not assign an explicitly cast
496 Frequency : only for rhel5
498 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
499 Details : Modify the target file & which_kernel.
501 Severity : enhancement
503 Description: omit set fsid for export NFS
504 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
505 as FS_REQUIRES_DEV for avoid problems with generate fsid.
509 Frequency : Always, for kernels after 2.6.16
510 Description: Fix warning idr_remove called for id=.. which is not allocated.
511 Details : Last kernels save old s_dev before kill super and not allow
512 to restore from callback - restore it before call kill_anon_super.
516 Description: buffer overruns could theoretically occur
517 Details : llapi_semantic_traverse() modifies the "path" argument by
518 appending values to the end of the origin string, and a
519 overrun may occur. Adding buffer overrun check in liblustreapi.
523 Description: Fix error on 'ls .' at the top of the Lustre mount.
524 Details : Don't revalidate dentry if it is a root dentry.
528 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
529 Details : Modify vm-tunables-rhel4.patch.
533 Description: Kernel config for 2.6.18-vanilla.
534 Details : Modify targets/2.6-vanilla.target.in.
535 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
536 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
537 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
538 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
542 Description: improve handling recoverable errors
543 Details : if request processig with error which can be recoverable on server
544 request should be resend, otherwise page released from cache and
549 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
550 Details : Modify target file & which_patch.
551 A flaw was found in the IA32 system call emulation provided
552 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
553 value could be stored in the %RAX register, which could trigger an
554 out-of-bounds system call table access. An untrusted local user
555 could exploit this flaw to run code in the kernel
556 (ie a root privilege escalation). (CVE-2007-4573).
560 Description: O_DIRECT bypasses client statistics.
561 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
562 read_ahead_stats were updated. copied the stats section from
563 osc_send_oap_rpc() into async_internal().
567 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
571 Description: change order of libsysio includes
572 Details : '#include sysio.h' should always come before '#include xtio.h'
576 Description: handle "serious error: objid * already exists" more gracefully
577 Details : If LAST_ID value on disk is smaller than the objects existing in
578 the O/0/d* directories, it indicates disk corruption and causes an
579 LBUG(). If the object is 0-length, then we should use the existing
580 object. This will help to avoid a full fsck in most cases.
582 Severity : enhancement
584 Description: adapt the lustre_config script to support the upgrade case
585 Details : Add "-u" option for lustre_config script to support upgrading 1.4
586 server targets to 1.6 in parallel.
590 Description: To avoid grant space > avaible space when the disk is almost
591 full. Without this patch you might see the error "grant XXXX >
592 available" or some LBUG about grant, when the disk is almost
594 Details : In filter_check_grant, for non_grant cache write, we should
595 check the left space by if (*left > ungranted + bytes), instead
596 of (*left > ungranted), because only we are sure the left space
597 is enough for another "bytes", then the ungrant space should be
598 increase. In client, we should update cl_avail_grant only there
599 is OBD_MD_FLGRANT in the reply.
603 Description: Update RHEL 4 kernel to fix local root privilege escalation.
604 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
605 described in CVE-2007-4573. This problem could allow untrusted
606 local users to gain root access.
609 Frequency : when using O_DIRECT and quotas
611 Description: Incorrect file ownership on O_DIRECT output files
612 Details : block usage reported by 'lfs quota' does not take into account
613 files that have been written with O_DIRECT.
618 Description: touch file failed when fs is not full
619 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
620 otherwise we can get ENOSP while fs is not full.
624 Description: parallel lock callbacks
625 Details : Instead of sending blocking and completion callbacks as separated
626 requests, adding them to a set and sending in parallel.
629 Frequency : only for Cray XT3
630 Bugzilla : 12829/13455
631 Description: Changing primary group doesn't change the group lustre assigns to
633 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
634 overridden with the primary group provided by the group upcall,
635 whereas the supplied fsgid can be trusted if it is in the list of
636 supplementary groups returned by the group upcall.
638 Severity : enhancement
640 Description: Allow masking D_WARNING, D_ERROR messages from console
641 Details : Console messages can now be disabled via lnet.printk.
645 Description: User code with malformed file open parameter crashes client node
646 Details : Before packing join_file req, all the related reference should be
647 checked carefully in case some malformed flags cause fake
648 join_file req on client.
652 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
653 Details : ldlm_completion_ast() assumes that a lock is granted when the req
654 mode is equal to the granted mode. However, it should also check
655 that LDLM_FL_CP_REQD is not set.
659 Description: Heavy nfs access might result in deadlocks
660 Details : After ELC code landed, it is now improper to enqueue any mds
661 locks under och_sem, because enqueue might want to decide to
662 cancel open locks for same inode we are holding och_sem for.
666 Description: Client eviction while running blogbench
667 Details : A lot of unlink operations with concurrent I/O can lead to a
668 deadlock causing evictions. To address the problem, the number of
669 oustanding OST_DESTROY requests is now throttled to
670 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
671 callbacks are processed in priority.
673 --------------------------------------------------------------------------------
675 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
677 * Support for kernels:
678 2.6.5-7.283 (SLES 9),
679 2.6.9-55.EL (RHEL 4),
680 2.6.16.46-0.14 (SLES 10),
681 2.6.18.8 vanilla (kernel.org)
682 * Client support for unpatched kernels:
683 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
684 2.6.16 - 2.6.22 vanilla (kernel.org)
685 * Due to recently discovered recovery problems, we do not recommend
686 using patchless RHEL 4 clients with this or any earlier release.
687 * Recommended e2fsprogs version: 1.39.cfs8
688 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
689 * Starting with this release, the ldiskfs backing filesystem required
690 by Lustre is now in its own package, lustre-ldiskfs. This package
691 should be installed. It is versioned separately from Lustre and
692 may be released separately in future.
696 Description: block reactivating mgc import until all deactivates complete
697 Details : Fix race when failing back MDT/MGS to itself (testing)
700 Frequency : at statup only
702 Description: mds_lov_synchronize race leads to various problems
703 Details : simultaneous MDT->OST connections at startup can cause the
704 sync to abort, leaving the OSC in a bad state.
706 Severity : enhancement
708 Description: add optional extra BUILD_VERSION info
709 Details : add a new environment variable (namely LUSTRE_VERS) which allows
710 to override the lustre version.
713 Frequency : 2.6.18 servers only
715 Description: ll_kern_mount() doesn't release the module reference
716 Details : The ldiskfs module reference count never drops down to 0
717 because ll_kern_mount() doesn't release the module reference.
722 Description: server LBUG when using old ost_num_threads parameter
723 Details : Accept the old ost_num_threads parameter but warn that it
724 is deprecated, and fix an off-by-one error that caused an LBUG.
729 Description: Transient SCSI error results in persistent IO issue
730 Details : iobuf->dr_error is not reinitialized to 0 between two
734 Frequency : sometimes when underlying device returns I/O errors
736 Description: OSTs not going read-only during write failures
737 Details : OSTs are not remounted read-only when the journal commit threads
738 get I/O errors because fsfilt_ext3 calls journal_start/stop()
739 instead of the ext3 wrappers.
743 Description: poor connect scaling with increasing client count
744 Details : Don't run filter_grant_sanity_check for more than 100 exports
745 to improve scaling for large numbers of clients.
748 Frequency : SLES10 only
750 Description: sanity-quota.sh quotacheck failed: rc = -22
751 Details : Quotas cannot be enabled on SLES10.
754 Frequency : liblustre clients only
756 Description: getdirentries does not give error when run on compute nodes
757 Details : getdirentries does not fail when the size specified as an argument
758 is too small to contain at least one entry
760 Severity : enhancement
762 Description: Add LNET router traceability for debug purposes
763 Details : If a checksum failure occurs with a router as part of the
764 IO path, the NID of the last router that forwarded the bulk data
765 is printed so it can be identified.
770 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
771 Details : Due to a race condition, liblustre clients were occasionally
774 Severity : enhancement
776 Description: lfs setstripe use optional parameters instead of postional
779 Severity : enhancement
781 Description: Nanosecond timestamp support for ldiskfs
782 Details : The on-disk ldiskfs filesystem has added support for nanosecond
783 resolution timestamps. There is not yet support for this at
784 the Lustre filesystem level.
787 Frequency : during server recovery
789 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
790 Details : request with rq_no_resend flag not awake l_wait_event if they get a
794 Frequency : nfs export on patchless client
796 Description: connectathon hang when test nfs export over patchless client
797 Details : Disconnected dentry cannot be found with lookup, so we do not need
798 to unhash it or make it invalid
801 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
802 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
803 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
808 Description: open req refcounting wrong on reconnect
809 Details : If reconnect happened between getting open reply from server and
810 call to mdc_set_replay_data in ll_file_open, we will schedule
811 replay for unreferenced request that we are about to free.
812 Subsequent close will crash in variety of ways.
813 Check that request is still eligible for replay in
814 mdc_set_replay_data().
819 Description: disable writes to filesystem when reading health_check file
820 Details : the default for reading the health_check proc file has changed
821 to NOT do a journal transaction and write to disk, because this
822 can cause reads of the /proc file to hang and block HA state
823 checking on a healthy but otherwise heavily loaded system. It
824 is possible to return to the previous behaviour during configure
825 with --enable-health-write.
827 Severity : enhancement
829 Description: 64-bit inode version
830 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
831 to the inode. This will be required for version-based recovery.
836 Description: MDS fails to start if a duplicate client export is detected
837 Details : in some rare cases it was possible for a client to connect to
838 an MDS multiple times. Upon recovery the MDS would detect this
839 and fail during startup. Handle this more gracefully.
841 Severity : enhancement
843 Description: Add -o localflock option to simulate old noflock
845 Details : This will achieve local-only flock/fcntl locks
851 Description: log_commit_thread vs filter_destroy race leads to crash
852 Details : Take import reference before releasing llog record semaphore
857 Description: Wrong request locking in request set processing
858 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
859 imp_delayed_list, in this place should be used imp_lock.
862 Frequency : when reconnection
864 Description: Grant Leak when osc reconnect to OST
865 Details : When osc reconnect ost, OST(filter) should check whether it
866 should grant more space to client by comparing fed_grant and
867 cl_avail_grant, and return the granted space to client instead
868 of "new granted" space, because client will call osc_init_grant
869 to update the client grant space info.
872 Frequency : when client reconnect to OST
874 Description: Grant Leak when osc do resend and replay bulk write
875 Details : When osc reconnect to OST, OST(filter)should clear grant info of
876 bulk write request, because the grant info while be sync between
877 OSC and OST when reconnect, and we should ignore the grant info
878 these of resend/replay write req.
883 Description: Grant space more than avaiable left space sometimes.
884 Details : When then OST is about to be full, if two bulk writing from
885 different clients came to OST. Accord the avaliable space of the
886 OST, the first req should be permitted, and the second one
887 should be denied by ENOSPC. But if the seconde arrived before
888 the first one is commited. The OST might wrongly permit second
889 writing, which will cause grant space > avaiable space.
892 Frequency : when client is evicted
894 Description: Grant might be wrongly erased when osc is evicted by OST
895 Details : when the import is evicted by server, it will fork another
896 thread ptlrpc_invalidate_import_thread to invalidate the
897 import, where the grant will be set to 0. While the original
898 thread will update the grant it got when connecting. So if
899 the former happened latter, the grant will be wrongly errased
900 because of this race.
905 Description: Checking Stale with correct fid
906 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
907 to check whether it is stale, because sometimes, we want the
908 enqueue happened anyway, and op_data.fid2 will not be initialized.
910 Severity : enhancement
912 Description: update patchless client
913 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
916 Frequency : only with 2.4 kernel
918 Description: random memory corruption
919 Details : size of struct ll_inode_info is to big for union inode.u and this
920 can be cause of random memory corruption.
925 Description: Memory leak in recovery
926 Details : Lov_mds_md was not free in an error handler in mds_create_object.
927 It should also check obd_fail before fsfilt_start, otherwise if
928 fsfilt_start return -EROFS,(failover mds during mds recovery).
929 then the req will return with repmsg->transno = 0 and rc = EROFS.
930 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
931 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
932 be freed no matter whether fsfilt_commit success or not.
935 Frequency : only with huge count clients
937 Description: Prevents from taking the superblock lock in llap_from_page for
939 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
940 for llap_from_page call in ll_removepage prevents from taking the
941 superblock lock for a soon died page.
946 Description: Not check open intent error before release open handle
947 Details : in some rare cases, the open intent error is not checked before
948 release open handle, which may cause
949 ASSERTION(open_req->rq_transno != 0), because it tries to release
950 the failed open handle.
955 Description: Set cat log bitmap only after create log success.
956 Details : in some rare cases, the cat log bitmap is set too early. and it
957 should be set only after create log success.
961 Description: Accessing a block bevice can re-enable I/O when Lustre is
962 tearing down a device.
963 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
968 Description: service threads may hog cpus when there are a lot of requests
969 Details : Insert cond_resched to give other threads a chance to use some CPU
974 Description: the cat log was not initialized in recovery
975 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
976 unlink log on mds will not be initialized until mds post
977 recovery. And also in mds post recovery, the unlink log will
978 initialization will be done asynchronausly, so there will be race
979 between add unlink log and unlink log initialization.
983 Description: brw_stats were being printed incorrectly
984 Details : brw_stats were being printed as log2 but all of them were not
985 recorded as log2. Also remove some code duplication arising from
986 filter_tally_{read,write}.
990 Frequency : rare, only in recovery.
991 Description: ASSERTION(req->rq_type != LI_POISON) failed
992 Details : imp_lock should be held while iterating over imp_sending_list for
993 prevent destroy request after get timeout in ptlrpc_queue_wait.
997 Description: replay-single.sh test 52 fails
998 Details : A lock's skiplist need to be cleanup when it being unlinked
999 from its resource list.
1003 Description: Short directio read returns full requested size rather than
1005 Details : Direct I/O operations should return actual amount of bytes
1006 transferred rather than requested size.
1008 Severity : enhancement
1010 Description: metadata RPC reduction (e.g. for rm performance)
1011 Details : decrease the amount of synchronous RPC between clients and servers
1012 by canceling conflicing lock before the operation on the client
1013 and packing thier handles into the main operation RPC to server.
1015 Severity : enhancement
1017 Description: Async OSC create to avoid the blocking unnecessarily.
1018 Details : If a OST has no remain object, system will block on the creating
1019 when need to create a new object on this OST. Now, ways use
1020 pre-created objects when available, instead of blocking on an
1021 empty osc while others are not empty. If we must block, we block
1022 for the shortest possible period of time.
1024 Severity : enhancement
1026 Description: refine locking for avoid write wrong info into lov_objid file
1027 Details : fix possible races with add new target and write/update data in
1030 --------------------------------------------------------------------------------
1032 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1037 Frequency : on some architectures
1039 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1040 Details : Uninitialized flags sometimes cause configuration commands to
1044 Frequency : patchless clients only
1046 Description: missing __iget() symbol export
1047 Details : The __iget() symbol export is missing. To avoid the need for
1048 this on patchless clients the deathrow inode reaper is turned
1049 off, and we depend on the VM to clean up old inodes. This
1050 dependency was during via the fix for bug 12181.
1052 --------------------------------------------------------------------------------
1054 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1056 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1057 INTEROPERATE with older versions automatically. In many cases a
1058 special upgrade step is needed. Please read the
1059 user documentation before upgrading any part of a live system.
1060 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
1061 version will not interoperate with 1.6 betas before beta5 (1.5.95).
1062 * WARNING: Lustre configuration and startup changes are required with
1063 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1069 Severity : enhancement
1071 Description: MountConf
1072 Details : Lustre configuration is now managed via mkfs and mount
1073 commands instead of lmc and lconf. New obd types (MGS, MGC)
1074 are added for dynamic configuration management. See
1075 https://mail.clusterfs.com/wikis/lustre/MountConf for
1078 Severity : enhancement
1080 Description: dynamic OST addition
1081 Details : OSTs can now be added to a live filesystem
1083 Severity : enhancement
1085 Description: startup order invariance
1086 Details : MDTs and OSTs can be started in any order. Clients only
1087 require the MDT to complete startup.
1089 Severity : enhancement
1091 Description: parallel, asynchronous orphan cleanup
1092 Details : orphan cleanup is now performed in separate threads for each
1093 OST, allowing parallel non-blocking operation.
1095 Severity : enhancement
1097 Description: optimized stripe assignment
1098 Details : stripe assignments are now made based on ost space available,
1099 ost previous usage, and OSS previous usage, in order to try
1100 to optimize storage space and networking resources.
1102 Severity : enhancement
1104 Description: Permanently set tunables
1105 Details : All writable /proc/fs/lustre tunables can now be permanently
1106 set on a per-server basis, at mkfs time or on a live system.
1108 Severity : enhancement
1110 Description: Lustre message v2
1111 Details : Add lustre message format v2.
1113 Severity : enhancement
1115 Description: client OST exclusion list
1116 Details : Clients can be started with a list of OSTs that should be
1117 declared "inactive" for known non-responsive OSTs.
1121 Description: SPEC SFS validation failure on NFS v2 over lustre.
1122 Details : Changes the blocksize for regular files to be 2x RPC size,
1123 and not depend on stripe size.
1125 Severity : enhancement
1127 Description: Multiple MD RPCs in flight.
1128 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1129 To avoid overly-overloading MDS, introduce a limit on number of
1130 MDS RPCs in flight for a single client and add /proc controls
1131 to adjust this limit.
1133 Severity : enhancement
1135 Description: client read/write statistics
1136 Details : Add client read/write call usage stats for performance
1137 analysis of user processes.
1138 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1139 file access. extents_stats shows chunk size distribution.
1140 extents_stats_per_process show chunk size distribution per
1143 Severity : enhancement
1145 Description: mds statistics
1146 Details : Add detailed mds operations statistics in
1147 /proc/fs/lustre/mds/*/stats.
1151 Description: Failure of copying files with lustre special EAs.
1152 Details : Client side always return success for setxattr call for lustre
1153 special xattr (currently only "trusted.lov").
1158 Description: Refcount LNET uuids
1159 Details : The global LNET uuid list grew linearly with every startup;
1160 refcount repeated list entries instead of always adding to
1163 Severity : enhancement
1165 Description: Dynamic service threads
1166 Details : Within a small range, start extra service threads
1167 automatically when the request queue builds up.
1170 Frequency : mixed-endian client/server environments
1172 Description: mixed-endian crashes
1173 Details : The new msg_v2 system had some failures in mixed-endian
1176 Severity : enhancement
1178 Description: Easy OST removal
1179 Details : OSTs can be permanently deactivated with e.g. 'lctl
1180 conf_param lustre-OST0001.osc.active=0'
1182 Severity : enhancement
1184 Description: MGS proc entries
1185 Details : Added basic proc entries for the MGS showing what filesystems
1188 Severity : enhancement
1190 Description: provide MGS failover
1191 Details : Added config lock reacquisition after MGS server failover.
1193 Severity : enhancement
1195 Description: add Linux 2.4 support
1196 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1200 Description: a large application tries to do I/O to the same resource and dies
1201 in the middle of it.
1202 Details : Check the req->rq_arrival time after the call to
1203 ost_brw_lock_get(), but before we do anything about
1204 processing it & sending the BULK transfer request. This
1205 should help move old stale pending locks off the queue as
1206 quickly as obd_timeout.
1209 Frequency : when an incorrect nid is specified during startup
1211 Description: ptlrpc connect to non-existant node causes kernel crash
1212 Details : LNET can't be re-entered from an event callback, which
1213 happened when we expire a message after the export has been
1214 cleaned up. Instead, hand the zombie cleanup off to another
1217 Severity : enhancement
1219 Description: plain/inodebits lock performance improvement
1220 Details : Grouping plain/inodebits in granted list by their request modes
1221 and bits policy, thus improving the performance of search through
1225 Frequency : only if OST filesystem is corrupted
1227 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1228 Details : for a short time RPCs with bulk IO are in the replay list,
1229 but replay of bulk IOs is unimplemented. If the OST filesystem
1230 is corrupted due to disk cache incoherency and then replay is
1231 started it is possible to trip an assertion. Avoid putting
1232 committed RPCs into the replay list at all to avoid this issue.
1235 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1237 Description: System hang on startup
1238 Details : This bug allowed the liblustre (e.g. catamount) client to
1239 return to the app before handling all startup RPCs. This
1240 could leave the node unresponsive to lustre network traffic
1241 and manifested as a server ptllnd timeout.
1243 Severity : enhancement
1245 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1246 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1247 boolean to control whether to print peer debug info when a
1248 client's RPC times out.
1251 Frequency : only for kernels with patches from Lustre below 1.4.3
1253 Description: Remove old rdonly API
1254 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1257 Frequency : only for devices with external journals
1259 Description: Set external device read-only also
1260 Details : During a commanded failover stop, we set the disk device
1261 read-only while the server shuts down. We now also set any
1262 external journal device read-only at the same time.
1265 Frequency : when upgrading from 1.4 while trying to change parameters
1267 Description: The wrong (new) MDC name was used when setting parameters for
1268 upgraded MDT's. Also allows changing of OSC (and MDC)
1269 parameters if --writeconf is specified at tunefs upgrade time.
1272 Frequency : when setting specific ost indicies
1274 Description: QOS code breaks on skipped indicies
1275 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1276 created with --index need not be sequential.
1278 Severity : enhancement
1280 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1281 Details : The uninit_groups feature works in conjunction with the kernel
1282 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1283 up the pass1 processing of e2fsck. This is a read-only feature
1284 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1285 mount filesystems that have had this feature enabled.
1287 Severity : enhancement
1289 Description: Improve multi-block allocation algorithm to avoid fragmentation
1290 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1291 allocation locality and avoid filesystem fragmentation.
1293 ------------------------------------------------------------------------------
1295 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1297 * Support for kernels:
1298 2.4.21-47.0.1.EL (RHEL 3)
1299 2.6.5-7.283 (SLES 9)
1300 2.6.9-42.0.10.EL (RHEL 4)
1301 2.6.12.6 vanilla (kernel.org)
1302 2.6.16.27-0.9 (SLES 10)
1303 * Recommended e2fsprogs version: 1.39.cfs5
1305 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1309 Frequency : occasional, depends on client load and configuration
1310 Bugzilla : 12181, 12203
1311 Description: data loss for recently-modified files
1313 Details : In some cases it is possible that recently written or created
1314 files may not be written to disk in a timely manner (this should
1315 normally be within 30s unless client IO load is very high).
1316 The problem appears as zero-length files or files that are a
1317 multiple of 1MB in size after a client crash or client eviction
1318 that are missing data at the end of the file.
1320 This problem is more likely to be hit on clients where files are
1321 repeatedly created and unlinked in the same directory, clients
1322 have a large amount of RAM, have many CPUs, the filesystem has
1323 many OSTs, the clients are rebooted frequently, and/or the files
1324 are not accessed by other nodes after being written.
1326 The presence of the problem can be detected by looking at
1327 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1328 smaller than the second (nr_unused) then dirty files will not
1329 be flushed automatically to disk. "sync; sleep 10" should be
1330 run several times on the node before unmounting it to update
1331 Lustre (this is also safe to run on nodes without this problem).
1333 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1334 that can cause this same problem, so customers using that kernel
1335 also need to update the kernel in addition to Lustre. In order
1336 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1338 It is normal that files written just before a client crash (less
1339 than 30s) may not yet have been flushed to disk, even for local
1343 Frequency : frequent on thin XT3 nodes
1345 Description: UUID collision on thin XT3 Linux nodes
1346 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1347 randomly, since we relied on an insufficiently-seeded PRNG.
1352 Description: OSS hangs after "All ost request buffers busy"
1353 Details : A deadlock between quota and journal operations caused OSS
1354 hangs after printing "All ost request buffers busy."
1357 Frequency : always on liblustre builds
1359 Description: Cleanup compiler warnings on liblustre
1362 Frequency : always on liblustre builds on XT3
1364 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1368 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1369 > 32000 subdirectories
1370 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1371 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1372 subdirectory count crosses 32000. This will aid e2fsck to
1373 correctly handle more than 32000 subdirectories.
1376 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1378 Description: System hang on startup
1379 Details : This bug allowed the liblustre (e.g. catamount) client to
1380 return to the app before handling all startup RPCs. This
1381 could leave the node unresponsive to lustre network traffic
1382 and manifested as a server ptllnd timeout.
1384 Severity : enhancement
1386 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1387 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1388 boolean to control whether to print peer debug info when a
1389 client's RPC times out.
1394 Description: make O_SYNC working on 2.6 kernels
1395 Details : 2.6 kernels use different method for mark pages for write,
1396 so need add a code to lustre for O_SYNC work.
1401 Description: Failure to close file and release space on NFS
1402 Details : Put inode details into lock acquired in ll_intent_file_open.
1403 Use mdc_intent_lock in ll_intent_open to properly
1404 detect all kind of errors unhandled by mdc_enqueue.
1409 Description: proc file read during shutdown sometimes raced obd removal,
1411 Details : Add lock to prevent obd access after proc file removal.
1414 Frequency : Only for files larger than 4GB on 32-bit clients.
1416 Description: improperly doing page alignment of locks
1417 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1418 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1423 Description: RPCs being resent when they shouldn't be
1424 Details : Some RPCs that should not be resent are being resent. This
1425 can cause inconsistencies in the RPC state machine. Do not
1426 resend such requests.
1429 Frequency : rare, only with NFS export
1431 Description: Crash on NFS re-export node
1432 Details : under very unusual load conditions an assertion is hit in
1433 ll_intent_file_open()
1436 Frequency : only if OST filesystem is corrupted
1438 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1439 Details : for a short time RPCs with bulk IO are in the replay list,
1440 but replay of bulk IOs is unimplemented. If the OST filesystem
1441 is corrupted due to disk cache incoherency and then replay is
1442 started it is possible to trip an assertion. Avoid putting
1443 committed RPCs into the replay list at all to avoid this issue.
1448 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1449 Details : Large single O_DIRECT read and write calls can fail to allocate
1450 a sufficiently large buffer to process the request. In case of
1451 allocation failure the allocation is retried with a smaller
1452 buffer and broken into smaller requests.
1454 Severity : enhancement
1456 Description: Add -o localflock option to simulate old noflock behaviour.
1457 Details : This will achieve local-only flock/fcntl locks coherentness.
1462 Description: versioning check is incomplete
1463 Details : Checking the version difference of client vs. server, report
1464 error if the gap is too big.
1469 Description: add support PG_writeback bit
1470 Details : add support for PG_writeback bit for Lustre, for more carefull
1471 work with page cache in 2.6 kernel. This also fix some deadlocks
1472 and remove hack for work O_SYNC with 2.6 kernel.
1474 Severity : enhancement
1476 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1477 Details : The uninit_groups feature works in conjunction with the kernel
1478 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1479 up the pass1 processing of e2fsck. This is a read-only feature
1480 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1481 mount filesystems that have had this feature enabled.
1483 Severity : enhancement
1485 Description: Improve multi-block allocation algorithm to avoid fragmentation
1486 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1487 allocation locality and avoid filesystem fragmentation.
1489 ------------------------------------------------------------------------------
1491 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1493 * Support for kernels:
1494 2.6.9-42.0.3.EL (RHEL 4)
1495 2.6.5-7.276 (SLES 9)
1496 2.4.21-47.0.1.EL (RHEL 3)
1497 2.6.12.6 vanilla (kernel.org)
1498 2.6.16.21-0.8 (SLES10)
1499 * Recommended e2fsprogs version: 1.39.cfs2-0
1501 * The backwards-compatible /proc/sys/portals symlink has been removed
1502 in this release. Before upgrading, please ensure that you change
1503 any configuration scripts or /etc/sysctl.conf files that access
1504 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1505 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1506 in advance of the upgrade on any system running Lustre 1.4.6 or
1507 newer, since /proc/sys/lnet was added in that version.
1508 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1512 Frequency : only when quota is used
1514 Description: avoid scanning export list for quota master
1515 Details : Change the algorithms to avoid scanning export list in order
1516 to improve the efficiency.
1519 Frequency : MDS failover only, very rarely
1521 Description: "went back in time" messages on mds failover
1522 Details : The greatest transno may be lost when the current operation
1523 finishes with an error (transno==0) and the client's last_rcvd
1524 record is over-written. Save the greatest transno in the
1525 mds_last_transno for this case.
1528 Frequency : always for specific kernels and striping counts
1530 Description: client may get "Matching packet too big" without ACL support
1531 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1532 when trying to access files in certain configurations. The
1533 clients should in fact be denied when mounting because they do
1534 not understand ACLs.
1537 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1539 Description: many clients connecting with IO in progress causes connect timeouts
1540 Details : Avoid synchronous journal commits to avoid delays caused by many
1541 clients connecting/disconnecting when bulk IO is in progress.
1542 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1543 OST_IO_PORTAL to avoid delays behind potentially many pending
1547 Frequency : occasionally with multiple writers to a single file
1549 Description: shared writes to file may result in wrong size reported by stat()
1550 Details : Allow growing of kms when extent lock is cancelled
1553 Frequency : always with random mmap IO to multi-striped file
1555 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1556 Details : Only if the hole is at the end of OST object so that kms is too
1557 small. Fix is to increase kms accordingly in ll_nopage.
1560 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1562 Description: writes to a missing object would leak memory on the OST
1563 Details : If there is an inconsistency between the MDS and OST filesystems,
1564 such that the MDS references an object that doesn't exist, writes
1565 to that object will leak memory due to incorrect cleanup in the
1566 error handling path, eventually running out of memory on the OST.
1571 Description: Creating too long symlink causes lustre errors
1572 Details : Check symlink and name lengths before sending requests to MDS.
1575 Frequency : only if flock is enabled (not on by default)
1577 Description: posix locks not released on fd closure on 2.6.9+
1578 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1579 kernels, this caused such locks not to be released on fd close and
1580 then assertions on fs unmount about still used locks.
1583 Frequency : MDS failover only, very rarely
1585 Description: clients may get ASSERTION(granted_lock != NULL)
1586 Details : When request was taking a long time, and a client was resending
1587 a getattr by name lock request. The were multiple lock requests
1588 with the same client lock handle and
1589 mds_getattr_name->fixup_handle_for_resent_request found one of the
1590 lock handles but later failed with ASSERTION(granted_lock != NULL).
1595 Description: handle->h_buffer_credits > 0, assertion failure
1596 Details : h_buffer_credits is zero after truncate, causing assertion
1597 failure. This patch extends the transaction or creates a new
1601 Frequency : NFS re-export or patchless client
1602 Bugzilla : 11179, 10796
1603 Description: Crash on NFS re-export node (__d_move)
1604 Details : We do not want to hash the dentry if we don't have a lock.
1605 But if this dentry is later used in d_move, we'd hit uninitialised
1606 list head d_hash, so we just do this to init d_hash field but
1607 leave dentry unhashed.
1610 Frequency : NFS re-export or patchless client
1612 Description: NFS exports has problem with symbolic link
1613 Details : lustre client didn't properly install dentry when re-exported
1614 to NFS or running patchless client.
1617 Frequency : NFS re-export or patchless client
1619 Description: Various nfs/patchless fixes.
1620 Details : fixes reuse disconected alias for lookup process - this fixes
1621 warning "find_exported_dentry: npd != pd",
1622 fix permission error with open files at nfs.
1623 fix apply umask when do revalidate.
1626 Frequency : occasional
1628 Description: Crash on NFS re-export node
1629 Details : calling clear_page() on the wrong pointer triggered oops in
1630 generic_mapping_read().
1633 Frequency : rarely, using O_DIRECT IO
1635 Description: unaligned directio crashes client with LASSERT
1636 Details : check for unaligned buffers before trying any requests.
1639 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1641 Description: stale data returned from RAID cache
1642 Details : If only a small amount of IO is done to the RAID device before
1643 reading it again it is possible to get stale data from the RAID
1644 cache instead of reading it from disk.
1647 Frequency : always for sles10 kernel
1649 Description: sles10 support
1650 Details : ll_follow_link: compile fixes and using of nd_set_link
1651 under newer kernels.
1654 Frequency : depends on arch, kernel and compiler version, always on sles10
1657 Description: recursive or deep enough symlinks cause stack overflow
1658 Details : getting rid of large stack-allocated variable in
1662 Frequency : depends on hardware
1664 Description: lustre write performance loss in the SLES10 kernel
1665 Details : the performance loss is caused by using of write barriers in the
1666 ext3 code. The SLES10 kernel turns barrier support on by
1667 default. The fix is to undo that change for ldiskfs.
1669 ------------------------------------------------------------------------------
1671 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1673 * Support for kernels:
1674 2.6.9-42.0.3EL (RHEL 4)
1675 2.6.5-7.276 (SLES 9)
1676 2.4.21-47.0.1.EL (RHEL 3)
1677 2.6.12.6 vanilla (kernel.org)
1681 Frequency : quota enabled and large files being deleted
1683 Description: releasing more than 4GB of quota at once hangs OST
1684 Details : If a user deletes more than 4GB of files on a single OST it
1685 will cause the OST to spin in an infinite loop. Release
1686 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1691 Description: statfs data retrieved from /proc may be stale or zero
1692 Details : When reading per-device statfs data from /proc, in the
1693 {kbytes,files}_{total,free,avail} files, it may appear
1694 as zero or be out of date.
1697 Frequency : systems with MD RAID1 external journal devices
1699 Description: lconf's call to blkid is confused by RAID1 journal devices
1700 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1701 of returning all block devices that match the journal UUID.
1704 Frequency : always, for aggregate stripe size over 4GB
1706 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1707 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1708 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1711 Frequency : always if "lfs find" used on a local file/directory
1713 Description: "lfs find" segfaults if used on a local file/directory
1714 Details : The case where a directory component was not specified wasn't
1715 handled correctly. Handle this properly.
1718 Frequency : always on ppc64
1720 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1721 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1722 perform endianness conversion.
1725 Frequency : rarely (truncate to non-zero file size after write under load)
1726 Bugzilla : 10730, 10687
1727 Description: Files padded with zeros to next 4K multiple
1728 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1729 it is possible that files that are truncated to a non-zero size
1730 immediately after being written are filled with zero bytes beyond
1731 the truncated size. No file data is lost.
1733 Severity : enhancement
1735 Description: Allow recovery/failover for liblustre clients.
1736 Details : liblustre clients were unaware of failover configurations until
1739 Severity : enhancement
1741 Description: user file locks should fail when not mounting with flock option
1742 Details : Set up an error-returning stub in ll_file_operations.lock field
1743 to prevent incorrect behaviour when client is mounted without
1744 flock option. Also, set up properly f_op->flock field for
1748 Frequency : always on ia64
1750 Description: "lfs df" loops on printing out MDS statfs information
1751 Details : The obd_ioctl_data was not initialized and in some systems
1752 this caused a failure during the ioctl that did not return
1753 an error. Initialize the struct and return an error on failure.
1756 Frequency : SLES 9 only
1758 Description: Error of copying files with lustre special EAs as root
1759 Details : Client side always return success for setxattr call for lustre
1760 special xattr (currently only "trusted.lov").
1763 Frequency : rarely on clusters with both ia64+i386 clients
1765 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1766 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1767 clients are doing concurrent writes to the same file it is
1768 possible that the ia64 clients may LASSERT because the OST
1769 extent locks are not PAGE_SIZE aligned. Ensure that grown
1770 locks are always aligned on the request boundary.
1773 Frequency : specific use, occasional
1775 Description: Overwriting in use executable truncates on-disk binary image
1776 Details : If one node attempts to overwrite an executable in use by
1777 another node, we now correctly return ETXTBSY instead of
1778 truncating the file.
1780 Severity : enhancement
1782 Description: Async OSC create to avoid the blocking unnecessarily.
1783 Details : If a OST has no remain object, system will block on the creating
1784 when need to create a new object on this OST. Now, ways use
1785 pre-created objects when available, instead of blocking on an
1786 empty osc while others are not empty. If we must block, we block
1787 for the shortest possible period of time.
1792 Description: chmod on Lustre root is propagated to other clients
1793 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1794 invalid by the follow_mount time.
1797 Frequency : liblustre clients only
1799 Description: Race in 'instant cancel' lock handling could lead to such locks
1800 never to be granted in case of SMP MDS
1801 Details : Do not destroy not yet granted but cbpending locks in
1805 Frequency : replay/resend of open
1807 Description: non null lock assetion failure in mds_intent_policy
1808 Details : Trying to replay/resend lockless open requests resulted in
1809 mds_open() returning 0 with no lock. Now it sets a flag if
1810 a lock is going to be returned.
1812 Severity : enhancement
1814 Description: Checksum enhancements
1815 Details : New checksum enhancements allow for resending RPCs that failed
1818 Severity : enhancement
1820 Description: Tunables on number of dirty pages in cacche
1821 Details : Allow to set limit on number of dirty pages cached.
1826 Description: client crash on unmount - lock still has references
1827 Details : In some error handling cases it was possible to leak a lock
1828 reference on a client while accessing a file. This was not
1829 harmful to the client during operation, but would cause the
1830 client to crash when the filesystem is unmounted.
1833 Frequency : specific case, rare
1835 Description: ETXTBSY on mds though file not in use by client
1836 Details : ETXTBSY is no longer incorrectly returned when attempting to
1837 chmod or chown a directory that the user previously tried to
1838 execute or a currently-executing binary.
1841 Frequency : extremely rare except on liblustre-based clients
1843 Description: Lustre space not freed when files are deleted
1844 Details : Clean up open-unlinked files after client eviction. Previously
1845 the unlink was skipped and the files remained as orphans.
1850 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1851 Details : In some cases it was possible to send a blocking callback to
1852 a client doing a glimpse, even though that client didn't get
1853 a lock granted. When the glimpse lock is cancelled on the OST
1854 the freed lock is left on the waiting list and corrupted the list.
1857 Frequency : all core dumps
1859 Description: Broke core dumps to lustre
1860 Details : Negative dentry may be unhashed if parent does not have UPDATE
1861 lock, but some callers, e.g. do_coredump, expect dentry to be
1862 hashed after successful create, hash it in ll_create_it.
1864 ------------------------------------------------------------------------------
1866 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1868 * Support for kernels:
1869 2.6.9-42.0.2.EL (RHEL 4)
1870 2.6.5-7.276 (SLES 9)
1871 2.4.21-40.EL (RHEL 3)
1872 2.6.12.6 vanilla (kernel.org)
1876 Frequency : always on RHEL 3
1878 Description: Number of open files grows over time
1879 Details : The number of open files grows over time, whether or not
1880 Lustre is started. This was due to a filp leak introduced
1881 by one of our kernel patches.
1883 ------------------------------------------------------------------------------
1885 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1887 * Support for kernels:
1888 2.6.9-42.EL (RHEL 4)
1889 2.6.5-7.276 (SLES 9)
1890 2.4.21-40.EL (RHEL 3)
1891 2.6.12.6 vanilla (kernel.org)
1896 Bugzilla : 5719, 9635, 9792, 9684
1897 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1898 Details : If a server is under heavy load and cannot reply to new
1899 connection requests before the client resends the (re)connect,
1900 the connection handling code can behave badly if two service
1901 threads are concurrently handing separate (re)connections from
1902 the same client. Add better locking to the connection handling
1903 code, and ensure that only a single connection will be processed
1904 for a given client UUID, even if the lock is dropped.
1906 Severity : enhancement
1908 Description: add TCP zero-copy support to kernel
1909 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1910 sends if the hardware supports scatter-gather and checksumming.
1911 This allows socklnd to do client-write and server-read more
1912 efficiently and reduce CPU utilization from skbuf copying.
1915 Frequency : only if NFS exporting from client
1917 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1918 Details : If setting attributes on a file created under NFS that had
1919 never been opened it would be possible to oops the client
1920 if the file had no objects.
1923 Frequency : always for liblustre
1925 Description: liblustre client does MDS+OSTs setattr RPC for each write
1926 Details : When doing a write from a liblustre client, the client
1927 incorrectly issued an RPC to the MDS and each OST the file was
1928 striped over in order to update the timestamps. When writing
1929 with small chunks and many clients this could overwhelm the MDS
1930 with RPCs. In all cases it would slow down the write because
1931 these RPCs are unnecessary.
1933 Severity : enhancement
1935 Description: allow number of MDS service threads to be changed at module load
1936 Details : It is now possible to change the number of MDS service threads
1937 running. Adding "options mds mds_num_threads={N}" to the MDS's
1938 /etc/modprobe.conf will set the number of threads for the next
1939 time Lustre is restarted (assuming the "mds" module is also
1940 reloaded at that time). The default number of threads will
1941 stay the same, 32 for most systems.
1946 Description: OST crash if filesystem is unformatted or corrupt
1947 Details : If an OST is started on a device that has never been formatted
1948 or if the filesystem is corrupt and cannot even mount then the
1949 error handling cleanup routines would dereference a NULL pointer.
1954 Description: NULL pointer deref in llap_from_page.
1955 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1956 incorrect) mapping if the page was truncated/reclaimed while it was
1957 searched for. Check for this condition and skip such pages when
1958 doing readahead. Introduce extra check to llap_from_page() to
1959 verify page->mapping->host is non-NULL (so page is not anonymous).
1962 Frequency : Sometimes when using sys_sendfile
1964 Description: "page not covered by a lock" warnings from ll_readpage
1965 Details : sendfile called ll_readpage without right page locks present.
1966 Now we introduced ll_file_sendfile that does necessary locking
1967 around call to generic_file_sendfile() much like we do in
1971 Frequency : with certain MDS communication failures at client mount time
1973 Description: NULL pointer deref after failed client mount
1974 Details : a client connection request may delayed by the network layer
1975 and not be sent until after the PTLRPC layer has timed out the
1976 request. If the client fails the mount immediately it will try
1977 to clean up before the network times out the request. Add a
1978 reference from the request import to the obd device and delay
1979 the cleanup until the network drops the request.
1982 Frequency : occasionally during client (re)connect
1984 Description: assertion failure during client (re)connect
1985 Details : processing a client connection request may be delayed by the
1986 client or server longer than the client connect timeout. This
1987 causes the client to resend the connection request. If the
1988 original connection request is replied in this interval, the
1989 client may trip an assertion failure in ptlrpc_connect_interpret()
1990 which thought it would be the only running connect process.
1993 Frequency : only with obd_echo servers and clients that are rebooted
1995 Description: kernel BUG accessing uninitialized data structure
1996 Details : When running an obd_echo server it did not start the ping_evictor
1997 thread, and when a client was evicted an uninitialized data
1998 structure was accessed. Start the ping_evictor in the RPC
1999 service startup instead of the OBD startup.
2001 Severity : enhancement
2002 Bugzilla : 10193 (patchless)
2003 Description: Remove dependency on various unexported kernel interfaces.
2004 Details : No longer need reparent_to_init, exit_mm, exit_files,
2005 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2008 Frequency : rare (only users of deprecated and unsupported LDAP config)
2010 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2011 Details : LDAP apparently contains 'lustreName' attributes instead of
2012 'name'. A simple remapping of the name is sufficient.
2015 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2017 Description: waiting_locks_callback trips kernel BUG if client is evicted
2018 Details : Running with the dump_on_timeout debug flag turned on makes
2019 it possible that the waiting_locks_callback() can try to dump
2020 the Lustre kernel debug logs from an interrupt handler. Defer
2021 this log dumping to the expired_lock_main() thread.
2023 Severity : enhancement
2025 Description: Support NFS exporting on 2.6 kernels.
2026 Details : Implement non-rawops metadata methods for NFS server to use without
2027 changing NFS server code.
2030 Frequency : very rare (synthetic metadata workload only)
2032 Description: two racing renames might cause an MDS thread to deadlock
2033 Details : Running the "racer" program may cause one MDS thread to rename
2034 a file from being the source of a rename to being the target of
2035 a rename at exactly the same time that another thread is doing
2036 so, and the second thread has already enqueued these locks after
2037 doing a lookup of the target and is trying to relock them in
2038 order. Ensure that we don't try to re-lock the same resource.
2041 Frequency : only very large systems with liblustre clients
2043 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2044 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2045 all OSTs in parallel. This allows the eviction of stale liblustre
2046 clients to proceed much faster than if they were done in series,
2047 and also offers similar improvements for other set_info RPCs.
2052 Description: excessive CPU usage during initial read phase on client
2053 Details : During the initial read phase on a client, it would agressively
2054 retry readahead on the file, consuming too much CPU and impacting
2055 performance (since 1.4.5.8). Improve the readahead algorithm
2056 to avoid this, and also improve some other common cases (read
2057 of small files in particular, where "small" is files smaller than
2058 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2063 Description: MDS crash when receiving packet with unknown intent.
2064 Details : Do not LBUG in unknown intent case, just return -EFAULT
2066 Severity : enhancement
2067 Bugzilla : 9293, 9385
2068 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2069 Details : Do not serialize getattr (non-intent version) and statfs.
2072 Frequency : occasional, when OST network is overloaded/intermittent
2074 Description: client evicted by OST after bulk IO timeout
2075 Details : If a client sends a bulk IO request (read or write) the OST
2076 may evict the client if it is unresposive to its data GET/PUT
2077 request. This is incorrect if the network is overloaded (takes
2078 too long to transfer the RPC data) or dropped the OST GET/PUT
2079 request. There is no need to evict the client at all, since
2080 the pinger and/or lock callbacks will handle this, and the
2081 client can restart the bulk request.
2084 Frequency : Always when mmapping file with no objects
2086 Description: client crashes when mmapping file with no objects
2087 Details : Check that we actually have objects in a file before doing any
2088 operations on objects in ll_vm_open, ll_vm_close and
2094 Description: Request leak when working with deleted CWD
2095 Details : Introduce advanced request refcount tracking for requests
2096 referenced from lustre intent.
2098 Severity : Enhancement
2100 Description: Cache open file handles on client.
2101 Details : MDS now will return special lock along with openhandle, if
2102 requested and client is allowed to hold openhandle, even if unused,
2103 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2104 closing files for every read/write openration.
2106 Severity : Enhancement
2108 Description: Cache open negative dentries on client when possible.
2109 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2110 negative dentries on lock revocation.
2115 Description: Remounting a client read-only wasn't possible with a zconf mount
2116 Details : It wasn't possible to remount a client read-only with llmount.
2118 Severity : enhancement
2119 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2120 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2121 MPICH is included to add Lustre-specific ADIO interfaces.
2122 This is based closely on the UFS ADIO layer and only differs
2123 in file creation, in order to allow the OST striping to be set.
2124 This is user-contributed code and not supported by CFS.
2129 Description: extended inode attributes (immutable, append-only) work improperly
2130 when 2.4 and 2.6 kernels are used on client/server or vice versa
2131 Details : Introduce kernel-independent values for these flags.
2133 Severity : enhancement
2136 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2137 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2138 granularity. Now, Fractional MB granularity is be supported,
2139 this is very useful for low memory system.
2141 Severity : enhancement
2143 Description: Getattr by fid
2144 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2145 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2146 have locks and avoids getattr rpc requests for it on every stat.
2149 Frequency : Always, for filesystems larger than 2TB
2151 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2152 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2153 filesystems larger than 2TB to be mounted with mballoc enabled.
2156 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2158 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2159 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2160 filesystems larger than 2TB is mounted then the kernel will
2161 silently corrupt the start of the filesystem. CONFIG_LBD is
2162 enabled for all CFS-supported kernels, but the possibility of
2163 this happening with a modified kernel config exists.
2165 Severity : enhancement
2167 Description: add client O_DIRECT support for 2.6 kernels
2168 Details : It is now possible to do O_DIRECT reads and writes to files
2169 in the Lustre client mountpoint on 2.6 kernel clients.
2171 Severity : enhancement
2173 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2174 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2175 parallel, not waiting for response from every OST before sending
2176 a rpc to the next OST.
2181 Description: setattr vs write race when updating file timestamps
2182 Details : Client processes that update a file timestamp into the past
2183 right after writing to the file (e.g. tar) it is possible that
2184 the updated file modification time can be reset to the current
2185 time due to a race between processing the setattr and write RPC.
2187 Severity : enhancement
2189 Description: Bring 'lfs find' closer in line with regular Linux find.
2190 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2191 -print0 options and obtains all the needed info through the lustre
2194 Severity : enhancement
2196 Description: support up to 1024 configured devices on one node
2197 Details : change obd_dev array from statically allocated to dynamically
2198 allocated structs as they are first used to reduce memory usage
2203 Description: Flush dirty partially truncated pages during truncate
2204 Details : Immediatelly flush partially truncated pages in filter_setattr,
2205 this way we completely avoid having any pages in page cache on OST
2206 and can retire ugly workarounds during writes to flush such pages.
2211 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2212 Details : protect inode from truncation within vfs_unlink() context
2213 just take a reference before calling vfs_unlink() and release it
2214 when parent's i_sem is free.
2219 Description: last_id value checked outside lock on OST caused LASSERT failure
2220 Details : If there were multiple MDS->OST object precreate requests in
2221 flight, it was possible that the OST's last object id was checked
2222 outside a lock and incorrectly tripped an assertion. Move checks
2223 inside locks, and discard old precreate requests.
2226 Frequency : always, if extents are used on OSTs
2228 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2229 Details : OSTs using the extents format would not zero the high 16 bits of
2230 the index physical block number. This is not a problem for any
2231 OST filesystems smaller than 16TB, and no kernels support ext3
2232 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2233 new/modified files) and can be fixed for existing filesystems
2234 with e2fsprogs-1.39-cfs1.
2239 Description: import connection selection may be incorrect if timer wraps
2240 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2241 import connections to be ignored if the 32-bit jiffies counter
2242 wraps. Use a 64-bit jiffies counter.
2245 Frequency : very large clusters immediately after boot
2247 Description: LNET request buffers exhausted under heavy short-term load
2248 Details : If a large number of client requests are generated on a service
2249 that has previously never seen so many requests it is possible
2250 that the request buffer growth cannot keep up with the spike in
2251 demand. Instead of dropping incoming requests, they are held in
2252 the LND until the RPC service can accept more requests.
2255 Frequency : Sometimes during replay
2257 Description: Assertion failure in ll_local_open after replay.
2258 Details : If replay happened on an open request reply before we were able
2259 to set replay handler, reply will become not swabbed tripping the
2260 assertion in ll_local_open. Now we set the handler right after
2261 recognising of open request
2264 Frequency : very rare
2266 Description: kernel reports "badness in vsnprintf"
2267 Details : Reading from the "recovery_status" /proc file in small chunks
2268 may cause a negative length in lprocfs_obd_rd_recovery_status()
2269 call to vsnprintf() (which is otherwise harmless). Exit early
2270 if there is no more space in the output buffer.
2272 Severity : enhancement
2274 Description: clear OBD RPC statistics by writing to them
2275 Details : It is now possible to clear the OBD RPC statistics by writing
2276 to the "stats" file.
2281 Description: Client mtime is not the same on different clients after utimes
2282 Details : In some cases, the client was using the utimes() syscall on
2283 a file cached on another node. The clients now validate the
2284 ctime from the MDS + OSTs to determine which one is right.
2289 Description: Inability to activate failout mode
2290 Details : lconf script incorrectly assumed that in pythong string's numeric
2291 value is used in comparisons.
2294 Frequency : always with multiple stripes per file
2296 Description: Inefficient object allocation for mutli-stripe files
2297 Details : When selecting which OSTs to stripe files over, for files with
2298 a stripe count that divides evenly into the number of OSTs,
2299 the MDS is always picking the same starting OST for each file.
2300 Return the OST selection heuristic to the original design.
2305 Description: mount failures may take full timeout to return an error
2306 Details : Under some heavy load conditions it is possible that a
2307 failed mount can wait for the full obd_timeout interval,
2308 possibly several minutes, before reporting an error.
2309 Instead return an error as soon as the status is known.
2311 Frequency : quota enabled and large files being deleted
2313 Description: releasing more than 4GB of quota at once hangs OST
2314 Details : If a user deletes more than 4GB of files on a single OST it
2315 will cause the OST to spin in an infinite loop. Release
2316 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2321 Description: statfs data retrieved from /proc may be stale or zero
2322 Details : When reading per-device statfs data from /proc, in the
2323 {kbytes,files}_{total,free,avail} files, it may appear
2324 as zero or be out of date.
2327 Frequency : systems with MD RAID1 external journal devices
2329 Description: lconf's call to blkid is confused by RAID1 journal devices
2330 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2331 of returning all block devices that match the journal UUID.
2334 Frequency : always, for aggregate stripe size over 4GB
2336 Description: assertion fails when trying to use 4GB stripe size
2337 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2338 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2341 Frequency : always on ppc64
2343 Description: the first write on an ext3 filesystem with mballoc got stuck
2344 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2345 perform endianness conversion.
2347 ------------------------------------------------------------------------------
2349 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2351 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2352 INTEROPERATE with older versions automatically. Please read the
2353 user documentation before upgrading any part of a live system.
2354 * WARNING: Lustre networking configuration changes are required with
2355 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2358 * Support for kernels:
2359 2.6.9-22.0.2.EL (RHEL 4)
2360 2.6.5-7.244 (SLES 9)
2361 2.6.12.6 vanilla (kernel.org)
2364 Severity : enhancement
2365 Bugzilla : 7981/8208
2366 Description: Introduced Lustre Networking (LNET)
2367 Details : LNET is new networking infrastructure for Lustre, it includes
2368 a reorganized network configuration mode (see the user
2369 documentation for full details) as well as support for routing
2370 between different network fabrics. Lustre Networking Devices
2371 (LNDs) for the supported network fabrics have also been
2372 created for this new infrastructure.
2374 Severity : enhancement
2375 Description: Introduced Access control lists
2376 Details : clients can set ACLs on files and directories in order to have
2377 more fine-grained permissions than the standard Unix UGO+RWX.
2378 The MDS must be started with the "-o acl" mount option.
2380 Severity : enhancement
2381 Description: Introduced filesystem quotas
2382 Details : Administrators may now establish per-user quotas on the
2385 Severity : enhancement
2387 Description: Configuration change for the XT3
2388 The PTLLND is now used to run Lustre over Portals on the XT3
2389 The configure option(s) --with-cray-portals are no longer used.
2390 Rather --with-portals=<path-to-portals-includes> is used to
2391 enable building on the XT3. In addition to enable XT3 specific
2392 features the option --enable-cray-xt3 must be used.
2397 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2398 Details : A race between allocating a new llcd and re-getting the llcd_lock
2399 allowed another thread to grab newly-allocated llcd.
2401 Severity : enhancement
2403 Description: 2.6 OST async journal commit and locking fix to improve performance
2404 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2405 2.6 kernels are now async as they already were in 2.4 kernels so
2406 that they can commit concurrently with the network bulk transfer.
2407 For block-allocated files the filter allocation semaphore is held
2408 to avoid filesystem fragmentation during allocation. BKL lock
2409 removed for 2.6 xattr operations where it is no longer needed.
2414 Description: lconf incorrectly determined whether two IP networks could talk
2415 Details : In some more complicated routing and multiple-network
2416 configurations, lconf will avoid trying to make a network
2417 connection to a disjoint part of the IP space. It was doing the
2418 math incorrectly for one set of cases.
2423 Description: Fix for potential infinite loop processing records in an llog.
2424 Details : If an llog record is corrupted/zeroed, it is possible to loop
2425 forever in llog_process(). Validate the llog record length
2426 and skip the remainder of the block on error.
2429 Frequency : occasional (liblustre only)
2431 Description: liblustre could not open files whose last component is a symlink
2432 Details : sysio_path_walk() would incorrectly pass the open intent to
2433 intermediate path components.
2436 Frequency : rare (liblustre only with non-standard tuning)
2437 Bugzilla : 7201 (7350)
2438 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2439 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2440 after it has been released, only do this for referenced locks
2442 Severity : enhancement
2444 Description: specifying an (invalid) directory default stripe_size of -1
2445 would reset the directory default striping
2446 Details : stripe_size -1 was used internally to signal directory stripe
2447 removal, now use "all default" to signal dir stripe removal
2448 as a directory striping of "all default" is not useful
2451 Frequency : common for large clusters running liblustre clients
2453 Description: doing an ls when liblustre clients are running is slow
2454 Details : sending a glimpse AST to a liblustre client waits for every AST
2455 to time out, as liblustre clients will not respond. Since they
2456 cannot cache data we refresh the OST lock LVB from disk instead.
2458 Severity : enhancement
2460 Description: doing an ls at the same time as file IO can be slow
2461 Details : enqueue and other "small" requests can be blocked behind many
2462 large IO requests. Create a new OST IO portal for non-IO
2463 requests so they can be processed faster.
2466 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2468 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2469 created file with mode 000
2470 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2471 allow an ftruncate() as a non-root user to a file with mode 000.
2472 Lustre now respects this flag to disable mode checks when
2473 truncating a file owned by the user
2476 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2479 Description: Revoking locks from clients that went dead or catatonic might take
2481 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2482 cancellation of such locks instant on servers without waiting for
2483 any reply from clients. Clients drops these locks when cancel
2484 notification from server is received without replying.
2487 Frequency : liblustre-only, when liblustre client dies or becomes busy
2489 Description: Doing ls on Linux clients can take a long time with active
2491 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2492 granting such locks to it in the first place if possible. Locks
2493 are taken by proxy on the OST during the read or write and
2494 dropped immediately afterward. Add connect flags handling, do
2495 not grant locks to liblustre clients for glimpse ASTs.
2497 Severity : enhancement
2499 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2500 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2501 sequential reads and behaves badly with applications doing random
2502 reads. Improve it by reading ahead at least read region, and
2503 avoiding excessive large RPC for small reads.
2505 Severity : enhancement
2507 Description: Creating more than 1000 files for a single job may cause a load
2508 imbalance on the OSTs if there are also a large number of OSTs.
2509 Details : qos_prep_create() uses an OST index reseed value that is an
2510 even multiple of the number of available OSTs so that if the
2511 reseed happens in the middle of the object allocation it will
2512 still utilize the OSTs as uniformly as possible.
2517 Description: OST or MDS may oops in ping_evictor_main()
2518 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2519 but doesn't restart at beginning of obd_exports_timed list
2522 Severity : enhancement
2524 Description: improve by-nid export eviction on the MDS and OST
2525 Details : allow multiple exports with the same NID to be evicted at one
2526 time without re-searching the exports list.
2529 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2531 Description: MDS may oops in groups_free()
2532 Details : in rare race conditions a newly allocated group_info struct is
2533 freed again, and this can be NULL. The 2.4 compatibility code
2534 for groups_free() checked for a NULL pointer, but 2.6 did not.
2537 Frequency : common for liblustre clients doing little filesystem IO
2538 Bugzilla : 9352, 7313
2539 Description: server may evict liblustre clients accessing contended locks
2540 Details : if a client is granted a lock or receives a completion AST
2541 with a blocking AST already set it would not reply to the AST
2542 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2545 Frequency : lfs setstripe, only systems with more than 160 OSTs
2547 Description: unable to set striping with a starting offset beyond OST 160
2548 Details : llapi_create_file() incorrectly limited the starting stripe
2549 index to the maximum single-file stripe count.
2552 Frequency : LDAP users only
2554 Description: lconf did not handle in-kernel recovery with LDAP properly
2555 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2557 Severity : enhancement
2559 Description: bind OST threads to NUMA nodes to improve performance
2560 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2561 node and do their allocations there to localize memory access
2563 Severity : enhancement
2565 Description: llmount can determine client NID directly from Myrinet (GM)
2566 Details : the client NID code from gmnalnid was moved directly into
2567 llmount, removing the need to use this or specifying the
2568 client NID explicitly when mounting GM clients with zeroconf
2571 Frequency : if client is started with down MDS
2573 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2574 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2575 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2576 ptlrpc_import_delayed_req() because we didn't check if the
2577 request was interrupted, and we also didn't break out of the
2578 event loop if there was no timeout
2583 Description: data loss during non-page-aligned writes to a single file from
2584 both multiple nodes and multiple threads on one node at same time
2585 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2586 inconsistency led to false short-reads, that were cached and later
2587 used by ->prepare_write() to fill in partially written page,
2588 leading to data loss.
2591 Frequency : always, if lconf --abort_recovery used
2593 Description: lconf --abort_recovery fails with 'Operation not supported'
2594 Details : lconf was attempting to abort recovery on the MDT device and not
2597 Severity : enhancement
2599 Description: remove cleanup logs
2600 Details : replace lconf-generated cleanup logs with lustre internal
2601 cleanup routines. Eliminates the need for client-cleanup and
2604 Severity : enhancement
2606 Description: add support for EAs (user and system) on lustre filesystems
2607 Details : it is now possible to store extended attributes in the Lustre
2608 client filesystem, and with the user_xattr mount option it
2609 is possible to allow users to store EAs on their files also
2611 Severity : enhancement
2613 Description: Add possibility (config option) to show minimal available OST free
2615 Details : When compiled with --enable-mindf configure option, statfs(2)
2616 (and so, df) will return least minimal free space available from
2617 all OSTs as amount of free space on FS, instead of summary of
2618 free spaces of all OSTs.
2620 Severity : enhancement
2622 Description: do not expand extent locks acquired on OST-side
2623 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2624 by server: they are not cached anyway.
2627 Frequency : when mmap is used/binaries executed from Lustre
2629 Description: Unmmap pages before throwing them away from read cache.
2630 Details : llap_shrink cache now attempts to unmap pages before discarding
2631 them (if unmapping failed - do not discard). SLES9 kernel has
2632 extra checks that trigger if this unmapping is not done first.
2637 Description: lconf didn't resolve symlinks before checking to see whether a
2638 given mountpoint was already in use
2641 Frequency : when migrating failover services
2642 Bugzilla : 6395, 9514
2643 Description: When migrating a subset of services from a node (e.g. failback
2644 from a failover service node) the remaining services would
2645 time out and evict clients.
2646 Details : lconf --force (implied by --failover) sets the global obd_timeout
2647 to 5 seconds in order to quickly disconnect, but this caused
2648 other RPCs to time out too quickly. Do not change the global
2649 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2651 Severity : enhancement
2652 Frequency : if MDS is started with down OST
2653 Bugzilla : 9439,5706
2654 Description: Allow startup/shutdown of an MDS without depending on the
2655 availability of the OSTs.
2656 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2657 Add appropriate locking and lov-osc refcounts for safe
2658 cleaning. Add osc abort_inflight calls in case the
2659 synchronize never started.
2662 Frequency : occasional (Cray XT3 only)
2664 Description: root not authorized to access files in CRAY_PORTALS environment
2665 Details : The client process capabilities were not honoured on the MDS in
2666 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2667 been accessed by an authorized user then root was able to access
2668 the file on the local client also. The root user capabilities
2669 are now allowed on the MDS, as this environment has secure UID.
2672 Frequency : occasional
2674 Description: ldiskfs "too long searching" message happens too often
2675 Details : A debugging message (otherwise harmless) prints too often on
2676 the OST console. This has been reduced to only happen when
2677 there are fragmentation problems on the filesystem.
2682 Description: Division by zero in statfs when all OSCs are inactive
2683 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2684 lov_statfs divides by value returned by lov_get_stripecnt().
2688 Bugzilla : 9489, 3273
2689 Description: First write from each client to each OST was only 4kB in size,
2690 to initialize client writeback cache, which caused sub-optimal
2691 RPCs and poor layout on disk for the first writen file.
2692 Details : Clients now request an initial cache grant at (re)connect time
2693 and so that they can start streaming writes to the cache right
2694 away and always do full-sized RPCs if there is enough data.
2695 If the OST is rebooted the client also re-establishes its grant
2696 so that client cached writes will be honoured under the grant.
2701 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2702 Details : Now I/O RPCs go to different portal number and (presumably) fast
2703 lock requests (and glimses) and other RPCs get their own service
2704 threads pool that should be able to service those RPCs
2707 Severity : enhancement
2709 Description: Ability to exchange lustre version between client and servers and
2710 issue warnings at client side if client is too old. Also for
2711 liblustre clients there is ability to refuse connection of too old
2713 Details : New 'version' field is added to connect data structure that is
2714 filled with version info. That info is later checked by server and
2718 Frequency : rare, liblustre only.
2719 Bugzilla : 9296, 9581
2720 Description: Two simultaneous writes from liblustre at offset within same page
2721 might proceed at the same time overwriting eachother with stale
2723 Details : I/O lock withing llu_file_prwv was released too early, before data
2724 actually was hitting the wire. Extended lock-holding time until
2725 server acknowledges receiving data.
2728 Frequency : extremely rare. Never observed in practice.
2730 Description: avoid generating lustre_handle cookie of 0.
2731 Details : class_handle_hash() generates handle cookies by incrementing
2732 global counter, and can hit 0 occasionaly (this is unlikely, but
2733 not impossible, because initial value of cookie counter is
2734 selected randonly). Value of 0 is used as a sentinel meaning
2735 "unassigned handle" --- avoid it. Also coalesce two critical
2736 sections in this function into one.
2738 Severity : enhancement
2740 Description: allow liblustre clients to delegate truncate locking to OST
2741 Details : To avoid overhead of locking, liblustre client instructs OST to
2742 take extent lock in ost_punch() on client's behalf. New connection
2743 flag is added to handle backward compatibility.
2745 Severity : enhancement
2746 Bugzilla : 4928, 7341, 9758
2747 Description: allow number of OST service threads to be specified
2748 Details : a module parameter allows the number of OST service threads
2749 to be specified via "options ost ost_num_threads={N}" in the
2750 OSS's /etc/modules.conf or /etc/modprobe.conf.
2754 Bugzilla : 6146, 9635, 9895
2755 Description: servers crash with bad pointer in target_handle_connect()
2756 Details : In rare cases when a client is reconnecting it was possible that
2757 the connection request was the last reference for that export.
2758 We would temporarily drop the export reference and get a new
2759 one, but this may have been the last reference and the export
2760 was just destroyed. Get new reference before dropping old one.
2762 Severity : enhancement
2763 Frequency : if client is started with failover MDS
2765 Description: Allow multiple MDS hostnames in the mount command
2766 Details : Try to read the configuration from all specified MDS
2767 hostnames during a client mount in case the "primary"
2770 Severity : enhancement
2772 Description: Stop sending data to evicted clients as soon as possible.
2773 Details : Check if the client we are about to send or are sending data to
2774 was evicted already. (Check is done every second of waiting,
2775 for which l_wait_event interface was extended to allow checking
2776 of exit condition at specified intervals).
2779 Frequency : rare, normally only when NFS exporting is done from client
2781 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2783 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2786 Frequency : always, if srand() is called before liblustre initialization
2788 Description: Liblustre uses system PRNG disturbing its usage by user application
2789 Details : Introduce internal to lustre fast and high-quality PRNG for
2790 lustre usage and make liblustre and some other places in generic
2791 lustre code to use it.
2793 Severity : enhancement
2794 Bugzilla : 9477, 9557, 9870
2795 Description: Verify that the MDS configuration logs are updated when xml is
2796 Details : Check if the .xml configuration logs are newer than the config
2797 logs stored on the MDS and report an error if this is the case.
2798 Request --write-conf, or allow starting with --old_conf.
2800 Severity : enhancement
2802 Description: Handle symlinks in the path when checking if Lustre is mounted.
2803 Details : Resolve intermediate symlinks when checking if a client has
2804 mounted a filesystem to avoid duplicate client mounts.
2809 Description: lconf can hit an error exception but still return success.
2810 Details : The lconf command catches the Command error exception at the top
2811 level script context and will exit with the associated exit
2812 status, but doesn't ensure that this exit status is non-zero.
2817 Description: failure of ptlrpc thread startup can cause oops
2818 Details : Starting a ptlrpc service thread can fail if there are a large
2819 number of threads or the server memory is very fragmented.
2820 Handle this without oopsing.
2823 Frequency : always, only if liblustre and non-default acceptor port was used
2825 Description: liblustre cannot connect to servers with non-default acceptor port
2826 Details : tcpnal_set_default_params() was not called and was therefore
2827 ignoring the environment varaible TCPNAL_PORT, as well as other
2828 TCPNAL_ environment variables
2833 Description: two objects could be created on the same OST for a single file
2834 Details : If an OST is down, in some cases it was possible to create two
2835 objects on a single OST for a single file. No problems other
2836 than potential performance impact and spurious error messages.
2840 Bugzilla : 5681, 9562
2841 Description: Client may oops in ll_unhash_aliases
2842 Details : Client dcache may become inconsistent in race condition.
2843 In some cases "getcwd" can fail if the current directory is
2849 Description: Inode refcounting problems in NFS export code
2850 Details : link_raw functions used to call d_instantiate without obtaining
2851 extra inode reference first.
2855 Bugzilla : 9942, 9903
2856 Description: Referencing freed requests leading to crash, memleaks with NFS.
2857 Details : We used to require that call to ll_revalidate_it was always
2858 followed by ll_lookup_it. Also with revalidate_special() it is
2859 possible to call ll_revalidate_it() twice for the same dentry
2860 even if first occurence returned success. This fix changes semantic
2861 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2862 reference on a request referred from the intent.
2863 ll_intent_release() then releases such a request.
2866 Frequency : rare, normally benchmark loads only
2868 Description: unlinked inodes were kept in memory on the client
2869 Details : If a client is repeatedly creating and unlinking files it
2870 can accumulate a lot of stale inodes in the inode slab cache.
2871 If there is no other client load running this can cause the
2872 client node to run out of memory. Instead flush old inodes
2873 from client cache that have the same inode number as a new inode.
2876 Frequency : SLES9 2.6.5 kernel and long filenames only
2877 Bugzilla : 9969, 10379
2878 Description: utime reports stale NFS file handle
2879 Details : SLES9 uses out-of-dentry names in some cases, which confused
2880 the lustre dentry revalidation. Change it to always use the
2884 Frequency : rare, unless heavy write-truncate concurrency is continuous
2885 Bugzilla : 4180, 6984, 7171, 9963, 9331
2886 Description: OST becomes very slow and/or deadlocked during object unlink
2887 Details : filter_destroy() was holding onto the parent directory lock
2888 while truncating+unlinking objects. For very large objects this
2889 may block other threads for a long time and slow overall OST
2890 responsiveness. It may also be possible to get a lock ordering
2891 deadlock in this case, or run out of journal credits because of
2892 the combined truncate+unlink. Solution is to do object truncate
2893 first in one transaction without parent lock, and then do the
2894 final unlink in a new transaction with the parent lock. This
2895 reduces the lock hold time dramatically.
2898 Frequency : rare, 2.4 kernels only
2900 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2901 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2902 before unmounting their filesystem in order to avoid deadlock.
2903 The kernel_locked() function in 2.4 kernels only checks whether
2904 the kernel lock is held, not whether it is this process that is
2905 holding it as 2.6 kernels do.
2910 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2911 Details : The client ptlrpc code may be trying to reconnect to a down
2912 server before a previous connection attempt has timed out.
2913 Increase the reconnect interval to be longer than the connection
2914 timeout interval to avoid sending duplicate connections to
2918 Frequency : echo_client brw_test command
2920 Description: fix echo_client to work with OST preallocated code
2921 Details : OST preallocation code (5137) didn't take echo_client IO path
2922 into account: echo_client calls filter methods outside of any
2923 OST thread and, hence, there is no per-thread preallocated
2924 pages and buffers to use. Solution: hijack pga pages for IO. As
2925 a byproduct, this avoids unnecessary data copying.
2929 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2930 Description: Client can oops in mdc_commit_close() after open replay
2931 Details : It was possible for the MDS to return an open request with no
2932 transaction number in mds_finish_transno() if the client was
2933 evicted, but without actually returning an error. Clients
2934 would later try to replay that open and may trip an assertion
2935 Simplify the client close codepath, and always return an error
2936 from the MDS in case the open is not successful.
2939 Frequency : rare, 2.6 OSTs only
2941 Description: OST may deadlock under high load on fragmented files
2942 Details : If there was a heavy load and highly-fragmented OST filesystems
2943 it was possible to have all the OST threads deadlock waiting on
2944 allocation of biovecs, because the biovecs were not released
2945 until the entire RPC IO was completed. Instead, release biovecs
2946 as soon as they are complete to ensure forward IO progress.
2948 Severity : enhancement
2950 Description: Support for specifying external journal device at mount
2951 Details : If an OST or MDS device is formatted with an external journal
2952 device, this device major/minor is stored in the ext3 superblock
2953 and may not be valid for failover. Allow detecting and
2954 specifying the external journal at mount time.
2959 Description: Mounting an MDS with pending unlinked files may cause oops
2960 Details : target_finish_recovery() calls mds_postrecov() which returned
2961 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2962 considers this an error and immediately begins cleaning up the
2963 lov, just after starting the mds_lov process
2965 Severity : enhancement
2967 Description: Implement 'lfs df' to report actual free space on per-OST basis
2968 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2969 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2970 usage of objects; '-h' to report in human readable format.
2972 ------------------------------------------------------------------------------
2974 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2981 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2982 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2983 at startup during orphan unlinking. Instead, skip buddy update
2984 before setup, as it will be regenerated after recovery is complete.
2989 Description: If an OST is inactive, its locks might reference stale inodes.
2990 Details : lov_change_cbdata() must iterate over all namespaces, even if
2991 they are inactive to clear inode references from the lock.
2993 Severity : enhancement
2994 Frequency : occasional, if non-standard max_dirty_mb used
2996 Description: Client will block write RPCs if not enough grant
2997 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2998 then the client will block writes while waiting for another RPC
2999 to complete instead of consuming its dirty limit. With change
3000 we get improved performance when max_dirty_mb is small.
3002 Severity : enhancement
3003 Bugzilla : 3389, 6253
3004 Description: Add support for supplementary groups on the MDS.
3005 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3006 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3007 lookups for user supplementary groups into a cache.
3011 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3012 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3013 for liblustre O_EXCL is enforced by MDS.
3017 Description: readdir never returns NULL in liblustre.
3018 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3019 dentry in struct dirent.
3023 Frequency : liblustre only, depends on application IO pattern
3024 Description: liblustre clients evicted if not contacting servers
3025 Details : Don't put liblustre clients into the ping_evictor list, so
3026 they will not be evicted by the pinger ever.
3028 Severity : enhancement
3030 Description: Add ability to evict clients by NID from MDS.
3031 Details : By echoing "nid:$NID" string into
3032 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3033 $NID would be instantly evicted from this MDS and from all active
3034 OSTs connected to it.
3038 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3039 Details : lookup_it_finish() used to query file size from OSTs that was not
3044 Description: service threads change working directory to that of init
3045 Details : Starting lustre service threads may pin the working directory
3046 of the parent thread, making that filesystem busy. Threads
3047 now change to the working directory of init to avoid this.
3051 Frequency : during shutdown only
3052 Description: shutdown with a failed MDS or OST can cause unmount to hang
3053 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3058 Frequency : 2.6 only
3059 Description: chmod/chown may include an extra supplementary group
3060 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3061 supplementary group and if none is specified this is used.
3064 Bugzilla : 5479 (6816)
3066 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3067 Details : If lookup is in progress on a file that is unlinked we might try
3068 to revalidate the inode and fail in revalidate after lookup is
3069 complete and ll_file_open() enqueues the open again but
3070 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3073 Frequency : always, if lconf --abort_recovery used
3075 Description: lconf --abort_recovery fails with 'Operation not supported'
3076 Details : lconf was attempting to abort recovery on the MDT device and not
3079 ------------------------------------------------------------------------------
3081 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3086 Frequency : rare (only unsupported configurations with a node running as an
3088 Bugzilla : 6514, 5137
3089 Description: Mounting a Lustre file system on a node running as an OST could
3091 Details : OSTs now preallocates memory needed to write out data at
3092 startup, instead of when needed, to avoid having to
3093 allocate memory in possibly low memory situations.
3094 Specifically, if the file system is mounted on on OST,
3095 memory pressure could force it to try to write out data,
3096 which it needed to allocate memory to do. Due to the low
3097 memory, it would be unable to do so and the node would
3098 become unresponsive.
3100 Severity : enhancement
3102 Description: Addition of lconf --service command line option
3103 Details : lconf now accepts a '--service <arg>' option, which is
3104 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3106 Severity : enhancement
3108 Description: Failover mode is now the default for OSTs.
3109 Details : By default, OSTs will now run in failover mode. To return to
3110 the old behaviour, add '--failout' to the lmc line for OSTs.
3112 Severity : enhancement
3114 Description: Health checks are now provided for MDS and OSTs
3115 Details : Additional detailed health check information on MSD and OSTs
3116 is now provided through the procfs health_check value.
3119 Frequency : occasional, depends on IO load
3121 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3122 after numerous create/delete cycles
3123 Details : The ext3 inode allocation policy would not allocate new inodes
3124 very well on the OSTs because there are no new directories
3125 being created. Instead we look for groups with free space if
3126 the parent directories are nearly full.
3131 Description: Network or server problems during mount may cause partially
3132 mounted clients instead of returning an error.
3133 Details : The config llog parsing code may overwrite the error return
3134 code during mount error handling, returning success instead
3140 Description: MDS can fail to allocate large reply buffers
3141 Details : After long uptimes the MDS can fail to allocate large reply
3142 buffers (e.g. zconf client mount config records) due to memory
3143 fragmentation or consumption by the buffer cache. Preallocate
3144 some large reply buffers so that these replies can be sent even
3145 under memory pressure.
3149 Frequency : rare (liblustre)
3150 Description: fsx running with liblustre complained that using truncate() to
3151 extend the file doesn't work. This patch corrects that issue.
3152 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3153 ATTR_SIZE and lsm use in llu_setattr_raw.
3157 Frequency : rare, only 2.6 kernels
3158 Description: Unusual file access patterns on the MDS may result in inode
3159 data being lost in very rare circumstances.
3160 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3161 code in the 2.6 kernel caused the inode and/or EA data not to
3162 be read from disk, causing single-file corruption.
3166 Frequency : rare, only 2.6 filesystems using extents
3167 Description: Heavy concurrent write and delete load may cause data corruption.
3168 Details : It was possible under high-load situations to have an extent
3169 metadata block in the block device cache from a just-unlinked
3170 file overwrite a newly-allocated data block. We now unmap any
3171 metadata buffers that alias just-allocated data blocks.
3175 Frequency : filesystems with default stripe_count larger than 77
3176 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3177 Details : lconf specifies an inode size of 4096 bytes when the default
3178 stripe_count is larger than 77. This conflicts with the default
3179 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3180 this case to avoid pinning too much memory for large EAs.
3182 ------------------------------------------------------------------------------
3184 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3189 Frequency : rare (extremely heavy IO load with hundreds of clients)
3191 Description: Client is evicted, gets IO error writing to file
3192 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3193 caused clients to be evicted for AST timeouts. The fixes in
3194 bug 5192 mean we no longer need to have such short AST timeouts
3195 so ldlm_timeout has been increased.
3198 Frequency : occasional during --force or --failover shutdown under load
3199 Bugzilla : 5949, 4834
3200 Description: Server oops/LBUG if stopped with --force or --failover under load
3201 Details : a collection of import/export refcount and cleanup ordering
3202 issues fixed for safer force cleanup
3205 Frequency : only filesystems larger than 120 OSTs
3206 Bugzilla : 5990, 6223
3207 Description: lfs getstripe would oops on a very large filesystem
3208 Details : lov_getconfig used kfree on vmalloc'd memory
3211 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3212 Bugzilla : 6242, 6243
3213 Description: reading from files that had been truncated to a non-zero size
3214 but never opened returned no data
3215 Details : ll_file_read() reads zeros from no-object files to EOF
3220 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3221 Details : There's little harm in aborting MDS/OSS recovery and letting it
3222 try again, so I removed the LASSERT and return an error instead.
3224 Severity : enhancement
3226 Description: New debugging infrastructure for tracking down data corruption
3227 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3228 (b) cover more of the client-side code path, and (c) try to narrow
3229 down where problems occurred
3233 Bugzilla : 3819, 4364, 4397, 6313
3234 Description: Racing close and eviction MDS could cause assertion in mds_close
3235 Details : It was possible to get multiple mfd references during close and
3236 client eviction, leading to one thread referencing a freed mfd.
3238 Severity: : enhancement
3239 Bugzilla : 3262, 6359
3240 Description: Attempts to reconnect to servers are now more aggressive.
3241 Details : This builds on the enhanced upcall-less recovery that was added
3242 in 1.4.2. When trying to reconnect to servers, clients will
3243 now try each server in the failover group every 10 seconds. By
3244 default, clients would previously try one server every 25 seconds.
3249 Description: After recovery, certain operations trigger a failed
3250 assertion on a client.
3251 Details : Failing over an mds, using lconf -d --failover, while a
3252 client was doing a readdir() call would cause the client to
3253 LBUG after recovery completed and the readdir() was resent.
3255 Severity : enhancement
3257 Description: Default groups are now added by lconf
3258 Details : You can now run lconf --group <servicename> without having to
3259 manually add groups with lmc.
3262 Frequency : occasional
3264 Description: Nodes with an elan id of 0 trigger a failed assertion
3267 Frequency : always when accessing e.g. tty/console device nodes
3269 Description: tty and some other devices nodes cannot be used on lustre
3270 Details : file's private_data field is used by device data and lustre
3271 values in there got lost. New field was added to struct file to
3272 store fs-specific private data.
3275 Frequency : when exporting Lustre via NFS
3277 Description: NFSD failed occasionally when looking up a path component
3278 Details : NFSD is looking up ".." which was broken in ext3 directories
3279 that had grown large enough to become hashed.
3282 Frequency : Clusters with multiple interfaces not on the same subnet
3284 Description: Nodes will repeatedly try to reconnect to an interface which it
3285 cannot reach and report an error to the log.
3286 Details : Extra peer list entries will be created by lconf with some peers
3287 unreachable. lconf now validates the peer before adding it.
3290 Frequency : Only if a default stripe is set on the filesystem root.
3292 Description: Setting a default stripe on the filesystem root prevented the
3293 filesystem from being remounted.
3294 Details : The client was sending extra request flags in the root getattr
3295 request and did not allocate a reply buffer for the dir EA.
3298 Frequency : occasional, higher if lots of files are accessed by one client
3299 Bugzilla : 6159, 6097
3300 Description: Client trips assertion regarding lsm mismatch/magic
3301 Details : While revalidating inodes the VFS looks up inodes with ifind()
3302 and in rare cases can find an inode that is being freed.
3303 The ll_test_inode() code will free the lsm during ifind()
3304 when it finds an existing inode and then the VFS later attaches
3305 this free lsm to a new inode.
3309 Bugzilla : 6422, 7030
3310 Description: MDS deadlock between mkdir and client eviction
3311 Details : Creating a new file via mkdir or mknod (starting a transaction
3312 and getting the ns lock) can deadlock with client eviction
3313 (gets ns lock and trying to finish a synchronous transaction).
3316 Frequency : occasional
3317 Description: While starting a server, the fsfilt_ext3 module could not be
3319 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3320 kernels. Previously, lconf would still use the ext3 name
3321 when trying to load modules. Now, it will correctly use
3322 ext3 on 2.4 and ldiskfs on 2.6.
3324 Severity : enhancement
3325 Description: The default stripe count has been changed to 1
3326 Details : The interpretation of the default stripe count (0, to lfs
3327 or lmc) has been changed to mean striping across a single
3328 OST, rather than all available. For general usage we have
3329 found a stripe count of 1 or 2 works best.
3331 Severity : enhancement
3332 Description: Add support for compiling against Cray portals.
3333 Details : Conditional compiling for some areas that are different
3337 Frequency : occasional
3338 Bugzilla : 6409, 6834
3339 Description: Creating files with an explicit stripe count may lead to
3340 a failed assertion on the MDS
3341 Details : If some OSTs are full or unavailable, creating files may
3342 trigger a failed assertion on the MDS. Now, Lustre will
3343 try to use other servers or return an error to the
3347 Frequency : occasional
3349 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3350 caused lock timeout during readahead (since 1.4.2).
3351 Details : Processes doing readahead might match a lock that hasn't been
3352 granted yet if there are overlapping and conflicting lock
3353 requests. The readahead process waits on ungranted lock
3354 (original lock is CBPENDING), while OST waits for that process
3355 to cancel CBPENDING read lock and eventually evicts client.
3357 Severity : enhancement
3359 Description: Initial enabling of flock support for clients
3360 Details : Implements fcntl advisory locking and file status functions.
3361 This feature is provided as an optional mount flag (default
3362 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3363 locking are implemented yet, and those that are are not guaranteed
3364 to be completely correct in production environments.
3365 mount -t lustre -o [flock|noflock] ...
3368 Frequency : occasional
3370 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3371 assertion in the ext3 JBD (journaling) layer.
3372 Details : The b_committed_data struct is protected by the big kernel lock
3373 in 2.4 kernels, serializing journal_commit_transaction() and
3374 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3375 access to this struct. In 2.6 kernels there is finer grained
3376 locking to improve SMP performance of the JBD layer.
3380 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3385 Description: merge and cleanup kernel patches.
3388 -----------------------------------------------------------------------------
3390 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3392 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3393 versions of Lustre. Please update all servers and clients to
3394 version 1.4.2 or later at the same time. You must also run
3395 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3396 to update the configuration logs.
3398 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3399 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3400 - swab llog records of type '0' so we get proper header size/idx (5861)
3401 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3402 - fix rename of one directory over another leaking an inode (5953)
3403 - avoid SetPageDirty on 2.6 (5981)
3404 - don't re-add just-being-destroyed locks to the waiting list (5653)
3405 - when creating new directories, inherit the parent's custom
3406 striping settings if present parent (3048)
3407 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3408 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3409 - don't hold client locks on temporary worklist from l_lru (5666)
3410 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3411 - automatically evict dead clients (5921)
3412 - Update file size properly in create+truncate+fstat case (6196)
3413 - Do not unhash mountpoint dentries, do not allow removal of
3415 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3416 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3417 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3418 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3419 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3420 - Correctly handle reads of files with no objects (6243)
3421 - lctl recover will also mark a device active if deactivate used (5933)
3423 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3424 - allow --write-conf on an MDS with different nettype than client (5619)
3425 - don't write config llogs to MDS for mounts not from that MDS (5617)
3426 - lconf should create multiple TCP connections from a client (5201)
3427 - init scripts are now turned off by default; run chkconfig --on
3428 lustre and chkconfig --on lustrefs to use them
3429 - upcalls are no longer needed for clients to recover to failover
3431 - add --abort-recovery option to lconf to abort recovery on device
3433 - add support for an arbitrary number of OSTs (3026)
3434 - Quota support protocol changes.
3435 - forward compatibility changes to wire structs (6007)
3436 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3437 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3438 - improved "lustre status" script
3439 - initialize blocksize for non-regular files (6062)
3440 - added --disable-server and --disable-client configure options (5782)
3441 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3442 - Vanilla 2.4.29 support
3443 - increase maximum number of obd devices to 520 (6242)
3444 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3445 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3446 SLES 9 2.6.5 kernels
3447 - limit stripes per file to 160 (the maximum EA size) (6093)
3449 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3452 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3453 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3454 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3455 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3456 - small ext3 extents cleanups and fixes (5733)
3457 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3458 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3459 - avoid needless client->OST connect, fix handle mismatch (5317)
3460 - fix DLM error path that led to out-of-sync client, long delays (5779)
3461 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3462 - fix several locking issues related to i_size (5492,5624,5654,5672)
3463 - don't move pending lock onto export if it is already evicted (5683)
3464 - fix kernel oops when creating .foo in unlinked directory (5548)
3465 - fix deadlock in obdfilter statistics vs. object create (5811)
3466 - use time_{before,after} to avoid timer jiffies wrap (5882)
3467 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3468 - Do not leak request if server was not able to process it (5154)
3469 - If mds_open unable to find parent dir, make that negative lookup(5154)
3470 - don't create new directories with extent-mapping (5909, 5936)
3472 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3473 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3474 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3475 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3476 - support for RHEL4 kernel on i686 (5773)
3477 - provide error messages when incompatible logs are encountered (5898)
3479 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3480 * version 1.4.0.10 (1.4.1 release candidate 1)
3482 - don't keep a lock reference when lock is not granted (4238)
3483 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3484 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3485 - drop import inflight refcount on signal_completed_replay error (5255)
3486 - unlock page after async write error during send (3677)
3487 - handle missing objects in filter_preprw_read properly (5265)
3488 - no transno return for symlink open, don't save no-trasno open (3440)
3489 - don't try to complete elan receive that already failed (4012)
3490 - free RPC server reply state on error (5406)
3491 - clean up thread from ptlrpc_start_thread() on error (5160)
3492 - readahead could read extra page into cache that wasn't ejected (5388)
3493 - prevent races in class_attach/setup/cleanup/detach (5260)
3494 - don't dereference de->d_inode after l_dput of de (5458)
3495 - use "int" for stripe value returned from lock_to_stripe (5544)
3496 - mballoc allocation and error-checking fixes in 2.6 (5504)
3497 - block device patches to fix I/O request sizes in 2.6 (5482)
3498 - look up hostnames for IB nals (5602)
3499 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3500 - don't start multiple acceptors for the same port (5277)
3501 - fix incorrect LASSERT in mds_getattr_name (5635)
3502 - export a proc file for general "ping" checking (5628)
3503 - fix "lfs check" to not block when the MDS is down (5628)
3505 - service request history (4965)
3506 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3507 - create an "evict_client" /proc entry on OSTs, like the MDS has
3508 - fix mount usage message, return errors per mount(8) (5168)
3509 - change grep [] to grep "[]" in tests so they work in more UMLs
3510 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3511 - remove ancient LOV_MAGIC_V0 EA support (5047)
3512 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3513 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3514 - allow readahead allocations to fail when low on memory (5383)
3515 - mmap locking landed again, after considerable improvement (2828)
3516 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3518 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3521 - send OST transaction number in read/write reply to free req (4966)
3522 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3523 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3524 - the watchdog thread now runs as interruptible (5246)
3525 - flock/lockf fixes (but it's still disabled, pending 5135)
3526 - don't use EXT3 constants in llite code (5094)
3527 - memory shortage at startup could cause assertion (5176)
3529 - reorganization of lov code
3530 - single portals codebase
3532 - add extents/mballoc support (5025)
3533 - direct I/O reads in the obdfilter (4048)
3534 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3536 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3539 - send OST transaction number in read/write reply to free req (4966)
3540 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3541 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3542 - the watchdog thread now runs as interruptible (5246)
3543 - handle missing objects in filter_preprw_read properly (5265)
3544 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3545 - drop import inflight refcount on signal_completed_replay error (5255)
3546 - unlock page after async write error during send (3677)
3547 - return original error code on reconstructed replies (3761)
3548 - no transno return for symlink open, don't save no-trasno open (3440)
3550 - add pid to ldlm debugging output (4922)
3551 - bump the watchdog timeouts -- we can't handle 30sec yet
3552 - extra debugging for orphan dentry/inode bug (5259)
3554 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3557 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3558 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3559 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3560 - fix NULL dereference in /proc/sys/portals/routes (4827)
3561 - allow failed mdc_close() operations to be interrupted (4561)
3562 - stop precreate on OST before MDS would time out on it (4778)
3563 - don't send partial-page writes before EOF from client (4410)
3564 - discard client grant for sub-page writes on large-page clients (4520)
3565 - don't free dentries not owned by NFS code, check generation (4806)
3566 - fix lsm leak if mds_create_objects() fails (4801)
3567 - limit debug_daemon file size, always print CERROR messages (4789)
3568 - use transno after validating reply (3892)
3569 - process timed out requests if import state changes (3754)
3570 - update mtime on OST during writes, return in glimpse (4829)
3571 - add mkfsoptions to LDAP (4679)
3572 - use ->max_readahead method instead of zapping global ra (5039)
3573 - don't interrupt __l_wait_event() during strace
3575 - add software watchdogs to catch hung threads quickly (4941)
3576 - make lustrefs init script start after nfs is mounted
3577 - fix CWARN/ERROR duplication (4930)
3578 - return async write errors to application if possible (2248)
3579 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3580 - print NAL number in %x format (4645)
3581 - update barely-supported suse-2.4.21-171 series (4842)
3582 - support for sles 9 %post scripts
3583 - support for building 2.6 kernel-source packages
3584 - support for sles km_* packages
3586 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3589 - ignore -ENOENT errors in osc_destroy (3639)
3590 - notify osc create thread that OSC is being cleaned up (4600)
3591 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3592 - reconstruct ost_handle() like mds_handle() (4657)
3593 - create a new thread to do import eviction to avoid deadlock (3969)
3594 - let lconf resolve symlinked-to devices (4629)
3595 - don't unlink "objects" from directory with default EA (4554)
3596 - hold socknal file ref over connect in case target is down (4394)
3597 - allow more than 32000 subdirectories in a single directory (3244)
3598 - fix blocks count for O_DIRECT writes (3751)
3599 - OST returns ENOSPC from object create when no space left (4539)
3600 - don't send truncate RPC if file size isn't changing (4410)
3601 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3602 - bind to privileged port in socknal and tcpnal (3689)
3604 - rate limit CERROR/CWARN console message to avoid overload (4519)
3605 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3606 - basic mmap support (3918)
3607 - kernel patch series update from b1_4 (4711)
3609 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3612 - avoid crash during MDS cleanup with OST shut down (2775)
3613 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3614 - don't use bad inodes on the MDS (3744)
3615 - dynamic object preallocation to improve recovery speed (4236)
3616 - don't hold spinlock over lock dumping or change debug flags (4401)
3617 - don't zero obd_dev when it is force cleaned (3651)
3618 - print grants to console if they go negative (4431)
3619 - "lctl deactivate" will stop automatic recovery attempts (3406)
3620 - look for existing locks in ldlm_handle_enqueue() (3764)
3621 - don't resolve lock handle twice in recovery avoiding race (4401)
3622 - revalidate should check working dir is a directory (4134)
3624 - don't always mark "slow" obdfilter messages as errors (4418)
3626 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3629 - don't close LustreDB during write_conf until it is done (3860)
3630 - fix typo in lconf for_each_profile (3821)
3631 - allow dumping logs from multiple threads at one time (3820)
3632 - don't allow multiple threads in OSC recovery (3812)
3633 - fix debug_size parameters (3864)
3634 - fix mds_postrecov to initialize import for llog ctxt (3121)
3635 - replace config semaphore with spinlock (3306)
3636 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3637 - don't allow enqueue to complete on a destroyed export (3822)
3638 - down write_lock before checking llog header bitmap (3825)
3639 - recover from lock replay timeout (3764)
3640 - up llog sem before sending rpc (3652)
3641 - reduce ns lock hold times when setting kms (3267)
3642 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3643 - fix NULL deref and obd_dev leak on setup error (3312)
3644 - replace some LBUG about llog ops with error handling (3841)
3645 - don't match INVALID dentries from d_lookup and spin (3784)
3646 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3647 - fix invalid assertion in ptlrpc_set_wait (3880)
3649 - add libwrap support for the TCP acceptor (3996)
3650 - add /proc/sys/portals/routes for non-root route listing (3994)
3651 - allow setting MDS UUID in .xml (2580)
3652 - print the stack of a process that LBUGs (4228)
3654 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3657 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3658 - make sure to unset replay flag from failed open requests (3440)
3659 - if default stripe count is 0, use OST count for inode size (3636)
3660 - update parent mtime/ctime on client for create/unlink (2611)
3661 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3662 - free recovery state on server during a forced cleanup (3571)
3663 - unregister_reply for resent reqs (3063)
3664 - loop back devices mounting and status check on 2.6 (3563)
3665 - fix resource-creation race that can provoke i_size == 0 (3513)
3666 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3667 - more debugging for page-accounting assertion (3746)
3668 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3669 - avoid deadlock after precreation failure (3758)
3670 - fix race and lock order deadlock in orphan handling (3450, 3750)
3671 - add validity checks when grabbing inodes from l_ast_data (3599)
3673 - add /proc/.../recovery_status to obdfilter (3428)
3674 - lightweight CDEBUG infrastructure, debug daemon (3668)
3675 - change default OSC RPC parameters to be better on small clusters
3676 - turn off OST read cache for files smaller than 32MB
3677 - install man pages and include them in rpms (3100)
3678 - add new init script for (un)mounting lustre filesystems (2593)
3679 - run chkconfig in %post for init scripts (3701)
3680 - drop scimac NAL (unmaintained)
3682 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3685 - clean kiobufs before and after use (3485)
3686 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3687 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3688 - change init script to fail more clearly if not run as root (1528)
3689 - allow clients to reconnect during replay (1742)
3690 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3691 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3692 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3693 - fix cond_resched (3554)
3694 - only evict unfinished clients after recovery (3515)
3695 - allow bulk resend, prevent data loss (3570)
3696 - dynamic ptlrpc request buffer allocation (2102)
3697 - don't allow unlinking open directory if it isn't empty (2904)
3698 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3699 - remove extraneous obd dereference causing LASSERT failure (3334)
3700 - don't use get_cycles() when creating temp. files on the mds (3156)
3701 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3702 - handle EEXIST for set-stripe, set proper directory name (3336)
3704 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3705 - fix ksocknal_fmb_callback() error messages (2918)
3707 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3710 - don't copy lvb into (possibly NULL) reply on error (2983)
3711 - don't deref dentry after dput, don't free lvb on error (2922)
3712 - use the kms to determine writeback rpc length (2947)
3713 - increment oti_logcookies when osc is inactive (2948)
3714 - update client's i_blocks count via lvb messages (2543)
3715 - handle intent open/close of special files properly (1557)
3716 - mount MDS with errors=remount-ro, like obdfilter (2009)
3717 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3718 - don't use cancelling-locks' kms values (2947)
3719 - use highest lock extent for kms, not last one (2925)
3720 - don't dereference ERR_PTR() dentry in error handling path (3107)
3721 - fix thread race in portals_debug_dumplog() (3122)
3722 - create lprocfs device entries at setup instead of at attach (1519)
3723 - common AST error handler, don't evict client on completion race (3145)
3724 - zero nameidata in detach_mnt in 2.6 (3118)
3725 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3726 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3727 - initialize RPC timeout timer earlier for 2.6 (3219)
3728 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3729 - print nal/nid for unknown nid (3258)
3730 - additional checks for oscc recovery before doing precreate (3284)
3731 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3732 - don't crash in mdc_close for bad permissions on open (3285)
3733 - zero i_rdev for non-device files (3147)
3734 - clear page->private before handing to FS, better assertion (3119)
3735 - tune the read pipeline (3236)
3736 - fix incorrect decref of invalidated dentry (2350)
3737 - provide read-ahead stats and refine rpc in flight stats (3328)
3738 - don't hold journal transaction open across create RPC (3313)
3739 - update atime on MDS at close time (3265)
3740 - close LDAP connection when recovering to avoid server load (3315)
3741 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3742 - don't leak open file on MDS after open resend (3325)
3743 - serialize filter_precreate and filter_destroy_precreated (3329)
3744 - loop device shouldn't call sync_dev() for nul device (3092)
3745 - clear page cache after eviction (2766)
3746 - resynchronize MDS->OST in background (2824)
3747 - refuse to mount the same filesystem twice on same mountpoint (3394)
3748 - allow llmount to create routes for mounting behind routers (3320)
3749 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3750 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3751 - fix rare problem with rename on htree directories (3417)
3753 - allow default OST striping configuration per directory (1414)
3754 - fix compilation for qswnal for 2.6 kernels (3125)
3755 - increase maximum number of MDS request buffers for large systems
3756 - change liblustreapi to be useful for external progs like lfsck (3098)
3757 - increase local configuration timeout for slow disks (3353)
3758 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3760 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3763 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3764 - don't overwrite extent policy data in reply if lock was blocked (2901)
3765 - drop filter export grants atomically with removal from device (2663)
3766 - del obd_self_export from work_list in class_disconnect_exports (2908)
3767 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3768 - swab reply message in mdc_close, other PPC fixes (2464)
3769 - fix destroying of named logs (2325)
3770 - overwrite old logs when running lconf --write_conf (2264)
3771 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3772 - fix race in target_handle_connect (2898)
3773 - mds_reint_create() should take same inode create lock (2926)
3774 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3775 - don't close files for self_export to avoid uninitialized obd (2936)
3776 - allow MDS with the same name as client node (2939)
3777 - hold dentry reference for closed log files for unlink (2325)
3778 - reserve space for all logs during transactions (2059)
3779 - don't evict page beyond end of stripe extent (2925)
3780 - don't oops on a deleted current working directory (2399)
3781 - handle hard links to targets without a parent properly (2517)
3782 - don't dereference NULL lock when racing during eviction (2867)
3783 - don't grow lock extents when lots of conflicting locks (2919)
3785 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3788 - account for cache space usage on clients to avoid data loss (974)
3789 - lfsck support in lustre kernel code (2349)
3790 - reduce journal credits needed for BRW writes (2370)
3791 - orphan handling to avoid losing space on client/server crashes
3792 - ptlrpcd can be blocked, stopping ALL progress (2477)
3793 - use lock value blocks to assist in proper KMS, faster stat (1021)
3794 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3795 - recovery for initial connections (2355)
3796 - fixes for mds_cleanup_orphans (1934)
3797 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3798 - block all file creations until orphan recovery completes (1901)
3799 - client remove rq_connection from request struct (2423)
3800 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3801 - recovery timer race (2670)
3802 - mdc_close recovey bug (2532)
3803 - ptlrpc cleanup bug (2710)
3804 - mds timeout on local locks (2588)
3805 - namespace lock held during RPCs (2431)
3806 - handle interrupted sync write properly (2503)
3807 - don't try to handle a message that hasn't been replied to (2699)
3808 - client assert failure during cleanup after abort recovery (2701)
3809 - leak mdc device after failed mount (2712)
3810 - ptlrpc_check_set allows timedout requests to complete (2714)
3811 - wait for inflight reqs when ptlrpcd finishes (2710)
3812 - make sure unregistered services are removed from the srv_list
3813 - reset bulk XID's when resending them (caught by 1138 test)
3814 - unregister_bulk after timeout
3815 - fix lconf error (2694)
3816 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3817 - readahead locks pages, leaves pending causing memory pressure (2673)
3818 - increase OST request buffers to 4096 on large machines (2729)
3819 - fix up permission of existing directories in simple_mkdir (2661)
3820 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3821 - don't assign transno to errored transactions (2742)
3822 - don't delete objects on OST if given a bogus objid from MDS (2751)
3823 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3824 - if rq_no_resend, then timeout request after recovery (2432)
3825 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3826 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3827 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3828 - MDS recovery shouldn't delete objects that it has given out (2730)
3829 - if enqueue arrives after completion, don't clobber LVB (2819)
3830 - don't unlock pages twice when trigger_group_io returns error (2814)
3831 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3832 - don't write pages to disk if there was an error (1450)
3833 - don't ping imports that have recovery disabled (2676)
3834 - take buffered bytes into account when balancing socknal conn (2817)
3835 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3836 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3837 - remove little-endian swabbing of llog records (1987)
3838 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3839 - retry reposting request buffers if they fail (1191)
3840 - grow extent at grant time to avoid granting a revoked lock (2809)
3841 - lock revoke doesn't evict page if covered by a second lock (2765)
3842 - disable VM readahead to avoid reading outside lock extents (2805)
3844 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3845 - updated kernel patches for hp-2.4.20 kernel (2681)
3847 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3850 - fix truncated write corruption (2366)
3851 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3853 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3854 - ensure len(uuid) < 37 in lmc (1171)
3855 - fix ia64 OOPS in llog_test (2255)
3856 - zero end of page at obdfilter for partial page writes (2648)
3857 - don't leave stale dentries around after renames (bug 2428)
3858 - fix timeouts when evicting a client with a single lock held (2642)
3859 - set deadline for the initial HELLO message to drain (2634)
3860 - print out dotted-quad IP addresses in the socknal (2302)
3862 - additional debugging for MDS client eviction problem (2443)
3863 - fix mkfsoptions support for osts (2603, 2604)
3865 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3868 - add series for the vanilla 2.6.0 kernel
3869 - add series for the vanilla 2.4.24 kernel
3870 - add series for a cray x86/64 UL kernel drop
3871 - fix xattr patches for the vanilla 2.4.19 series
3873 - generate true UUIDs in lmc (1171)
3874 - have portals stack dumping break in UML (2466)
3875 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3876 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3877 - wake callback waiting threads on client eviction (2460)
3878 - Add --ptldebug and --subsystem to lmc (1719)
3879 - update assertion to allow safe interrupt allocation
3880 - set rq_no_resend for cancel requests (2432)
3881 - recalculate ptlrpcd timeout after resend (2494)
3882 - call vfs_rmdir when removing pending directories (2368)
3883 - fix renaming a file to itself (2429)
3884 - lmc creates a default one-stripe lov (2454)
3885 - expand procfs space to handle large clusters (2326)
3886 - increase UML stack to avoid overflow
3887 - update lconf's list of debug and subsystem masks
3888 - fix lfs find --obd (2510)
3889 - /proc tunable for disabling filter read caching (2591)
3890 - stop rpm packages from altering slapd.conf (2301)
3891 - disable nagle in the socknal under 0conf (2578)
3892 - choose mds inode size based on stripe count (2572)
3893 - fix kernel-source rpm problems (2516)
3895 - add --disable-doc to avoid pdf generation (2421)
3896 - update documentation, tests, type-os, comments
3897 - avoid format warnings on ia64
3898 - remove the TOE NAL
3899 - tiny code cleanups by removing unused fields
3901 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3904 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3905 - remove the most blatant lies from BUILDING file (2371)
3906 - change default debug level to reasonable production setting
3907 - reduce client side cache size to reduce cache flush time
3908 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3909 - make TCP zerocopy and pinger support enabled by default (2476)
3910 - sync writes completed after process exits caused crashes (2319)
3911 - maintain correct mount count on the MDS (2356)
3912 - backout 1557, because 2316 wasn't really fixed
3913 - better file I/O statistics gathering in /proc
3914 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3915 - another kernel patch to fix zero-copy TCP function export
3916 - don't take duplicate lock when processing re-sent getattr (2420)
3917 - lctl uses obd_self_export instead of creating new conn (2353)
3918 - MDS/OST recovery case which requires object creation asserted (2425)
3919 - move lfs from /usr/sbin to /usr/bin in packages
3920 - fix race between mds_client_add and mds_client_free (2417)
3921 - use kmalloc instead of slabs in portals (2430)
3922 - don't create duplicate records when a failover MDS is present (2442)
3923 - remove unnecessary mount age check (2332)
3924 - don't remove directory inodes from locks prematurely (2451)
3925 - don't break if MDS service name is the same as hostname (2103)
3926 - fix races in client write RPC generation when cache full (2482)
3928 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3931 - remove now-unused request->rq_obd (278)
3932 - if an allocation fails, print out how much memory we've used (1933)
3933 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3934 - add the "configurable stack size" patch to most series files (1256)
3935 - ability to write large log records, for 100+ OST configs (2306)
3936 - fix NULL deref when filter_prep fails (2314)
3937 - fix operator precedence error in filter_sync
3938 - dynamic allocation of socknal TX descriptors (2315)
3939 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3940 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3941 - ensure that CWARN messages in Portals always get to the syslog
3942 - __init/__exit are not for prototype decls (ldlm_init/exit)
3943 - x86-64 compile warning fixes
3944 - fix gateway LMC keyword conflict (2318)
3945 - fix MDS lock inversions in getattr/reint paths (1844)
3946 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3947 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3948 - DLM race condition prevented some lock evictions (2328)
3949 - ENOMEM detection and retry on socknal sends (2230)
3950 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3951 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3952 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3953 - better error messages when a client is rejected during recovery (1505)
3954 - avoid cancelling locks which were never granted, after failure (2330)
3955 - fix i_sem/journal inversion in mds_client_add (2333)
3956 - fix truncate/getattr lock cycle deadlock (2334)
3957 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3958 - fix two rare exit paths which could leak an l_lock() ref (2321)
3959 - fencepost error in MDS/OST orphan recovery (2226)
3960 - make log record alignment 8 bytes (1988)
3961 - lstripe now fails when requested offset > ost_count (2237)
3962 - ensure that all kernel series have a complete list.h (1607)
3963 - fix crashes in special-file operations (2316)
3964 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3966 - allow configurable automake binary, for testing new versions
3967 - small update to the lfs documentation
3969 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3971 * fix negative export reference count in fsfilt_sync (2312)
3973 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3974 * release candidate 0.9.1
3976 - orphans are moved into the PENDING directory for possible recovery
3977 - replayed opens now open by fid for orphan/rename safety (1042)
3978 - last close of an orphan inode generates a transno (683)
3979 - chdir() and mount() now pin the directory entry (1020)
3980 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3981 - discard very old requests without processing them (1502)
3982 - remove some common, well-understood CERRORs (1505)
3983 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3984 - clear "grant" flags in OST replies until OST grant code lands (1644)
3985 - fix read performance by not clobbering i_blksize on client (1598)
3986 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3987 - make LRU size a /proc tunable, clears locks when reduced (707)
3988 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3989 - server locks take references on exports now (1558)
3990 - build fixes for 2.4.20-rh trees (1663)
3991 - return an error from lov_create if all OSCs are inactive (1751)
3992 - fix import levels when a reconnect happens without a timeout (1597)
3993 - exit early from mds_open if we get a lookup error (1749)
3994 - partial page read at EOF wouldn't wait for disk before sending (1642)
3995 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3996 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3997 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3998 - write/truncate lock inversion (1639)
3999 - Don't auto-load obdclass, portals modules during cleanup (1495)
4000 - fix timestamps from jumping to "now" (1763)
4001 - extra journal assertions (1648)
4002 - add an extra multiunlink test (1771)
4003 - fix read_record/write_record API (1776)
4004 - fix leak of offset_extent, possible incorrect i_size later (1772)
4005 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4006 - Debugging for the kqswnal_get_idle_tx problems (1820)
4007 - Allow recovery to be attempted multiple times (1536)
4008 - Write out MDS last_rcvd file after it is first created (1600)
4009 - Fix tx_descriptor leak in failed transmit situations (1827)
4010 - ext3 journaling fixes for assertion failure after IO error (1871)
4011 - class_export_put() on freed export after completion AST error (1896)
4012 - Fix revalidate looping in VFS (1322)
4013 - Don't access a freed export during MDS_REINT timeout (1521)
4014 - Add open-unlink recovery support on the MDS (1673,1764)
4015 - Return an error if no MDS data was read from last_rcvd (1946)
4016 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4017 - Don't setup a disk device that doesn't match exported UUID (317)
4018 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4019 - avoid committing NULL handle in force close
4020 - local.sh is now a one-stripe LOV configuration
4021 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4022 - fix ext3 htree duplicate directory entry corruption (1516)
4023 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4024 - update to new LOV EA format (2097)
4025 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4027 2003-06-15 Phil Schwan <phil@clusterfs.com>
4030 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4031 - per-import/export recovery handling (958, 931, 959)
4032 - multiple last-rcvd slots, for serving multiple FSes (949)
4033 - connections are again shared between multiple imp/exports (963, 964)
4034 - "umount -f" would hang if any requests needed to be sent (393, 978)
4035 - avoid pinning large req buffer by copying for queued messages (989)
4036 - add "uuid" to "lctl device" command to help upcalls (991)
4037 - "open" RPCs with transnos would confuse recovery counters (1037)
4038 - do proper endian conversion of all wire messages (288, 340, 891)
4039 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4040 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4041 - fix ptlrpc_connection leak in target_handle_connect (1174)
4042 - fix import refcounting bug in OST and MDS cleanup (1134)
4043 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4044 - fix very unlikely obd_types race condition (501)
4045 - remove osc_open hack for echo_client (1187)
4046 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4047 - a failure in read_inode2 leads to deadlock (1139)
4048 - cancel ack-locks as soon as transaction is committed (1072)
4049 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4050 - make sure to commitrw after any preprw to avoid deadlock (1162)
4051 - failing to execute a file in a lustre FS would lock inode (1203)
4052 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4053 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4054 - obd_uuid2tgt would walk off the end of the list (1255)
4055 - on IA64 the osc would give portals incorrect bulk size (1258)
4056 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4057 - fix lock inversion caused by new llite matching code (1282)
4058 - limit the number of dirty pages on a client to 10MB (1286)
4059 - timed out locks were not being corrected cancelled (1289)
4060 - fix O_DIRECT above 4GB on IA-32 (1292)
4061 * major user-visible changes
4062 - fail out/fail over policy now controlled by the upcall (993)
4064 - add OBD_PING to check server availability and failure (954)
4065 - lustre messages are now sent in sending host order (288, 340, 891)
4066 - add eadatalen to MDS getattr reply (340)
4067 - OST read replies may contain second buffer, with per-page status (593)
4069 2003-03-11 Phil Schwan <phil@clusterfs.com>
4072 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4073 - failed open()s could cause deadlock; fixed (867, 869)
4074 - stop cancelling OST locks when files are closed (481)
4075 - overlapping XID spaces caused network corruption (851, 853)
4076 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4077 - setattr_raw added, to do single-RPC, server-side setattrs
4078 - lmc/lconf syntax change for OST UUIDs
4079 - fix crashy race condition between ptlrpc_free_req and osc_close
4080 - don't use request in mdc_enqueue if we hit a timeout (889)
4081 - don't set the inode i_size for regular files from the MDS (896)
4082 - handle out of order completion AST (842)
4083 - don't LBUG if a lock request times out after receiving AST (913)
4084 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4085 - if a bad lock AST arrives, send an error instead of dropping entirely
4086 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4087 - fix leak in bulk IO when only partially completed (899, 900, 926)
4088 - fix O_DIRECT for ia64 (55)
4089 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4090 - C-z after timeout could hang a process forever; fixed (977)
4092 - client-side I/O cache (678, 924, 929, 941, 970)
4094 - READPAGE and SETATTRs which don't take server-side locks get
4097 2003-02-11 Phil Schwan <phil@clusterfs.com>
4100 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4101 - Fix fsfilt_extN_readpage() to read a full page of directory
4102 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4103 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4104 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4105 - Most intent operations were removed, in favour of a new RPC mode
4106 that does a single RPC to the server and bypasses most of the VFS
4107 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4108 - Aggressively cancel local locks on DLM servers
4109 - mds_reint_unlink sends EA to the client if it's the last nlink.
4110 client uses that EA to unlink OST objects.
4111 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4112 - recursive symlinks were fixed (439)
4113 - fixed NULL deref in DEBUG_REQ
4114 - filter_update_lastobjid no longer calls sync, which annoyed extN
4115 - fixed multi-client small-writes to a single file problem (445)
4116 - fixed mtime updates during file writes (607)
4117 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4118 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4119 - fixed Linux OST size reporting problem (444, 656)
4120 - OST now updates object mtime with writes or setattr (607, 619)
4121 - client verifies file size before zeroing page past EOF (445)
4122 - OST now writes last allocated objid to disk with allocation (108)
4123 - LOV on echo now works (409)
4125 - mds_reint_unlink sends a new buffer, with the EA included. this
4126 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4127 set if a regular file was being unlinked, and it was the last link
4128 - use PtlGet from the target for bulk writes (315)
4129 - OST now updates object mtime with writes or setattr (607, 619)
4130 - LDLM now has a grant-time callback to revalidate locked items, if
4132 - Many MDS operations were reorganized to combat race conditions
4134 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4135 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4137 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4140 - Fully reactivate OST imports after reconnection (512, others)
4141 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4142 - More graceful error handling for truncating on dead OST (515)
4143 - Don't error out unless we're actually accessing dead stripes (474)
4144 - Fix garbage sizes when stripes are missing (410)
4145 - LRU counters were broken, causing constant lock purge (433, 432)
4146 - garbage on read from stripes with failed OSTs (441)
4147 - mark OSCs as active before reconnecting during recovery (438)
4148 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4149 - lfind did not preserve OST order in output (443)
4150 - symlinks cause hung clients, incorrect data (439)
4151 - stop dereferencing request after dropping refcount (457)
4152 - don't LASSERT(spin_is_locked) on non-SMP (455)
4153 - fixes for many rename() bugs
4154 - fstat didn't correctly synchronize attributes (399)
4155 - server must handle lock cancellation during blocking AST prep (487)
4156 - bulk descriptors were free()d too soon (511)
4157 - fix paths in lconf, which would load incorrect modules (451, 507)
4158 - fix confusing lconf 'host not found' error message (386)
4159 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4160 - fix race condition in mdc_blocking_ast() for inode access (526)
4161 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4162 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4163 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4164 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4165 - fix lstripe in several ways (kernel side) (527)
4166 - fix request leak in ldlm_cli_enqueue (262)
4167 - incorrect OSC was marked inactive after OST failure
4168 - call mds_fs_cleanup before unmounting filesystem (524)
4169 - fix races between taking ns_lock and ldlm_lock_change_resource
4170 - fix races updating LOV export open file list
4171 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4172 - fix recovery NULL deref in ldlm_cli_cancel_unused
4173 - fix some DLM races by using new hash table for lock handles (419)
4174 - permit the client to specify desired inodes, at replay
4175 - duplicate requests when we queue them for replay reintegration
4176 - fix last_rcvd offset calculation
4177 - sync after each recovered transaction, so we always make progress
4178 - never, not always, ERESTART requests without transnos
4179 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4181 - skip replay if the MDS says that the client is already connected
4182 - don't check for a recovery-enabled export to match lctl's UUID
4183 - don't INC_USE_COUNT for phantom exports
4184 - don't crash when cleaning up phantom exports (567)
4185 - don't double-finish or set replay data for errored mdc_open requests
4186 - abort requests when they time out, so we don't get old replies
4187 - send/receive replies for AST messages again
4188 - if the client says that it doesn't have the lock, cancel it on the
4190 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4191 mark it as destroyed, it will all work out when decref is called
4192 - fix module use counts (22, 581)
4194 - ASTs now expect a reply (server cancels lock on error reply)
4196 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4199 - fix many simultaneous client startup (392)
4200 - fix dentry->d_it clobbering
4201 - credentials weren't being shipped for readdir/getattr operations
4202 - remove invalid assertions triggered during some concurrent MD
4204 - proper Lustre versions added (336, 389)
4205 - fix memory leak for create error case (398)
4206 - fix LOV locking bug that would get cli/srv out of sync
4207 - fix echo client over LOV (409)
4208 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4209 - fix double-O_EXCL intent crash (424)
4210 - avoid sending multiple lock CANCELs (352)
4212 - MDS can do multi-client recovery (modulo bugs in new code)
4214 - many updates, edits, cleanups
4216 2002-11-18 Phil Schwan <phil@clusterfs.com>
4219 - fix null d_it dereference (346)
4220 - fix full OST/dbench hang (333)
4221 - fix permission problem with file removal (286)
4222 - fix removal of OSCs from LOV when they fail
4223 - fix NULL deref during bulk timeout (214)
4224 - fix problems related to multiple filesystems on one MDS (241)
4225 - fixed serious subtle metadata locking bugs
4226 - free locks on clients when inodes are removed due to memory
4228 - fix inode pointer in lock data (285)
4229 - partial support for multiple MDS on a single host (241)
4230 - data locks weren't cancelled at clear_inode time (290, 311)
4231 - intent locks could lead to unbounded lock growth (205)
4232 - added a maximum lock count, an LRU list, and a flusher
4233 - fix multiple rename (365)
4234 - properly abstracted the echo client
4235 - OSC locked 1 byte too many; fixed
4236 - rewrote brw callback code:
4237 - fixed recovery bugs related to LOVs (306)
4238 - fixed too-many-pages-in-one-write crash (191)
4239 - fixed (again) crash in sync_io_timeout (214)
4240 - probably fixed callback-related race (385)
4242 - Add capability to MDS protocol
4243 - LDLM cancellations and callbacks on different portals
4245 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4248 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4249 - defer open object destruction to close time (601981, 138)
4250 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4251 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4252 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4253 - don't call obd_setattr in ll_file_release for destroyed objects
4254 * protocol change to lustre_msg: move |version| and add |flags|
4255 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4256 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4257 * added replay of create, unlink, link and rename operations during
4258 MDS failover; recovery should be much more robust now
4259 * remove failed OSCs from LOVs (only lov_create uses this so far)
4260 * the lustre-HOWTO was brought (more) up to date (582544)
4262 2002-10-23 Phil Schwan <phil@clusterfs.com>
4265 - in-use dentries weren't being reused properly (617851)
4266 - prevent multiple LDLM setup (599178)
4267 - fix LOV size calculations for truncate (617853)
4268 - fix client handling of MDS intent errors (POSIX)
4269 - fix permission bug in lovstripe.c test (624321)
4270 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4271 - truncate past end of file could corrupt data
4272 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4273 - a race in recovery could return ETIMEDOUT to apps (623947)
4274 - building outside the source directory was fixed
4275 * the lustre-HOWTO was brought (more) up to date (582544)
4276 * major progress was made on recovery functionality
4278 2002-10-10 Phil Schwan <phil@clusterfs.com>
4281 - recovery deadlock fix
4282 - rm -rf causes LBUG fix (617817)
4283 - file open by multiple tasks fix (618962)
4284 - directory permissions bugs (602707 and 620007)
4285 - journal_stop fixed with locking (611313)
4286 - O_APPEND failures resolved (618273, perhaps 614459)
4287 - lconf PATH fix (619770)
4288 - IA64 build fix (621450)
4289 - RPC buffer sizes scale with amount of memory
4291 2002-10-01 Phil Schwan <phil@clusterfs.com>
4294 - locks would be cancelled without throwing away data pages,
4295 resulting in inconsistent data (605627)
4296 - inode attributes were not always being refreshed (605627, 612449)
4297 - lconf now continues to cleanup after lctl reports an error
4298 - MDS now enforces user permissions (602707)
4299 - lprocfs cleanup fixed, but not yet enabled (614157)
4300 - fixed infinite server hang, should a client not respond to an AST
4301 - avoid going into recovery if user calls readlink() with a buffer
4302 that's too small (613941)
4303 - AST RPCs no longer require replies (614867) -- this may be changed
4304 - don't crash server if client sends an IOV that's too big (611336)
4305 - fixed lock conversion deadlock (611892)
4306 - fixed the following of symlinks (614622)
4307 * recovery: the server can remove locks from a client that dies, other
4308 clients can make progress
4309 * more extN patch fixes
4310 * compile-time configurable ptlrpc buffer allocations
4312 - collaborative read cache document
4313 - Lustre Lite Performance CDR document-in-progress
4315 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4318 - fix typo in patch-2.4.18
4320 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4323 - clear ptlrpc request each time in handle_incoming_request()
4324 - unlink of files now destroys the object on the OST
4326 2002-09-19 Peter Braam <braam@clusterfs.com>
4328 * add hard link support
4329 * change obdfile creation method
4330 * kernel patch changed
4332 2002-09-19 Peter Braam <braam@clusterfs.com>
4335 - stack overflow bug in extN fixed
4337 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4339 * documentation updates
4340 - add man pages for config tools
4341 - update tests/README to describe testing with new config tools
4342 - finish metadata API descriptions
4343 * bug fixes and cleanups
4344 - statfs workaround for 16TB limit
4345 - LOV stripe allocation improved, can stripe on subset of OSTs
4346 - LOV file size/IO offset was wrong for files > 4GB in size
4347 - object EA data was being dropped, caused files to be unreadable
4348 - memory overflow with non-LOV OST caused memory corruption
4349 - fixed regression tests to work with new config tools, obdfilter
4350 - fixed bug when directory size became larger than 1 block
4351 - fixed bug (for single client case) when PWD was deleted
4352 - invalidate local directory pages when doing intent-based ops
4353 - avoid LDLM oops when lock callback contained bad data
4355 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4357 * documentation updates
4358 * bug fixes and cleanups
4359 - configuration tools
4362 - 64-bit compile warnings
4363 - 64-bit internal statfs data
4365 * test_brw on persistent OST devices
4367 * lprocfs (disabled)
4369 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4371 * documentation updates
4372 * bug fixes and cleanups
4373 * configuration tools
4375 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4378 * many small fixes to 0_5_4
4379 * io/network handling
4380 * thinkos in MDS operations
4382 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4385 * crucial basic fixes to 0.5.3
4386 * IOR, Iozone work over Elan
4389 2002-08-07 Phil Schwan <phil@clusterfs.com>
4390 * version 0_5_3, our first alpha
4391 * we use the new Portals iovs
4392 * documentation updates
4393 * bug fixes and cleanups
4394 * small changes in the DLM wire protocol
4396 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4397 * version 0_5_1 with some initial stability,
4398 * locking on MD and file I/O.
4399 * documentation updates
4400 * several bug fixes since 0.5.0
4401 * small changes in wire protocol
4403 2002-07-18 Phil Schwan <phil@clusterfs.com>
4405 * delivered as Lustre Light Alpha
4406 * fixed a crash after handling invalid MDS requests
4407 * fixed directory pages for architectures with non-4k pages sizes
4409 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4410 * release version v0_4_4
4411 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4413 * Fixes a number of interruption problems with OST operations.
4414 * Update documentation for portals header changes
4415 * Move all wire protocol structs/defines to lustre_idl.h
4416 * Fixes symlink length bug.
4417 * Add tcpdump to repository.
4419 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4420 * release version v0_4_3
4421 * Fixes statfs for inodes on extN.
4422 * Fixes bug in runtests which would delete /etc/hosts.
4423 * Use 64-bit object IDs wherever possible (not into VFS though)
4424 Remove ost_get_info, which is unused by lustre, and out of date.
4426 2002-07-03 Peter Braam <braam@clusterfs.com>
4427 * release version v0_4_2 Fixes a lookup error (type not passed)
4428 * move forward to head of Portals
4429 * move forward to latest Lustre kernel
4431 2002-06-25 Peter Braam <braam@clusterfs.com>
4432 * release version v0_4_1. Hopefully stable on single node use.