3 2.6.5-7.287.3 (SLES 9),
4 2.6.9-55.0.12.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 2.6.18.8 vanilla (kernel.org)
8 * Client support for unpatched kernels:
9 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
10 2.6.9-42.0.10.EL (RHEL 4),
11 2.6.16 - 2.6.21 vanilla (kernel.org)
12 * Recommended e2fsprogs version: 1.40.2-cfs5
13 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
15 Severity : enhancement
17 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
21 Description: make mgs_setparam() handle fsname containing dash
22 Details : fsname containing a dash does not work with lctl conf_param
24 Severity : enhancement
26 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
28 Severity : enhancement
30 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
34 Frequency : rare, at shutdown
35 Description: access already free / zero obd_namespace.
36 Details : if client_disconnect_export was called without force flag set,
37 and exist connect request in flight, this can produce access to
38 NULL pointer (or already free pointer) when connect_interpret
39 store ocd flags in obd_namespace.
43 Frequency : only at startup
44 Description: not alloc memory with spinlock held.
45 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
48 Severity : enhancement
50 Description: make lustre randomly fail allocating memory
51 Details : Make lustre randomly failed allocating memory for testing purpose.
53 Severity : enhancement
55 Description: lost problems with lov objid file
56 Details : Fixes some scability and access to not inited memory problems
57 in work with lov objdid file.
62 Description: lfs find does not continue on file error
63 Details : Continue other files processing when a file/dir is absent.
67 Description: Inconsistent usage of lustre_pack_reply()
68 Details : Standardize the usage of lustre_pack_reply() such that it
69 always generate a CERROR on failure.
72 Frequency : occasional
74 Description: MDS hang or stay in waiting lock
75 Details : If client receive lock with CBPENDING flag ldlm need send lock
76 cancel as separate rpc, to avoid situation when cancel request
77 can't processed due all i/o threads stay in wait lock.
81 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
85 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
89 Description: Update to SLES9 kernel-2.6.5-7.287.3.
93 Description: Update to RHEL4 latest kernel.
95 Severity : enhancement
97 Description: Build SLES10 patchless client fails
98 Details : The configure was broken by run ./configure with
99 --with-linux-obj=.... argument for patchless client. When the
100 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
101 can't search header adequately. Use absolute path such as
102 -I($LINUX)/include instead.
106 Description: interrupt oig_wait produce painc on resend.
107 Details : brw_redo_request can be used for resend requests from ptlrpcd and
108 private set, and this produce situation when rq_ptlrpcd_data not
109 copyed to new allocated request and triggered LBUG on assert
110 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
111 ptlrpcd set if request is changed and can be safety changed to use
116 Description: LASSERT_{REQ,REP}SWAB macros are buggy
117 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
118 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
119 these macros and replace them with their difinitions instead.
123 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
124 Details : Update which_patch & target file for SLES10 latest kernel.
127 Description: Remove client patches from SLES 10 kernel.
128 Details : This causes SLES 10 clients to behave as patchless clients
129 even on a Lustre-patched (server) kernel.
131 Severity : enhancement
133 Description: self-adjustable client's lru lists
134 Details : use adaptive algorithm for managing client cached locks lru
135 lists according to current server load, other client's work
136 pattern, memory activities, etc. Both, server and client
137 side namespaces provide number of proc tunables for controlling
140 Severity : enhancement
142 Description: light-weight GSS support
143 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
144 and reduce performance overhead.
146 Severity : enhancement
148 Description: Linux keyring support
149 Details : Support using service of Linux keyring for Lustre GSS internal
150 context refresh/cache mechanism.
154 Description: Fix errors in lfs documentation
155 Details : Fixes man pages
159 Description: don't use GFP_* in generic Lustre code.
160 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
164 Description: obdclass is limited by single OBD_ALLOC(idarray)
165 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
169 Description: warnings with build patchless client with vanila 2.6.19 and up
170 Details : change old ctl_table style and replace ctl_table/ctl_table_header
171 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
175 Frequency : Only for SLES
176 Description: sanity_quota fail test_1
177 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
178 file, which makes TSTID[2] inunique.
182 Frequency : Only for SLES9
183 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
184 Details : Update target/ChangeLog/which_patch .
188 Frequency : Only for test_10 in sanity-quota.sh
189 Description: a bug in quota test code
190 Details : checking a test flag in a wrong place
194 Frequency : Only for test_13 in sanity-quota.sh
195 Description: a bug in quota test script
196 Details : shouldn't check overfull allocated quota
200 Description: jbd statistics
201 Details : Port older jbd statistics patch for sles10
203 Severity : enhancement
205 Description: Add printing inode info into message about error in writepage.
210 Description: reply_lock_interpret crash due to race with it and lock cancel.
211 Details : Do not replay locks that are being cancelled. Do not reference
212 locks by their address during replay, just by their handle.
216 Frequency : When flocks are used.
217 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
218 Details : Flock locks might destroy just granted lock if it could be merged
219 with another existing flock, this is done in completion handler,
220 so teach ldlm_cli_enquque_fini that this is a valid case for
226 Description: Oops in read and write path when failing to allocate lock.
227 Details : Check if lock allocation failed and return error back.
231 Description: lstripe command fails for valid OST index
232 Details : The stripe offset is compared to lov->desc.ld_tgt_count
233 instead of lov->desc.ld_active_tgt_count.
237 Description: sanity.sh failed test 103
238 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
243 Description: df doesn't work properly if diskfs blocksize != 4K
244 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
248 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
249 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
253 Description: obd_health_check_timeout too short
254 Details : set obd_health_check_timeout as 1.5x of obd_timeout
258 Description: llapi_file_create() does not allow some changes
259 Details : add llapi_file_open() that allows specifying the mode and
260 open flags, and also returns an open file handle.
264 Description: merge and cleanup kernel patches.
265 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
269 Description: Add journal checksum support.(Kernel part)
270 Details : The journal checksum feature adds two new flags i.e
271 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
272 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
273 indicates that the commit block contains the checksum for
274 the blocks described by the descriptor blocks. Now commit
275 record can be sent to disk without waiting for descriptor
276 blocks to be written to disk. This behavior is controlled
277 using JBD2_FEATURE_ASYNC_COMMIT flag.
281 Description: OSS needs mutliple precreate threads
282 Details : Add ability to start more than one create thread per OSS.
286 Description: Sanity.sh test_65k failed: setstripe should have succeeded
287 Details : execute lfs setstripe on client
291 Description: mds_obd_create error creating tmp object
292 Details : When the user sets quota on root, llog will be affected and can't
293 create files and write files.
296 Frequency : Always on ia64 patchless client, and possibly others.
298 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
299 Details : This allows the patchless client to be loaded on architectures
304 Description: RedHat Update kernel for RHEL5
305 Details : Add the kernel config file for RHEL5.
309 Description: RedHat Update kernel for RHEL5
310 Details : Modify the kernel config file more closer RHEL5.
314 Description: Build failure against Centos5 (RHEL5)
315 Details : Define PAGE_SIZE when it isn't present.
319 Description: test_7 on sanity-quota.sh can't be used on separate machines
320 Details : add facet to handle it
324 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
325 Details : with NFS, the anon dentry's parent was set to itself in
326 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
327 corresponding dentry other than use rec->ur_name.
329 Severity : enhancement
331 Description: lfs setstripe enhancement
332 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
336 Description: enable data checksumming by default
337 Details : enable checksum by default, allow --disable-checksum
338 configure option and "-o nochecksum" mount option. Checksums
339 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
343 Description: lustre support for RHEL5
344 Details : Add support for RHEL5.
348 Description: Client eviction due to failover config
349 Details : after a connection loss, the lustre client should attempt to
350 reconnect to the last active server first before trying the
351 other potential connections.
355 Description: when mds and osts use different quota unit(32bit and 64bit),
356 quota will be released repeatly.
357 Details : void sending multiple quota reqs to mds, which will keep the
358 status between the reqs.
362 Description: osts not allocated evenly to files
363 Details : change the condition to increase offset_idx
366 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
368 Description: Data corruption for OSTs that are formatted larger than 2TB
370 Details : When generating the bio request for lustre file writes the
371 sector number would overflow a temporary variable before being
372 used for the IO. The data reads correctly from Lustre (which
373 will overflow in a similar manner) but other file data or
374 filesystem metadata may be corrupted in some cases.
378 Description: Tune the kernel for good SCSI performance.
379 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
380 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
385 Description: rewrite ext2-derived code in obdclass/uuid.c
386 Details : rewrite inherited code (uuid parsing code from ext2 utils)
387 from scratch preserving functionality.
391 Description: "lfs find -obd UUID" prints directories
392 Details : "lfs find -obd UUID" will return all directory names instead
393 of just file names. It is incorrect because the directories
394 do not reside on the OSTs.
398 Description: use i_size_read and i_size_write in 2.6 port
399 Details : replace inode->i_size access with i_size_read/write()
402 Frequency : when removing large files
404 Description: scheduling issue during removal of large Lustre files
405 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
406 It causes scheduling issues when removing large files (17TB in the
410 Frequency : only with liblustre clients on XT3
412 Description: evictions taking too long
413 Details : allow llrd to evict clients directly on OSTs
416 Frequency : only on ppc
418 Description: /proc/fs/lustre/devices broken on ppc
419 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
420 We should make sure the type of 'index' is loff_t and then cast
421 explicitly as needed below. Do not assign an explicitly cast
425 Frequency : only for rhel5
427 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
428 Details : Modify the target file & which_kernel.
430 Severity : enhancement
432 Description: omit set fsid for export NFS
433 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
434 as FS_REQUIRES_DEV for avoid problems with generate fsid.
438 Frequency : Always, for kernels after 2.6.16
439 Description: Fix warning idr_remove called for id=.. which is not allocated.
440 Details : Last kernels save old s_dev before kill super and not allow
441 to restore from callback - restore it before call kill_anon_super.
445 Description: buffer overruns could theoretically occur
446 Details : llapi_semantic_traverse() modifies the "path" argument by
447 appending values to the end of the origin string, and a
448 overrun may occur. Adding buffer overrun check in liblustreapi.
452 Description: Fix error on 'ls .' at the top of the Lustre mount.
453 Details : Don't revalidate dentry if it is a root dentry.
457 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
458 Details : Modify vm-tunables-rhel4.patch.
462 Description: Kernel config for 2.6.18-vanilla.
463 Details : Modify targets/2.6-vanilla.target.in.
464 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
465 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
466 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
467 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
471 Description: improve handling recoverable errors
472 Details : if request processig with error which can be recoverable on server
473 request should be resend, otherwise page released from cache and
478 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
479 Details : Modify target file & which_patch.
480 A flaw was found in the IA32 system call emulation provided
481 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
482 value could be stored in the %RAX register, which could trigger an
483 out-of-bounds system call table access. An untrusted local user
484 could exploit this flaw to run code in the kernel
485 (ie a root privilege escalation). (CVE-2007-4573).
489 Description: O_DIRECT bypasses client statistics.
490 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
491 read_ahead_stats were updated. copied the stats section from
492 osc_send_oap_rpc() into async_internal().
496 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
500 Description: change order of libsysio includes
501 Details : '#include sysio.h' should always come before '#include xtio.h'
505 Description: handle "serious error: objid * already exists" more gracefully
506 Details : If LAST_ID value on disk is smaller than the objects existing in
507 the O/0/d* directories, it indicates disk corruption and causes an
508 LBUG(). If the object is 0-length, then we should use the existing
509 object. This will help to avoid a full fsck in most cases.
511 Severity : enhancement
513 Description: adapt the lustre_config script to support the upgrade case
514 Details : Add "-u" option for lustre_config script to support upgrading 1.4
515 server targets to 1.6 in parallel.
519 Description: To avoid grant space > avaible space when the disk is almost
520 full. Without this patch you might see the error "grant XXXX >
521 available" or some LBUG about grant, when the disk is almost
523 Details : In filter_check_grant, for non_grant cache write, we should
524 check the left space by if (*left > ungranted + bytes), instead
525 of (*left > ungranted), because only we are sure the left space
526 is enough for another "bytes", then the ungrant space should be
527 increase. In client, we should update cl_avail_grant only there
528 is OBD_MD_FLGRANT in the reply.
532 Description: Update RHEL 4 kernel to fix local root privilege escalation.
533 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
534 described in CVE-2007-4573. This problem could allow untrusted
535 local users to gain root access.
538 Frequency : when using O_DIRECT and quotas
540 Description: Incorrect file ownership on O_DIRECT output files
541 Details : block usage reported by 'lfs quota' does not take into account
542 files that have been written with O_DIRECT.
547 Description: touch file failed when fs is not full
548 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
549 otherwise we can get ENOSP while fs is not full.
553 Description: parallel lock callbacks
554 Details : Instead of sending blocking and completion callbacks as separated
555 requests, adding them to a set and sending in parallel.
558 Frequency : only for Cray XT3
559 Bugzilla : 12829/13455
560 Description: Changing primary group doesn't change the group lustre assigns to
562 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
563 overridden with the primary group provided by the group upcall,
564 whereas the supplied fsgid can be trusted if it is in the list of
565 supplementary groups returned by the group upcall.
567 Severity : enhancement
569 Description: Allow masking D_WARNING, D_ERROR messages from console
570 Details : Console messages can now be disabled via lnet.printk.
572 --------------------------------------------------------------------------------
574 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
576 * Support for kernels:
577 2.6.5-7.283 (SLES 9),
578 2.6.9-55.EL (RHEL 4),
579 2.6.16.46-0.14 (SLES 10),
580 2.6.18.8 vanilla (kernel.org)
581 * Client support for unpatched kernels:
582 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
583 2.6.16 - 2.6.22 vanilla (kernel.org)
584 * Due to recently discovered recovery problems, we do not recommend
585 using patchless RHEL 4 clients with this or any earlier release.
586 * Recommended e2fsprogs version: 1.39.cfs8
587 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
588 * Starting with this release, the ldiskfs backing filesystem required
589 by Lustre is now in its own package, lustre-ldiskfs. This package
590 should be installed. It is versioned separately from Lustre and
591 may be released separately in future.
595 Description: block reactivating mgc import until all deactivates complete
596 Details : Fix race when failing back MDT/MGS to itself (testing)
599 Frequency : at statup only
601 Description: mds_lov_synchronize race leads to various problems
602 Details : simultaneous MDT->OST connections at startup can cause the
603 sync to abort, leaving the OSC in a bad state.
605 Severity : enhancement
607 Description: add optional extra BUILD_VERSION info
608 Details : add a new environment variable (namely LUSTRE_VERS) which allows
609 to override the lustre version.
612 Frequency : 2.6.18 servers only
614 Description: ll_kern_mount() doesn't release the module reference
615 Details : The ldiskfs module reference count never drops down to 0
616 because ll_kern_mount() doesn't release the module reference.
621 Description: server LBUG when using old ost_num_threads parameter
622 Details : Accept the old ost_num_threads parameter but warn that it
623 is deprecated, and fix an off-by-one error that caused an LBUG.
628 Description: Transient SCSI error results in persistent IO issue
629 Details : iobuf->dr_error is not reinitialized to 0 between two
633 Frequency : sometimes when underlying device returns I/O errors
635 Description: OSTs not going read-only during write failures
636 Details : OSTs are not remounted read-only when the journal commit threads
637 get I/O errors because fsfilt_ext3 calls journal_start/stop()
638 instead of the ext3 wrappers.
642 Description: poor connect scaling with increasing client count
643 Details : Don't run filter_grant_sanity_check for more than 100 exports
644 to improve scaling for large numbers of clients.
647 Frequency : SLES10 only
649 Description: sanity-quota.sh quotacheck failed: rc = -22
650 Details : Quotas cannot be enabled on SLES10.
653 Frequency : liblustre clients only
655 Description: getdirentries does not give error when run on compute nodes
656 Details : getdirentries does not fail when the size specified as an argument
657 is too small to contain at least one entry
659 Severity : enhancement
661 Description: Add LNET router traceability for debug purposes
662 Details : If a checksum failure occurs with a router as part of the
663 IO path, the NID of the last router that forwarded the bulk data
664 is printed so it can be identified.
669 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
670 Details : Due to a race condition, liblustre clients were occasionally
673 Severity : enhancement
675 Description: lfs setstripe use optional parameters instead of postional
678 Severity : enhancement
680 Description: Nanosecond timestamp support for ldiskfs
681 Details : The on-disk ldiskfs filesystem has added support for nanosecond
682 resolution timestamps. There is not yet support for this at
683 the Lustre filesystem level.
686 Frequency : during server recovery
688 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
689 Details : request with rq_no_resend flag not awake l_wait_event if they get a
693 Frequency : nfs export on patchless client
695 Description: connectathon hang when test nfs export over patchless client
696 Details : Disconnected dentry cannot be found with lookup, so we do not need
697 to unhash it or make it invalid
700 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
701 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
702 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
707 Description: open req refcounting wrong on reconnect
708 Details : If reconnect happened between getting open reply from server and
709 call to mdc_set_replay_data in ll_file_open, we will schedule
710 replay for unreferenced request that we are about to free.
711 Subsequent close will crash in variety of ways.
712 Check that request is still eligible for replay in
713 mdc_set_replay_data().
718 Description: disable writes to filesystem when reading health_check file
719 Details : the default for reading the health_check proc file has changed
720 to NOT do a journal transaction and write to disk, because this
721 can cause reads of the /proc file to hang and block HA state
722 checking on a healthy but otherwise heavily loaded system. It
723 is possible to return to the previous behaviour during configure
724 with --enable-health-write.
726 Severity : enhancement
728 Description: 64-bit inode version
729 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
730 to the inode. This will be required for version-based recovery.
735 Description: MDS fails to start if a duplicate client export is detected
736 Details : in some rare cases it was possible for a client to connect to
737 an MDS multiple times. Upon recovery the MDS would detect this
738 and fail during startup. Handle this more gracefully.
740 Severity : enhancement
742 Description: Add -o localflock option to simulate old noflock
744 Details : This will achieve local-only flock/fcntl locks
750 Description: log_commit_thread vs filter_destroy race leads to crash
751 Details : Take import reference before releasing llog record semaphore
756 Description: Wrong request locking in request set processing
757 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
758 imp_delayed_list, in this place should be used imp_lock.
761 Frequency : when reconnection
763 Description: Grant Leak when osc reconnect to OST
764 Details : When osc reconnect ost, OST(filter) should check whether it
765 should grant more space to client by comparing fed_grant and
766 cl_avail_grant, and return the granted space to client instead
767 of "new granted" space, because client will call osc_init_grant
768 to update the client grant space info.
771 Frequency : when client reconnect to OST
773 Description: Grant Leak when osc do resend and replay bulk write
774 Details : When osc reconnect to OST, OST(filter)should clear grant info of
775 bulk write request, because the grant info while be sync between
776 OSC and OST when reconnect, and we should ignore the grant info
777 these of resend/replay write req.
782 Description: Grant space more than avaiable left space sometimes.
783 Details : When then OST is about to be full, if two bulk writing from
784 different clients came to OST. Accord the avaliable space of the
785 OST, the first req should be permitted, and the second one
786 should be denied by ENOSPC. But if the seconde arrived before
787 the first one is commited. The OST might wrongly permit second
788 writing, which will cause grant space > avaiable space.
791 Frequency : when client is evicted
793 Description: Grant might be wrongly erased when osc is evicted by OST
794 Details : when the import is evicted by server, it will fork another
795 thread ptlrpc_invalidate_import_thread to invalidate the
796 import, where the grant will be set to 0. While the original
797 thread will update the grant it got when connecting. So if
798 the former happened latter, the grant will be wrongly errased
799 because of this race.
804 Description: Checking Stale with correct fid
805 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
806 to check whether it is stale, because sometimes, we want the
807 enqueue happened anyway, and op_data.fid2 will not be initialized.
809 Severity : enhancement
811 Description: update patchless client
812 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
815 Frequency : only with 2.4 kernel
817 Description: random memory corruption
818 Details : size of struct ll_inode_info is to big for union inode.u and this
819 can be cause of random memory corruption.
824 Description: Memory leak in recovery
825 Details : Lov_mds_md was not free in an error handler in mds_create_object.
826 It should also check obd_fail before fsfilt_start, otherwise if
827 fsfilt_start return -EROFS,(failover mds during mds recovery).
828 then the req will return with repmsg->transno = 0 and rc = EROFS.
829 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
830 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
831 be freed no matter whether fsfilt_commit success or not.
834 Frequency : only with huge count clients
836 Description: Prevents from taking the superblock lock in llap_from_page for
838 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
839 for llap_from_page call in ll_removepage prevents from taking the
840 superblock lock for a soon died page.
845 Description: Not check open intent error before release open handle
846 Details : in some rare cases, the open intent error is not checked before
847 release open handle, which may cause
848 ASSERTION(open_req->rq_transno != 0), because it tries to release
849 the failed open handle.
854 Description: Set cat log bitmap only after create log success.
855 Details : in some rare cases, the cat log bitmap is set too early. and it
856 should be set only after create log success.
860 Description: Accessing a block bevice can re-enable I/O when Lustre is
861 tearing down a device.
862 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
867 Description: service threads may hog cpus when there are a lot of requests
868 Details : Insert cond_resched to give other threads a chance to use some CPU
873 Description: the cat log was not initialized in recovery
874 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
875 unlink log on mds will not be initialized until mds post
876 recovery. And also in mds post recovery, the unlink log will
877 initialization will be done asynchronausly, so there will be race
878 between add unlink log and unlink log initialization.
882 Description: brw_stats were being printed incorrectly
883 Details : brw_stats were being printed as log2 but all of them were not
884 recorded as log2. Also remove some code duplication arising from
885 filter_tally_{read,write}.
889 Frequency : rare, only in recovery.
890 Description: ASSERTION(req->rq_type != LI_POISON) failed
891 Details : imp_lock should be held while iterating over imp_sending_list for
892 prevent destroy request after get timeout in ptlrpc_queue_wait.
896 Description: replay-single.sh test 52 fails
897 Details : A lock's skiplist need to be cleanup when it being unlinked
898 from its resource list.
902 Description: Short directio read returns full requested size rather than
904 Details : Direct I/O operations should return actual amount of bytes
905 transferred rather than requested size.
907 Severity : enhancement
909 Description: metadata RPC reduction (e.g. for rm performance)
910 Details : decrease the amount of synchronous RPC between clients and servers
911 by canceling conflicing lock before the operation on the client
912 and packing thier handles into the main operation RPC to server.
914 Severity : enhancement
916 Description: Async OSC create to avoid the blocking unnecessarily.
917 Details : If a OST has no remain object, system will block on the creating
918 when need to create a new object on this OST. Now, ways use
919 pre-created objects when available, instead of blocking on an
920 empty osc while others are not empty. If we must block, we block
921 for the shortest possible period of time.
923 Severity : enhancement
925 Description: refine locking for avoid write wrong info into lov_objid file
926 Details : fix possible races with add new target and write/update data in
929 --------------------------------------------------------------------------------
931 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
936 Frequency : on some architectures
938 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
939 Details : Uninitialized flags sometimes cause configuration commands to
943 Frequency : patchless clients only
945 Description: missing __iget() symbol export
946 Details : The __iget() symbol export is missing. To avoid the need for
947 this on patchless clients the deathrow inode reaper is turned
948 off, and we depend on the VM to clean up old inodes. This
949 dependency was during via the fix for bug 12181.
951 --------------------------------------------------------------------------------
953 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
955 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
956 INTEROPERATE with older versions automatically. In many cases a
957 special upgrade step is needed. Please read the
958 user documentation before upgrading any part of a live system.
959 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
960 version will not interoperate with 1.6 betas before beta5 (1.5.95).
961 * WARNING: Lustre configuration and startup changes are required with
962 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
968 Severity : enhancement
970 Description: MountConf
971 Details : Lustre configuration is now managed via mkfs and mount
972 commands instead of lmc and lconf. New obd types (MGS, MGC)
973 are added for dynamic configuration management. See
974 https://mail.clusterfs.com/wikis/lustre/MountConf for
977 Severity : enhancement
979 Description: dynamic OST addition
980 Details : OSTs can now be added to a live filesystem
982 Severity : enhancement
984 Description: startup order invariance
985 Details : MDTs and OSTs can be started in any order. Clients only
986 require the MDT to complete startup.
988 Severity : enhancement
990 Description: parallel, asynchronous orphan cleanup
991 Details : orphan cleanup is now performed in separate threads for each
992 OST, allowing parallel non-blocking operation.
994 Severity : enhancement
996 Description: optimized stripe assignment
997 Details : stripe assignments are now made based on ost space available,
998 ost previous usage, and OSS previous usage, in order to try
999 to optimize storage space and networking resources.
1001 Severity : enhancement
1003 Description: Permanently set tunables
1004 Details : All writable /proc/fs/lustre tunables can now be permanently
1005 set on a per-server basis, at mkfs time or on a live system.
1007 Severity : enhancement
1009 Description: Lustre message v2
1010 Details : Add lustre message format v2.
1012 Severity : enhancement
1014 Description: client OST exclusion list
1015 Details : Clients can be started with a list of OSTs that should be
1016 declared "inactive" for known non-responsive OSTs.
1020 Description: SPEC SFS validation failure on NFS v2 over lustre.
1021 Details : Changes the blocksize for regular files to be 2x RPC size,
1022 and not depend on stripe size.
1024 Severity : enhancement
1026 Description: Multiple MD RPCs in flight.
1027 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1028 To avoid overly-overloading MDS, introduce a limit on number of
1029 MDS RPCs in flight for a single client and add /proc controls
1030 to adjust this limit.
1032 Severity : enhancement
1034 Description: client read/write statistics
1035 Details : Add client read/write call usage stats for performance
1036 analysis of user processes.
1037 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1038 file access. extents_stats shows chunk size distribution.
1039 extents_stats_per_process show chunk size distribution per
1042 Severity : enhancement
1044 Description: mds statistics
1045 Details : Add detailed mds operations statistics in
1046 /proc/fs/lustre/mds/*/stats.
1050 Description: Failure of copying files with lustre special EAs.
1051 Details : Client side always return success for setxattr call for lustre
1052 special xattr (currently only "trusted.lov").
1057 Description: Refcount LNET uuids
1058 Details : The global LNET uuid list grew linearly with every startup;
1059 refcount repeated list entries instead of always adding to
1062 Severity : enhancement
1064 Description: Dynamic service threads
1065 Details : Within a small range, start extra service threads
1066 automatically when the request queue builds up.
1069 Frequency : mixed-endian client/server environments
1071 Description: mixed-endian crashes
1072 Details : The new msg_v2 system had some failures in mixed-endian
1075 Severity : enhancement
1077 Description: Easy OST removal
1078 Details : OSTs can be permanently deactivated with e.g. 'lctl
1079 conf_param lustre-OST0001.osc.active=0'
1081 Severity : enhancement
1083 Description: MGS proc entries
1084 Details : Added basic proc entries for the MGS showing what filesystems
1087 Severity : enhancement
1089 Description: provide MGS failover
1090 Details : Added config lock reacquisition after MGS server failover.
1092 Severity : enhancement
1094 Description: add Linux 2.4 support
1095 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1099 Description: a large application tries to do I/O to the same resource and dies
1100 in the middle of it.
1101 Details : Check the req->rq_arrival time after the call to
1102 ost_brw_lock_get(), but before we do anything about
1103 processing it & sending the BULK transfer request. This
1104 should help move old stale pending locks off the queue as
1105 quickly as obd_timeout.
1108 Frequency : when an incorrect nid is specified during startup
1110 Description: ptlrpc connect to non-existant node causes kernel crash
1111 Details : LNET can't be re-entered from an event callback, which
1112 happened when we expire a message after the export has been
1113 cleaned up. Instead, hand the zombie cleanup off to another
1116 Severity : enhancement
1118 Description: plain/inodebits lock performance improvement
1119 Details : Grouping plain/inodebits in granted list by their request modes
1120 and bits policy, thus improving the performance of search through
1124 Frequency : only if OST filesystem is corrupted
1126 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1127 Details : for a short time RPCs with bulk IO are in the replay list,
1128 but replay of bulk IOs is unimplemented. If the OST filesystem
1129 is corrupted due to disk cache incoherency and then replay is
1130 started it is possible to trip an assertion. Avoid putting
1131 committed RPCs into the replay list at all to avoid this issue.
1134 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1136 Description: System hang on startup
1137 Details : This bug allowed the liblustre (e.g. catamount) client to
1138 return to the app before handling all startup RPCs. This
1139 could leave the node unresponsive to lustre network traffic
1140 and manifested as a server ptllnd timeout.
1142 Severity : enhancement
1144 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1145 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1146 boolean to control whether to print peer debug info when a
1147 client's RPC times out.
1150 Frequency : only for kernels with patches from Lustre below 1.4.3
1152 Description: Remove old rdonly API
1153 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1156 Frequency : only for devices with external journals
1158 Description: Set external device read-only also
1159 Details : During a commanded failover stop, we set the disk device
1160 read-only while the server shuts down. We now also set any
1161 external journal device read-only at the same time.
1164 Frequency : when upgrading from 1.4 while trying to change parameters
1166 Description: The wrong (new) MDC name was used when setting parameters for
1167 upgraded MDT's. Also allows changing of OSC (and MDC)
1168 parameters if --writeconf is specified at tunefs upgrade time.
1171 Frequency : when setting specific ost indicies
1173 Description: QOS code breaks on skipped indicies
1174 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1175 created with --index need not be sequential.
1177 Severity : enhancement
1179 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1180 Details : The uninit_groups feature works in conjunction with the kernel
1181 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1182 up the pass1 processing of e2fsck. This is a read-only feature
1183 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1184 mount filesystems that have had this feature enabled.
1186 Severity : enhancement
1188 Description: Improve multi-block allocation algorithm to avoid fragmentation
1189 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1190 allocation locality and avoid filesystem fragmentation.
1192 ------------------------------------------------------------------------------
1194 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1196 * Support for kernels:
1197 2.4.21-47.0.1.EL (RHEL 3)
1198 2.6.5-7.283 (SLES 9)
1199 2.6.9-42.0.10.EL (RHEL 4)
1200 2.6.12.6 vanilla (kernel.org)
1201 2.6.16.27-0.9 (SLES 10)
1202 * Recommended e2fsprogs version: 1.39.cfs5
1204 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1208 Frequency : occasional, depends on client load and configuration
1209 Bugzilla : 12181, 12203
1210 Description: data loss for recently-modified files
1212 Details : In some cases it is possible that recently written or created
1213 files may not be written to disk in a timely manner (this should
1214 normally be within 30s unless client IO load is very high).
1215 The problem appears as zero-length files or files that are a
1216 multiple of 1MB in size after a client crash or client eviction
1217 that are missing data at the end of the file.
1219 This problem is more likely to be hit on clients where files are
1220 repeatedly created and unlinked in the same directory, clients
1221 have a large amount of RAM, have many CPUs, the filesystem has
1222 many OSTs, the clients are rebooted frequently, and/or the files
1223 are not accessed by other nodes after being written.
1225 The presence of the problem can be detected by looking at
1226 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1227 smaller than the second (nr_unused) then dirty files will not
1228 be flushed automatically to disk. "sync; sleep 10" should be
1229 run several times on the node before unmounting it to update
1230 Lustre (this is also safe to run on nodes without this problem).
1232 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1233 that can cause this same problem, so customers using that kernel
1234 also need to update the kernel in addition to Lustre. In order
1235 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1237 It is normal that files written just before a client crash (less
1238 than 30s) may not yet have been flushed to disk, even for local
1242 Frequency : frequent on thin XT3 nodes
1244 Description: UUID collision on thin XT3 Linux nodes
1245 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1246 randomly, since we relied on an insufficiently-seeded PRNG.
1251 Description: OSS hangs after "All ost request buffers busy"
1252 Details : A deadlock between quota and journal operations caused OSS
1253 hangs after printing "All ost request buffers busy."
1256 Frequency : always on liblustre builds
1258 Description: Cleanup compiler warnings on liblustre
1261 Frequency : always on liblustre builds on XT3
1263 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1267 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1268 > 32000 subdirectories
1269 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1270 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1271 subdirectory count crosses 32000. This will aid e2fsck to
1272 correctly handle more than 32000 subdirectories.
1275 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1277 Description: System hang on startup
1278 Details : This bug allowed the liblustre (e.g. catamount) client to
1279 return to the app before handling all startup RPCs. This
1280 could leave the node unresponsive to lustre network traffic
1281 and manifested as a server ptllnd timeout.
1283 Severity : enhancement
1285 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1286 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1287 boolean to control whether to print peer debug info when a
1288 client's RPC times out.
1293 Description: make O_SYNC working on 2.6 kernels
1294 Details : 2.6 kernels use different method for mark pages for write,
1295 so need add a code to lustre for O_SYNC work.
1300 Description: Failure to close file and release space on NFS
1301 Details : Put inode details into lock acquired in ll_intent_file_open.
1302 Use mdc_intent_lock in ll_intent_open to properly
1303 detect all kind of errors unhandled by mdc_enqueue.
1308 Description: proc file read during shutdown sometimes raced obd removal,
1310 Details : Add lock to prevent obd access after proc file removal.
1313 Frequency : Only for files larger than 4GB on 32-bit clients.
1315 Description: improperly doing page alignment of locks
1316 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1317 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1322 Description: RPCs being resent when they shouldn't be
1323 Details : Some RPCs that should not be resent are being resent. This
1324 can cause inconsistencies in the RPC state machine. Do not
1325 resend such requests.
1328 Frequency : rare, only with NFS export
1330 Description: Crash on NFS re-export node
1331 Details : under very unusual load conditions an assertion is hit in
1332 ll_intent_file_open()
1335 Frequency : only if OST filesystem is corrupted
1337 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1338 Details : for a short time RPCs with bulk IO are in the replay list,
1339 but replay of bulk IOs is unimplemented. If the OST filesystem
1340 is corrupted due to disk cache incoherency and then replay is
1341 started it is possible to trip an assertion. Avoid putting
1342 committed RPCs into the replay list at all to avoid this issue.
1347 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1348 Details : Large single O_DIRECT read and write calls can fail to allocate
1349 a sufficiently large buffer to process the request. In case of
1350 allocation failure the allocation is retried with a smaller
1351 buffer and broken into smaller requests.
1353 Severity : enhancement
1355 Description: Add -o localflock option to simulate old noflock behaviour.
1356 Details : This will achieve local-only flock/fcntl locks coherentness.
1361 Description: versioning check is incomplete
1362 Details : Checking the version difference of client vs. server, report
1363 error if the gap is too big.
1368 Description: add support PG_writeback bit
1369 Details : add support for PG_writeback bit for Lustre, for more carefull
1370 work with page cache in 2.6 kernel. This also fix some deadlocks
1371 and remove hack for work O_SYNC with 2.6 kernel.
1373 Severity : enhancement
1375 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1376 Details : The uninit_groups feature works in conjunction with the kernel
1377 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1378 up the pass1 processing of e2fsck. This is a read-only feature
1379 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1380 mount filesystems that have had this feature enabled.
1382 Severity : enhancement
1384 Description: Improve multi-block allocation algorithm to avoid fragmentation
1385 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1386 allocation locality and avoid filesystem fragmentation.
1388 ------------------------------------------------------------------------------
1390 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1392 * Support for kernels:
1393 2.6.9-42.0.3.EL (RHEL 4)
1394 2.6.5-7.276 (SLES 9)
1395 2.4.21-47.0.1.EL (RHEL 3)
1396 2.6.12.6 vanilla (kernel.org)
1397 2.6.16.21-0.8 (SLES10)
1398 * Recommended e2fsprogs version: 1.39.cfs2-0
1400 * The backwards-compatible /proc/sys/portals symlink has been removed
1401 in this release. Before upgrading, please ensure that you change
1402 any configuration scripts or /etc/sysctl.conf files that access
1403 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1404 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1405 in advance of the upgrade on any system running Lustre 1.4.6 or
1406 newer, since /proc/sys/lnet was added in that version.
1407 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1411 Frequency : only when quota is used
1413 Description: avoid scanning export list for quota master
1414 Details : Change the algorithms to avoid scanning export list in order
1415 to improve the efficiency.
1418 Frequency : MDS failover only, very rarely
1420 Description: "went back in time" messages on mds failover
1421 Details : The greatest transno may be lost when the current operation
1422 finishes with an error (transno==0) and the client's last_rcvd
1423 record is over-written. Save the greatest transno in the
1424 mds_last_transno for this case.
1427 Frequency : always for specific kernels and striping counts
1429 Description: client may get "Matching packet too big" without ACL support
1430 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1431 when trying to access files in certain configurations. The
1432 clients should in fact be denied when mounting because they do
1433 not understand ACLs.
1436 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1438 Description: many clients connecting with IO in progress causes connect timeouts
1439 Details : Avoid synchronous journal commits to avoid delays caused by many
1440 clients connecting/disconnecting when bulk IO is in progress.
1441 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1442 OST_IO_PORTAL to avoid delays behind potentially many pending
1446 Frequency : occasionally with multiple writers to a single file
1448 Description: shared writes to file may result in wrong size reported by stat()
1449 Details : Allow growing of kms when extent lock is cancelled
1452 Frequency : always with random mmap IO to multi-striped file
1454 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1455 Details : Only if the hole is at the end of OST object so that kms is too
1456 small. Fix is to increase kms accordingly in ll_nopage.
1459 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1461 Description: writes to a missing object would leak memory on the OST
1462 Details : If there is an inconsistency between the MDS and OST filesystems,
1463 such that the MDS references an object that doesn't exist, writes
1464 to that object will leak memory due to incorrect cleanup in the
1465 error handling path, eventually running out of memory on the OST.
1470 Description: Creating too long symlink causes lustre errors
1471 Details : Check symlink and name lengths before sending requests to MDS.
1474 Frequency : only if flock is enabled (not on by default)
1476 Description: posix locks not released on fd closure on 2.6.9+
1477 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1478 kernels, this caused such locks not to be released on fd close and
1479 then assertions on fs unmount about still used locks.
1482 Frequency : MDS failover only, very rarely
1484 Description: clients may get ASSERTION(granted_lock != NULL)
1485 Details : When request was taking a long time, and a client was resending
1486 a getattr by name lock request. The were multiple lock requests
1487 with the same client lock handle and
1488 mds_getattr_name->fixup_handle_for_resent_request found one of the
1489 lock handles but later failed with ASSERTION(granted_lock != NULL).
1494 Description: handle->h_buffer_credits > 0, assertion failure
1495 Details : h_buffer_credits is zero after truncate, causing assertion
1496 failure. This patch extends the transaction or creates a new
1500 Frequency : NFS re-export or patchless client
1501 Bugzilla : 11179, 10796
1502 Description: Crash on NFS re-export node (__d_move)
1503 Details : We do not want to hash the dentry if we don't have a lock.
1504 But if this dentry is later used in d_move, we'd hit uninitialised
1505 list head d_hash, so we just do this to init d_hash field but
1506 leave dentry unhashed.
1509 Frequency : NFS re-export or patchless client
1511 Description: NFS exports has problem with symbolic link
1512 Details : lustre client didn't properly install dentry when re-exported
1513 to NFS or running patchless client.
1516 Frequency : NFS re-export or patchless client
1518 Description: Various nfs/patchless fixes.
1519 Details : fixes reuse disconected alias for lookup process - this fixes
1520 warning "find_exported_dentry: npd != pd",
1521 fix permission error with open files at nfs.
1522 fix apply umask when do revalidate.
1525 Frequency : occasional
1527 Description: Crash on NFS re-export node
1528 Details : calling clear_page() on the wrong pointer triggered oops in
1529 generic_mapping_read().
1532 Frequency : rarely, using O_DIRECT IO
1534 Description: unaligned directio crashes client with LASSERT
1535 Details : check for unaligned buffers before trying any requests.
1538 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1540 Description: stale data returned from RAID cache
1541 Details : If only a small amount of IO is done to the RAID device before
1542 reading it again it is possible to get stale data from the RAID
1543 cache instead of reading it from disk.
1546 Frequency : always for sles10 kernel
1548 Description: sles10 support
1549 Details : ll_follow_link: compile fixes and using of nd_set_link
1550 under newer kernels.
1553 Frequency : depends on arch, kernel and compiler version, always on sles10
1556 Description: recursive or deep enough symlinks cause stack overflow
1557 Details : getting rid of large stack-allocated variable in
1561 Frequency : depends on hardware
1563 Description: lustre write performance loss in the SLES10 kernel
1564 Details : the performance loss is caused by using of write barriers in the
1565 ext3 code. The SLES10 kernel turns barrier support on by
1566 default. The fix is to undo that change for ldiskfs.
1568 ------------------------------------------------------------------------------
1570 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1572 * Support for kernels:
1573 2.6.9-42.0.3EL (RHEL 4)
1574 2.6.5-7.276 (SLES 9)
1575 2.4.21-47.0.1.EL (RHEL 3)
1576 2.6.12.6 vanilla (kernel.org)
1580 Frequency : quota enabled and large files being deleted
1582 Description: releasing more than 4GB of quota at once hangs OST
1583 Details : If a user deletes more than 4GB of files on a single OST it
1584 will cause the OST to spin in an infinite loop. Release
1585 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1590 Description: statfs data retrieved from /proc may be stale or zero
1591 Details : When reading per-device statfs data from /proc, in the
1592 {kbytes,files}_{total,free,avail} files, it may appear
1593 as zero or be out of date.
1596 Frequency : systems with MD RAID1 external journal devices
1598 Description: lconf's call to blkid is confused by RAID1 journal devices
1599 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1600 of returning all block devices that match the journal UUID.
1603 Frequency : always, for aggregate stripe size over 4GB
1605 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1606 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1607 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1610 Frequency : always if "lfs find" used on a local file/directory
1612 Description: "lfs find" segfaults if used on a local file/directory
1613 Details : The case where a directory component was not specified wasn't
1614 handled correctly. Handle this properly.
1617 Frequency : always on ppc64
1619 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1620 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1621 perform endianness conversion.
1624 Frequency : rarely (truncate to non-zero file size after write under load)
1625 Bugzilla : 10730, 10687
1626 Description: Files padded with zeros to next 4K multiple
1627 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1628 it is possible that files that are truncated to a non-zero size
1629 immediately after being written are filled with zero bytes beyond
1630 the truncated size. No file data is lost.
1632 Severity : enhancement
1634 Description: Allow recovery/failover for liblustre clients.
1635 Details : liblustre clients were unaware of failover configurations until
1638 Severity : enhancement
1640 Description: user file locks should fail when not mounting with flock option
1641 Details : Set up an error-returning stub in ll_file_operations.lock field
1642 to prevent incorrect behaviour when client is mounted without
1643 flock option. Also, set up properly f_op->flock field for
1647 Frequency : always on ia64
1649 Description: "lfs df" loops on printing out MDS statfs information
1650 Details : The obd_ioctl_data was not initialized and in some systems
1651 this caused a failure during the ioctl that did not return
1652 an error. Initialize the struct and return an error on failure.
1655 Frequency : SLES 9 only
1657 Description: Error of copying files with lustre special EAs as root
1658 Details : Client side always return success for setxattr call for lustre
1659 special xattr (currently only "trusted.lov").
1662 Frequency : rarely on clusters with both ia64+i386 clients
1664 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1665 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1666 clients are doing concurrent writes to the same file it is
1667 possible that the ia64 clients may LASSERT because the OST
1668 extent locks are not PAGE_SIZE aligned. Ensure that grown
1669 locks are always aligned on the request boundary.
1672 Frequency : specific use, occasional
1674 Description: Overwriting in use executable truncates on-disk binary image
1675 Details : If one node attempts to overwrite an executable in use by
1676 another node, we now correctly return ETXTBSY instead of
1677 truncating the file.
1679 Severity : enhancement
1681 Description: Async OSC create to avoid the blocking unnecessarily.
1682 Details : If a OST has no remain object, system will block on the creating
1683 when need to create a new object on this OST. Now, ways use
1684 pre-created objects when available, instead of blocking on an
1685 empty osc while others are not empty. If we must block, we block
1686 for the shortest possible period of time.
1691 Description: chmod on Lustre root is propagated to other clients
1692 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1693 invalid by the follow_mount time.
1696 Frequency : liblustre clients only
1698 Description: Race in 'instant cancel' lock handling could lead to such locks
1699 never to be granted in case of SMP MDS
1700 Details : Do not destroy not yet granted but cbpending locks in
1704 Frequency : replay/resend of open
1706 Description: non null lock assetion failure in mds_intent_policy
1707 Details : Trying to replay/resend lockless open requests resulted in
1708 mds_open() returning 0 with no lock. Now it sets a flag if
1709 a lock is going to be returned.
1711 Severity : enhancement
1713 Description: Checksum enhancements
1714 Details : New checksum enhancements allow for resending RPCs that failed
1717 Severity : enhancement
1719 Description: Tunables on number of dirty pages in cacche
1720 Details : Allow to set limit on number of dirty pages cached.
1725 Description: client crash on unmount - lock still has references
1726 Details : In some error handling cases it was possible to leak a lock
1727 reference on a client while accessing a file. This was not
1728 harmful to the client during operation, but would cause the
1729 client to crash when the filesystem is unmounted.
1732 Frequency : specific case, rare
1734 Description: ETXTBSY on mds though file not in use by client
1735 Details : ETXTBSY is no longer incorrectly returned when attempting to
1736 chmod or chown a directory that the user previously tried to
1737 execute or a currently-executing binary.
1740 Frequency : extremely rare except on liblustre-based clients
1742 Description: Lustre space not freed when files are deleted
1743 Details : Clean up open-unlinked files after client eviction. Previously
1744 the unlink was skipped and the files remained as orphans.
1749 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1750 Details : In some cases it was possible to send a blocking callback to
1751 a client doing a glimpse, even though that client didn't get
1752 a lock granted. When the glimpse lock is cancelled on the OST
1753 the freed lock is left on the waiting list and corrupted the list.
1756 Frequency : all core dumps
1758 Description: Broke core dumps to lustre
1759 Details : Negative dentry may be unhashed if parent does not have UPDATE
1760 lock, but some callers, e.g. do_coredump, expect dentry to be
1761 hashed after successful create, hash it in ll_create_it.
1763 ------------------------------------------------------------------------------
1765 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1767 * Support for kernels:
1768 2.6.9-42.0.2.EL (RHEL 4)
1769 2.6.5-7.276 (SLES 9)
1770 2.4.21-40.EL (RHEL 3)
1771 2.6.12.6 vanilla (kernel.org)
1775 Frequency : always on RHEL 3
1777 Description: Number of open files grows over time
1778 Details : The number of open files grows over time, whether or not
1779 Lustre is started. This was due to a filp leak introduced
1780 by one of our kernel patches.
1782 ------------------------------------------------------------------------------
1784 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1786 * Support for kernels:
1787 2.6.9-42.EL (RHEL 4)
1788 2.6.5-7.276 (SLES 9)
1789 2.4.21-40.EL (RHEL 3)
1790 2.6.12.6 vanilla (kernel.org)
1795 Bugzilla : 5719, 9635, 9792, 9684
1796 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1797 Details : If a server is under heavy load and cannot reply to new
1798 connection requests before the client resends the (re)connect,
1799 the connection handling code can behave badly if two service
1800 threads are concurrently handing separate (re)connections from
1801 the same client. Add better locking to the connection handling
1802 code, and ensure that only a single connection will be processed
1803 for a given client UUID, even if the lock is dropped.
1805 Severity : enhancement
1807 Description: add TCP zero-copy support to kernel
1808 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1809 sends if the hardware supports scatter-gather and checksumming.
1810 This allows socklnd to do client-write and server-read more
1811 efficiently and reduce CPU utilization from skbuf copying.
1814 Frequency : only if NFS exporting from client
1816 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1817 Details : If setting attributes on a file created under NFS that had
1818 never been opened it would be possible to oops the client
1819 if the file had no objects.
1822 Frequency : always for liblustre
1824 Description: liblustre client does MDS+OSTs setattr RPC for each write
1825 Details : When doing a write from a liblustre client, the client
1826 incorrectly issued an RPC to the MDS and each OST the file was
1827 striped over in order to update the timestamps. When writing
1828 with small chunks and many clients this could overwhelm the MDS
1829 with RPCs. In all cases it would slow down the write because
1830 these RPCs are unnecessary.
1832 Severity : enhancement
1834 Description: allow number of MDS service threads to be changed at module load
1835 Details : It is now possible to change the number of MDS service threads
1836 running. Adding "options mds mds_num_threads={N}" to the MDS's
1837 /etc/modprobe.conf will set the number of threads for the next
1838 time Lustre is restarted (assuming the "mds" module is also
1839 reloaded at that time). The default number of threads will
1840 stay the same, 32 for most systems.
1845 Description: OST crash if filesystem is unformatted or corrupt
1846 Details : If an OST is started on a device that has never been formatted
1847 or if the filesystem is corrupt and cannot even mount then the
1848 error handling cleanup routines would dereference a NULL pointer.
1853 Description: NULL pointer deref in llap_from_page.
1854 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1855 incorrect) mapping if the page was truncated/reclaimed while it was
1856 searched for. Check for this condition and skip such pages when
1857 doing readahead. Introduce extra check to llap_from_page() to
1858 verify page->mapping->host is non-NULL (so page is not anonymous).
1861 Frequency : Sometimes when using sys_sendfile
1863 Description: "page not covered by a lock" warnings from ll_readpage
1864 Details : sendfile called ll_readpage without right page locks present.
1865 Now we introduced ll_file_sendfile that does necessary locking
1866 around call to generic_file_sendfile() much like we do in
1870 Frequency : with certain MDS communication failures at client mount time
1872 Description: NULL pointer deref after failed client mount
1873 Details : a client connection request may delayed by the network layer
1874 and not be sent until after the PTLRPC layer has timed out the
1875 request. If the client fails the mount immediately it will try
1876 to clean up before the network times out the request. Add a
1877 reference from the request import to the obd device and delay
1878 the cleanup until the network drops the request.
1881 Frequency : occasionally during client (re)connect
1883 Description: assertion failure during client (re)connect
1884 Details : processing a client connection request may be delayed by the
1885 client or server longer than the client connect timeout. This
1886 causes the client to resend the connection request. If the
1887 original connection request is replied in this interval, the
1888 client may trip an assertion failure in ptlrpc_connect_interpret()
1889 which thought it would be the only running connect process.
1892 Frequency : only with obd_echo servers and clients that are rebooted
1894 Description: kernel BUG accessing uninitialized data structure
1895 Details : When running an obd_echo server it did not start the ping_evictor
1896 thread, and when a client was evicted an uninitialized data
1897 structure was accessed. Start the ping_evictor in the RPC
1898 service startup instead of the OBD startup.
1900 Severity : enhancement
1901 Bugzilla : 10193 (patchless)
1902 Description: Remove dependency on various unexported kernel interfaces.
1903 Details : No longer need reparent_to_init, exit_mm, exit_files,
1904 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1907 Frequency : rare (only users of deprecated and unsupported LDAP config)
1909 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1910 Details : LDAP apparently contains 'lustreName' attributes instead of
1911 'name'. A simple remapping of the name is sufficient.
1914 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1916 Description: waiting_locks_callback trips kernel BUG if client is evicted
1917 Details : Running with the dump_on_timeout debug flag turned on makes
1918 it possible that the waiting_locks_callback() can try to dump
1919 the Lustre kernel debug logs from an interrupt handler. Defer
1920 this log dumping to the expired_lock_main() thread.
1922 Severity : enhancement
1924 Description: Support NFS exporting on 2.6 kernels.
1925 Details : Implement non-rawops metadata methods for NFS server to use without
1926 changing NFS server code.
1929 Frequency : very rare (synthetic metadata workload only)
1931 Description: two racing renames might cause an MDS thread to deadlock
1932 Details : Running the "racer" program may cause one MDS thread to rename
1933 a file from being the source of a rename to being the target of
1934 a rename at exactly the same time that another thread is doing
1935 so, and the second thread has already enqueued these locks after
1936 doing a lookup of the target and is trying to relock them in
1937 order. Ensure that we don't try to re-lock the same resource.
1940 Frequency : only very large systems with liblustre clients
1942 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1943 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1944 all OSTs in parallel. This allows the eviction of stale liblustre
1945 clients to proceed much faster than if they were done in series,
1946 and also offers similar improvements for other set_info RPCs.
1951 Description: excessive CPU usage during initial read phase on client
1952 Details : During the initial read phase on a client, it would agressively
1953 retry readahead on the file, consuming too much CPU and impacting
1954 performance (since 1.4.5.8). Improve the readahead algorithm
1955 to avoid this, and also improve some other common cases (read
1956 of small files in particular, where "small" is files smaller than
1957 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1962 Description: MDS crash when receiving packet with unknown intent.
1963 Details : Do not LBUG in unknown intent case, just return -EFAULT
1965 Severity : enhancement
1966 Bugzilla : 9293, 9385
1967 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1968 Details : Do not serialize getattr (non-intent version) and statfs.
1971 Frequency : occasional, when OST network is overloaded/intermittent
1973 Description: client evicted by OST after bulk IO timeout
1974 Details : If a client sends a bulk IO request (read or write) the OST
1975 may evict the client if it is unresposive to its data GET/PUT
1976 request. This is incorrect if the network is overloaded (takes
1977 too long to transfer the RPC data) or dropped the OST GET/PUT
1978 request. There is no need to evict the client at all, since
1979 the pinger and/or lock callbacks will handle this, and the
1980 client can restart the bulk request.
1983 Frequency : Always when mmapping file with no objects
1985 Description: client crashes when mmapping file with no objects
1986 Details : Check that we actually have objects in a file before doing any
1987 operations on objects in ll_vm_open, ll_vm_close and
1993 Description: Request leak when working with deleted CWD
1994 Details : Introduce advanced request refcount tracking for requests
1995 referenced from lustre intent.
1997 Severity : Enhancement
1999 Description: Cache open file handles on client.
2000 Details : MDS now will return special lock along with openhandle, if
2001 requested and client is allowed to hold openhandle, even if unused,
2002 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2003 closing files for every read/write openration.
2005 Severity : Enhancement
2007 Description: Cache open negative dentries on client when possible.
2008 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2009 negative dentries on lock revocation.
2014 Description: Remounting a client read-only wasn't possible with a zconf mount
2015 Details : It wasn't possible to remount a client read-only with llmount.
2017 Severity : enhancement
2018 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2019 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2020 MPICH is included to add Lustre-specific ADIO interfaces.
2021 This is based closely on the UFS ADIO layer and only differs
2022 in file creation, in order to allow the OST striping to be set.
2023 This is user-contributed code and not supported by CFS.
2028 Description: extended inode attributes (immutable, append-only) work improperly
2029 when 2.4 and 2.6 kernels are used on client/server or vice versa
2030 Details : Introduce kernel-independent values for these flags.
2032 Severity : enhancement
2035 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2036 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2037 granularity. Now, Fractional MB granularity is be supported,
2038 this is very useful for low memory system.
2040 Severity : enhancement
2042 Description: Getattr by fid
2043 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2044 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2045 have locks and avoids getattr rpc requests for it on every stat.
2048 Frequency : Always, for filesystems larger than 2TB
2050 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2051 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2052 filesystems larger than 2TB to be mounted with mballoc enabled.
2055 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2057 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2058 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2059 filesystems larger than 2TB is mounted then the kernel will
2060 silently corrupt the start of the filesystem. CONFIG_LBD is
2061 enabled for all CFS-supported kernels, but the possibility of
2062 this happening with a modified kernel config exists.
2064 Severity : enhancement
2066 Description: add client O_DIRECT support for 2.6 kernels
2067 Details : It is now possible to do O_DIRECT reads and writes to files
2068 in the Lustre client mountpoint on 2.6 kernel clients.
2070 Severity : enhancement
2072 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2073 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2074 parallel, not waiting for response from every OST before sending
2075 a rpc to the next OST.
2080 Description: setattr vs write race when updating file timestamps
2081 Details : Client processes that update a file timestamp into the past
2082 right after writing to the file (e.g. tar) it is possible that
2083 the updated file modification time can be reset to the current
2084 time due to a race between processing the setattr and write RPC.
2086 Severity : enhancement
2088 Description: Bring 'lfs find' closer in line with regular Linux find.
2089 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2090 -print0 options and obtains all the needed info through the lustre
2093 Severity : enhancement
2095 Description: support up to 1024 configured devices on one node
2096 Details : change obd_dev array from statically allocated to dynamically
2097 allocated structs as they are first used to reduce memory usage
2102 Description: Flush dirty partially truncated pages during truncate
2103 Details : Immediatelly flush partially truncated pages in filter_setattr,
2104 this way we completely avoid having any pages in page cache on OST
2105 and can retire ugly workarounds during writes to flush such pages.
2110 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2111 Details : protect inode from truncation within vfs_unlink() context
2112 just take a reference before calling vfs_unlink() and release it
2113 when parent's i_sem is free.
2118 Description: last_id value checked outside lock on OST caused LASSERT failure
2119 Details : If there were multiple MDS->OST object precreate requests in
2120 flight, it was possible that the OST's last object id was checked
2121 outside a lock and incorrectly tripped an assertion. Move checks
2122 inside locks, and discard old precreate requests.
2125 Frequency : always, if extents are used on OSTs
2127 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2128 Details : OSTs using the extents format would not zero the high 16 bits of
2129 the index physical block number. This is not a problem for any
2130 OST filesystems smaller than 16TB, and no kernels support ext3
2131 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2132 new/modified files) and can be fixed for existing filesystems
2133 with e2fsprogs-1.39-cfs1.
2138 Description: import connection selection may be incorrect if timer wraps
2139 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2140 import connections to be ignored if the 32-bit jiffies counter
2141 wraps. Use a 64-bit jiffies counter.
2144 Frequency : very large clusters immediately after boot
2146 Description: LNET request buffers exhausted under heavy short-term load
2147 Details : If a large number of client requests are generated on a service
2148 that has previously never seen so many requests it is possible
2149 that the request buffer growth cannot keep up with the spike in
2150 demand. Instead of dropping incoming requests, they are held in
2151 the LND until the RPC service can accept more requests.
2154 Frequency : Sometimes during replay
2156 Description: Assertion failure in ll_local_open after replay.
2157 Details : If replay happened on an open request reply before we were able
2158 to set replay handler, reply will become not swabbed tripping the
2159 assertion in ll_local_open. Now we set the handler right after
2160 recognising of open request
2163 Frequency : very rare
2165 Description: kernel reports "badness in vsnprintf"
2166 Details : Reading from the "recovery_status" /proc file in small chunks
2167 may cause a negative length in lprocfs_obd_rd_recovery_status()
2168 call to vsnprintf() (which is otherwise harmless). Exit early
2169 if there is no more space in the output buffer.
2171 Severity : enhancement
2173 Description: clear OBD RPC statistics by writing to them
2174 Details : It is now possible to clear the OBD RPC statistics by writing
2175 to the "stats" file.
2180 Description: Client mtime is not the same on different clients after utimes
2181 Details : In some cases, the client was using the utimes() syscall on
2182 a file cached on another node. The clients now validate the
2183 ctime from the MDS + OSTs to determine which one is right.
2188 Description: Inability to activate failout mode
2189 Details : lconf script incorrectly assumed that in pythong string's numeric
2190 value is used in comparisons.
2193 Frequency : always with multiple stripes per file
2195 Description: Inefficient object allocation for mutli-stripe files
2196 Details : When selecting which OSTs to stripe files over, for files with
2197 a stripe count that divides evenly into the number of OSTs,
2198 the MDS is always picking the same starting OST for each file.
2199 Return the OST selection heuristic to the original design.
2204 Description: mount failures may take full timeout to return an error
2205 Details : Under some heavy load conditions it is possible that a
2206 failed mount can wait for the full obd_timeout interval,
2207 possibly several minutes, before reporting an error.
2208 Instead return an error as soon as the status is known.
2210 Frequency : quota enabled and large files being deleted
2212 Description: releasing more than 4GB of quota at once hangs OST
2213 Details : If a user deletes more than 4GB of files on a single OST it
2214 will cause the OST to spin in an infinite loop. Release
2215 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2220 Description: statfs data retrieved from /proc may be stale or zero
2221 Details : When reading per-device statfs data from /proc, in the
2222 {kbytes,files}_{total,free,avail} files, it may appear
2223 as zero or be out of date.
2226 Frequency : systems with MD RAID1 external journal devices
2228 Description: lconf's call to blkid is confused by RAID1 journal devices
2229 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2230 of returning all block devices that match the journal UUID.
2233 Frequency : always, for aggregate stripe size over 4GB
2235 Description: assertion fails when trying to use 4GB stripe size
2236 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2237 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2240 Frequency : always on ppc64
2242 Description: the first write on an ext3 filesystem with mballoc got stuck
2243 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2244 perform endianness conversion.
2246 ------------------------------------------------------------------------------
2248 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2250 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2251 INTEROPERATE with older versions automatically. Please read the
2252 user documentation before upgrading any part of a live system.
2253 * WARNING: Lustre networking configuration changes are required with
2254 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2257 * Support for kernels:
2258 2.6.9-22.0.2.EL (RHEL 4)
2259 2.6.5-7.244 (SLES 9)
2260 2.6.12.6 vanilla (kernel.org)
2263 Severity : enhancement
2264 Bugzilla : 7981/8208
2265 Description: Introduced Lustre Networking (LNET)
2266 Details : LNET is new networking infrastructure for Lustre, it includes
2267 a reorganized network configuration mode (see the user
2268 documentation for full details) as well as support for routing
2269 between different network fabrics. Lustre Networking Devices
2270 (LNDs) for the supported network fabrics have also been
2271 created for this new infrastructure.
2273 Severity : enhancement
2274 Description: Introduced Access control lists
2275 Details : clients can set ACLs on files and directories in order to have
2276 more fine-grained permissions than the standard Unix UGO+RWX.
2277 The MDS must be started with the "-o acl" mount option.
2279 Severity : enhancement
2280 Description: Introduced filesystem quotas
2281 Details : Administrators may now establish per-user quotas on the
2284 Severity : enhancement
2286 Description: Configuration change for the XT3
2287 The PTLLND is now used to run Lustre over Portals on the XT3
2288 The configure option(s) --with-cray-portals are no longer used.
2289 Rather --with-portals=<path-to-portals-includes> is used to
2290 enable building on the XT3. In addition to enable XT3 specific
2291 features the option --enable-cray-xt3 must be used.
2296 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2297 Details : A race between allocating a new llcd and re-getting the llcd_lock
2298 allowed another thread to grab newly-allocated llcd.
2300 Severity : enhancement
2302 Description: 2.6 OST async journal commit and locking fix to improve performance
2303 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2304 2.6 kernels are now async as they already were in 2.4 kernels so
2305 that they can commit concurrently with the network bulk transfer.
2306 For block-allocated files the filter allocation semaphore is held
2307 to avoid filesystem fragmentation during allocation. BKL lock
2308 removed for 2.6 xattr operations where it is no longer needed.
2313 Description: lconf incorrectly determined whether two IP networks could talk
2314 Details : In some more complicated routing and multiple-network
2315 configurations, lconf will avoid trying to make a network
2316 connection to a disjoint part of the IP space. It was doing the
2317 math incorrectly for one set of cases.
2322 Description: Fix for potential infinite loop processing records in an llog.
2323 Details : If an llog record is corrupted/zeroed, it is possible to loop
2324 forever in llog_process(). Validate the llog record length
2325 and skip the remainder of the block on error.
2328 Frequency : occasional (liblustre only)
2330 Description: liblustre could not open files whose last component is a symlink
2331 Details : sysio_path_walk() would incorrectly pass the open intent to
2332 intermediate path components.
2335 Frequency : rare (liblustre only with non-standard tuning)
2336 Bugzilla : 7201 (7350)
2337 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2338 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2339 after it has been released, only do this for referenced locks
2341 Severity : enhancement
2343 Description: specifying an (invalid) directory default stripe_size of -1
2344 would reset the directory default striping
2345 Details : stripe_size -1 was used internally to signal directory stripe
2346 removal, now use "all default" to signal dir stripe removal
2347 as a directory striping of "all default" is not useful
2350 Frequency : common for large clusters running liblustre clients
2352 Description: doing an ls when liblustre clients are running is slow
2353 Details : sending a glimpse AST to a liblustre client waits for every AST
2354 to time out, as liblustre clients will not respond. Since they
2355 cannot cache data we refresh the OST lock LVB from disk instead.
2357 Severity : enhancement
2359 Description: doing an ls at the same time as file IO can be slow
2360 Details : enqueue and other "small" requests can be blocked behind many
2361 large IO requests. Create a new OST IO portal for non-IO
2362 requests so they can be processed faster.
2365 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2367 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2368 created file with mode 000
2369 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2370 allow an ftruncate() as a non-root user to a file with mode 000.
2371 Lustre now respects this flag to disable mode checks when
2372 truncating a file owned by the user
2375 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2378 Description: Revoking locks from clients that went dead or catatonic might take
2380 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2381 cancellation of such locks instant on servers without waiting for
2382 any reply from clients. Clients drops these locks when cancel
2383 notification from server is received without replying.
2386 Frequency : liblustre-only, when liblustre client dies or becomes busy
2388 Description: Doing ls on Linux clients can take a long time with active
2390 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2391 granting such locks to it in the first place if possible. Locks
2392 are taken by proxy on the OST during the read or write and
2393 dropped immediately afterward. Add connect flags handling, do
2394 not grant locks to liblustre clients for glimpse ASTs.
2396 Severity : enhancement
2398 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2399 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2400 sequential reads and behaves badly with applications doing random
2401 reads. Improve it by reading ahead at least read region, and
2402 avoiding excessive large RPC for small reads.
2404 Severity : enhancement
2406 Description: Creating more than 1000 files for a single job may cause a load
2407 imbalance on the OSTs if there are also a large number of OSTs.
2408 Details : qos_prep_create() uses an OST index reseed value that is an
2409 even multiple of the number of available OSTs so that if the
2410 reseed happens in the middle of the object allocation it will
2411 still utilize the OSTs as uniformly as possible.
2416 Description: OST or MDS may oops in ping_evictor_main()
2417 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2418 but doesn't restart at beginning of obd_exports_timed list
2421 Severity : enhancement
2423 Description: improve by-nid export eviction on the MDS and OST
2424 Details : allow multiple exports with the same NID to be evicted at one
2425 time without re-searching the exports list.
2428 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2430 Description: MDS may oops in groups_free()
2431 Details : in rare race conditions a newly allocated group_info struct is
2432 freed again, and this can be NULL. The 2.4 compatibility code
2433 for groups_free() checked for a NULL pointer, but 2.6 did not.
2436 Frequency : common for liblustre clients doing little filesystem IO
2437 Bugzilla : 9352, 7313
2438 Description: server may evict liblustre clients accessing contended locks
2439 Details : if a client is granted a lock or receives a completion AST
2440 with a blocking AST already set it would not reply to the AST
2441 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2444 Frequency : lfs setstripe, only systems with more than 160 OSTs
2446 Description: unable to set striping with a starting offset beyond OST 160
2447 Details : llapi_create_file() incorrectly limited the starting stripe
2448 index to the maximum single-file stripe count.
2451 Frequency : LDAP users only
2453 Description: lconf did not handle in-kernel recovery with LDAP properly
2454 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2456 Severity : enhancement
2458 Description: bind OST threads to NUMA nodes to improve performance
2459 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2460 node and do their allocations there to localize memory access
2462 Severity : enhancement
2464 Description: llmount can determine client NID directly from Myrinet (GM)
2465 Details : the client NID code from gmnalnid was moved directly into
2466 llmount, removing the need to use this or specifying the
2467 client NID explicitly when mounting GM clients with zeroconf
2470 Frequency : if client is started with down MDS
2472 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2473 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2474 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2475 ptlrpc_import_delayed_req() because we didn't check if the
2476 request was interrupted, and we also didn't break out of the
2477 event loop if there was no timeout
2482 Description: data loss during non-page-aligned writes to a single file from
2483 both multiple nodes and multiple threads on one node at same time
2484 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2485 inconsistency led to false short-reads, that were cached and later
2486 used by ->prepare_write() to fill in partially written page,
2487 leading to data loss.
2490 Frequency : always, if lconf --abort_recovery used
2492 Description: lconf --abort_recovery fails with 'Operation not supported'
2493 Details : lconf was attempting to abort recovery on the MDT device and not
2496 Severity : enhancement
2498 Description: remove cleanup logs
2499 Details : replace lconf-generated cleanup logs with lustre internal
2500 cleanup routines. Eliminates the need for client-cleanup and
2503 Severity : enhancement
2505 Description: add support for EAs (user and system) on lustre filesystems
2506 Details : it is now possible to store extended attributes in the Lustre
2507 client filesystem, and with the user_xattr mount option it
2508 is possible to allow users to store EAs on their files also
2510 Severity : enhancement
2512 Description: Add possibility (config option) to show minimal available OST free
2514 Details : When compiled with --enable-mindf configure option, statfs(2)
2515 (and so, df) will return least minimal free space available from
2516 all OSTs as amount of free space on FS, instead of summary of
2517 free spaces of all OSTs.
2519 Severity : enhancement
2521 Description: do not expand extent locks acquired on OST-side
2522 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2523 by server: they are not cached anyway.
2526 Frequency : when mmap is used/binaries executed from Lustre
2528 Description: Unmmap pages before throwing them away from read cache.
2529 Details : llap_shrink cache now attempts to unmap pages before discarding
2530 them (if unmapping failed - do not discard). SLES9 kernel has
2531 extra checks that trigger if this unmapping is not done first.
2536 Description: lconf didn't resolve symlinks before checking to see whether a
2537 given mountpoint was already in use
2540 Frequency : when migrating failover services
2541 Bugzilla : 6395, 9514
2542 Description: When migrating a subset of services from a node (e.g. failback
2543 from a failover service node) the remaining services would
2544 time out and evict clients.
2545 Details : lconf --force (implied by --failover) sets the global obd_timeout
2546 to 5 seconds in order to quickly disconnect, but this caused
2547 other RPCs to time out too quickly. Do not change the global
2548 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2550 Severity : enhancement
2551 Frequency : if MDS is started with down OST
2552 Bugzilla : 9439,5706
2553 Description: Allow startup/shutdown of an MDS without depending on the
2554 availability of the OSTs.
2555 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2556 Add appropriate locking and lov-osc refcounts for safe
2557 cleaning. Add osc abort_inflight calls in case the
2558 synchronize never started.
2561 Frequency : occasional (Cray XT3 only)
2563 Description: root not authorized to access files in CRAY_PORTALS environment
2564 Details : The client process capabilities were not honoured on the MDS in
2565 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2566 been accessed by an authorized user then root was able to access
2567 the file on the local client also. The root user capabilities
2568 are now allowed on the MDS, as this environment has secure UID.
2571 Frequency : occasional
2573 Description: ldiskfs "too long searching" message happens too often
2574 Details : A debugging message (otherwise harmless) prints too often on
2575 the OST console. This has been reduced to only happen when
2576 there are fragmentation problems on the filesystem.
2581 Description: Division by zero in statfs when all OSCs are inactive
2582 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2583 lov_statfs divides by value returned by lov_get_stripecnt().
2587 Bugzilla : 9489, 3273
2588 Description: First write from each client to each OST was only 4kB in size,
2589 to initialize client writeback cache, which caused sub-optimal
2590 RPCs and poor layout on disk for the first writen file.
2591 Details : Clients now request an initial cache grant at (re)connect time
2592 and so that they can start streaming writes to the cache right
2593 away and always do full-sized RPCs if there is enough data.
2594 If the OST is rebooted the client also re-establishes its grant
2595 so that client cached writes will be honoured under the grant.
2600 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2601 Details : Now I/O RPCs go to different portal number and (presumably) fast
2602 lock requests (and glimses) and other RPCs get their own service
2603 threads pool that should be able to service those RPCs
2606 Severity : enhancement
2608 Description: Ability to exchange lustre version between client and servers and
2609 issue warnings at client side if client is too old. Also for
2610 liblustre clients there is ability to refuse connection of too old
2612 Details : New 'version' field is added to connect data structure that is
2613 filled with version info. That info is later checked by server and
2617 Frequency : rare, liblustre only.
2618 Bugzilla : 9296, 9581
2619 Description: Two simultaneous writes from liblustre at offset within same page
2620 might proceed at the same time overwriting eachother with stale
2622 Details : I/O lock withing llu_file_prwv was released too early, before data
2623 actually was hitting the wire. Extended lock-holding time until
2624 server acknowledges receiving data.
2627 Frequency : extremely rare. Never observed in practice.
2629 Description: avoid generating lustre_handle cookie of 0.
2630 Details : class_handle_hash() generates handle cookies by incrementing
2631 global counter, and can hit 0 occasionaly (this is unlikely, but
2632 not impossible, because initial value of cookie counter is
2633 selected randonly). Value of 0 is used as a sentinel meaning
2634 "unassigned handle" --- avoid it. Also coalesce two critical
2635 sections in this function into one.
2637 Severity : enhancement
2639 Description: allow liblustre clients to delegate truncate locking to OST
2640 Details : To avoid overhead of locking, liblustre client instructs OST to
2641 take extent lock in ost_punch() on client's behalf. New connection
2642 flag is added to handle backward compatibility.
2644 Severity : enhancement
2645 Bugzilla : 4928, 7341, 9758
2646 Description: allow number of OST service threads to be specified
2647 Details : a module parameter allows the number of OST service threads
2648 to be specified via "options ost ost_num_threads={N}" in the
2649 OSS's /etc/modules.conf or /etc/modprobe.conf.
2653 Bugzilla : 6146, 9635, 9895
2654 Description: servers crash with bad pointer in target_handle_connect()
2655 Details : In rare cases when a client is reconnecting it was possible that
2656 the connection request was the last reference for that export.
2657 We would temporarily drop the export reference and get a new
2658 one, but this may have been the last reference and the export
2659 was just destroyed. Get new reference before dropping old one.
2661 Severity : enhancement
2662 Frequency : if client is started with failover MDS
2664 Description: Allow multiple MDS hostnames in the mount command
2665 Details : Try to read the configuration from all specified MDS
2666 hostnames during a client mount in case the "primary"
2669 Severity : enhancement
2671 Description: Stop sending data to evicted clients as soon as possible.
2672 Details : Check if the client we are about to send or are sending data to
2673 was evicted already. (Check is done every second of waiting,
2674 for which l_wait_event interface was extended to allow checking
2675 of exit condition at specified intervals).
2678 Frequency : rare, normally only when NFS exporting is done from client
2680 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2682 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2685 Frequency : always, if srand() is called before liblustre initialization
2687 Description: Liblustre uses system PRNG disturbing its usage by user application
2688 Details : Introduce internal to lustre fast and high-quality PRNG for
2689 lustre usage and make liblustre and some other places in generic
2690 lustre code to use it.
2692 Severity : enhancement
2693 Bugzilla : 9477, 9557, 9870
2694 Description: Verify that the MDS configuration logs are updated when xml is
2695 Details : Check if the .xml configuration logs are newer than the config
2696 logs stored on the MDS and report an error if this is the case.
2697 Request --write-conf, or allow starting with --old_conf.
2699 Severity : enhancement
2701 Description: Handle symlinks in the path when checking if Lustre is mounted.
2702 Details : Resolve intermediate symlinks when checking if a client has
2703 mounted a filesystem to avoid duplicate client mounts.
2708 Description: lconf can hit an error exception but still return success.
2709 Details : The lconf command catches the Command error exception at the top
2710 level script context and will exit with the associated exit
2711 status, but doesn't ensure that this exit status is non-zero.
2716 Description: failure of ptlrpc thread startup can cause oops
2717 Details : Starting a ptlrpc service thread can fail if there are a large
2718 number of threads or the server memory is very fragmented.
2719 Handle this without oopsing.
2722 Frequency : always, only if liblustre and non-default acceptor port was used
2724 Description: liblustre cannot connect to servers with non-default acceptor port
2725 Details : tcpnal_set_default_params() was not called and was therefore
2726 ignoring the environment varaible TCPNAL_PORT, as well as other
2727 TCPNAL_ environment variables
2732 Description: two objects could be created on the same OST for a single file
2733 Details : If an OST is down, in some cases it was possible to create two
2734 objects on a single OST for a single file. No problems other
2735 than potential performance impact and spurious error messages.
2739 Bugzilla : 5681, 9562
2740 Description: Client may oops in ll_unhash_aliases
2741 Details : Client dcache may become inconsistent in race condition.
2742 In some cases "getcwd" can fail if the current directory is
2748 Description: Inode refcounting problems in NFS export code
2749 Details : link_raw functions used to call d_instantiate without obtaining
2750 extra inode reference first.
2754 Bugzilla : 9942, 9903
2755 Description: Referencing freed requests leading to crash, memleaks with NFS.
2756 Details : We used to require that call to ll_revalidate_it was always
2757 followed by ll_lookup_it. Also with revalidate_special() it is
2758 possible to call ll_revalidate_it() twice for the same dentry
2759 even if first occurence returned success. This fix changes semantic
2760 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2761 reference on a request referred from the intent.
2762 ll_intent_release() then releases such a request.
2765 Frequency : rare, normally benchmark loads only
2767 Description: unlinked inodes were kept in memory on the client
2768 Details : If a client is repeatedly creating and unlinking files it
2769 can accumulate a lot of stale inodes in the inode slab cache.
2770 If there is no other client load running this can cause the
2771 client node to run out of memory. Instead flush old inodes
2772 from client cache that have the same inode number as a new inode.
2775 Frequency : SLES9 2.6.5 kernel and long filenames only
2776 Bugzilla : 9969, 10379
2777 Description: utime reports stale NFS file handle
2778 Details : SLES9 uses out-of-dentry names in some cases, which confused
2779 the lustre dentry revalidation. Change it to always use the
2783 Frequency : rare, unless heavy write-truncate concurrency is continuous
2784 Bugzilla : 4180, 6984, 7171, 9963, 9331
2785 Description: OST becomes very slow and/or deadlocked during object unlink
2786 Details : filter_destroy() was holding onto the parent directory lock
2787 while truncating+unlinking objects. For very large objects this
2788 may block other threads for a long time and slow overall OST
2789 responsiveness. It may also be possible to get a lock ordering
2790 deadlock in this case, or run out of journal credits because of
2791 the combined truncate+unlink. Solution is to do object truncate
2792 first in one transaction without parent lock, and then do the
2793 final unlink in a new transaction with the parent lock. This
2794 reduces the lock hold time dramatically.
2797 Frequency : rare, 2.4 kernels only
2799 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2800 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2801 before unmounting their filesystem in order to avoid deadlock.
2802 The kernel_locked() function in 2.4 kernels only checks whether
2803 the kernel lock is held, not whether it is this process that is
2804 holding it as 2.6 kernels do.
2809 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2810 Details : The client ptlrpc code may be trying to reconnect to a down
2811 server before a previous connection attempt has timed out.
2812 Increase the reconnect interval to be longer than the connection
2813 timeout interval to avoid sending duplicate connections to
2817 Frequency : echo_client brw_test command
2819 Description: fix echo_client to work with OST preallocated code
2820 Details : OST preallocation code (5137) didn't take echo_client IO path
2821 into account: echo_client calls filter methods outside of any
2822 OST thread and, hence, there is no per-thread preallocated
2823 pages and buffers to use. Solution: hijack pga pages for IO. As
2824 a byproduct, this avoids unnecessary data copying.
2828 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2829 Description: Client can oops in mdc_commit_close() after open replay
2830 Details : It was possible for the MDS to return an open request with no
2831 transaction number in mds_finish_transno() if the client was
2832 evicted, but without actually returning an error. Clients
2833 would later try to replay that open and may trip an assertion
2834 Simplify the client close codepath, and always return an error
2835 from the MDS in case the open is not successful.
2838 Frequency : rare, 2.6 OSTs only
2840 Description: OST may deadlock under high load on fragmented files
2841 Details : If there was a heavy load and highly-fragmented OST filesystems
2842 it was possible to have all the OST threads deadlock waiting on
2843 allocation of biovecs, because the biovecs were not released
2844 until the entire RPC IO was completed. Instead, release biovecs
2845 as soon as they are complete to ensure forward IO progress.
2847 Severity : enhancement
2849 Description: Support for specifying external journal device at mount
2850 Details : If an OST or MDS device is formatted with an external journal
2851 device, this device major/minor is stored in the ext3 superblock
2852 and may not be valid for failover. Allow detecting and
2853 specifying the external journal at mount time.
2858 Description: Mounting an MDS with pending unlinked files may cause oops
2859 Details : target_finish_recovery() calls mds_postrecov() which returned
2860 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2861 considers this an error and immediately begins cleaning up the
2862 lov, just after starting the mds_lov process
2864 Severity : enhancement
2866 Description: Implement 'lfs df' to report actual free space on per-OST basis
2867 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2868 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2869 usage of objects; '-h' to report in human readable format.
2871 ------------------------------------------------------------------------------
2873 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2880 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2881 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2882 at startup during orphan unlinking. Instead, skip buddy update
2883 before setup, as it will be regenerated after recovery is complete.
2888 Description: If an OST is inactive, its locks might reference stale inodes.
2889 Details : lov_change_cbdata() must iterate over all namespaces, even if
2890 they are inactive to clear inode references from the lock.
2892 Severity : enhancement
2893 Frequency : occasional, if non-standard max_dirty_mb used
2895 Description: Client will block write RPCs if not enough grant
2896 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2897 then the client will block writes while waiting for another RPC
2898 to complete instead of consuming its dirty limit. With change
2899 we get improved performance when max_dirty_mb is small.
2901 Severity : enhancement
2902 Bugzilla : 3389, 6253
2903 Description: Add support for supplementary groups on the MDS.
2904 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2905 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2906 lookups for user supplementary groups into a cache.
2910 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2911 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2912 for liblustre O_EXCL is enforced by MDS.
2916 Description: readdir never returns NULL in liblustre.
2917 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2918 dentry in struct dirent.
2922 Frequency : liblustre only, depends on application IO pattern
2923 Description: liblustre clients evicted if not contacting servers
2924 Details : Don't put liblustre clients into the ping_evictor list, so
2925 they will not be evicted by the pinger ever.
2927 Severity : enhancement
2929 Description: Add ability to evict clients by NID from MDS.
2930 Details : By echoing "nid:$NID" string into
2931 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2932 $NID would be instantly evicted from this MDS and from all active
2933 OSTs connected to it.
2937 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2938 Details : lookup_it_finish() used to query file size from OSTs that was not
2943 Description: service threads change working directory to that of init
2944 Details : Starting lustre service threads may pin the working directory
2945 of the parent thread, making that filesystem busy. Threads
2946 now change to the working directory of init to avoid this.
2950 Frequency : during shutdown only
2951 Description: shutdown with a failed MDS or OST can cause unmount to hang
2952 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2957 Frequency : 2.6 only
2958 Description: chmod/chown may include an extra supplementary group
2959 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2960 supplementary group and if none is specified this is used.
2963 Bugzilla : 5479 (6816)
2965 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2966 Details : If lookup is in progress on a file that is unlinked we might try
2967 to revalidate the inode and fail in revalidate after lookup is
2968 complete and ll_file_open() enqueues the open again but
2969 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2972 Frequency : always, if lconf --abort_recovery used
2974 Description: lconf --abort_recovery fails with 'Operation not supported'
2975 Details : lconf was attempting to abort recovery on the MDT device and not
2978 ------------------------------------------------------------------------------
2980 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2985 Frequency : rare (only unsupported configurations with a node running as an
2987 Bugzilla : 6514, 5137
2988 Description: Mounting a Lustre file system on a node running as an OST could
2990 Details : OSTs now preallocates memory needed to write out data at
2991 startup, instead of when needed, to avoid having to
2992 allocate memory in possibly low memory situations.
2993 Specifically, if the file system is mounted on on OST,
2994 memory pressure could force it to try to write out data,
2995 which it needed to allocate memory to do. Due to the low
2996 memory, it would be unable to do so and the node would
2997 become unresponsive.
2999 Severity : enhancement
3001 Description: Addition of lconf --service command line option
3002 Details : lconf now accepts a '--service <arg>' option, which is
3003 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3005 Severity : enhancement
3007 Description: Failover mode is now the default for OSTs.
3008 Details : By default, OSTs will now run in failover mode. To return to
3009 the old behaviour, add '--failout' to the lmc line for OSTs.
3011 Severity : enhancement
3013 Description: Health checks are now provided for MDS and OSTs
3014 Details : Additional detailed health check information on MSD and OSTs
3015 is now provided through the procfs health_check value.
3018 Frequency : occasional, depends on IO load
3020 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3021 after numerous create/delete cycles
3022 Details : The ext3 inode allocation policy would not allocate new inodes
3023 very well on the OSTs because there are no new directories
3024 being created. Instead we look for groups with free space if
3025 the parent directories are nearly full.
3030 Description: Network or server problems during mount may cause partially
3031 mounted clients instead of returning an error.
3032 Details : The config llog parsing code may overwrite the error return
3033 code during mount error handling, returning success instead
3039 Description: MDS can fail to allocate large reply buffers
3040 Details : After long uptimes the MDS can fail to allocate large reply
3041 buffers (e.g. zconf client mount config records) due to memory
3042 fragmentation or consumption by the buffer cache. Preallocate
3043 some large reply buffers so that these replies can be sent even
3044 under memory pressure.
3048 Frequency : rare (liblustre)
3049 Description: fsx running with liblustre complained that using truncate() to
3050 extend the file doesn't work. This patch corrects that issue.
3051 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3052 ATTR_SIZE and lsm use in llu_setattr_raw.
3056 Frequency : rare, only 2.6 kernels
3057 Description: Unusual file access patterns on the MDS may result in inode
3058 data being lost in very rare circumstances.
3059 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3060 code in the 2.6 kernel caused the inode and/or EA data not to
3061 be read from disk, causing single-file corruption.
3065 Frequency : rare, only 2.6 filesystems using extents
3066 Description: Heavy concurrent write and delete load may cause data corruption.
3067 Details : It was possible under high-load situations to have an extent
3068 metadata block in the block device cache from a just-unlinked
3069 file overwrite a newly-allocated data block. We now unmap any
3070 metadata buffers that alias just-allocated data blocks.
3074 Frequency : filesystems with default stripe_count larger than 77
3075 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3076 Details : lconf specifies an inode size of 4096 bytes when the default
3077 stripe_count is larger than 77. This conflicts with the default
3078 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3079 this case to avoid pinning too much memory for large EAs.
3081 ------------------------------------------------------------------------------
3083 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3088 Frequency : rare (extremely heavy IO load with hundreds of clients)
3090 Description: Client is evicted, gets IO error writing to file
3091 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3092 caused clients to be evicted for AST timeouts. The fixes in
3093 bug 5192 mean we no longer need to have such short AST timeouts
3094 so ldlm_timeout has been increased.
3097 Frequency : occasional during --force or --failover shutdown under load
3098 Bugzilla : 5949, 4834
3099 Description: Server oops/LBUG if stopped with --force or --failover under load
3100 Details : a collection of import/export refcount and cleanup ordering
3101 issues fixed for safer force cleanup
3104 Frequency : only filesystems larger than 120 OSTs
3105 Bugzilla : 5990, 6223
3106 Description: lfs getstripe would oops on a very large filesystem
3107 Details : lov_getconfig used kfree on vmalloc'd memory
3110 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3111 Bugzilla : 6242, 6243
3112 Description: reading from files that had been truncated to a non-zero size
3113 but never opened returned no data
3114 Details : ll_file_read() reads zeros from no-object files to EOF
3119 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3120 Details : There's little harm in aborting MDS/OSS recovery and letting it
3121 try again, so I removed the LASSERT and return an error instead.
3123 Severity : enhancement
3125 Description: New debugging infrastructure for tracking down data corruption
3126 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3127 (b) cover more of the client-side code path, and (c) try to narrow
3128 down where problems occurred
3132 Bugzilla : 3819, 4364, 4397, 6313
3133 Description: Racing close and eviction MDS could cause assertion in mds_close
3134 Details : It was possible to get multiple mfd references during close and
3135 client eviction, leading to one thread referencing a freed mfd.
3137 Severity: : enhancement
3138 Bugzilla : 3262, 6359
3139 Description: Attempts to reconnect to servers are now more aggressive.
3140 Details : This builds on the enhanced upcall-less recovery that was added
3141 in 1.4.2. When trying to reconnect to servers, clients will
3142 now try each server in the failover group every 10 seconds. By
3143 default, clients would previously try one server every 25 seconds.
3148 Description: After recovery, certain operations trigger a failed
3149 assertion on a client.
3150 Details : Failing over an mds, using lconf -d --failover, while a
3151 client was doing a readdir() call would cause the client to
3152 LBUG after recovery completed and the readdir() was resent.
3154 Severity : enhancement
3156 Description: Default groups are now added by lconf
3157 Details : You can now run lconf --group <servicename> without having to
3158 manually add groups with lmc.
3161 Frequency : occasional
3163 Description: Nodes with an elan id of 0 trigger a failed assertion
3166 Frequency : always when accessing e.g. tty/console device nodes
3168 Description: tty and some other devices nodes cannot be used on lustre
3169 Details : file's private_data field is used by device data and lustre
3170 values in there got lost. New field was added to struct file to
3171 store fs-specific private data.
3174 Frequency : when exporting Lustre via NFS
3176 Description: NFSD failed occasionally when looking up a path component
3177 Details : NFSD is looking up ".." which was broken in ext3 directories
3178 that had grown large enough to become hashed.
3181 Frequency : Clusters with multiple interfaces not on the same subnet
3183 Description: Nodes will repeatedly try to reconnect to an interface which it
3184 cannot reach and report an error to the log.
3185 Details : Extra peer list entries will be created by lconf with some peers
3186 unreachable. lconf now validates the peer before adding it.
3189 Frequency : Only if a default stripe is set on the filesystem root.
3191 Description: Setting a default stripe on the filesystem root prevented the
3192 filesystem from being remounted.
3193 Details : The client was sending extra request flags in the root getattr
3194 request and did not allocate a reply buffer for the dir EA.
3197 Frequency : occasional, higher if lots of files are accessed by one client
3198 Bugzilla : 6159, 6097
3199 Description: Client trips assertion regarding lsm mismatch/magic
3200 Details : While revalidating inodes the VFS looks up inodes with ifind()
3201 and in rare cases can find an inode that is being freed.
3202 The ll_test_inode() code will free the lsm during ifind()
3203 when it finds an existing inode and then the VFS later attaches
3204 this free lsm to a new inode.
3208 Bugzilla : 6422, 7030
3209 Description: MDS deadlock between mkdir and client eviction
3210 Details : Creating a new file via mkdir or mknod (starting a transaction
3211 and getting the ns lock) can deadlock with client eviction
3212 (gets ns lock and trying to finish a synchronous transaction).
3215 Frequency : occasional
3216 Description: While starting a server, the fsfilt_ext3 module could not be
3218 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3219 kernels. Previously, lconf would still use the ext3 name
3220 when trying to load modules. Now, it will correctly use
3221 ext3 on 2.4 and ldiskfs on 2.6.
3223 Severity : enhancement
3224 Description: The default stripe count has been changed to 1
3225 Details : The interpretation of the default stripe count (0, to lfs
3226 or lmc) has been changed to mean striping across a single
3227 OST, rather than all available. For general usage we have
3228 found a stripe count of 1 or 2 works best.
3230 Severity : enhancement
3231 Description: Add support for compiling against Cray portals.
3232 Details : Conditional compiling for some areas that are different
3236 Frequency : occasional
3237 Bugzilla : 6409, 6834
3238 Description: Creating files with an explicit stripe count may lead to
3239 a failed assertion on the MDS
3240 Details : If some OSTs are full or unavailable, creating files may
3241 trigger a failed assertion on the MDS. Now, Lustre will
3242 try to use other servers or return an error to the
3246 Frequency : occasional
3248 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3249 caused lock timeout during readahead (since 1.4.2).
3250 Details : Processes doing readahead might match a lock that hasn't been
3251 granted yet if there are overlapping and conflicting lock
3252 requests. The readahead process waits on ungranted lock
3253 (original lock is CBPENDING), while OST waits for that process
3254 to cancel CBPENDING read lock and eventually evicts client.
3256 Severity : enhancement
3258 Description: Initial enabling of flock support for clients
3259 Details : Implements fcntl advisory locking and file status functions.
3260 This feature is provided as an optional mount flag (default
3261 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3262 locking are implemented yet, and those that are are not guaranteed
3263 to be completely correct in production environments.
3264 mount -t lustre -o [flock|noflock] ...
3267 Frequency : occasional
3269 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3270 assertion in the ext3 JBD (journaling) layer.
3271 Details : The b_committed_data struct is protected by the big kernel lock
3272 in 2.4 kernels, serializing journal_commit_transaction() and
3273 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3274 access to this struct. In 2.6 kernels there is finer grained
3275 locking to improve SMP performance of the JBD layer.
3279 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3284 Description: merge and cleanup kernel patches.
3287 -----------------------------------------------------------------------------
3289 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3291 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3292 versions of Lustre. Please update all servers and clients to
3293 version 1.4.2 or later at the same time. You must also run
3294 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3295 to update the configuration logs.
3297 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3298 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3299 - swab llog records of type '0' so we get proper header size/idx (5861)
3300 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3301 - fix rename of one directory over another leaking an inode (5953)
3302 - avoid SetPageDirty on 2.6 (5981)
3303 - don't re-add just-being-destroyed locks to the waiting list (5653)
3304 - when creating new directories, inherit the parent's custom
3305 striping settings if present parent (3048)
3306 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3307 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3308 - don't hold client locks on temporary worklist from l_lru (5666)
3309 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3310 - automatically evict dead clients (5921)
3311 - Update file size properly in create+truncate+fstat case (6196)
3312 - Do not unhash mountpoint dentries, do not allow removal of
3314 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3315 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3316 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3317 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3318 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3319 - Correctly handle reads of files with no objects (6243)
3320 - lctl recover will also mark a device active if deactivate used (5933)
3322 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3323 - allow --write-conf on an MDS with different nettype than client (5619)
3324 - don't write config llogs to MDS for mounts not from that MDS (5617)
3325 - lconf should create multiple TCP connections from a client (5201)
3326 - init scripts are now turned off by default; run chkconfig --on
3327 lustre and chkconfig --on lustrefs to use them
3328 - upcalls are no longer needed for clients to recover to failover
3330 - add --abort-recovery option to lconf to abort recovery on device
3332 - add support for an arbitrary number of OSTs (3026)
3333 - Quota support protocol changes.
3334 - forward compatibility changes to wire structs (6007)
3335 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3336 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3337 - improved "lustre status" script
3338 - initialize blocksize for non-regular files (6062)
3339 - added --disable-server and --disable-client configure options (5782)
3340 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3341 - Vanilla 2.4.29 support
3342 - increase maximum number of obd devices to 520 (6242)
3343 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3344 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3345 SLES 9 2.6.5 kernels
3346 - limit stripes per file to 160 (the maximum EA size) (6093)
3348 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3351 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3352 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3353 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3354 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3355 - small ext3 extents cleanups and fixes (5733)
3356 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3357 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3358 - avoid needless client->OST connect, fix handle mismatch (5317)
3359 - fix DLM error path that led to out-of-sync client, long delays (5779)
3360 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3361 - fix several locking issues related to i_size (5492,5624,5654,5672)
3362 - don't move pending lock onto export if it is already evicted (5683)
3363 - fix kernel oops when creating .foo in unlinked directory (5548)
3364 - fix deadlock in obdfilter statistics vs. object create (5811)
3365 - use time_{before,after} to avoid timer jiffies wrap (5882)
3366 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3367 - Do not leak request if server was not able to process it (5154)
3368 - If mds_open unable to find parent dir, make that negative lookup(5154)
3369 - don't create new directories with extent-mapping (5909, 5936)
3371 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3372 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3373 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3374 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3375 - support for RHEL4 kernel on i686 (5773)
3376 - provide error messages when incompatible logs are encountered (5898)
3378 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3379 * version 1.4.0.10 (1.4.1 release candidate 1)
3381 - don't keep a lock reference when lock is not granted (4238)
3382 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3383 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3384 - drop import inflight refcount on signal_completed_replay error (5255)
3385 - unlock page after async write error during send (3677)
3386 - handle missing objects in filter_preprw_read properly (5265)
3387 - no transno return for symlink open, don't save no-trasno open (3440)
3388 - don't try to complete elan receive that already failed (4012)
3389 - free RPC server reply state on error (5406)
3390 - clean up thread from ptlrpc_start_thread() on error (5160)
3391 - readahead could read extra page into cache that wasn't ejected (5388)
3392 - prevent races in class_attach/setup/cleanup/detach (5260)
3393 - don't dereference de->d_inode after l_dput of de (5458)
3394 - use "int" for stripe value returned from lock_to_stripe (5544)
3395 - mballoc allocation and error-checking fixes in 2.6 (5504)
3396 - block device patches to fix I/O request sizes in 2.6 (5482)
3397 - look up hostnames for IB nals (5602)
3398 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3399 - don't start multiple acceptors for the same port (5277)
3400 - fix incorrect LASSERT in mds_getattr_name (5635)
3401 - export a proc file for general "ping" checking (5628)
3402 - fix "lfs check" to not block when the MDS is down (5628)
3404 - service request history (4965)
3405 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3406 - create an "evict_client" /proc entry on OSTs, like the MDS has
3407 - fix mount usage message, return errors per mount(8) (5168)
3408 - change grep [] to grep "[]" in tests so they work in more UMLs
3409 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3410 - remove ancient LOV_MAGIC_V0 EA support (5047)
3411 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3412 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3413 - allow readahead allocations to fail when low on memory (5383)
3414 - mmap locking landed again, after considerable improvement (2828)
3415 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3417 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3420 - send OST transaction number in read/write reply to free req (4966)
3421 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3422 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3423 - the watchdog thread now runs as interruptible (5246)
3424 - flock/lockf fixes (but it's still disabled, pending 5135)
3425 - don't use EXT3 constants in llite code (5094)
3426 - memory shortage at startup could cause assertion (5176)
3428 - reorganization of lov code
3429 - single portals codebase
3431 - add extents/mballoc support (5025)
3432 - direct I/O reads in the obdfilter (4048)
3433 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3435 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3438 - send OST transaction number in read/write reply to free req (4966)
3439 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3440 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3441 - the watchdog thread now runs as interruptible (5246)
3442 - handle missing objects in filter_preprw_read properly (5265)
3443 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3444 - drop import inflight refcount on signal_completed_replay error (5255)
3445 - unlock page after async write error during send (3677)
3446 - return original error code on reconstructed replies (3761)
3447 - no transno return for symlink open, don't save no-trasno open (3440)
3449 - add pid to ldlm debugging output (4922)
3450 - bump the watchdog timeouts -- we can't handle 30sec yet
3451 - extra debugging for orphan dentry/inode bug (5259)
3453 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3456 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3457 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3458 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3459 - fix NULL dereference in /proc/sys/portals/routes (4827)
3460 - allow failed mdc_close() operations to be interrupted (4561)
3461 - stop precreate on OST before MDS would time out on it (4778)
3462 - don't send partial-page writes before EOF from client (4410)
3463 - discard client grant for sub-page writes on large-page clients (4520)
3464 - don't free dentries not owned by NFS code, check generation (4806)
3465 - fix lsm leak if mds_create_objects() fails (4801)
3466 - limit debug_daemon file size, always print CERROR messages (4789)
3467 - use transno after validating reply (3892)
3468 - process timed out requests if import state changes (3754)
3469 - update mtime on OST during writes, return in glimpse (4829)
3470 - add mkfsoptions to LDAP (4679)
3471 - use ->max_readahead method instead of zapping global ra (5039)
3472 - don't interrupt __l_wait_event() during strace
3474 - add software watchdogs to catch hung threads quickly (4941)
3475 - make lustrefs init script start after nfs is mounted
3476 - fix CWARN/ERROR duplication (4930)
3477 - return async write errors to application if possible (2248)
3478 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3479 - print NAL number in %x format (4645)
3480 - update barely-supported suse-2.4.21-171 series (4842)
3481 - support for sles 9 %post scripts
3482 - support for building 2.6 kernel-source packages
3483 - support for sles km_* packages
3485 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3488 - ignore -ENOENT errors in osc_destroy (3639)
3489 - notify osc create thread that OSC is being cleaned up (4600)
3490 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3491 - reconstruct ost_handle() like mds_handle() (4657)
3492 - create a new thread to do import eviction to avoid deadlock (3969)
3493 - let lconf resolve symlinked-to devices (4629)
3494 - don't unlink "objects" from directory with default EA (4554)
3495 - hold socknal file ref over connect in case target is down (4394)
3496 - allow more than 32000 subdirectories in a single directory (3244)
3497 - fix blocks count for O_DIRECT writes (3751)
3498 - OST returns ENOSPC from object create when no space left (4539)
3499 - don't send truncate RPC if file size isn't changing (4410)
3500 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3501 - bind to privileged port in socknal and tcpnal (3689)
3503 - rate limit CERROR/CWARN console message to avoid overload (4519)
3504 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3505 - basic mmap support (3918)
3506 - kernel patch series update from b1_4 (4711)
3508 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3511 - avoid crash during MDS cleanup with OST shut down (2775)
3512 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3513 - don't use bad inodes on the MDS (3744)
3514 - dynamic object preallocation to improve recovery speed (4236)
3515 - don't hold spinlock over lock dumping or change debug flags (4401)
3516 - don't zero obd_dev when it is force cleaned (3651)
3517 - print grants to console if they go negative (4431)
3518 - "lctl deactivate" will stop automatic recovery attempts (3406)
3519 - look for existing locks in ldlm_handle_enqueue() (3764)
3520 - don't resolve lock handle twice in recovery avoiding race (4401)
3521 - revalidate should check working dir is a directory (4134)
3523 - don't always mark "slow" obdfilter messages as errors (4418)
3525 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3528 - don't close LustreDB during write_conf until it is done (3860)
3529 - fix typo in lconf for_each_profile (3821)
3530 - allow dumping logs from multiple threads at one time (3820)
3531 - don't allow multiple threads in OSC recovery (3812)
3532 - fix debug_size parameters (3864)
3533 - fix mds_postrecov to initialize import for llog ctxt (3121)
3534 - replace config semaphore with spinlock (3306)
3535 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3536 - don't allow enqueue to complete on a destroyed export (3822)
3537 - down write_lock before checking llog header bitmap (3825)
3538 - recover from lock replay timeout (3764)
3539 - up llog sem before sending rpc (3652)
3540 - reduce ns lock hold times when setting kms (3267)
3541 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3542 - fix NULL deref and obd_dev leak on setup error (3312)
3543 - replace some LBUG about llog ops with error handling (3841)
3544 - don't match INVALID dentries from d_lookup and spin (3784)
3545 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3546 - fix invalid assertion in ptlrpc_set_wait (3880)
3548 - add libwrap support for the TCP acceptor (3996)
3549 - add /proc/sys/portals/routes for non-root route listing (3994)
3550 - allow setting MDS UUID in .xml (2580)
3551 - print the stack of a process that LBUGs (4228)
3553 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3556 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3557 - make sure to unset replay flag from failed open requests (3440)
3558 - if default stripe count is 0, use OST count for inode size (3636)
3559 - update parent mtime/ctime on client for create/unlink (2611)
3560 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3561 - free recovery state on server during a forced cleanup (3571)
3562 - unregister_reply for resent reqs (3063)
3563 - loop back devices mounting and status check on 2.6 (3563)
3564 - fix resource-creation race that can provoke i_size == 0 (3513)
3565 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3566 - more debugging for page-accounting assertion (3746)
3567 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3568 - avoid deadlock after precreation failure (3758)
3569 - fix race and lock order deadlock in orphan handling (3450, 3750)
3570 - add validity checks when grabbing inodes from l_ast_data (3599)
3572 - add /proc/.../recovery_status to obdfilter (3428)
3573 - lightweight CDEBUG infrastructure, debug daemon (3668)
3574 - change default OSC RPC parameters to be better on small clusters
3575 - turn off OST read cache for files smaller than 32MB
3576 - install man pages and include them in rpms (3100)
3577 - add new init script for (un)mounting lustre filesystems (2593)
3578 - run chkconfig in %post for init scripts (3701)
3579 - drop scimac NAL (unmaintained)
3581 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3584 - clean kiobufs before and after use (3485)
3585 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3586 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3587 - change init script to fail more clearly if not run as root (1528)
3588 - allow clients to reconnect during replay (1742)
3589 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3590 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3591 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3592 - fix cond_resched (3554)
3593 - only evict unfinished clients after recovery (3515)
3594 - allow bulk resend, prevent data loss (3570)
3595 - dynamic ptlrpc request buffer allocation (2102)
3596 - don't allow unlinking open directory if it isn't empty (2904)
3597 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3598 - remove extraneous obd dereference causing LASSERT failure (3334)
3599 - don't use get_cycles() when creating temp. files on the mds (3156)
3600 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3601 - handle EEXIST for set-stripe, set proper directory name (3336)
3603 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3604 - fix ksocknal_fmb_callback() error messages (2918)
3606 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3609 - don't copy lvb into (possibly NULL) reply on error (2983)
3610 - don't deref dentry after dput, don't free lvb on error (2922)
3611 - use the kms to determine writeback rpc length (2947)
3612 - increment oti_logcookies when osc is inactive (2948)
3613 - update client's i_blocks count via lvb messages (2543)
3614 - handle intent open/close of special files properly (1557)
3615 - mount MDS with errors=remount-ro, like obdfilter (2009)
3616 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3617 - don't use cancelling-locks' kms values (2947)
3618 - use highest lock extent for kms, not last one (2925)
3619 - don't dereference ERR_PTR() dentry in error handling path (3107)
3620 - fix thread race in portals_debug_dumplog() (3122)
3621 - create lprocfs device entries at setup instead of at attach (1519)
3622 - common AST error handler, don't evict client on completion race (3145)
3623 - zero nameidata in detach_mnt in 2.6 (3118)
3624 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3625 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3626 - initialize RPC timeout timer earlier for 2.6 (3219)
3627 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3628 - print nal/nid for unknown nid (3258)
3629 - additional checks for oscc recovery before doing precreate (3284)
3630 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3631 - don't crash in mdc_close for bad permissions on open (3285)
3632 - zero i_rdev for non-device files (3147)
3633 - clear page->private before handing to FS, better assertion (3119)
3634 - tune the read pipeline (3236)
3635 - fix incorrect decref of invalidated dentry (2350)
3636 - provide read-ahead stats and refine rpc in flight stats (3328)
3637 - don't hold journal transaction open across create RPC (3313)
3638 - update atime on MDS at close time (3265)
3639 - close LDAP connection when recovering to avoid server load (3315)
3640 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3641 - don't leak open file on MDS after open resend (3325)
3642 - serialize filter_precreate and filter_destroy_precreated (3329)
3643 - loop device shouldn't call sync_dev() for nul device (3092)
3644 - clear page cache after eviction (2766)
3645 - resynchronize MDS->OST in background (2824)
3646 - refuse to mount the same filesystem twice on same mountpoint (3394)
3647 - allow llmount to create routes for mounting behind routers (3320)
3648 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3649 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3650 - fix rare problem with rename on htree directories (3417)
3652 - allow default OST striping configuration per directory (1414)
3653 - fix compilation for qswnal for 2.6 kernels (3125)
3654 - increase maximum number of MDS request buffers for large systems
3655 - change liblustreapi to be useful for external progs like lfsck (3098)
3656 - increase local configuration timeout for slow disks (3353)
3657 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3659 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3662 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3663 - don't overwrite extent policy data in reply if lock was blocked (2901)
3664 - drop filter export grants atomically with removal from device (2663)
3665 - del obd_self_export from work_list in class_disconnect_exports (2908)
3666 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3667 - swab reply message in mdc_close, other PPC fixes (2464)
3668 - fix destroying of named logs (2325)
3669 - overwrite old logs when running lconf --write_conf (2264)
3670 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3671 - fix race in target_handle_connect (2898)
3672 - mds_reint_create() should take same inode create lock (2926)
3673 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3674 - don't close files for self_export to avoid uninitialized obd (2936)
3675 - allow MDS with the same name as client node (2939)
3676 - hold dentry reference for closed log files for unlink (2325)
3677 - reserve space for all logs during transactions (2059)
3678 - don't evict page beyond end of stripe extent (2925)
3679 - don't oops on a deleted current working directory (2399)
3680 - handle hard links to targets without a parent properly (2517)
3681 - don't dereference NULL lock when racing during eviction (2867)
3682 - don't grow lock extents when lots of conflicting locks (2919)
3684 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3687 - account for cache space usage on clients to avoid data loss (974)
3688 - lfsck support in lustre kernel code (2349)
3689 - reduce journal credits needed for BRW writes (2370)
3690 - orphan handling to avoid losing space on client/server crashes
3691 - ptlrpcd can be blocked, stopping ALL progress (2477)
3692 - use lock value blocks to assist in proper KMS, faster stat (1021)
3693 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3694 - recovery for initial connections (2355)
3695 - fixes for mds_cleanup_orphans (1934)
3696 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3697 - block all file creations until orphan recovery completes (1901)
3698 - client remove rq_connection from request struct (2423)
3699 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3700 - recovery timer race (2670)
3701 - mdc_close recovey bug (2532)
3702 - ptlrpc cleanup bug (2710)
3703 - mds timeout on local locks (2588)
3704 - namespace lock held during RPCs (2431)
3705 - handle interrupted sync write properly (2503)
3706 - don't try to handle a message that hasn't been replied to (2699)
3707 - client assert failure during cleanup after abort recovery (2701)
3708 - leak mdc device after failed mount (2712)
3709 - ptlrpc_check_set allows timedout requests to complete (2714)
3710 - wait for inflight reqs when ptlrpcd finishes (2710)
3711 - make sure unregistered services are removed from the srv_list
3712 - reset bulk XID's when resending them (caught by 1138 test)
3713 - unregister_bulk after timeout
3714 - fix lconf error (2694)
3715 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3716 - readahead locks pages, leaves pending causing memory pressure (2673)
3717 - increase OST request buffers to 4096 on large machines (2729)
3718 - fix up permission of existing directories in simple_mkdir (2661)
3719 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3720 - don't assign transno to errored transactions (2742)
3721 - don't delete objects on OST if given a bogus objid from MDS (2751)
3722 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3723 - if rq_no_resend, then timeout request after recovery (2432)
3724 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3725 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3726 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3727 - MDS recovery shouldn't delete objects that it has given out (2730)
3728 - if enqueue arrives after completion, don't clobber LVB (2819)
3729 - don't unlock pages twice when trigger_group_io returns error (2814)
3730 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3731 - don't write pages to disk if there was an error (1450)
3732 - don't ping imports that have recovery disabled (2676)
3733 - take buffered bytes into account when balancing socknal conn (2817)
3734 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3735 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3736 - remove little-endian swabbing of llog records (1987)
3737 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3738 - retry reposting request buffers if they fail (1191)
3739 - grow extent at grant time to avoid granting a revoked lock (2809)
3740 - lock revoke doesn't evict page if covered by a second lock (2765)
3741 - disable VM readahead to avoid reading outside lock extents (2805)
3743 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3744 - updated kernel patches for hp-2.4.20 kernel (2681)
3746 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3749 - fix truncated write corruption (2366)
3750 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3752 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3753 - ensure len(uuid) < 37 in lmc (1171)
3754 - fix ia64 OOPS in llog_test (2255)
3755 - zero end of page at obdfilter for partial page writes (2648)
3756 - don't leave stale dentries around after renames (bug 2428)
3757 - fix timeouts when evicting a client with a single lock held (2642)
3758 - set deadline for the initial HELLO message to drain (2634)
3759 - print out dotted-quad IP addresses in the socknal (2302)
3761 - additional debugging for MDS client eviction problem (2443)
3762 - fix mkfsoptions support for osts (2603, 2604)
3764 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3767 - add series for the vanilla 2.6.0 kernel
3768 - add series for the vanilla 2.4.24 kernel
3769 - add series for a cray x86/64 UL kernel drop
3770 - fix xattr patches for the vanilla 2.4.19 series
3772 - generate true UUIDs in lmc (1171)
3773 - have portals stack dumping break in UML (2466)
3774 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3775 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3776 - wake callback waiting threads on client eviction (2460)
3777 - Add --ptldebug and --subsystem to lmc (1719)
3778 - update assertion to allow safe interrupt allocation
3779 - set rq_no_resend for cancel requests (2432)
3780 - recalculate ptlrpcd timeout after resend (2494)
3781 - call vfs_rmdir when removing pending directories (2368)
3782 - fix renaming a file to itself (2429)
3783 - lmc creates a default one-stripe lov (2454)
3784 - expand procfs space to handle large clusters (2326)
3785 - increase UML stack to avoid overflow
3786 - update lconf's list of debug and subsystem masks
3787 - fix lfs find --obd (2510)
3788 - /proc tunable for disabling filter read caching (2591)
3789 - stop rpm packages from altering slapd.conf (2301)
3790 - disable nagle in the socknal under 0conf (2578)
3791 - choose mds inode size based on stripe count (2572)
3792 - fix kernel-source rpm problems (2516)
3794 - add --disable-doc to avoid pdf generation (2421)
3795 - update documentation, tests, type-os, comments
3796 - avoid format warnings on ia64
3797 - remove the TOE NAL
3798 - tiny code cleanups by removing unused fields
3800 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3803 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3804 - remove the most blatant lies from BUILDING file (2371)
3805 - change default debug level to reasonable production setting
3806 - reduce client side cache size to reduce cache flush time
3807 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3808 - make TCP zerocopy and pinger support enabled by default (2476)
3809 - sync writes completed after process exits caused crashes (2319)
3810 - maintain correct mount count on the MDS (2356)
3811 - backout 1557, because 2316 wasn't really fixed
3812 - better file I/O statistics gathering in /proc
3813 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3814 - another kernel patch to fix zero-copy TCP function export
3815 - don't take duplicate lock when processing re-sent getattr (2420)
3816 - lctl uses obd_self_export instead of creating new conn (2353)
3817 - MDS/OST recovery case which requires object creation asserted (2425)
3818 - move lfs from /usr/sbin to /usr/bin in packages
3819 - fix race between mds_client_add and mds_client_free (2417)
3820 - use kmalloc instead of slabs in portals (2430)
3821 - don't create duplicate records when a failover MDS is present (2442)
3822 - remove unnecessary mount age check (2332)
3823 - don't remove directory inodes from locks prematurely (2451)
3824 - don't break if MDS service name is the same as hostname (2103)
3825 - fix races in client write RPC generation when cache full (2482)
3827 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3830 - remove now-unused request->rq_obd (278)
3831 - if an allocation fails, print out how much memory we've used (1933)
3832 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3833 - add the "configurable stack size" patch to most series files (1256)
3834 - ability to write large log records, for 100+ OST configs (2306)
3835 - fix NULL deref when filter_prep fails (2314)
3836 - fix operator precedence error in filter_sync
3837 - dynamic allocation of socknal TX descriptors (2315)
3838 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3839 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3840 - ensure that CWARN messages in Portals always get to the syslog
3841 - __init/__exit are not for prototype decls (ldlm_init/exit)
3842 - x86-64 compile warning fixes
3843 - fix gateway LMC keyword conflict (2318)
3844 - fix MDS lock inversions in getattr/reint paths (1844)
3845 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3846 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3847 - DLM race condition prevented some lock evictions (2328)
3848 - ENOMEM detection and retry on socknal sends (2230)
3849 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3850 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3851 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3852 - better error messages when a client is rejected during recovery (1505)
3853 - avoid cancelling locks which were never granted, after failure (2330)
3854 - fix i_sem/journal inversion in mds_client_add (2333)
3855 - fix truncate/getattr lock cycle deadlock (2334)
3856 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3857 - fix two rare exit paths which could leak an l_lock() ref (2321)
3858 - fencepost error in MDS/OST orphan recovery (2226)
3859 - make log record alignment 8 bytes (1988)
3860 - lstripe now fails when requested offset > ost_count (2237)
3861 - ensure that all kernel series have a complete list.h (1607)
3862 - fix crashes in special-file operations (2316)
3863 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3865 - allow configurable automake binary, for testing new versions
3866 - small update to the lfs documentation
3868 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3870 * fix negative export reference count in fsfilt_sync (2312)
3872 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3873 * release candidate 0.9.1
3875 - orphans are moved into the PENDING directory for possible recovery
3876 - replayed opens now open by fid for orphan/rename safety (1042)
3877 - last close of an orphan inode generates a transno (683)
3878 - chdir() and mount() now pin the directory entry (1020)
3879 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3880 - discard very old requests without processing them (1502)
3881 - remove some common, well-understood CERRORs (1505)
3882 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3883 - clear "grant" flags in OST replies until OST grant code lands (1644)
3884 - fix read performance by not clobbering i_blksize on client (1598)
3885 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3886 - make LRU size a /proc tunable, clears locks when reduced (707)
3887 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3888 - server locks take references on exports now (1558)
3889 - build fixes for 2.4.20-rh trees (1663)
3890 - return an error from lov_create if all OSCs are inactive (1751)
3891 - fix import levels when a reconnect happens without a timeout (1597)
3892 - exit early from mds_open if we get a lookup error (1749)
3893 - partial page read at EOF wouldn't wait for disk before sending (1642)
3894 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3895 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3896 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3897 - write/truncate lock inversion (1639)
3898 - Don't auto-load obdclass, portals modules during cleanup (1495)
3899 - fix timestamps from jumping to "now" (1763)
3900 - extra journal assertions (1648)
3901 - add an extra multiunlink test (1771)
3902 - fix read_record/write_record API (1776)
3903 - fix leak of offset_extent, possible incorrect i_size later (1772)
3904 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3905 - Debugging for the kqswnal_get_idle_tx problems (1820)
3906 - Allow recovery to be attempted multiple times (1536)
3907 - Write out MDS last_rcvd file after it is first created (1600)
3908 - Fix tx_descriptor leak in failed transmit situations (1827)
3909 - ext3 journaling fixes for assertion failure after IO error (1871)
3910 - class_export_put() on freed export after completion AST error (1896)
3911 - Fix revalidate looping in VFS (1322)
3912 - Don't access a freed export during MDS_REINT timeout (1521)
3913 - Add open-unlink recovery support on the MDS (1673,1764)
3914 - Return an error if no MDS data was read from last_rcvd (1946)
3915 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3916 - Don't setup a disk device that doesn't match exported UUID (317)
3917 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3918 - avoid committing NULL handle in force close
3919 - local.sh is now a one-stripe LOV configuration
3920 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3921 - fix ext3 htree duplicate directory entry corruption (1516)
3922 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3923 - update to new LOV EA format (2097)
3924 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3926 2003-06-15 Phil Schwan <phil@clusterfs.com>
3929 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3930 - per-import/export recovery handling (958, 931, 959)
3931 - multiple last-rcvd slots, for serving multiple FSes (949)
3932 - connections are again shared between multiple imp/exports (963, 964)
3933 - "umount -f" would hang if any requests needed to be sent (393, 978)
3934 - avoid pinning large req buffer by copying for queued messages (989)
3935 - add "uuid" to "lctl device" command to help upcalls (991)
3936 - "open" RPCs with transnos would confuse recovery counters (1037)
3937 - do proper endian conversion of all wire messages (288, 340, 891)
3938 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3939 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3940 - fix ptlrpc_connection leak in target_handle_connect (1174)
3941 - fix import refcounting bug in OST and MDS cleanup (1134)
3942 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3943 - fix very unlikely obd_types race condition (501)
3944 - remove osc_open hack for echo_client (1187)
3945 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3946 - a failure in read_inode2 leads to deadlock (1139)
3947 - cancel ack-locks as soon as transaction is committed (1072)
3948 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3949 - make sure to commitrw after any preprw to avoid deadlock (1162)
3950 - failing to execute a file in a lustre FS would lock inode (1203)
3951 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3952 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3953 - obd_uuid2tgt would walk off the end of the list (1255)
3954 - on IA64 the osc would give portals incorrect bulk size (1258)
3955 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3956 - fix lock inversion caused by new llite matching code (1282)
3957 - limit the number of dirty pages on a client to 10MB (1286)
3958 - timed out locks were not being corrected cancelled (1289)
3959 - fix O_DIRECT above 4GB on IA-32 (1292)
3960 * major user-visible changes
3961 - fail out/fail over policy now controlled by the upcall (993)
3963 - add OBD_PING to check server availability and failure (954)
3964 - lustre messages are now sent in sending host order (288, 340, 891)
3965 - add eadatalen to MDS getattr reply (340)
3966 - OST read replies may contain second buffer, with per-page status (593)
3968 2003-03-11 Phil Schwan <phil@clusterfs.com>
3971 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3972 - failed open()s could cause deadlock; fixed (867, 869)
3973 - stop cancelling OST locks when files are closed (481)
3974 - overlapping XID spaces caused network corruption (851, 853)
3975 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3976 - setattr_raw added, to do single-RPC, server-side setattrs
3977 - lmc/lconf syntax change for OST UUIDs
3978 - fix crashy race condition between ptlrpc_free_req and osc_close
3979 - don't use request in mdc_enqueue if we hit a timeout (889)
3980 - don't set the inode i_size for regular files from the MDS (896)
3981 - handle out of order completion AST (842)
3982 - don't LBUG if a lock request times out after receiving AST (913)
3983 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3984 - if a bad lock AST arrives, send an error instead of dropping entirely
3985 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3986 - fix leak in bulk IO when only partially completed (899, 900, 926)
3987 - fix O_DIRECT for ia64 (55)
3988 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3989 - C-z after timeout could hang a process forever; fixed (977)
3991 - client-side I/O cache (678, 924, 929, 941, 970)
3993 - READPAGE and SETATTRs which don't take server-side locks get
3996 2003-02-11 Phil Schwan <phil@clusterfs.com>
3999 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4000 - Fix fsfilt_extN_readpage() to read a full page of directory
4001 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4002 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4003 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4004 - Most intent operations were removed, in favour of a new RPC mode
4005 that does a single RPC to the server and bypasses most of the VFS
4006 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4007 - Aggressively cancel local locks on DLM servers
4008 - mds_reint_unlink sends EA to the client if it's the last nlink.
4009 client uses that EA to unlink OST objects.
4010 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4011 - recursive symlinks were fixed (439)
4012 - fixed NULL deref in DEBUG_REQ
4013 - filter_update_lastobjid no longer calls sync, which annoyed extN
4014 - fixed multi-client small-writes to a single file problem (445)
4015 - fixed mtime updates during file writes (607)
4016 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4017 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4018 - fixed Linux OST size reporting problem (444, 656)
4019 - OST now updates object mtime with writes or setattr (607, 619)
4020 - client verifies file size before zeroing page past EOF (445)
4021 - OST now writes last allocated objid to disk with allocation (108)
4022 - LOV on echo now works (409)
4024 - mds_reint_unlink sends a new buffer, with the EA included. this
4025 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4026 set if a regular file was being unlinked, and it was the last link
4027 - use PtlGet from the target for bulk writes (315)
4028 - OST now updates object mtime with writes or setattr (607, 619)
4029 - LDLM now has a grant-time callback to revalidate locked items, if
4031 - Many MDS operations were reorganized to combat race conditions
4033 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4034 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4036 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4039 - Fully reactivate OST imports after reconnection (512, others)
4040 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4041 - More graceful error handling for truncating on dead OST (515)
4042 - Don't error out unless we're actually accessing dead stripes (474)
4043 - Fix garbage sizes when stripes are missing (410)
4044 - LRU counters were broken, causing constant lock purge (433, 432)
4045 - garbage on read from stripes with failed OSTs (441)
4046 - mark OSCs as active before reconnecting during recovery (438)
4047 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4048 - lfind did not preserve OST order in output (443)
4049 - symlinks cause hung clients, incorrect data (439)
4050 - stop dereferencing request after dropping refcount (457)
4051 - don't LASSERT(spin_is_locked) on non-SMP (455)
4052 - fixes for many rename() bugs
4053 - fstat didn't correctly synchronize attributes (399)
4054 - server must handle lock cancellation during blocking AST prep (487)
4055 - bulk descriptors were free()d too soon (511)
4056 - fix paths in lconf, which would load incorrect modules (451, 507)
4057 - fix confusing lconf 'host not found' error message (386)
4058 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4059 - fix race condition in mdc_blocking_ast() for inode access (526)
4060 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4061 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4062 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4063 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4064 - fix lstripe in several ways (kernel side) (527)
4065 - fix request leak in ldlm_cli_enqueue (262)
4066 - incorrect OSC was marked inactive after OST failure
4067 - call mds_fs_cleanup before unmounting filesystem (524)
4068 - fix races between taking ns_lock and ldlm_lock_change_resource
4069 - fix races updating LOV export open file list
4070 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4071 - fix recovery NULL deref in ldlm_cli_cancel_unused
4072 - fix some DLM races by using new hash table for lock handles (419)
4073 - permit the client to specify desired inodes, at replay
4074 - duplicate requests when we queue them for replay reintegration
4075 - fix last_rcvd offset calculation
4076 - sync after each recovered transaction, so we always make progress
4077 - never, not always, ERESTART requests without transnos
4078 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4080 - skip replay if the MDS says that the client is already connected
4081 - don't check for a recovery-enabled export to match lctl's UUID
4082 - don't INC_USE_COUNT for phantom exports
4083 - don't crash when cleaning up phantom exports (567)
4084 - don't double-finish or set replay data for errored mdc_open requests
4085 - abort requests when they time out, so we don't get old replies
4086 - send/receive replies for AST messages again
4087 - if the client says that it doesn't have the lock, cancel it on the
4089 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4090 mark it as destroyed, it will all work out when decref is called
4091 - fix module use counts (22, 581)
4093 - ASTs now expect a reply (server cancels lock on error reply)
4095 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4098 - fix many simultaneous client startup (392)
4099 - fix dentry->d_it clobbering
4100 - credentials weren't being shipped for readdir/getattr operations
4101 - remove invalid assertions triggered during some concurrent MD
4103 - proper Lustre versions added (336, 389)
4104 - fix memory leak for create error case (398)
4105 - fix LOV locking bug that would get cli/srv out of sync
4106 - fix echo client over LOV (409)
4107 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4108 - fix double-O_EXCL intent crash (424)
4109 - avoid sending multiple lock CANCELs (352)
4111 - MDS can do multi-client recovery (modulo bugs in new code)
4113 - many updates, edits, cleanups
4115 2002-11-18 Phil Schwan <phil@clusterfs.com>
4118 - fix null d_it dereference (346)
4119 - fix full OST/dbench hang (333)
4120 - fix permission problem with file removal (286)
4121 - fix removal of OSCs from LOV when they fail
4122 - fix NULL deref during bulk timeout (214)
4123 - fix problems related to multiple filesystems on one MDS (241)
4124 - fixed serious subtle metadata locking bugs
4125 - free locks on clients when inodes are removed due to memory
4127 - fix inode pointer in lock data (285)
4128 - partial support for multiple MDS on a single host (241)
4129 - data locks weren't cancelled at clear_inode time (290, 311)
4130 - intent locks could lead to unbounded lock growth (205)
4131 - added a maximum lock count, an LRU list, and a flusher
4132 - fix multiple rename (365)
4133 - properly abstracted the echo client
4134 - OSC locked 1 byte too many; fixed
4135 - rewrote brw callback code:
4136 - fixed recovery bugs related to LOVs (306)
4137 - fixed too-many-pages-in-one-write crash (191)
4138 - fixed (again) crash in sync_io_timeout (214)
4139 - probably fixed callback-related race (385)
4141 - Add capability to MDS protocol
4142 - LDLM cancellations and callbacks on different portals
4144 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4147 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4148 - defer open object destruction to close time (601981, 138)
4149 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4150 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4151 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4152 - don't call obd_setattr in ll_file_release for destroyed objects
4153 * protocol change to lustre_msg: move |version| and add |flags|
4154 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4155 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4156 * added replay of create, unlink, link and rename operations during
4157 MDS failover; recovery should be much more robust now
4158 * remove failed OSCs from LOVs (only lov_create uses this so far)
4159 * the lustre-HOWTO was brought (more) up to date (582544)
4161 2002-10-23 Phil Schwan <phil@clusterfs.com>
4164 - in-use dentries weren't being reused properly (617851)
4165 - prevent multiple LDLM setup (599178)
4166 - fix LOV size calculations for truncate (617853)
4167 - fix client handling of MDS intent errors (POSIX)
4168 - fix permission bug in lovstripe.c test (624321)
4169 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4170 - truncate past end of file could corrupt data
4171 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4172 - a race in recovery could return ETIMEDOUT to apps (623947)
4173 - building outside the source directory was fixed
4174 * the lustre-HOWTO was brought (more) up to date (582544)
4175 * major progress was made on recovery functionality
4177 2002-10-10 Phil Schwan <phil@clusterfs.com>
4180 - recovery deadlock fix
4181 - rm -rf causes LBUG fix (617817)
4182 - file open by multiple tasks fix (618962)
4183 - directory permissions bugs (602707 and 620007)
4184 - journal_stop fixed with locking (611313)
4185 - O_APPEND failures resolved (618273, perhaps 614459)
4186 - lconf PATH fix (619770)
4187 - IA64 build fix (621450)
4188 - RPC buffer sizes scale with amount of memory
4190 2002-10-01 Phil Schwan <phil@clusterfs.com>
4193 - locks would be cancelled without throwing away data pages,
4194 resulting in inconsistent data (605627)
4195 - inode attributes were not always being refreshed (605627, 612449)
4196 - lconf now continues to cleanup after lctl reports an error
4197 - MDS now enforces user permissions (602707)
4198 - lprocfs cleanup fixed, but not yet enabled (614157)
4199 - fixed infinite server hang, should a client not respond to an AST
4200 - avoid going into recovery if user calls readlink() with a buffer
4201 that's too small (613941)
4202 - AST RPCs no longer require replies (614867) -- this may be changed
4203 - don't crash server if client sends an IOV that's too big (611336)
4204 - fixed lock conversion deadlock (611892)
4205 - fixed the following of symlinks (614622)
4206 * recovery: the server can remove locks from a client that dies, other
4207 clients can make progress
4208 * more extN patch fixes
4209 * compile-time configurable ptlrpc buffer allocations
4211 - collaborative read cache document
4212 - Lustre Lite Performance CDR document-in-progress
4214 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4217 - fix typo in patch-2.4.18
4219 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4222 - clear ptlrpc request each time in handle_incoming_request()
4223 - unlink of files now destroys the object on the OST
4225 2002-09-19 Peter Braam <braam@clusterfs.com>
4227 * add hard link support
4228 * change obdfile creation method
4229 * kernel patch changed
4231 2002-09-19 Peter Braam <braam@clusterfs.com>
4234 - stack overflow bug in extN fixed
4236 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4238 * documentation updates
4239 - add man pages for config tools
4240 - update tests/README to describe testing with new config tools
4241 - finish metadata API descriptions
4242 * bug fixes and cleanups
4243 - statfs workaround for 16TB limit
4244 - LOV stripe allocation improved, can stripe on subset of OSTs
4245 - LOV file size/IO offset was wrong for files > 4GB in size
4246 - object EA data was being dropped, caused files to be unreadable
4247 - memory overflow with non-LOV OST caused memory corruption
4248 - fixed regression tests to work with new config tools, obdfilter
4249 - fixed bug when directory size became larger than 1 block
4250 - fixed bug (for single client case) when PWD was deleted
4251 - invalidate local directory pages when doing intent-based ops
4252 - avoid LDLM oops when lock callback contained bad data
4254 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4256 * documentation updates
4257 * bug fixes and cleanups
4258 - configuration tools
4261 - 64-bit compile warnings
4262 - 64-bit internal statfs data
4264 * test_brw on persistent OST devices
4266 * lprocfs (disabled)
4268 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4270 * documentation updates
4271 * bug fixes and cleanups
4272 * configuration tools
4274 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4277 * many small fixes to 0_5_4
4278 * io/network handling
4279 * thinkos in MDS operations
4281 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4284 * crucial basic fixes to 0.5.3
4285 * IOR, Iozone work over Elan
4288 2002-08-07 Phil Schwan <phil@clusterfs.com>
4289 * version 0_5_3, our first alpha
4290 * we use the new Portals iovs
4291 * documentation updates
4292 * bug fixes and cleanups
4293 * small changes in the DLM wire protocol
4295 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4296 * version 0_5_1 with some initial stability,
4297 * locking on MD and file I/O.
4298 * documentation updates
4299 * several bug fixes since 0.5.0
4300 * small changes in wire protocol
4302 2002-07-18 Phil Schwan <phil@clusterfs.com>
4304 * delivered as Lustre Light Alpha
4305 * fixed a crash after handling invalid MDS requests
4306 * fixed directory pages for architectures with non-4k pages sizes
4308 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4309 * release version v0_4_4
4310 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4312 * Fixes a number of interruption problems with OST operations.
4313 * Update documentation for portals header changes
4314 * Move all wire protocol structs/defines to lustre_idl.h
4315 * Fixes symlink length bug.
4316 * Add tcpdump to repository.
4318 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4319 * release version v0_4_3
4320 * Fixes statfs for inodes on extN.
4321 * Fixes bug in runtests which would delete /etc/hosts.
4322 * Use 64-bit object IDs wherever possible (not into VFS though)
4323 Remove ost_get_info, which is unused by lustre, and out of date.
4325 2002-07-03 Peter Braam <braam@clusterfs.com>
4326 * release version v0_4_2 Fixes a lookup error (type not passed)
4327 * move forward to head of Portals
4328 * move forward to latest Lustre kernel
4330 2002-06-25 Peter Braam <braam@clusterfs.com>
4331 * release version v0_4_1. Hopefully stable on single node use.