1 tbd Sun Microsystems, Inc.
4 2.6.5-7.287.3 (SLES 9),
6 2.6.16.54-0.2.3 (SLES 10),
7 2.6.18-53.1.4.el5 (RHEL 5).
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.9-42.0.10.EL (RHEL 4),
12 2.6.16 - 2.6.21 vanilla (kernel.org)
13 * Recommended e2fsprogs version: 1.40.2-cfs5
14 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
16 Severity : enhancement
18 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
20 Severity : enhancement
22 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
26 Description: make mgs_setparam() handle fsname containing dash
27 Details : fsname containing a dash does not work with lctl conf_param
29 Severity : enhancement
31 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
33 Severity : enhancement
35 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
39 Frequency : rare, at shutdown
40 Description: access already free / zero obd_namespace.
41 Details : if client_disconnect_export was called without force flag set,
42 and exist connect request in flight, this can produce access to
43 NULL pointer (or already free pointer) when connect_interpret
44 store ocd flags in obd_namespace.
48 Frequency : only at startup
49 Description: not alloc memory with spinlock held.
50 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
53 Severity : enhancement
55 Description: make lustre randomly fail allocating memory
56 Details : Make lustre randomly failed allocating memory for testing purpose.
58 Severity : enhancement
60 Description: lost problems with lov objid file
61 Details : Fixes some scability and access to not inited memory problems
62 in work with lov objdid file.
67 Description: lfs find does not continue on file error
68 Details : Continue other files processing when a file/dir is absent.
72 Description: Inconsistent usage of lustre_pack_reply()
73 Details : Standardize the usage of lustre_pack_reply() such that it
74 always generate a CERROR on failure.
77 Frequency : occasional
79 Description: MDS hang or stay in waiting lock
80 Details : If client receive lock with CBPENDING flag ldlm need send lock
81 cancel as separate rpc, to avoid situation when cancel request
82 can't processed due all i/o threads stay in wait lock.
86 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
90 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
94 Description: Update to SLES9 kernel-2.6.5-7.287.3.
98 Description: Update to RHEL4 latest kernel.
100 Severity : enhancement
102 Description: Build SLES10 patchless client fails
103 Details : The configure was broken by run ./configure with
104 --with-linux-obj=.... argument for patchless client. When the
105 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
106 can't search header adequately. Use absolute path such as
107 -I($LINUX)/include instead.
111 Description: interrupt oig_wait produce painc on resend.
112 Details : brw_redo_request can be used for resend requests from ptlrpcd and
113 private set, and this produce situation when rq_ptlrpcd_data not
114 copyed to new allocated request and triggered LBUG on assert
115 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
116 ptlrpcd set if request is changed and can be safety changed to use
121 Description: LASSERT_{REQ,REP}SWAB macros are buggy
122 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
123 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
124 these macros and replace them with their difinitions instead.
128 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
129 Details : Update which_patch & target file for SLES10 latest kernel.
132 Description: Remove client patches from SLES 10 kernel.
133 Details : This causes SLES 10 clients to behave as patchless clients
134 even on a Lustre-patched (server) kernel.
136 Severity : enhancement
138 Description: self-adjustable client's lru lists
139 Details : use adaptive algorithm for managing client cached locks lru
140 lists according to current server load, other client's work
141 pattern, memory activities, etc. Both, server and client
142 side namespaces provide number of proc tunables for controlling
145 Severity : enhancement
147 Description: light-weight GSS support
148 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
149 and reduce performance overhead.
151 Severity : enhancement
153 Description: Linux keyring support
154 Details : Support using service of Linux keyring for Lustre GSS internal
155 context refresh/cache mechanism.
159 Description: Fix errors in lfs documentation
160 Details : Fixes man pages
164 Description: don't use GFP_* in generic Lustre code.
165 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
169 Description: obdclass is limited by single OBD_ALLOC(idarray)
170 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
174 Description: warnings with build patchless client with vanila 2.6.19 and up
175 Details : change old ctl_table style and replace ctl_table/ctl_table_header
176 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
180 Frequency : Only for SLES
181 Description: sanity_quota fail test_1
182 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
183 file, which makes TSTID[2] inunique.
187 Frequency : Only for SLES9
188 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
189 Details : Update target/ChangeLog/which_patch .
193 Frequency : Only for test_10 in sanity-quota.sh
194 Description: a bug in quota test code
195 Details : checking a test flag in a wrong place
199 Frequency : Only for test_13 in sanity-quota.sh
200 Description: a bug in quota test script
201 Details : shouldn't check overfull allocated quota
205 Description: jbd statistics
206 Details : Port older jbd statistics patch for sles10
208 Severity : enhancement
210 Description: Add printing inode info into message about error in writepage.
215 Description: reply_lock_interpret crash due to race with it and lock cancel.
216 Details : Do not replay locks that are being cancelled. Do not reference
217 locks by their address during replay, just by their handle.
221 Frequency : When flocks are used.
222 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
223 Details : Flock locks might destroy just granted lock if it could be merged
224 with another existing flock, this is done in completion handler,
225 so teach ldlm_cli_enquque_fini that this is a valid case for
231 Description: Oops in read and write path when failing to allocate lock.
232 Details : Check if lock allocation failed and return error back.
236 Description: lstripe command fails for valid OST index
237 Details : The stripe offset is compared to lov->desc.ld_tgt_count
238 instead of lov->desc.ld_active_tgt_count.
242 Description: sanity.sh failed test 103
243 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
248 Description: df doesn't work properly if diskfs blocksize != 4K
249 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
253 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
254 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
258 Description: obd_health_check_timeout too short
259 Details : set obd_health_check_timeout as 1.5x of obd_timeout
263 Description: llapi_file_create() does not allow some changes
264 Details : add llapi_file_open() that allows specifying the mode and
265 open flags, and also returns an open file handle.
269 Description: merge and cleanup kernel patches.
270 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
274 Description: Add journal checksum support.(Kernel part)
275 Details : The journal checksum feature adds two new flags i.e
276 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
277 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
278 indicates that the commit block contains the checksum for
279 the blocks described by the descriptor blocks. Now commit
280 record can be sent to disk without waiting for descriptor
281 blocks to be written to disk. This behavior is controlled
282 using JBD2_FEATURE_ASYNC_COMMIT flag.
286 Description: OSS needs mutliple precreate threads
287 Details : Add ability to start more than one create thread per OSS.
291 Description: Sanity.sh test_65k failed: setstripe should have succeeded
292 Details : execute lfs setstripe on client
296 Description: mds_obd_create error creating tmp object
297 Details : When the user sets quota on root, llog will be affected and can't
298 create files and write files.
301 Frequency : Always on ia64 patchless client, and possibly others.
303 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
304 Details : This allows the patchless client to be loaded on architectures
309 Description: RedHat Update kernel for RHEL5
310 Details : Add the kernel config file for RHEL5.
314 Description: RedHat Update kernel for RHEL5
315 Details : Modify the kernel config file more closer RHEL5.
319 Description: Build failure against Centos5 (RHEL5)
320 Details : Define PAGE_SIZE when it isn't present.
324 Description: test_7 on sanity-quota.sh can't be used on separate machines
325 Details : add facet to handle it
329 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
330 Details : with NFS, the anon dentry's parent was set to itself in
331 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
332 corresponding dentry other than use rec->ur_name.
334 Severity : enhancement
336 Description: lfs setstripe enhancement
337 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
341 Description: enable data checksumming by default
342 Details : enable checksum by default, allow --disable-checksum
343 configure option and "-o nochecksum" mount option. Checksums
344 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
348 Description: lustre support for RHEL5
349 Details : Add support for RHEL5.
353 Description: Client eviction due to failover config
354 Details : after a connection loss, the lustre client should attempt to
355 reconnect to the last active server first before trying the
356 other potential connections.
360 Description: when mds and osts use different quota unit(32bit and 64bit),
361 quota will be released repeatly.
362 Details : void sending multiple quota reqs to mds, which will keep the
363 status between the reqs.
367 Description: osts not allocated evenly to files
368 Details : change the condition to increase offset_idx
371 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
373 Description: Data corruption for OSTs that are formatted larger than 2TB
375 Details : When generating the bio request for lustre file writes the
376 sector number would overflow a temporary variable before being
377 used for the IO. The data reads correctly from Lustre (which
378 will overflow in a similar manner) but other file data or
379 filesystem metadata may be corrupted in some cases.
383 Description: Tune the kernel for good SCSI performance.
384 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
385 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
390 Description: rewrite ext2-derived code in obdclass/uuid.c
391 Details : rewrite inherited code (uuid parsing code from ext2 utils)
392 from scratch preserving functionality.
396 Description: "lfs find -obd UUID" prints directories
397 Details : "lfs find -obd UUID" will return all directory names instead
398 of just file names. It is incorrect because the directories
399 do not reside on the OSTs.
403 Description: use i_size_read and i_size_write in 2.6 port
404 Details : replace inode->i_size access with i_size_read/write()
407 Frequency : when removing large files
409 Description: scheduling issue during removal of large Lustre files
410 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
411 It causes scheduling issues when removing large files (17TB in the
415 Frequency : only with liblustre clients on XT3
417 Description: evictions taking too long
418 Details : allow llrd to evict clients directly on OSTs
421 Frequency : only on ppc
423 Description: /proc/fs/lustre/devices broken on ppc
424 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
425 We should make sure the type of 'index' is loff_t and then cast
426 explicitly as needed below. Do not assign an explicitly cast
430 Frequency : only for rhel5
432 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
433 Details : Modify the target file & which_kernel.
435 Severity : enhancement
437 Description: omit set fsid for export NFS
438 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
439 as FS_REQUIRES_DEV for avoid problems with generate fsid.
443 Frequency : Always, for kernels after 2.6.16
444 Description: Fix warning idr_remove called for id=.. which is not allocated.
445 Details : Last kernels save old s_dev before kill super and not allow
446 to restore from callback - restore it before call kill_anon_super.
450 Description: buffer overruns could theoretically occur
451 Details : llapi_semantic_traverse() modifies the "path" argument by
452 appending values to the end of the origin string, and a
453 overrun may occur. Adding buffer overrun check in liblustreapi.
457 Description: Fix error on 'ls .' at the top of the Lustre mount.
458 Details : Don't revalidate dentry if it is a root dentry.
462 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
463 Details : Modify vm-tunables-rhel4.patch.
467 Description: Kernel config for 2.6.18-vanilla.
468 Details : Modify targets/2.6-vanilla.target.in.
469 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
470 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
471 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
472 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
476 Description: improve handling recoverable errors
477 Details : if request processig with error which can be recoverable on server
478 request should be resend, otherwise page released from cache and
483 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
484 Details : Modify target file & which_patch.
485 A flaw was found in the IA32 system call emulation provided
486 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
487 value could be stored in the %RAX register, which could trigger an
488 out-of-bounds system call table access. An untrusted local user
489 could exploit this flaw to run code in the kernel
490 (ie a root privilege escalation). (CVE-2007-4573).
494 Description: O_DIRECT bypasses client statistics.
495 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
496 read_ahead_stats were updated. copied the stats section from
497 osc_send_oap_rpc() into async_internal().
501 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
505 Description: change order of libsysio includes
506 Details : '#include sysio.h' should always come before '#include xtio.h'
510 Description: handle "serious error: objid * already exists" more gracefully
511 Details : If LAST_ID value on disk is smaller than the objects existing in
512 the O/0/d* directories, it indicates disk corruption and causes an
513 LBUG(). If the object is 0-length, then we should use the existing
514 object. This will help to avoid a full fsck in most cases.
516 Severity : enhancement
518 Description: adapt the lustre_config script to support the upgrade case
519 Details : Add "-u" option for lustre_config script to support upgrading 1.4
520 server targets to 1.6 in parallel.
524 Description: To avoid grant space > avaible space when the disk is almost
525 full. Without this patch you might see the error "grant XXXX >
526 available" or some LBUG about grant, when the disk is almost
528 Details : In filter_check_grant, for non_grant cache write, we should
529 check the left space by if (*left > ungranted + bytes), instead
530 of (*left > ungranted), because only we are sure the left space
531 is enough for another "bytes", then the ungrant space should be
532 increase. In client, we should update cl_avail_grant only there
533 is OBD_MD_FLGRANT in the reply.
537 Description: Update RHEL 4 kernel to fix local root privilege escalation.
538 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
539 described in CVE-2007-4573. This problem could allow untrusted
540 local users to gain root access.
543 Frequency : when using O_DIRECT and quotas
545 Description: Incorrect file ownership on O_DIRECT output files
546 Details : block usage reported by 'lfs quota' does not take into account
547 files that have been written with O_DIRECT.
552 Description: touch file failed when fs is not full
553 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
554 otherwise we can get ENOSP while fs is not full.
558 Description: parallel lock callbacks
559 Details : Instead of sending blocking and completion callbacks as separated
560 requests, adding them to a set and sending in parallel.
563 Frequency : only for Cray XT3
564 Bugzilla : 12829/13455
565 Description: Changing primary group doesn't change the group lustre assigns to
567 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
568 overridden with the primary group provided by the group upcall,
569 whereas the supplied fsgid can be trusted if it is in the list of
570 supplementary groups returned by the group upcall.
572 Severity : enhancement
574 Description: Allow masking D_WARNING, D_ERROR messages from console
575 Details : Console messages can now be disabled via lnet.printk.
577 --------------------------------------------------------------------------------
579 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
581 * Support for kernels:
582 2.6.5-7.283 (SLES 9),
583 2.6.9-55.EL (RHEL 4),
584 2.6.16.46-0.14 (SLES 10),
585 2.6.18.8 vanilla (kernel.org)
586 * Client support for unpatched kernels:
587 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
588 2.6.16 - 2.6.22 vanilla (kernel.org)
589 * Due to recently discovered recovery problems, we do not recommend
590 using patchless RHEL 4 clients with this or any earlier release.
591 * Recommended e2fsprogs version: 1.39.cfs8
592 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
593 * Starting with this release, the ldiskfs backing filesystem required
594 by Lustre is now in its own package, lustre-ldiskfs. This package
595 should be installed. It is versioned separately from Lustre and
596 may be released separately in future.
600 Description: block reactivating mgc import until all deactivates complete
601 Details : Fix race when failing back MDT/MGS to itself (testing)
604 Frequency : at statup only
606 Description: mds_lov_synchronize race leads to various problems
607 Details : simultaneous MDT->OST connections at startup can cause the
608 sync to abort, leaving the OSC in a bad state.
610 Severity : enhancement
612 Description: add optional extra BUILD_VERSION info
613 Details : add a new environment variable (namely LUSTRE_VERS) which allows
614 to override the lustre version.
617 Frequency : 2.6.18 servers only
619 Description: ll_kern_mount() doesn't release the module reference
620 Details : The ldiskfs module reference count never drops down to 0
621 because ll_kern_mount() doesn't release the module reference.
626 Description: server LBUG when using old ost_num_threads parameter
627 Details : Accept the old ost_num_threads parameter but warn that it
628 is deprecated, and fix an off-by-one error that caused an LBUG.
633 Description: Transient SCSI error results in persistent IO issue
634 Details : iobuf->dr_error is not reinitialized to 0 between two
638 Frequency : sometimes when underlying device returns I/O errors
640 Description: OSTs not going read-only during write failures
641 Details : OSTs are not remounted read-only when the journal commit threads
642 get I/O errors because fsfilt_ext3 calls journal_start/stop()
643 instead of the ext3 wrappers.
647 Description: poor connect scaling with increasing client count
648 Details : Don't run filter_grant_sanity_check for more than 100 exports
649 to improve scaling for large numbers of clients.
652 Frequency : SLES10 only
654 Description: sanity-quota.sh quotacheck failed: rc = -22
655 Details : Quotas cannot be enabled on SLES10.
658 Frequency : liblustre clients only
660 Description: getdirentries does not give error when run on compute nodes
661 Details : getdirentries does not fail when the size specified as an argument
662 is too small to contain at least one entry
664 Severity : enhancement
666 Description: Add LNET router traceability for debug purposes
667 Details : If a checksum failure occurs with a router as part of the
668 IO path, the NID of the last router that forwarded the bulk data
669 is printed so it can be identified.
674 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
675 Details : Due to a race condition, liblustre clients were occasionally
678 Severity : enhancement
680 Description: lfs setstripe use optional parameters instead of postional
683 Severity : enhancement
685 Description: Nanosecond timestamp support for ldiskfs
686 Details : The on-disk ldiskfs filesystem has added support for nanosecond
687 resolution timestamps. There is not yet support for this at
688 the Lustre filesystem level.
691 Frequency : during server recovery
693 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
694 Details : request with rq_no_resend flag not awake l_wait_event if they get a
698 Frequency : nfs export on patchless client
700 Description: connectathon hang when test nfs export over patchless client
701 Details : Disconnected dentry cannot be found with lookup, so we do not need
702 to unhash it or make it invalid
705 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
706 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
707 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
712 Description: open req refcounting wrong on reconnect
713 Details : If reconnect happened between getting open reply from server and
714 call to mdc_set_replay_data in ll_file_open, we will schedule
715 replay for unreferenced request that we are about to free.
716 Subsequent close will crash in variety of ways.
717 Check that request is still eligible for replay in
718 mdc_set_replay_data().
723 Description: disable writes to filesystem when reading health_check file
724 Details : the default for reading the health_check proc file has changed
725 to NOT do a journal transaction and write to disk, because this
726 can cause reads of the /proc file to hang and block HA state
727 checking on a healthy but otherwise heavily loaded system. It
728 is possible to return to the previous behaviour during configure
729 with --enable-health-write.
731 Severity : enhancement
733 Description: 64-bit inode version
734 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
735 to the inode. This will be required for version-based recovery.
740 Description: MDS fails to start if a duplicate client export is detected
741 Details : in some rare cases it was possible for a client to connect to
742 an MDS multiple times. Upon recovery the MDS would detect this
743 and fail during startup. Handle this more gracefully.
745 Severity : enhancement
747 Description: Add -o localflock option to simulate old noflock
749 Details : This will achieve local-only flock/fcntl locks
755 Description: log_commit_thread vs filter_destroy race leads to crash
756 Details : Take import reference before releasing llog record semaphore
761 Description: Wrong request locking in request set processing
762 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
763 imp_delayed_list, in this place should be used imp_lock.
766 Frequency : when reconnection
768 Description: Grant Leak when osc reconnect to OST
769 Details : When osc reconnect ost, OST(filter) should check whether it
770 should grant more space to client by comparing fed_grant and
771 cl_avail_grant, and return the granted space to client instead
772 of "new granted" space, because client will call osc_init_grant
773 to update the client grant space info.
776 Frequency : when client reconnect to OST
778 Description: Grant Leak when osc do resend and replay bulk write
779 Details : When osc reconnect to OST, OST(filter)should clear grant info of
780 bulk write request, because the grant info while be sync between
781 OSC and OST when reconnect, and we should ignore the grant info
782 these of resend/replay write req.
787 Description: Grant space more than avaiable left space sometimes.
788 Details : When then OST is about to be full, if two bulk writing from
789 different clients came to OST. Accord the avaliable space of the
790 OST, the first req should be permitted, and the second one
791 should be denied by ENOSPC. But if the seconde arrived before
792 the first one is commited. The OST might wrongly permit second
793 writing, which will cause grant space > avaiable space.
796 Frequency : when client is evicted
798 Description: Grant might be wrongly erased when osc is evicted by OST
799 Details : when the import is evicted by server, it will fork another
800 thread ptlrpc_invalidate_import_thread to invalidate the
801 import, where the grant will be set to 0. While the original
802 thread will update the grant it got when connecting. So if
803 the former happened latter, the grant will be wrongly errased
804 because of this race.
809 Description: Checking Stale with correct fid
810 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
811 to check whether it is stale, because sometimes, we want the
812 enqueue happened anyway, and op_data.fid2 will not be initialized.
814 Severity : enhancement
816 Description: update patchless client
817 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
820 Frequency : only with 2.4 kernel
822 Description: random memory corruption
823 Details : size of struct ll_inode_info is to big for union inode.u and this
824 can be cause of random memory corruption.
829 Description: Memory leak in recovery
830 Details : Lov_mds_md was not free in an error handler in mds_create_object.
831 It should also check obd_fail before fsfilt_start, otherwise if
832 fsfilt_start return -EROFS,(failover mds during mds recovery).
833 then the req will return with repmsg->transno = 0 and rc = EROFS.
834 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
835 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
836 be freed no matter whether fsfilt_commit success or not.
839 Frequency : only with huge count clients
841 Description: Prevents from taking the superblock lock in llap_from_page for
843 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
844 for llap_from_page call in ll_removepage prevents from taking the
845 superblock lock for a soon died page.
850 Description: Not check open intent error before release open handle
851 Details : in some rare cases, the open intent error is not checked before
852 release open handle, which may cause
853 ASSERTION(open_req->rq_transno != 0), because it tries to release
854 the failed open handle.
859 Description: Set cat log bitmap only after create log success.
860 Details : in some rare cases, the cat log bitmap is set too early. and it
861 should be set only after create log success.
865 Description: Accessing a block bevice can re-enable I/O when Lustre is
866 tearing down a device.
867 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
872 Description: service threads may hog cpus when there are a lot of requests
873 Details : Insert cond_resched to give other threads a chance to use some CPU
878 Description: the cat log was not initialized in recovery
879 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
880 unlink log on mds will not be initialized until mds post
881 recovery. And also in mds post recovery, the unlink log will
882 initialization will be done asynchronausly, so there will be race
883 between add unlink log and unlink log initialization.
887 Description: brw_stats were being printed incorrectly
888 Details : brw_stats were being printed as log2 but all of them were not
889 recorded as log2. Also remove some code duplication arising from
890 filter_tally_{read,write}.
894 Frequency : rare, only in recovery.
895 Description: ASSERTION(req->rq_type != LI_POISON) failed
896 Details : imp_lock should be held while iterating over imp_sending_list for
897 prevent destroy request after get timeout in ptlrpc_queue_wait.
901 Description: replay-single.sh test 52 fails
902 Details : A lock's skiplist need to be cleanup when it being unlinked
903 from its resource list.
907 Description: Short directio read returns full requested size rather than
909 Details : Direct I/O operations should return actual amount of bytes
910 transferred rather than requested size.
912 Severity : enhancement
914 Description: metadata RPC reduction (e.g. for rm performance)
915 Details : decrease the amount of synchronous RPC between clients and servers
916 by canceling conflicing lock before the operation on the client
917 and packing thier handles into the main operation RPC to server.
919 Severity : enhancement
921 Description: Async OSC create to avoid the blocking unnecessarily.
922 Details : If a OST has no remain object, system will block on the creating
923 when need to create a new object on this OST. Now, ways use
924 pre-created objects when available, instead of blocking on an
925 empty osc while others are not empty. If we must block, we block
926 for the shortest possible period of time.
928 Severity : enhancement
930 Description: refine locking for avoid write wrong info into lov_objid file
931 Details : fix possible races with add new target and write/update data in
934 --------------------------------------------------------------------------------
936 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
941 Frequency : on some architectures
943 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
944 Details : Uninitialized flags sometimes cause configuration commands to
948 Frequency : patchless clients only
950 Description: missing __iget() symbol export
951 Details : The __iget() symbol export is missing. To avoid the need for
952 this on patchless clients the deathrow inode reaper is turned
953 off, and we depend on the VM to clean up old inodes. This
954 dependency was during via the fix for bug 12181.
956 --------------------------------------------------------------------------------
958 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
960 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
961 INTEROPERATE with older versions automatically. In many cases a
962 special upgrade step is needed. Please read the
963 user documentation before upgrading any part of a live system.
964 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
965 version will not interoperate with 1.6 betas before beta5 (1.5.95).
966 * WARNING: Lustre configuration and startup changes are required with
967 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
973 Severity : enhancement
975 Description: MountConf
976 Details : Lustre configuration is now managed via mkfs and mount
977 commands instead of lmc and lconf. New obd types (MGS, MGC)
978 are added for dynamic configuration management. See
979 https://mail.clusterfs.com/wikis/lustre/MountConf for
982 Severity : enhancement
984 Description: dynamic OST addition
985 Details : OSTs can now be added to a live filesystem
987 Severity : enhancement
989 Description: startup order invariance
990 Details : MDTs and OSTs can be started in any order. Clients only
991 require the MDT to complete startup.
993 Severity : enhancement
995 Description: parallel, asynchronous orphan cleanup
996 Details : orphan cleanup is now performed in separate threads for each
997 OST, allowing parallel non-blocking operation.
999 Severity : enhancement
1001 Description: optimized stripe assignment
1002 Details : stripe assignments are now made based on ost space available,
1003 ost previous usage, and OSS previous usage, in order to try
1004 to optimize storage space and networking resources.
1006 Severity : enhancement
1008 Description: Permanently set tunables
1009 Details : All writable /proc/fs/lustre tunables can now be permanently
1010 set on a per-server basis, at mkfs time or on a live system.
1012 Severity : enhancement
1014 Description: Lustre message v2
1015 Details : Add lustre message format v2.
1017 Severity : enhancement
1019 Description: client OST exclusion list
1020 Details : Clients can be started with a list of OSTs that should be
1021 declared "inactive" for known non-responsive OSTs.
1025 Description: SPEC SFS validation failure on NFS v2 over lustre.
1026 Details : Changes the blocksize for regular files to be 2x RPC size,
1027 and not depend on stripe size.
1029 Severity : enhancement
1031 Description: Multiple MD RPCs in flight.
1032 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1033 To avoid overly-overloading MDS, introduce a limit on number of
1034 MDS RPCs in flight for a single client and add /proc controls
1035 to adjust this limit.
1037 Severity : enhancement
1039 Description: client read/write statistics
1040 Details : Add client read/write call usage stats for performance
1041 analysis of user processes.
1042 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1043 file access. extents_stats shows chunk size distribution.
1044 extents_stats_per_process show chunk size distribution per
1047 Severity : enhancement
1049 Description: mds statistics
1050 Details : Add detailed mds operations statistics in
1051 /proc/fs/lustre/mds/*/stats.
1055 Description: Failure of copying files with lustre special EAs.
1056 Details : Client side always return success for setxattr call for lustre
1057 special xattr (currently only "trusted.lov").
1062 Description: Refcount LNET uuids
1063 Details : The global LNET uuid list grew linearly with every startup;
1064 refcount repeated list entries instead of always adding to
1067 Severity : enhancement
1069 Description: Dynamic service threads
1070 Details : Within a small range, start extra service threads
1071 automatically when the request queue builds up.
1074 Frequency : mixed-endian client/server environments
1076 Description: mixed-endian crashes
1077 Details : The new msg_v2 system had some failures in mixed-endian
1080 Severity : enhancement
1082 Description: Easy OST removal
1083 Details : OSTs can be permanently deactivated with e.g. 'lctl
1084 conf_param lustre-OST0001.osc.active=0'
1086 Severity : enhancement
1088 Description: MGS proc entries
1089 Details : Added basic proc entries for the MGS showing what filesystems
1092 Severity : enhancement
1094 Description: provide MGS failover
1095 Details : Added config lock reacquisition after MGS server failover.
1097 Severity : enhancement
1099 Description: add Linux 2.4 support
1100 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1104 Description: a large application tries to do I/O to the same resource and dies
1105 in the middle of it.
1106 Details : Check the req->rq_arrival time after the call to
1107 ost_brw_lock_get(), but before we do anything about
1108 processing it & sending the BULK transfer request. This
1109 should help move old stale pending locks off the queue as
1110 quickly as obd_timeout.
1113 Frequency : when an incorrect nid is specified during startup
1115 Description: ptlrpc connect to non-existant node causes kernel crash
1116 Details : LNET can't be re-entered from an event callback, which
1117 happened when we expire a message after the export has been
1118 cleaned up. Instead, hand the zombie cleanup off to another
1121 Severity : enhancement
1123 Description: plain/inodebits lock performance improvement
1124 Details : Grouping plain/inodebits in granted list by their request modes
1125 and bits policy, thus improving the performance of search through
1129 Frequency : only if OST filesystem is corrupted
1131 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1132 Details : for a short time RPCs with bulk IO are in the replay list,
1133 but replay of bulk IOs is unimplemented. If the OST filesystem
1134 is corrupted due to disk cache incoherency and then replay is
1135 started it is possible to trip an assertion. Avoid putting
1136 committed RPCs into the replay list at all to avoid this issue.
1139 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1141 Description: System hang on startup
1142 Details : This bug allowed the liblustre (e.g. catamount) client to
1143 return to the app before handling all startup RPCs. This
1144 could leave the node unresponsive to lustre network traffic
1145 and manifested as a server ptllnd timeout.
1147 Severity : enhancement
1149 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1150 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1151 boolean to control whether to print peer debug info when a
1152 client's RPC times out.
1155 Frequency : only for kernels with patches from Lustre below 1.4.3
1157 Description: Remove old rdonly API
1158 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1161 Frequency : only for devices with external journals
1163 Description: Set external device read-only also
1164 Details : During a commanded failover stop, we set the disk device
1165 read-only while the server shuts down. We now also set any
1166 external journal device read-only at the same time.
1169 Frequency : when upgrading from 1.4 while trying to change parameters
1171 Description: The wrong (new) MDC name was used when setting parameters for
1172 upgraded MDT's. Also allows changing of OSC (and MDC)
1173 parameters if --writeconf is specified at tunefs upgrade time.
1176 Frequency : when setting specific ost indicies
1178 Description: QOS code breaks on skipped indicies
1179 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1180 created with --index need not be sequential.
1182 Severity : enhancement
1184 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1185 Details : The uninit_groups feature works in conjunction with the kernel
1186 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1187 up the pass1 processing of e2fsck. This is a read-only feature
1188 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1189 mount filesystems that have had this feature enabled.
1191 Severity : enhancement
1193 Description: Improve multi-block allocation algorithm to avoid fragmentation
1194 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1195 allocation locality and avoid filesystem fragmentation.
1197 ------------------------------------------------------------------------------
1199 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1201 * Support for kernels:
1202 2.4.21-47.0.1.EL (RHEL 3)
1203 2.6.5-7.283 (SLES 9)
1204 2.6.9-42.0.10.EL (RHEL 4)
1205 2.6.12.6 vanilla (kernel.org)
1206 2.6.16.27-0.9 (SLES 10)
1207 * Recommended e2fsprogs version: 1.39.cfs5
1209 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1213 Frequency : occasional, depends on client load and configuration
1214 Bugzilla : 12181, 12203
1215 Description: data loss for recently-modified files
1217 Details : In some cases it is possible that recently written or created
1218 files may not be written to disk in a timely manner (this should
1219 normally be within 30s unless client IO load is very high).
1220 The problem appears as zero-length files or files that are a
1221 multiple of 1MB in size after a client crash or client eviction
1222 that are missing data at the end of the file.
1224 This problem is more likely to be hit on clients where files are
1225 repeatedly created and unlinked in the same directory, clients
1226 have a large amount of RAM, have many CPUs, the filesystem has
1227 many OSTs, the clients are rebooted frequently, and/or the files
1228 are not accessed by other nodes after being written.
1230 The presence of the problem can be detected by looking at
1231 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1232 smaller than the second (nr_unused) then dirty files will not
1233 be flushed automatically to disk. "sync; sleep 10" should be
1234 run several times on the node before unmounting it to update
1235 Lustre (this is also safe to run on nodes without this problem).
1237 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1238 that can cause this same problem, so customers using that kernel
1239 also need to update the kernel in addition to Lustre. In order
1240 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1242 It is normal that files written just before a client crash (less
1243 than 30s) may not yet have been flushed to disk, even for local
1247 Frequency : frequent on thin XT3 nodes
1249 Description: UUID collision on thin XT3 Linux nodes
1250 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1251 randomly, since we relied on an insufficiently-seeded PRNG.
1256 Description: OSS hangs after "All ost request buffers busy"
1257 Details : A deadlock between quota and journal operations caused OSS
1258 hangs after printing "All ost request buffers busy."
1261 Frequency : always on liblustre builds
1263 Description: Cleanup compiler warnings on liblustre
1266 Frequency : always on liblustre builds on XT3
1268 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1272 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1273 > 32000 subdirectories
1274 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1275 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1276 subdirectory count crosses 32000. This will aid e2fsck to
1277 correctly handle more than 32000 subdirectories.
1280 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1282 Description: System hang on startup
1283 Details : This bug allowed the liblustre (e.g. catamount) client to
1284 return to the app before handling all startup RPCs. This
1285 could leave the node unresponsive to lustre network traffic
1286 and manifested as a server ptllnd timeout.
1288 Severity : enhancement
1290 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1291 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1292 boolean to control whether to print peer debug info when a
1293 client's RPC times out.
1298 Description: make O_SYNC working on 2.6 kernels
1299 Details : 2.6 kernels use different method for mark pages for write,
1300 so need add a code to lustre for O_SYNC work.
1305 Description: Failure to close file and release space on NFS
1306 Details : Put inode details into lock acquired in ll_intent_file_open.
1307 Use mdc_intent_lock in ll_intent_open to properly
1308 detect all kind of errors unhandled by mdc_enqueue.
1313 Description: proc file read during shutdown sometimes raced obd removal,
1315 Details : Add lock to prevent obd access after proc file removal.
1318 Frequency : Only for files larger than 4GB on 32-bit clients.
1320 Description: improperly doing page alignment of locks
1321 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1322 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1327 Description: RPCs being resent when they shouldn't be
1328 Details : Some RPCs that should not be resent are being resent. This
1329 can cause inconsistencies in the RPC state machine. Do not
1330 resend such requests.
1333 Frequency : rare, only with NFS export
1335 Description: Crash on NFS re-export node
1336 Details : under very unusual load conditions an assertion is hit in
1337 ll_intent_file_open()
1340 Frequency : only if OST filesystem is corrupted
1342 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1343 Details : for a short time RPCs with bulk IO are in the replay list,
1344 but replay of bulk IOs is unimplemented. If the OST filesystem
1345 is corrupted due to disk cache incoherency and then replay is
1346 started it is possible to trip an assertion. Avoid putting
1347 committed RPCs into the replay list at all to avoid this issue.
1352 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1353 Details : Large single O_DIRECT read and write calls can fail to allocate
1354 a sufficiently large buffer to process the request. In case of
1355 allocation failure the allocation is retried with a smaller
1356 buffer and broken into smaller requests.
1358 Severity : enhancement
1360 Description: Add -o localflock option to simulate old noflock behaviour.
1361 Details : This will achieve local-only flock/fcntl locks coherentness.
1366 Description: versioning check is incomplete
1367 Details : Checking the version difference of client vs. server, report
1368 error if the gap is too big.
1373 Description: add support PG_writeback bit
1374 Details : add support for PG_writeback bit for Lustre, for more carefull
1375 work with page cache in 2.6 kernel. This also fix some deadlocks
1376 and remove hack for work O_SYNC with 2.6 kernel.
1378 Severity : enhancement
1380 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1381 Details : The uninit_groups feature works in conjunction with the kernel
1382 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1383 up the pass1 processing of e2fsck. This is a read-only feature
1384 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1385 mount filesystems that have had this feature enabled.
1387 Severity : enhancement
1389 Description: Improve multi-block allocation algorithm to avoid fragmentation
1390 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1391 allocation locality and avoid filesystem fragmentation.
1393 ------------------------------------------------------------------------------
1395 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1397 * Support for kernels:
1398 2.6.9-42.0.3.EL (RHEL 4)
1399 2.6.5-7.276 (SLES 9)
1400 2.4.21-47.0.1.EL (RHEL 3)
1401 2.6.12.6 vanilla (kernel.org)
1402 2.6.16.21-0.8 (SLES10)
1403 * Recommended e2fsprogs version: 1.39.cfs2-0
1405 * The backwards-compatible /proc/sys/portals symlink has been removed
1406 in this release. Before upgrading, please ensure that you change
1407 any configuration scripts or /etc/sysctl.conf files that access
1408 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1409 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1410 in advance of the upgrade on any system running Lustre 1.4.6 or
1411 newer, since /proc/sys/lnet was added in that version.
1412 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1416 Frequency : only when quota is used
1418 Description: avoid scanning export list for quota master
1419 Details : Change the algorithms to avoid scanning export list in order
1420 to improve the efficiency.
1423 Frequency : MDS failover only, very rarely
1425 Description: "went back in time" messages on mds failover
1426 Details : The greatest transno may be lost when the current operation
1427 finishes with an error (transno==0) and the client's last_rcvd
1428 record is over-written. Save the greatest transno in the
1429 mds_last_transno for this case.
1432 Frequency : always for specific kernels and striping counts
1434 Description: client may get "Matching packet too big" without ACL support
1435 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1436 when trying to access files in certain configurations. The
1437 clients should in fact be denied when mounting because they do
1438 not understand ACLs.
1441 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1443 Description: many clients connecting with IO in progress causes connect timeouts
1444 Details : Avoid synchronous journal commits to avoid delays caused by many
1445 clients connecting/disconnecting when bulk IO is in progress.
1446 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1447 OST_IO_PORTAL to avoid delays behind potentially many pending
1451 Frequency : occasionally with multiple writers to a single file
1453 Description: shared writes to file may result in wrong size reported by stat()
1454 Details : Allow growing of kms when extent lock is cancelled
1457 Frequency : always with random mmap IO to multi-striped file
1459 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1460 Details : Only if the hole is at the end of OST object so that kms is too
1461 small. Fix is to increase kms accordingly in ll_nopage.
1464 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1466 Description: writes to a missing object would leak memory on the OST
1467 Details : If there is an inconsistency between the MDS and OST filesystems,
1468 such that the MDS references an object that doesn't exist, writes
1469 to that object will leak memory due to incorrect cleanup in the
1470 error handling path, eventually running out of memory on the OST.
1475 Description: Creating too long symlink causes lustre errors
1476 Details : Check symlink and name lengths before sending requests to MDS.
1479 Frequency : only if flock is enabled (not on by default)
1481 Description: posix locks not released on fd closure on 2.6.9+
1482 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1483 kernels, this caused such locks not to be released on fd close and
1484 then assertions on fs unmount about still used locks.
1487 Frequency : MDS failover only, very rarely
1489 Description: clients may get ASSERTION(granted_lock != NULL)
1490 Details : When request was taking a long time, and a client was resending
1491 a getattr by name lock request. The were multiple lock requests
1492 with the same client lock handle and
1493 mds_getattr_name->fixup_handle_for_resent_request found one of the
1494 lock handles but later failed with ASSERTION(granted_lock != NULL).
1499 Description: handle->h_buffer_credits > 0, assertion failure
1500 Details : h_buffer_credits is zero after truncate, causing assertion
1501 failure. This patch extends the transaction or creates a new
1505 Frequency : NFS re-export or patchless client
1506 Bugzilla : 11179, 10796
1507 Description: Crash on NFS re-export node (__d_move)
1508 Details : We do not want to hash the dentry if we don't have a lock.
1509 But if this dentry is later used in d_move, we'd hit uninitialised
1510 list head d_hash, so we just do this to init d_hash field but
1511 leave dentry unhashed.
1514 Frequency : NFS re-export or patchless client
1516 Description: NFS exports has problem with symbolic link
1517 Details : lustre client didn't properly install dentry when re-exported
1518 to NFS or running patchless client.
1521 Frequency : NFS re-export or patchless client
1523 Description: Various nfs/patchless fixes.
1524 Details : fixes reuse disconected alias for lookup process - this fixes
1525 warning "find_exported_dentry: npd != pd",
1526 fix permission error with open files at nfs.
1527 fix apply umask when do revalidate.
1530 Frequency : occasional
1532 Description: Crash on NFS re-export node
1533 Details : calling clear_page() on the wrong pointer triggered oops in
1534 generic_mapping_read().
1537 Frequency : rarely, using O_DIRECT IO
1539 Description: unaligned directio crashes client with LASSERT
1540 Details : check for unaligned buffers before trying any requests.
1543 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1545 Description: stale data returned from RAID cache
1546 Details : If only a small amount of IO is done to the RAID device before
1547 reading it again it is possible to get stale data from the RAID
1548 cache instead of reading it from disk.
1551 Frequency : always for sles10 kernel
1553 Description: sles10 support
1554 Details : ll_follow_link: compile fixes and using of nd_set_link
1555 under newer kernels.
1558 Frequency : depends on arch, kernel and compiler version, always on sles10
1561 Description: recursive or deep enough symlinks cause stack overflow
1562 Details : getting rid of large stack-allocated variable in
1566 Frequency : depends on hardware
1568 Description: lustre write performance loss in the SLES10 kernel
1569 Details : the performance loss is caused by using of write barriers in the
1570 ext3 code. The SLES10 kernel turns barrier support on by
1571 default. The fix is to undo that change for ldiskfs.
1573 ------------------------------------------------------------------------------
1575 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1577 * Support for kernels:
1578 2.6.9-42.0.3EL (RHEL 4)
1579 2.6.5-7.276 (SLES 9)
1580 2.4.21-47.0.1.EL (RHEL 3)
1581 2.6.12.6 vanilla (kernel.org)
1585 Frequency : quota enabled and large files being deleted
1587 Description: releasing more than 4GB of quota at once hangs OST
1588 Details : If a user deletes more than 4GB of files on a single OST it
1589 will cause the OST to spin in an infinite loop. Release
1590 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1595 Description: statfs data retrieved from /proc may be stale or zero
1596 Details : When reading per-device statfs data from /proc, in the
1597 {kbytes,files}_{total,free,avail} files, it may appear
1598 as zero or be out of date.
1601 Frequency : systems with MD RAID1 external journal devices
1603 Description: lconf's call to blkid is confused by RAID1 journal devices
1604 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1605 of returning all block devices that match the journal UUID.
1608 Frequency : always, for aggregate stripe size over 4GB
1610 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1611 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1612 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1615 Frequency : always if "lfs find" used on a local file/directory
1617 Description: "lfs find" segfaults if used on a local file/directory
1618 Details : The case where a directory component was not specified wasn't
1619 handled correctly. Handle this properly.
1622 Frequency : always on ppc64
1624 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1625 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1626 perform endianness conversion.
1629 Frequency : rarely (truncate to non-zero file size after write under load)
1630 Bugzilla : 10730, 10687
1631 Description: Files padded with zeros to next 4K multiple
1632 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1633 it is possible that files that are truncated to a non-zero size
1634 immediately after being written are filled with zero bytes beyond
1635 the truncated size. No file data is lost.
1637 Severity : enhancement
1639 Description: Allow recovery/failover for liblustre clients.
1640 Details : liblustre clients were unaware of failover configurations until
1643 Severity : enhancement
1645 Description: user file locks should fail when not mounting with flock option
1646 Details : Set up an error-returning stub in ll_file_operations.lock field
1647 to prevent incorrect behaviour when client is mounted without
1648 flock option. Also, set up properly f_op->flock field for
1652 Frequency : always on ia64
1654 Description: "lfs df" loops on printing out MDS statfs information
1655 Details : The obd_ioctl_data was not initialized and in some systems
1656 this caused a failure during the ioctl that did not return
1657 an error. Initialize the struct and return an error on failure.
1660 Frequency : SLES 9 only
1662 Description: Error of copying files with lustre special EAs as root
1663 Details : Client side always return success for setxattr call for lustre
1664 special xattr (currently only "trusted.lov").
1667 Frequency : rarely on clusters with both ia64+i386 clients
1669 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1670 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1671 clients are doing concurrent writes to the same file it is
1672 possible that the ia64 clients may LASSERT because the OST
1673 extent locks are not PAGE_SIZE aligned. Ensure that grown
1674 locks are always aligned on the request boundary.
1677 Frequency : specific use, occasional
1679 Description: Overwriting in use executable truncates on-disk binary image
1680 Details : If one node attempts to overwrite an executable in use by
1681 another node, we now correctly return ETXTBSY instead of
1682 truncating the file.
1684 Severity : enhancement
1686 Description: Async OSC create to avoid the blocking unnecessarily.
1687 Details : If a OST has no remain object, system will block on the creating
1688 when need to create a new object on this OST. Now, ways use
1689 pre-created objects when available, instead of blocking on an
1690 empty osc while others are not empty. If we must block, we block
1691 for the shortest possible period of time.
1696 Description: chmod on Lustre root is propagated to other clients
1697 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1698 invalid by the follow_mount time.
1701 Frequency : liblustre clients only
1703 Description: Race in 'instant cancel' lock handling could lead to such locks
1704 never to be granted in case of SMP MDS
1705 Details : Do not destroy not yet granted but cbpending locks in
1709 Frequency : replay/resend of open
1711 Description: non null lock assetion failure in mds_intent_policy
1712 Details : Trying to replay/resend lockless open requests resulted in
1713 mds_open() returning 0 with no lock. Now it sets a flag if
1714 a lock is going to be returned.
1716 Severity : enhancement
1718 Description: Checksum enhancements
1719 Details : New checksum enhancements allow for resending RPCs that failed
1722 Severity : enhancement
1724 Description: Tunables on number of dirty pages in cacche
1725 Details : Allow to set limit on number of dirty pages cached.
1730 Description: client crash on unmount - lock still has references
1731 Details : In some error handling cases it was possible to leak a lock
1732 reference on a client while accessing a file. This was not
1733 harmful to the client during operation, but would cause the
1734 client to crash when the filesystem is unmounted.
1737 Frequency : specific case, rare
1739 Description: ETXTBSY on mds though file not in use by client
1740 Details : ETXTBSY is no longer incorrectly returned when attempting to
1741 chmod or chown a directory that the user previously tried to
1742 execute or a currently-executing binary.
1745 Frequency : extremely rare except on liblustre-based clients
1747 Description: Lustre space not freed when files are deleted
1748 Details : Clean up open-unlinked files after client eviction. Previously
1749 the unlink was skipped and the files remained as orphans.
1754 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1755 Details : In some cases it was possible to send a blocking callback to
1756 a client doing a glimpse, even though that client didn't get
1757 a lock granted. When the glimpse lock is cancelled on the OST
1758 the freed lock is left on the waiting list and corrupted the list.
1761 Frequency : all core dumps
1763 Description: Broke core dumps to lustre
1764 Details : Negative dentry may be unhashed if parent does not have UPDATE
1765 lock, but some callers, e.g. do_coredump, expect dentry to be
1766 hashed after successful create, hash it in ll_create_it.
1768 ------------------------------------------------------------------------------
1770 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1772 * Support for kernels:
1773 2.6.9-42.0.2.EL (RHEL 4)
1774 2.6.5-7.276 (SLES 9)
1775 2.4.21-40.EL (RHEL 3)
1776 2.6.12.6 vanilla (kernel.org)
1780 Frequency : always on RHEL 3
1782 Description: Number of open files grows over time
1783 Details : The number of open files grows over time, whether or not
1784 Lustre is started. This was due to a filp leak introduced
1785 by one of our kernel patches.
1787 ------------------------------------------------------------------------------
1789 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1791 * Support for kernels:
1792 2.6.9-42.EL (RHEL 4)
1793 2.6.5-7.276 (SLES 9)
1794 2.4.21-40.EL (RHEL 3)
1795 2.6.12.6 vanilla (kernel.org)
1800 Bugzilla : 5719, 9635, 9792, 9684
1801 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1802 Details : If a server is under heavy load and cannot reply to new
1803 connection requests before the client resends the (re)connect,
1804 the connection handling code can behave badly if two service
1805 threads are concurrently handing separate (re)connections from
1806 the same client. Add better locking to the connection handling
1807 code, and ensure that only a single connection will be processed
1808 for a given client UUID, even if the lock is dropped.
1810 Severity : enhancement
1812 Description: add TCP zero-copy support to kernel
1813 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1814 sends if the hardware supports scatter-gather and checksumming.
1815 This allows socklnd to do client-write and server-read more
1816 efficiently and reduce CPU utilization from skbuf copying.
1819 Frequency : only if NFS exporting from client
1821 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1822 Details : If setting attributes on a file created under NFS that had
1823 never been opened it would be possible to oops the client
1824 if the file had no objects.
1827 Frequency : always for liblustre
1829 Description: liblustre client does MDS+OSTs setattr RPC for each write
1830 Details : When doing a write from a liblustre client, the client
1831 incorrectly issued an RPC to the MDS and each OST the file was
1832 striped over in order to update the timestamps. When writing
1833 with small chunks and many clients this could overwhelm the MDS
1834 with RPCs. In all cases it would slow down the write because
1835 these RPCs are unnecessary.
1837 Severity : enhancement
1839 Description: allow number of MDS service threads to be changed at module load
1840 Details : It is now possible to change the number of MDS service threads
1841 running. Adding "options mds mds_num_threads={N}" to the MDS's
1842 /etc/modprobe.conf will set the number of threads for the next
1843 time Lustre is restarted (assuming the "mds" module is also
1844 reloaded at that time). The default number of threads will
1845 stay the same, 32 for most systems.
1850 Description: OST crash if filesystem is unformatted or corrupt
1851 Details : If an OST is started on a device that has never been formatted
1852 or if the filesystem is corrupt and cannot even mount then the
1853 error handling cleanup routines would dereference a NULL pointer.
1858 Description: NULL pointer deref in llap_from_page.
1859 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1860 incorrect) mapping if the page was truncated/reclaimed while it was
1861 searched for. Check for this condition and skip such pages when
1862 doing readahead. Introduce extra check to llap_from_page() to
1863 verify page->mapping->host is non-NULL (so page is not anonymous).
1866 Frequency : Sometimes when using sys_sendfile
1868 Description: "page not covered by a lock" warnings from ll_readpage
1869 Details : sendfile called ll_readpage without right page locks present.
1870 Now we introduced ll_file_sendfile that does necessary locking
1871 around call to generic_file_sendfile() much like we do in
1875 Frequency : with certain MDS communication failures at client mount time
1877 Description: NULL pointer deref after failed client mount
1878 Details : a client connection request may delayed by the network layer
1879 and not be sent until after the PTLRPC layer has timed out the
1880 request. If the client fails the mount immediately it will try
1881 to clean up before the network times out the request. Add a
1882 reference from the request import to the obd device and delay
1883 the cleanup until the network drops the request.
1886 Frequency : occasionally during client (re)connect
1888 Description: assertion failure during client (re)connect
1889 Details : processing a client connection request may be delayed by the
1890 client or server longer than the client connect timeout. This
1891 causes the client to resend the connection request. If the
1892 original connection request is replied in this interval, the
1893 client may trip an assertion failure in ptlrpc_connect_interpret()
1894 which thought it would be the only running connect process.
1897 Frequency : only with obd_echo servers and clients that are rebooted
1899 Description: kernel BUG accessing uninitialized data structure
1900 Details : When running an obd_echo server it did not start the ping_evictor
1901 thread, and when a client was evicted an uninitialized data
1902 structure was accessed. Start the ping_evictor in the RPC
1903 service startup instead of the OBD startup.
1905 Severity : enhancement
1906 Bugzilla : 10193 (patchless)
1907 Description: Remove dependency on various unexported kernel interfaces.
1908 Details : No longer need reparent_to_init, exit_mm, exit_files,
1909 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1912 Frequency : rare (only users of deprecated and unsupported LDAP config)
1914 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1915 Details : LDAP apparently contains 'lustreName' attributes instead of
1916 'name'. A simple remapping of the name is sufficient.
1919 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1921 Description: waiting_locks_callback trips kernel BUG if client is evicted
1922 Details : Running with the dump_on_timeout debug flag turned on makes
1923 it possible that the waiting_locks_callback() can try to dump
1924 the Lustre kernel debug logs from an interrupt handler. Defer
1925 this log dumping to the expired_lock_main() thread.
1927 Severity : enhancement
1929 Description: Support NFS exporting on 2.6 kernels.
1930 Details : Implement non-rawops metadata methods for NFS server to use without
1931 changing NFS server code.
1934 Frequency : very rare (synthetic metadata workload only)
1936 Description: two racing renames might cause an MDS thread to deadlock
1937 Details : Running the "racer" program may cause one MDS thread to rename
1938 a file from being the source of a rename to being the target of
1939 a rename at exactly the same time that another thread is doing
1940 so, and the second thread has already enqueued these locks after
1941 doing a lookup of the target and is trying to relock them in
1942 order. Ensure that we don't try to re-lock the same resource.
1945 Frequency : only very large systems with liblustre clients
1947 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1948 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1949 all OSTs in parallel. This allows the eviction of stale liblustre
1950 clients to proceed much faster than if they were done in series,
1951 and also offers similar improvements for other set_info RPCs.
1956 Description: excessive CPU usage during initial read phase on client
1957 Details : During the initial read phase on a client, it would agressively
1958 retry readahead on the file, consuming too much CPU and impacting
1959 performance (since 1.4.5.8). Improve the readahead algorithm
1960 to avoid this, and also improve some other common cases (read
1961 of small files in particular, where "small" is files smaller than
1962 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1967 Description: MDS crash when receiving packet with unknown intent.
1968 Details : Do not LBUG in unknown intent case, just return -EFAULT
1970 Severity : enhancement
1971 Bugzilla : 9293, 9385
1972 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1973 Details : Do not serialize getattr (non-intent version) and statfs.
1976 Frequency : occasional, when OST network is overloaded/intermittent
1978 Description: client evicted by OST after bulk IO timeout
1979 Details : If a client sends a bulk IO request (read or write) the OST
1980 may evict the client if it is unresposive to its data GET/PUT
1981 request. This is incorrect if the network is overloaded (takes
1982 too long to transfer the RPC data) or dropped the OST GET/PUT
1983 request. There is no need to evict the client at all, since
1984 the pinger and/or lock callbacks will handle this, and the
1985 client can restart the bulk request.
1988 Frequency : Always when mmapping file with no objects
1990 Description: client crashes when mmapping file with no objects
1991 Details : Check that we actually have objects in a file before doing any
1992 operations on objects in ll_vm_open, ll_vm_close and
1998 Description: Request leak when working with deleted CWD
1999 Details : Introduce advanced request refcount tracking for requests
2000 referenced from lustre intent.
2002 Severity : Enhancement
2004 Description: Cache open file handles on client.
2005 Details : MDS now will return special lock along with openhandle, if
2006 requested and client is allowed to hold openhandle, even if unused,
2007 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2008 closing files for every read/write openration.
2010 Severity : Enhancement
2012 Description: Cache open negative dentries on client when possible.
2013 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2014 negative dentries on lock revocation.
2019 Description: Remounting a client read-only wasn't possible with a zconf mount
2020 Details : It wasn't possible to remount a client read-only with llmount.
2022 Severity : enhancement
2023 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2024 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2025 MPICH is included to add Lustre-specific ADIO interfaces.
2026 This is based closely on the UFS ADIO layer and only differs
2027 in file creation, in order to allow the OST striping to be set.
2028 This is user-contributed code and not supported by CFS.
2033 Description: extended inode attributes (immutable, append-only) work improperly
2034 when 2.4 and 2.6 kernels are used on client/server or vice versa
2035 Details : Introduce kernel-independent values for these flags.
2037 Severity : enhancement
2040 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2041 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2042 granularity. Now, Fractional MB granularity is be supported,
2043 this is very useful for low memory system.
2045 Severity : enhancement
2047 Description: Getattr by fid
2048 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2049 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2050 have locks and avoids getattr rpc requests for it on every stat.
2053 Frequency : Always, for filesystems larger than 2TB
2055 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2056 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2057 filesystems larger than 2TB to be mounted with mballoc enabled.
2060 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2062 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2063 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2064 filesystems larger than 2TB is mounted then the kernel will
2065 silently corrupt the start of the filesystem. CONFIG_LBD is
2066 enabled for all CFS-supported kernels, but the possibility of
2067 this happening with a modified kernel config exists.
2069 Severity : enhancement
2071 Description: add client O_DIRECT support for 2.6 kernels
2072 Details : It is now possible to do O_DIRECT reads and writes to files
2073 in the Lustre client mountpoint on 2.6 kernel clients.
2075 Severity : enhancement
2077 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2078 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2079 parallel, not waiting for response from every OST before sending
2080 a rpc to the next OST.
2085 Description: setattr vs write race when updating file timestamps
2086 Details : Client processes that update a file timestamp into the past
2087 right after writing to the file (e.g. tar) it is possible that
2088 the updated file modification time can be reset to the current
2089 time due to a race between processing the setattr and write RPC.
2091 Severity : enhancement
2093 Description: Bring 'lfs find' closer in line with regular Linux find.
2094 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2095 -print0 options and obtains all the needed info through the lustre
2098 Severity : enhancement
2100 Description: support up to 1024 configured devices on one node
2101 Details : change obd_dev array from statically allocated to dynamically
2102 allocated structs as they are first used to reduce memory usage
2107 Description: Flush dirty partially truncated pages during truncate
2108 Details : Immediatelly flush partially truncated pages in filter_setattr,
2109 this way we completely avoid having any pages in page cache on OST
2110 and can retire ugly workarounds during writes to flush such pages.
2115 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2116 Details : protect inode from truncation within vfs_unlink() context
2117 just take a reference before calling vfs_unlink() and release it
2118 when parent's i_sem is free.
2123 Description: last_id value checked outside lock on OST caused LASSERT failure
2124 Details : If there were multiple MDS->OST object precreate requests in
2125 flight, it was possible that the OST's last object id was checked
2126 outside a lock and incorrectly tripped an assertion. Move checks
2127 inside locks, and discard old precreate requests.
2130 Frequency : always, if extents are used on OSTs
2132 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2133 Details : OSTs using the extents format would not zero the high 16 bits of
2134 the index physical block number. This is not a problem for any
2135 OST filesystems smaller than 16TB, and no kernels support ext3
2136 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2137 new/modified files) and can be fixed for existing filesystems
2138 with e2fsprogs-1.39-cfs1.
2143 Description: import connection selection may be incorrect if timer wraps
2144 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2145 import connections to be ignored if the 32-bit jiffies counter
2146 wraps. Use a 64-bit jiffies counter.
2149 Frequency : very large clusters immediately after boot
2151 Description: LNET request buffers exhausted under heavy short-term load
2152 Details : If a large number of client requests are generated on a service
2153 that has previously never seen so many requests it is possible
2154 that the request buffer growth cannot keep up with the spike in
2155 demand. Instead of dropping incoming requests, they are held in
2156 the LND until the RPC service can accept more requests.
2159 Frequency : Sometimes during replay
2161 Description: Assertion failure in ll_local_open after replay.
2162 Details : If replay happened on an open request reply before we were able
2163 to set replay handler, reply will become not swabbed tripping the
2164 assertion in ll_local_open. Now we set the handler right after
2165 recognising of open request
2168 Frequency : very rare
2170 Description: kernel reports "badness in vsnprintf"
2171 Details : Reading from the "recovery_status" /proc file in small chunks
2172 may cause a negative length in lprocfs_obd_rd_recovery_status()
2173 call to vsnprintf() (which is otherwise harmless). Exit early
2174 if there is no more space in the output buffer.
2176 Severity : enhancement
2178 Description: clear OBD RPC statistics by writing to them
2179 Details : It is now possible to clear the OBD RPC statistics by writing
2180 to the "stats" file.
2185 Description: Client mtime is not the same on different clients after utimes
2186 Details : In some cases, the client was using the utimes() syscall on
2187 a file cached on another node. The clients now validate the
2188 ctime from the MDS + OSTs to determine which one is right.
2193 Description: Inability to activate failout mode
2194 Details : lconf script incorrectly assumed that in pythong string's numeric
2195 value is used in comparisons.
2198 Frequency : always with multiple stripes per file
2200 Description: Inefficient object allocation for mutli-stripe files
2201 Details : When selecting which OSTs to stripe files over, for files with
2202 a stripe count that divides evenly into the number of OSTs,
2203 the MDS is always picking the same starting OST for each file.
2204 Return the OST selection heuristic to the original design.
2209 Description: mount failures may take full timeout to return an error
2210 Details : Under some heavy load conditions it is possible that a
2211 failed mount can wait for the full obd_timeout interval,
2212 possibly several minutes, before reporting an error.
2213 Instead return an error as soon as the status is known.
2215 Frequency : quota enabled and large files being deleted
2217 Description: releasing more than 4GB of quota at once hangs OST
2218 Details : If a user deletes more than 4GB of files on a single OST it
2219 will cause the OST to spin in an infinite loop. Release
2220 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2225 Description: statfs data retrieved from /proc may be stale or zero
2226 Details : When reading per-device statfs data from /proc, in the
2227 {kbytes,files}_{total,free,avail} files, it may appear
2228 as zero or be out of date.
2231 Frequency : systems with MD RAID1 external journal devices
2233 Description: lconf's call to blkid is confused by RAID1 journal devices
2234 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2235 of returning all block devices that match the journal UUID.
2238 Frequency : always, for aggregate stripe size over 4GB
2240 Description: assertion fails when trying to use 4GB stripe size
2241 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2242 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2245 Frequency : always on ppc64
2247 Description: the first write on an ext3 filesystem with mballoc got stuck
2248 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2249 perform endianness conversion.
2251 ------------------------------------------------------------------------------
2253 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2255 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2256 INTEROPERATE with older versions automatically. Please read the
2257 user documentation before upgrading any part of a live system.
2258 * WARNING: Lustre networking configuration changes are required with
2259 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2262 * Support for kernels:
2263 2.6.9-22.0.2.EL (RHEL 4)
2264 2.6.5-7.244 (SLES 9)
2265 2.6.12.6 vanilla (kernel.org)
2268 Severity : enhancement
2269 Bugzilla : 7981/8208
2270 Description: Introduced Lustre Networking (LNET)
2271 Details : LNET is new networking infrastructure for Lustre, it includes
2272 a reorganized network configuration mode (see the user
2273 documentation for full details) as well as support for routing
2274 between different network fabrics. Lustre Networking Devices
2275 (LNDs) for the supported network fabrics have also been
2276 created for this new infrastructure.
2278 Severity : enhancement
2279 Description: Introduced Access control lists
2280 Details : clients can set ACLs on files and directories in order to have
2281 more fine-grained permissions than the standard Unix UGO+RWX.
2282 The MDS must be started with the "-o acl" mount option.
2284 Severity : enhancement
2285 Description: Introduced filesystem quotas
2286 Details : Administrators may now establish per-user quotas on the
2289 Severity : enhancement
2291 Description: Configuration change for the XT3
2292 The PTLLND is now used to run Lustre over Portals on the XT3
2293 The configure option(s) --with-cray-portals are no longer used.
2294 Rather --with-portals=<path-to-portals-includes> is used to
2295 enable building on the XT3. In addition to enable XT3 specific
2296 features the option --enable-cray-xt3 must be used.
2301 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2302 Details : A race between allocating a new llcd and re-getting the llcd_lock
2303 allowed another thread to grab newly-allocated llcd.
2305 Severity : enhancement
2307 Description: 2.6 OST async journal commit and locking fix to improve performance
2308 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2309 2.6 kernels are now async as they already were in 2.4 kernels so
2310 that they can commit concurrently with the network bulk transfer.
2311 For block-allocated files the filter allocation semaphore is held
2312 to avoid filesystem fragmentation during allocation. BKL lock
2313 removed for 2.6 xattr operations where it is no longer needed.
2318 Description: lconf incorrectly determined whether two IP networks could talk
2319 Details : In some more complicated routing and multiple-network
2320 configurations, lconf will avoid trying to make a network
2321 connection to a disjoint part of the IP space. It was doing the
2322 math incorrectly for one set of cases.
2327 Description: Fix for potential infinite loop processing records in an llog.
2328 Details : If an llog record is corrupted/zeroed, it is possible to loop
2329 forever in llog_process(). Validate the llog record length
2330 and skip the remainder of the block on error.
2333 Frequency : occasional (liblustre only)
2335 Description: liblustre could not open files whose last component is a symlink
2336 Details : sysio_path_walk() would incorrectly pass the open intent to
2337 intermediate path components.
2340 Frequency : rare (liblustre only with non-standard tuning)
2341 Bugzilla : 7201 (7350)
2342 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2343 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2344 after it has been released, only do this for referenced locks
2346 Severity : enhancement
2348 Description: specifying an (invalid) directory default stripe_size of -1
2349 would reset the directory default striping
2350 Details : stripe_size -1 was used internally to signal directory stripe
2351 removal, now use "all default" to signal dir stripe removal
2352 as a directory striping of "all default" is not useful
2355 Frequency : common for large clusters running liblustre clients
2357 Description: doing an ls when liblustre clients are running is slow
2358 Details : sending a glimpse AST to a liblustre client waits for every AST
2359 to time out, as liblustre clients will not respond. Since they
2360 cannot cache data we refresh the OST lock LVB from disk instead.
2362 Severity : enhancement
2364 Description: doing an ls at the same time as file IO can be slow
2365 Details : enqueue and other "small" requests can be blocked behind many
2366 large IO requests. Create a new OST IO portal for non-IO
2367 requests so they can be processed faster.
2370 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2372 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2373 created file with mode 000
2374 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2375 allow an ftruncate() as a non-root user to a file with mode 000.
2376 Lustre now respects this flag to disable mode checks when
2377 truncating a file owned by the user
2380 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2383 Description: Revoking locks from clients that went dead or catatonic might take
2385 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2386 cancellation of such locks instant on servers without waiting for
2387 any reply from clients. Clients drops these locks when cancel
2388 notification from server is received without replying.
2391 Frequency : liblustre-only, when liblustre client dies or becomes busy
2393 Description: Doing ls on Linux clients can take a long time with active
2395 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2396 granting such locks to it in the first place if possible. Locks
2397 are taken by proxy on the OST during the read or write and
2398 dropped immediately afterward. Add connect flags handling, do
2399 not grant locks to liblustre clients for glimpse ASTs.
2401 Severity : enhancement
2403 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2404 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2405 sequential reads and behaves badly with applications doing random
2406 reads. Improve it by reading ahead at least read region, and
2407 avoiding excessive large RPC for small reads.
2409 Severity : enhancement
2411 Description: Creating more than 1000 files for a single job may cause a load
2412 imbalance on the OSTs if there are also a large number of OSTs.
2413 Details : qos_prep_create() uses an OST index reseed value that is an
2414 even multiple of the number of available OSTs so that if the
2415 reseed happens in the middle of the object allocation it will
2416 still utilize the OSTs as uniformly as possible.
2421 Description: OST or MDS may oops in ping_evictor_main()
2422 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2423 but doesn't restart at beginning of obd_exports_timed list
2426 Severity : enhancement
2428 Description: improve by-nid export eviction on the MDS and OST
2429 Details : allow multiple exports with the same NID to be evicted at one
2430 time without re-searching the exports list.
2433 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2435 Description: MDS may oops in groups_free()
2436 Details : in rare race conditions a newly allocated group_info struct is
2437 freed again, and this can be NULL. The 2.4 compatibility code
2438 for groups_free() checked for a NULL pointer, but 2.6 did not.
2441 Frequency : common for liblustre clients doing little filesystem IO
2442 Bugzilla : 9352, 7313
2443 Description: server may evict liblustre clients accessing contended locks
2444 Details : if a client is granted a lock or receives a completion AST
2445 with a blocking AST already set it would not reply to the AST
2446 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2449 Frequency : lfs setstripe, only systems with more than 160 OSTs
2451 Description: unable to set striping with a starting offset beyond OST 160
2452 Details : llapi_create_file() incorrectly limited the starting stripe
2453 index to the maximum single-file stripe count.
2456 Frequency : LDAP users only
2458 Description: lconf did not handle in-kernel recovery with LDAP properly
2459 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2461 Severity : enhancement
2463 Description: bind OST threads to NUMA nodes to improve performance
2464 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2465 node and do their allocations there to localize memory access
2467 Severity : enhancement
2469 Description: llmount can determine client NID directly from Myrinet (GM)
2470 Details : the client NID code from gmnalnid was moved directly into
2471 llmount, removing the need to use this or specifying the
2472 client NID explicitly when mounting GM clients with zeroconf
2475 Frequency : if client is started with down MDS
2477 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2478 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2479 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2480 ptlrpc_import_delayed_req() because we didn't check if the
2481 request was interrupted, and we also didn't break out of the
2482 event loop if there was no timeout
2487 Description: data loss during non-page-aligned writes to a single file from
2488 both multiple nodes and multiple threads on one node at same time
2489 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2490 inconsistency led to false short-reads, that were cached and later
2491 used by ->prepare_write() to fill in partially written page,
2492 leading to data loss.
2495 Frequency : always, if lconf --abort_recovery used
2497 Description: lconf --abort_recovery fails with 'Operation not supported'
2498 Details : lconf was attempting to abort recovery on the MDT device and not
2501 Severity : enhancement
2503 Description: remove cleanup logs
2504 Details : replace lconf-generated cleanup logs with lustre internal
2505 cleanup routines. Eliminates the need for client-cleanup and
2508 Severity : enhancement
2510 Description: add support for EAs (user and system) on lustre filesystems
2511 Details : it is now possible to store extended attributes in the Lustre
2512 client filesystem, and with the user_xattr mount option it
2513 is possible to allow users to store EAs on their files also
2515 Severity : enhancement
2517 Description: Add possibility (config option) to show minimal available OST free
2519 Details : When compiled with --enable-mindf configure option, statfs(2)
2520 (and so, df) will return least minimal free space available from
2521 all OSTs as amount of free space on FS, instead of summary of
2522 free spaces of all OSTs.
2524 Severity : enhancement
2526 Description: do not expand extent locks acquired on OST-side
2527 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2528 by server: they are not cached anyway.
2531 Frequency : when mmap is used/binaries executed from Lustre
2533 Description: Unmmap pages before throwing them away from read cache.
2534 Details : llap_shrink cache now attempts to unmap pages before discarding
2535 them (if unmapping failed - do not discard). SLES9 kernel has
2536 extra checks that trigger if this unmapping is not done first.
2541 Description: lconf didn't resolve symlinks before checking to see whether a
2542 given mountpoint was already in use
2545 Frequency : when migrating failover services
2546 Bugzilla : 6395, 9514
2547 Description: When migrating a subset of services from a node (e.g. failback
2548 from a failover service node) the remaining services would
2549 time out and evict clients.
2550 Details : lconf --force (implied by --failover) sets the global obd_timeout
2551 to 5 seconds in order to quickly disconnect, but this caused
2552 other RPCs to time out too quickly. Do not change the global
2553 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2555 Severity : enhancement
2556 Frequency : if MDS is started with down OST
2557 Bugzilla : 9439,5706
2558 Description: Allow startup/shutdown of an MDS without depending on the
2559 availability of the OSTs.
2560 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2561 Add appropriate locking and lov-osc refcounts for safe
2562 cleaning. Add osc abort_inflight calls in case the
2563 synchronize never started.
2566 Frequency : occasional (Cray XT3 only)
2568 Description: root not authorized to access files in CRAY_PORTALS environment
2569 Details : The client process capabilities were not honoured on the MDS in
2570 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2571 been accessed by an authorized user then root was able to access
2572 the file on the local client also. The root user capabilities
2573 are now allowed on the MDS, as this environment has secure UID.
2576 Frequency : occasional
2578 Description: ldiskfs "too long searching" message happens too often
2579 Details : A debugging message (otherwise harmless) prints too often on
2580 the OST console. This has been reduced to only happen when
2581 there are fragmentation problems on the filesystem.
2586 Description: Division by zero in statfs when all OSCs are inactive
2587 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2588 lov_statfs divides by value returned by lov_get_stripecnt().
2592 Bugzilla : 9489, 3273
2593 Description: First write from each client to each OST was only 4kB in size,
2594 to initialize client writeback cache, which caused sub-optimal
2595 RPCs and poor layout on disk for the first writen file.
2596 Details : Clients now request an initial cache grant at (re)connect time
2597 and so that they can start streaming writes to the cache right
2598 away and always do full-sized RPCs if there is enough data.
2599 If the OST is rebooted the client also re-establishes its grant
2600 so that client cached writes will be honoured under the grant.
2605 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2606 Details : Now I/O RPCs go to different portal number and (presumably) fast
2607 lock requests (and glimses) and other RPCs get their own service
2608 threads pool that should be able to service those RPCs
2611 Severity : enhancement
2613 Description: Ability to exchange lustre version between client and servers and
2614 issue warnings at client side if client is too old. Also for
2615 liblustre clients there is ability to refuse connection of too old
2617 Details : New 'version' field is added to connect data structure that is
2618 filled with version info. That info is later checked by server and
2622 Frequency : rare, liblustre only.
2623 Bugzilla : 9296, 9581
2624 Description: Two simultaneous writes from liblustre at offset within same page
2625 might proceed at the same time overwriting eachother with stale
2627 Details : I/O lock withing llu_file_prwv was released too early, before data
2628 actually was hitting the wire. Extended lock-holding time until
2629 server acknowledges receiving data.
2632 Frequency : extremely rare. Never observed in practice.
2634 Description: avoid generating lustre_handle cookie of 0.
2635 Details : class_handle_hash() generates handle cookies by incrementing
2636 global counter, and can hit 0 occasionaly (this is unlikely, but
2637 not impossible, because initial value of cookie counter is
2638 selected randonly). Value of 0 is used as a sentinel meaning
2639 "unassigned handle" --- avoid it. Also coalesce two critical
2640 sections in this function into one.
2642 Severity : enhancement
2644 Description: allow liblustre clients to delegate truncate locking to OST
2645 Details : To avoid overhead of locking, liblustre client instructs OST to
2646 take extent lock in ost_punch() on client's behalf. New connection
2647 flag is added to handle backward compatibility.
2649 Severity : enhancement
2650 Bugzilla : 4928, 7341, 9758
2651 Description: allow number of OST service threads to be specified
2652 Details : a module parameter allows the number of OST service threads
2653 to be specified via "options ost ost_num_threads={N}" in the
2654 OSS's /etc/modules.conf or /etc/modprobe.conf.
2658 Bugzilla : 6146, 9635, 9895
2659 Description: servers crash with bad pointer in target_handle_connect()
2660 Details : In rare cases when a client is reconnecting it was possible that
2661 the connection request was the last reference for that export.
2662 We would temporarily drop the export reference and get a new
2663 one, but this may have been the last reference and the export
2664 was just destroyed. Get new reference before dropping old one.
2666 Severity : enhancement
2667 Frequency : if client is started with failover MDS
2669 Description: Allow multiple MDS hostnames in the mount command
2670 Details : Try to read the configuration from all specified MDS
2671 hostnames during a client mount in case the "primary"
2674 Severity : enhancement
2676 Description: Stop sending data to evicted clients as soon as possible.
2677 Details : Check if the client we are about to send or are sending data to
2678 was evicted already. (Check is done every second of waiting,
2679 for which l_wait_event interface was extended to allow checking
2680 of exit condition at specified intervals).
2683 Frequency : rare, normally only when NFS exporting is done from client
2685 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2687 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2690 Frequency : always, if srand() is called before liblustre initialization
2692 Description: Liblustre uses system PRNG disturbing its usage by user application
2693 Details : Introduce internal to lustre fast and high-quality PRNG for
2694 lustre usage and make liblustre and some other places in generic
2695 lustre code to use it.
2697 Severity : enhancement
2698 Bugzilla : 9477, 9557, 9870
2699 Description: Verify that the MDS configuration logs are updated when xml is
2700 Details : Check if the .xml configuration logs are newer than the config
2701 logs stored on the MDS and report an error if this is the case.
2702 Request --write-conf, or allow starting with --old_conf.
2704 Severity : enhancement
2706 Description: Handle symlinks in the path when checking if Lustre is mounted.
2707 Details : Resolve intermediate symlinks when checking if a client has
2708 mounted a filesystem to avoid duplicate client mounts.
2713 Description: lconf can hit an error exception but still return success.
2714 Details : The lconf command catches the Command error exception at the top
2715 level script context and will exit with the associated exit
2716 status, but doesn't ensure that this exit status is non-zero.
2721 Description: failure of ptlrpc thread startup can cause oops
2722 Details : Starting a ptlrpc service thread can fail if there are a large
2723 number of threads or the server memory is very fragmented.
2724 Handle this without oopsing.
2727 Frequency : always, only if liblustre and non-default acceptor port was used
2729 Description: liblustre cannot connect to servers with non-default acceptor port
2730 Details : tcpnal_set_default_params() was not called and was therefore
2731 ignoring the environment varaible TCPNAL_PORT, as well as other
2732 TCPNAL_ environment variables
2737 Description: two objects could be created on the same OST for a single file
2738 Details : If an OST is down, in some cases it was possible to create two
2739 objects on a single OST for a single file. No problems other
2740 than potential performance impact and spurious error messages.
2744 Bugzilla : 5681, 9562
2745 Description: Client may oops in ll_unhash_aliases
2746 Details : Client dcache may become inconsistent in race condition.
2747 In some cases "getcwd" can fail if the current directory is
2753 Description: Inode refcounting problems in NFS export code
2754 Details : link_raw functions used to call d_instantiate without obtaining
2755 extra inode reference first.
2759 Bugzilla : 9942, 9903
2760 Description: Referencing freed requests leading to crash, memleaks with NFS.
2761 Details : We used to require that call to ll_revalidate_it was always
2762 followed by ll_lookup_it. Also with revalidate_special() it is
2763 possible to call ll_revalidate_it() twice for the same dentry
2764 even if first occurence returned success. This fix changes semantic
2765 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2766 reference on a request referred from the intent.
2767 ll_intent_release() then releases such a request.
2770 Frequency : rare, normally benchmark loads only
2772 Description: unlinked inodes were kept in memory on the client
2773 Details : If a client is repeatedly creating and unlinking files it
2774 can accumulate a lot of stale inodes in the inode slab cache.
2775 If there is no other client load running this can cause the
2776 client node to run out of memory. Instead flush old inodes
2777 from client cache that have the same inode number as a new inode.
2780 Frequency : SLES9 2.6.5 kernel and long filenames only
2781 Bugzilla : 9969, 10379
2782 Description: utime reports stale NFS file handle
2783 Details : SLES9 uses out-of-dentry names in some cases, which confused
2784 the lustre dentry revalidation. Change it to always use the
2788 Frequency : rare, unless heavy write-truncate concurrency is continuous
2789 Bugzilla : 4180, 6984, 7171, 9963, 9331
2790 Description: OST becomes very slow and/or deadlocked during object unlink
2791 Details : filter_destroy() was holding onto the parent directory lock
2792 while truncating+unlinking objects. For very large objects this
2793 may block other threads for a long time and slow overall OST
2794 responsiveness. It may also be possible to get a lock ordering
2795 deadlock in this case, or run out of journal credits because of
2796 the combined truncate+unlink. Solution is to do object truncate
2797 first in one transaction without parent lock, and then do the
2798 final unlink in a new transaction with the parent lock. This
2799 reduces the lock hold time dramatically.
2802 Frequency : rare, 2.4 kernels only
2804 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2805 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2806 before unmounting their filesystem in order to avoid deadlock.
2807 The kernel_locked() function in 2.4 kernels only checks whether
2808 the kernel lock is held, not whether it is this process that is
2809 holding it as 2.6 kernels do.
2814 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2815 Details : The client ptlrpc code may be trying to reconnect to a down
2816 server before a previous connection attempt has timed out.
2817 Increase the reconnect interval to be longer than the connection
2818 timeout interval to avoid sending duplicate connections to
2822 Frequency : echo_client brw_test command
2824 Description: fix echo_client to work with OST preallocated code
2825 Details : OST preallocation code (5137) didn't take echo_client IO path
2826 into account: echo_client calls filter methods outside of any
2827 OST thread and, hence, there is no per-thread preallocated
2828 pages and buffers to use. Solution: hijack pga pages for IO. As
2829 a byproduct, this avoids unnecessary data copying.
2833 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2834 Description: Client can oops in mdc_commit_close() after open replay
2835 Details : It was possible for the MDS to return an open request with no
2836 transaction number in mds_finish_transno() if the client was
2837 evicted, but without actually returning an error. Clients
2838 would later try to replay that open and may trip an assertion
2839 Simplify the client close codepath, and always return an error
2840 from the MDS in case the open is not successful.
2843 Frequency : rare, 2.6 OSTs only
2845 Description: OST may deadlock under high load on fragmented files
2846 Details : If there was a heavy load and highly-fragmented OST filesystems
2847 it was possible to have all the OST threads deadlock waiting on
2848 allocation of biovecs, because the biovecs were not released
2849 until the entire RPC IO was completed. Instead, release biovecs
2850 as soon as they are complete to ensure forward IO progress.
2852 Severity : enhancement
2854 Description: Support for specifying external journal device at mount
2855 Details : If an OST or MDS device is formatted with an external journal
2856 device, this device major/minor is stored in the ext3 superblock
2857 and may not be valid for failover. Allow detecting and
2858 specifying the external journal at mount time.
2863 Description: Mounting an MDS with pending unlinked files may cause oops
2864 Details : target_finish_recovery() calls mds_postrecov() which returned
2865 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2866 considers this an error and immediately begins cleaning up the
2867 lov, just after starting the mds_lov process
2869 Severity : enhancement
2871 Description: Implement 'lfs df' to report actual free space on per-OST basis
2872 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2873 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2874 usage of objects; '-h' to report in human readable format.
2876 ------------------------------------------------------------------------------
2878 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2885 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2886 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2887 at startup during orphan unlinking. Instead, skip buddy update
2888 before setup, as it will be regenerated after recovery is complete.
2893 Description: If an OST is inactive, its locks might reference stale inodes.
2894 Details : lov_change_cbdata() must iterate over all namespaces, even if
2895 they are inactive to clear inode references from the lock.
2897 Severity : enhancement
2898 Frequency : occasional, if non-standard max_dirty_mb used
2900 Description: Client will block write RPCs if not enough grant
2901 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2902 then the client will block writes while waiting for another RPC
2903 to complete instead of consuming its dirty limit. With change
2904 we get improved performance when max_dirty_mb is small.
2906 Severity : enhancement
2907 Bugzilla : 3389, 6253
2908 Description: Add support for supplementary groups on the MDS.
2909 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2910 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2911 lookups for user supplementary groups into a cache.
2915 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2916 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2917 for liblustre O_EXCL is enforced by MDS.
2921 Description: readdir never returns NULL in liblustre.
2922 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2923 dentry in struct dirent.
2927 Frequency : liblustre only, depends on application IO pattern
2928 Description: liblustre clients evicted if not contacting servers
2929 Details : Don't put liblustre clients into the ping_evictor list, so
2930 they will not be evicted by the pinger ever.
2932 Severity : enhancement
2934 Description: Add ability to evict clients by NID from MDS.
2935 Details : By echoing "nid:$NID" string into
2936 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2937 $NID would be instantly evicted from this MDS and from all active
2938 OSTs connected to it.
2942 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2943 Details : lookup_it_finish() used to query file size from OSTs that was not
2948 Description: service threads change working directory to that of init
2949 Details : Starting lustre service threads may pin the working directory
2950 of the parent thread, making that filesystem busy. Threads
2951 now change to the working directory of init to avoid this.
2955 Frequency : during shutdown only
2956 Description: shutdown with a failed MDS or OST can cause unmount to hang
2957 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2962 Frequency : 2.6 only
2963 Description: chmod/chown may include an extra supplementary group
2964 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2965 supplementary group and if none is specified this is used.
2968 Bugzilla : 5479 (6816)
2970 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2971 Details : If lookup is in progress on a file that is unlinked we might try
2972 to revalidate the inode and fail in revalidate after lookup is
2973 complete and ll_file_open() enqueues the open again but
2974 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2977 Frequency : always, if lconf --abort_recovery used
2979 Description: lconf --abort_recovery fails with 'Operation not supported'
2980 Details : lconf was attempting to abort recovery on the MDT device and not
2983 ------------------------------------------------------------------------------
2985 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2990 Frequency : rare (only unsupported configurations with a node running as an
2992 Bugzilla : 6514, 5137
2993 Description: Mounting a Lustre file system on a node running as an OST could
2995 Details : OSTs now preallocates memory needed to write out data at
2996 startup, instead of when needed, to avoid having to
2997 allocate memory in possibly low memory situations.
2998 Specifically, if the file system is mounted on on OST,
2999 memory pressure could force it to try to write out data,
3000 which it needed to allocate memory to do. Due to the low
3001 memory, it would be unable to do so and the node would
3002 become unresponsive.
3004 Severity : enhancement
3006 Description: Addition of lconf --service command line option
3007 Details : lconf now accepts a '--service <arg>' option, which is
3008 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3010 Severity : enhancement
3012 Description: Failover mode is now the default for OSTs.
3013 Details : By default, OSTs will now run in failover mode. To return to
3014 the old behaviour, add '--failout' to the lmc line for OSTs.
3016 Severity : enhancement
3018 Description: Health checks are now provided for MDS and OSTs
3019 Details : Additional detailed health check information on MSD and OSTs
3020 is now provided through the procfs health_check value.
3023 Frequency : occasional, depends on IO load
3025 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3026 after numerous create/delete cycles
3027 Details : The ext3 inode allocation policy would not allocate new inodes
3028 very well on the OSTs because there are no new directories
3029 being created. Instead we look for groups with free space if
3030 the parent directories are nearly full.
3035 Description: Network or server problems during mount may cause partially
3036 mounted clients instead of returning an error.
3037 Details : The config llog parsing code may overwrite the error return
3038 code during mount error handling, returning success instead
3044 Description: MDS can fail to allocate large reply buffers
3045 Details : After long uptimes the MDS can fail to allocate large reply
3046 buffers (e.g. zconf client mount config records) due to memory
3047 fragmentation or consumption by the buffer cache. Preallocate
3048 some large reply buffers so that these replies can be sent even
3049 under memory pressure.
3053 Frequency : rare (liblustre)
3054 Description: fsx running with liblustre complained that using truncate() to
3055 extend the file doesn't work. This patch corrects that issue.
3056 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3057 ATTR_SIZE and lsm use in llu_setattr_raw.
3061 Frequency : rare, only 2.6 kernels
3062 Description: Unusual file access patterns on the MDS may result in inode
3063 data being lost in very rare circumstances.
3064 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3065 code in the 2.6 kernel caused the inode and/or EA data not to
3066 be read from disk, causing single-file corruption.
3070 Frequency : rare, only 2.6 filesystems using extents
3071 Description: Heavy concurrent write and delete load may cause data corruption.
3072 Details : It was possible under high-load situations to have an extent
3073 metadata block in the block device cache from a just-unlinked
3074 file overwrite a newly-allocated data block. We now unmap any
3075 metadata buffers that alias just-allocated data blocks.
3079 Frequency : filesystems with default stripe_count larger than 77
3080 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3081 Details : lconf specifies an inode size of 4096 bytes when the default
3082 stripe_count is larger than 77. This conflicts with the default
3083 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3084 this case to avoid pinning too much memory for large EAs.
3086 ------------------------------------------------------------------------------
3088 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3093 Frequency : rare (extremely heavy IO load with hundreds of clients)
3095 Description: Client is evicted, gets IO error writing to file
3096 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3097 caused clients to be evicted for AST timeouts. The fixes in
3098 bug 5192 mean we no longer need to have such short AST timeouts
3099 so ldlm_timeout has been increased.
3102 Frequency : occasional during --force or --failover shutdown under load
3103 Bugzilla : 5949, 4834
3104 Description: Server oops/LBUG if stopped with --force or --failover under load
3105 Details : a collection of import/export refcount and cleanup ordering
3106 issues fixed for safer force cleanup
3109 Frequency : only filesystems larger than 120 OSTs
3110 Bugzilla : 5990, 6223
3111 Description: lfs getstripe would oops on a very large filesystem
3112 Details : lov_getconfig used kfree on vmalloc'd memory
3115 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3116 Bugzilla : 6242, 6243
3117 Description: reading from files that had been truncated to a non-zero size
3118 but never opened returned no data
3119 Details : ll_file_read() reads zeros from no-object files to EOF
3124 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3125 Details : There's little harm in aborting MDS/OSS recovery and letting it
3126 try again, so I removed the LASSERT and return an error instead.
3128 Severity : enhancement
3130 Description: New debugging infrastructure for tracking down data corruption
3131 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3132 (b) cover more of the client-side code path, and (c) try to narrow
3133 down where problems occurred
3137 Bugzilla : 3819, 4364, 4397, 6313
3138 Description: Racing close and eviction MDS could cause assertion in mds_close
3139 Details : It was possible to get multiple mfd references during close and
3140 client eviction, leading to one thread referencing a freed mfd.
3142 Severity: : enhancement
3143 Bugzilla : 3262, 6359
3144 Description: Attempts to reconnect to servers are now more aggressive.
3145 Details : This builds on the enhanced upcall-less recovery that was added
3146 in 1.4.2. When trying to reconnect to servers, clients will
3147 now try each server in the failover group every 10 seconds. By
3148 default, clients would previously try one server every 25 seconds.
3153 Description: After recovery, certain operations trigger a failed
3154 assertion on a client.
3155 Details : Failing over an mds, using lconf -d --failover, while a
3156 client was doing a readdir() call would cause the client to
3157 LBUG after recovery completed and the readdir() was resent.
3159 Severity : enhancement
3161 Description: Default groups are now added by lconf
3162 Details : You can now run lconf --group <servicename> without having to
3163 manually add groups with lmc.
3166 Frequency : occasional
3168 Description: Nodes with an elan id of 0 trigger a failed assertion
3171 Frequency : always when accessing e.g. tty/console device nodes
3173 Description: tty and some other devices nodes cannot be used on lustre
3174 Details : file's private_data field is used by device data and lustre
3175 values in there got lost. New field was added to struct file to
3176 store fs-specific private data.
3179 Frequency : when exporting Lustre via NFS
3181 Description: NFSD failed occasionally when looking up a path component
3182 Details : NFSD is looking up ".." which was broken in ext3 directories
3183 that had grown large enough to become hashed.
3186 Frequency : Clusters with multiple interfaces not on the same subnet
3188 Description: Nodes will repeatedly try to reconnect to an interface which it
3189 cannot reach and report an error to the log.
3190 Details : Extra peer list entries will be created by lconf with some peers
3191 unreachable. lconf now validates the peer before adding it.
3194 Frequency : Only if a default stripe is set on the filesystem root.
3196 Description: Setting a default stripe on the filesystem root prevented the
3197 filesystem from being remounted.
3198 Details : The client was sending extra request flags in the root getattr
3199 request and did not allocate a reply buffer for the dir EA.
3202 Frequency : occasional, higher if lots of files are accessed by one client
3203 Bugzilla : 6159, 6097
3204 Description: Client trips assertion regarding lsm mismatch/magic
3205 Details : While revalidating inodes the VFS looks up inodes with ifind()
3206 and in rare cases can find an inode that is being freed.
3207 The ll_test_inode() code will free the lsm during ifind()
3208 when it finds an existing inode and then the VFS later attaches
3209 this free lsm to a new inode.
3213 Bugzilla : 6422, 7030
3214 Description: MDS deadlock between mkdir and client eviction
3215 Details : Creating a new file via mkdir or mknod (starting a transaction
3216 and getting the ns lock) can deadlock with client eviction
3217 (gets ns lock and trying to finish a synchronous transaction).
3220 Frequency : occasional
3221 Description: While starting a server, the fsfilt_ext3 module could not be
3223 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3224 kernels. Previously, lconf would still use the ext3 name
3225 when trying to load modules. Now, it will correctly use
3226 ext3 on 2.4 and ldiskfs on 2.6.
3228 Severity : enhancement
3229 Description: The default stripe count has been changed to 1
3230 Details : The interpretation of the default stripe count (0, to lfs
3231 or lmc) has been changed to mean striping across a single
3232 OST, rather than all available. For general usage we have
3233 found a stripe count of 1 or 2 works best.
3235 Severity : enhancement
3236 Description: Add support for compiling against Cray portals.
3237 Details : Conditional compiling for some areas that are different
3241 Frequency : occasional
3242 Bugzilla : 6409, 6834
3243 Description: Creating files with an explicit stripe count may lead to
3244 a failed assertion on the MDS
3245 Details : If some OSTs are full or unavailable, creating files may
3246 trigger a failed assertion on the MDS. Now, Lustre will
3247 try to use other servers or return an error to the
3251 Frequency : occasional
3253 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3254 caused lock timeout during readahead (since 1.4.2).
3255 Details : Processes doing readahead might match a lock that hasn't been
3256 granted yet if there are overlapping and conflicting lock
3257 requests. The readahead process waits on ungranted lock
3258 (original lock is CBPENDING), while OST waits for that process
3259 to cancel CBPENDING read lock and eventually evicts client.
3261 Severity : enhancement
3263 Description: Initial enabling of flock support for clients
3264 Details : Implements fcntl advisory locking and file status functions.
3265 This feature is provided as an optional mount flag (default
3266 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3267 locking are implemented yet, and those that are are not guaranteed
3268 to be completely correct in production environments.
3269 mount -t lustre -o [flock|noflock] ...
3272 Frequency : occasional
3274 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3275 assertion in the ext3 JBD (journaling) layer.
3276 Details : The b_committed_data struct is protected by the big kernel lock
3277 in 2.4 kernels, serializing journal_commit_transaction() and
3278 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3279 access to this struct. In 2.6 kernels there is finer grained
3280 locking to improve SMP performance of the JBD layer.
3284 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3289 Description: merge and cleanup kernel patches.
3292 -----------------------------------------------------------------------------
3294 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3296 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3297 versions of Lustre. Please update all servers and clients to
3298 version 1.4.2 or later at the same time. You must also run
3299 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3300 to update the configuration logs.
3302 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3303 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3304 - swab llog records of type '0' so we get proper header size/idx (5861)
3305 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3306 - fix rename of one directory over another leaking an inode (5953)
3307 - avoid SetPageDirty on 2.6 (5981)
3308 - don't re-add just-being-destroyed locks to the waiting list (5653)
3309 - when creating new directories, inherit the parent's custom
3310 striping settings if present parent (3048)
3311 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3312 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3313 - don't hold client locks on temporary worklist from l_lru (5666)
3314 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3315 - automatically evict dead clients (5921)
3316 - Update file size properly in create+truncate+fstat case (6196)
3317 - Do not unhash mountpoint dentries, do not allow removal of
3319 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3320 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3321 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3322 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3323 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3324 - Correctly handle reads of files with no objects (6243)
3325 - lctl recover will also mark a device active if deactivate used (5933)
3327 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3328 - allow --write-conf on an MDS with different nettype than client (5619)
3329 - don't write config llogs to MDS for mounts not from that MDS (5617)
3330 - lconf should create multiple TCP connections from a client (5201)
3331 - init scripts are now turned off by default; run chkconfig --on
3332 lustre and chkconfig --on lustrefs to use them
3333 - upcalls are no longer needed for clients to recover to failover
3335 - add --abort-recovery option to lconf to abort recovery on device
3337 - add support for an arbitrary number of OSTs (3026)
3338 - Quota support protocol changes.
3339 - forward compatibility changes to wire structs (6007)
3340 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3341 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3342 - improved "lustre status" script
3343 - initialize blocksize for non-regular files (6062)
3344 - added --disable-server and --disable-client configure options (5782)
3345 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3346 - Vanilla 2.4.29 support
3347 - increase maximum number of obd devices to 520 (6242)
3348 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3349 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3350 SLES 9 2.6.5 kernels
3351 - limit stripes per file to 160 (the maximum EA size) (6093)
3353 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3356 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3357 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3358 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3359 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3360 - small ext3 extents cleanups and fixes (5733)
3361 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3362 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3363 - avoid needless client->OST connect, fix handle mismatch (5317)
3364 - fix DLM error path that led to out-of-sync client, long delays (5779)
3365 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3366 - fix several locking issues related to i_size (5492,5624,5654,5672)
3367 - don't move pending lock onto export if it is already evicted (5683)
3368 - fix kernel oops when creating .foo in unlinked directory (5548)
3369 - fix deadlock in obdfilter statistics vs. object create (5811)
3370 - use time_{before,after} to avoid timer jiffies wrap (5882)
3371 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3372 - Do not leak request if server was not able to process it (5154)
3373 - If mds_open unable to find parent dir, make that negative lookup(5154)
3374 - don't create new directories with extent-mapping (5909, 5936)
3376 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3377 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3378 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3379 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3380 - support for RHEL4 kernel on i686 (5773)
3381 - provide error messages when incompatible logs are encountered (5898)
3383 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3384 * version 1.4.0.10 (1.4.1 release candidate 1)
3386 - don't keep a lock reference when lock is not granted (4238)
3387 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3388 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3389 - drop import inflight refcount on signal_completed_replay error (5255)
3390 - unlock page after async write error during send (3677)
3391 - handle missing objects in filter_preprw_read properly (5265)
3392 - no transno return for symlink open, don't save no-trasno open (3440)
3393 - don't try to complete elan receive that already failed (4012)
3394 - free RPC server reply state on error (5406)
3395 - clean up thread from ptlrpc_start_thread() on error (5160)
3396 - readahead could read extra page into cache that wasn't ejected (5388)
3397 - prevent races in class_attach/setup/cleanup/detach (5260)
3398 - don't dereference de->d_inode after l_dput of de (5458)
3399 - use "int" for stripe value returned from lock_to_stripe (5544)
3400 - mballoc allocation and error-checking fixes in 2.6 (5504)
3401 - block device patches to fix I/O request sizes in 2.6 (5482)
3402 - look up hostnames for IB nals (5602)
3403 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3404 - don't start multiple acceptors for the same port (5277)
3405 - fix incorrect LASSERT in mds_getattr_name (5635)
3406 - export a proc file for general "ping" checking (5628)
3407 - fix "lfs check" to not block when the MDS is down (5628)
3409 - service request history (4965)
3410 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3411 - create an "evict_client" /proc entry on OSTs, like the MDS has
3412 - fix mount usage message, return errors per mount(8) (5168)
3413 - change grep [] to grep "[]" in tests so they work in more UMLs
3414 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3415 - remove ancient LOV_MAGIC_V0 EA support (5047)
3416 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3417 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3418 - allow readahead allocations to fail when low on memory (5383)
3419 - mmap locking landed again, after considerable improvement (2828)
3420 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3422 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3425 - send OST transaction number in read/write reply to free req (4966)
3426 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3427 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3428 - the watchdog thread now runs as interruptible (5246)
3429 - flock/lockf fixes (but it's still disabled, pending 5135)
3430 - don't use EXT3 constants in llite code (5094)
3431 - memory shortage at startup could cause assertion (5176)
3433 - reorganization of lov code
3434 - single portals codebase
3436 - add extents/mballoc support (5025)
3437 - direct I/O reads in the obdfilter (4048)
3438 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3440 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3443 - send OST transaction number in read/write reply to free req (4966)
3444 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3445 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3446 - the watchdog thread now runs as interruptible (5246)
3447 - handle missing objects in filter_preprw_read properly (5265)
3448 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3449 - drop import inflight refcount on signal_completed_replay error (5255)
3450 - unlock page after async write error during send (3677)
3451 - return original error code on reconstructed replies (3761)
3452 - no transno return for symlink open, don't save no-trasno open (3440)
3454 - add pid to ldlm debugging output (4922)
3455 - bump the watchdog timeouts -- we can't handle 30sec yet
3456 - extra debugging for orphan dentry/inode bug (5259)
3458 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3461 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3462 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3463 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3464 - fix NULL dereference in /proc/sys/portals/routes (4827)
3465 - allow failed mdc_close() operations to be interrupted (4561)
3466 - stop precreate on OST before MDS would time out on it (4778)
3467 - don't send partial-page writes before EOF from client (4410)
3468 - discard client grant for sub-page writes on large-page clients (4520)
3469 - don't free dentries not owned by NFS code, check generation (4806)
3470 - fix lsm leak if mds_create_objects() fails (4801)
3471 - limit debug_daemon file size, always print CERROR messages (4789)
3472 - use transno after validating reply (3892)
3473 - process timed out requests if import state changes (3754)
3474 - update mtime on OST during writes, return in glimpse (4829)
3475 - add mkfsoptions to LDAP (4679)
3476 - use ->max_readahead method instead of zapping global ra (5039)
3477 - don't interrupt __l_wait_event() during strace
3479 - add software watchdogs to catch hung threads quickly (4941)
3480 - make lustrefs init script start after nfs is mounted
3481 - fix CWARN/ERROR duplication (4930)
3482 - return async write errors to application if possible (2248)
3483 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3484 - print NAL number in %x format (4645)
3485 - update barely-supported suse-2.4.21-171 series (4842)
3486 - support for sles 9 %post scripts
3487 - support for building 2.6 kernel-source packages
3488 - support for sles km_* packages
3490 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3493 - ignore -ENOENT errors in osc_destroy (3639)
3494 - notify osc create thread that OSC is being cleaned up (4600)
3495 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3496 - reconstruct ost_handle() like mds_handle() (4657)
3497 - create a new thread to do import eviction to avoid deadlock (3969)
3498 - let lconf resolve symlinked-to devices (4629)
3499 - don't unlink "objects" from directory with default EA (4554)
3500 - hold socknal file ref over connect in case target is down (4394)
3501 - allow more than 32000 subdirectories in a single directory (3244)
3502 - fix blocks count for O_DIRECT writes (3751)
3503 - OST returns ENOSPC from object create when no space left (4539)
3504 - don't send truncate RPC if file size isn't changing (4410)
3505 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3506 - bind to privileged port in socknal and tcpnal (3689)
3508 - rate limit CERROR/CWARN console message to avoid overload (4519)
3509 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3510 - basic mmap support (3918)
3511 - kernel patch series update from b1_4 (4711)
3513 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3516 - avoid crash during MDS cleanup with OST shut down (2775)
3517 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3518 - don't use bad inodes on the MDS (3744)
3519 - dynamic object preallocation to improve recovery speed (4236)
3520 - don't hold spinlock over lock dumping or change debug flags (4401)
3521 - don't zero obd_dev when it is force cleaned (3651)
3522 - print grants to console if they go negative (4431)
3523 - "lctl deactivate" will stop automatic recovery attempts (3406)
3524 - look for existing locks in ldlm_handle_enqueue() (3764)
3525 - don't resolve lock handle twice in recovery avoiding race (4401)
3526 - revalidate should check working dir is a directory (4134)
3528 - don't always mark "slow" obdfilter messages as errors (4418)
3530 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3533 - don't close LustreDB during write_conf until it is done (3860)
3534 - fix typo in lconf for_each_profile (3821)
3535 - allow dumping logs from multiple threads at one time (3820)
3536 - don't allow multiple threads in OSC recovery (3812)
3537 - fix debug_size parameters (3864)
3538 - fix mds_postrecov to initialize import for llog ctxt (3121)
3539 - replace config semaphore with spinlock (3306)
3540 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3541 - don't allow enqueue to complete on a destroyed export (3822)
3542 - down write_lock before checking llog header bitmap (3825)
3543 - recover from lock replay timeout (3764)
3544 - up llog sem before sending rpc (3652)
3545 - reduce ns lock hold times when setting kms (3267)
3546 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3547 - fix NULL deref and obd_dev leak on setup error (3312)
3548 - replace some LBUG about llog ops with error handling (3841)
3549 - don't match INVALID dentries from d_lookup and spin (3784)
3550 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3551 - fix invalid assertion in ptlrpc_set_wait (3880)
3553 - add libwrap support for the TCP acceptor (3996)
3554 - add /proc/sys/portals/routes for non-root route listing (3994)
3555 - allow setting MDS UUID in .xml (2580)
3556 - print the stack of a process that LBUGs (4228)
3558 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3561 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3562 - make sure to unset replay flag from failed open requests (3440)
3563 - if default stripe count is 0, use OST count for inode size (3636)
3564 - update parent mtime/ctime on client for create/unlink (2611)
3565 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3566 - free recovery state on server during a forced cleanup (3571)
3567 - unregister_reply for resent reqs (3063)
3568 - loop back devices mounting and status check on 2.6 (3563)
3569 - fix resource-creation race that can provoke i_size == 0 (3513)
3570 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3571 - more debugging for page-accounting assertion (3746)
3572 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3573 - avoid deadlock after precreation failure (3758)
3574 - fix race and lock order deadlock in orphan handling (3450, 3750)
3575 - add validity checks when grabbing inodes from l_ast_data (3599)
3577 - add /proc/.../recovery_status to obdfilter (3428)
3578 - lightweight CDEBUG infrastructure, debug daemon (3668)
3579 - change default OSC RPC parameters to be better on small clusters
3580 - turn off OST read cache for files smaller than 32MB
3581 - install man pages and include them in rpms (3100)
3582 - add new init script for (un)mounting lustre filesystems (2593)
3583 - run chkconfig in %post for init scripts (3701)
3584 - drop scimac NAL (unmaintained)
3586 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3589 - clean kiobufs before and after use (3485)
3590 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3591 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3592 - change init script to fail more clearly if not run as root (1528)
3593 - allow clients to reconnect during replay (1742)
3594 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3595 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3596 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3597 - fix cond_resched (3554)
3598 - only evict unfinished clients after recovery (3515)
3599 - allow bulk resend, prevent data loss (3570)
3600 - dynamic ptlrpc request buffer allocation (2102)
3601 - don't allow unlinking open directory if it isn't empty (2904)
3602 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3603 - remove extraneous obd dereference causing LASSERT failure (3334)
3604 - don't use get_cycles() when creating temp. files on the mds (3156)
3605 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3606 - handle EEXIST for set-stripe, set proper directory name (3336)
3608 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3609 - fix ksocknal_fmb_callback() error messages (2918)
3611 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3614 - don't copy lvb into (possibly NULL) reply on error (2983)
3615 - don't deref dentry after dput, don't free lvb on error (2922)
3616 - use the kms to determine writeback rpc length (2947)
3617 - increment oti_logcookies when osc is inactive (2948)
3618 - update client's i_blocks count via lvb messages (2543)
3619 - handle intent open/close of special files properly (1557)
3620 - mount MDS with errors=remount-ro, like obdfilter (2009)
3621 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3622 - don't use cancelling-locks' kms values (2947)
3623 - use highest lock extent for kms, not last one (2925)
3624 - don't dereference ERR_PTR() dentry in error handling path (3107)
3625 - fix thread race in portals_debug_dumplog() (3122)
3626 - create lprocfs device entries at setup instead of at attach (1519)
3627 - common AST error handler, don't evict client on completion race (3145)
3628 - zero nameidata in detach_mnt in 2.6 (3118)
3629 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3630 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3631 - initialize RPC timeout timer earlier for 2.6 (3219)
3632 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3633 - print nal/nid for unknown nid (3258)
3634 - additional checks for oscc recovery before doing precreate (3284)
3635 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3636 - don't crash in mdc_close for bad permissions on open (3285)
3637 - zero i_rdev for non-device files (3147)
3638 - clear page->private before handing to FS, better assertion (3119)
3639 - tune the read pipeline (3236)
3640 - fix incorrect decref of invalidated dentry (2350)
3641 - provide read-ahead stats and refine rpc in flight stats (3328)
3642 - don't hold journal transaction open across create RPC (3313)
3643 - update atime on MDS at close time (3265)
3644 - close LDAP connection when recovering to avoid server load (3315)
3645 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3646 - don't leak open file on MDS after open resend (3325)
3647 - serialize filter_precreate and filter_destroy_precreated (3329)
3648 - loop device shouldn't call sync_dev() for nul device (3092)
3649 - clear page cache after eviction (2766)
3650 - resynchronize MDS->OST in background (2824)
3651 - refuse to mount the same filesystem twice on same mountpoint (3394)
3652 - allow llmount to create routes for mounting behind routers (3320)
3653 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3654 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3655 - fix rare problem with rename on htree directories (3417)
3657 - allow default OST striping configuration per directory (1414)
3658 - fix compilation for qswnal for 2.6 kernels (3125)
3659 - increase maximum number of MDS request buffers for large systems
3660 - change liblustreapi to be useful for external progs like lfsck (3098)
3661 - increase local configuration timeout for slow disks (3353)
3662 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3664 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3667 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3668 - don't overwrite extent policy data in reply if lock was blocked (2901)
3669 - drop filter export grants atomically with removal from device (2663)
3670 - del obd_self_export from work_list in class_disconnect_exports (2908)
3671 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3672 - swab reply message in mdc_close, other PPC fixes (2464)
3673 - fix destroying of named logs (2325)
3674 - overwrite old logs when running lconf --write_conf (2264)
3675 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3676 - fix race in target_handle_connect (2898)
3677 - mds_reint_create() should take same inode create lock (2926)
3678 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3679 - don't close files for self_export to avoid uninitialized obd (2936)
3680 - allow MDS with the same name as client node (2939)
3681 - hold dentry reference for closed log files for unlink (2325)
3682 - reserve space for all logs during transactions (2059)
3683 - don't evict page beyond end of stripe extent (2925)
3684 - don't oops on a deleted current working directory (2399)
3685 - handle hard links to targets without a parent properly (2517)
3686 - don't dereference NULL lock when racing during eviction (2867)
3687 - don't grow lock extents when lots of conflicting locks (2919)
3689 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3692 - account for cache space usage on clients to avoid data loss (974)
3693 - lfsck support in lustre kernel code (2349)
3694 - reduce journal credits needed for BRW writes (2370)
3695 - orphan handling to avoid losing space on client/server crashes
3696 - ptlrpcd can be blocked, stopping ALL progress (2477)
3697 - use lock value blocks to assist in proper KMS, faster stat (1021)
3698 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3699 - recovery for initial connections (2355)
3700 - fixes for mds_cleanup_orphans (1934)
3701 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3702 - block all file creations until orphan recovery completes (1901)
3703 - client remove rq_connection from request struct (2423)
3704 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3705 - recovery timer race (2670)
3706 - mdc_close recovey bug (2532)
3707 - ptlrpc cleanup bug (2710)
3708 - mds timeout on local locks (2588)
3709 - namespace lock held during RPCs (2431)
3710 - handle interrupted sync write properly (2503)
3711 - don't try to handle a message that hasn't been replied to (2699)
3712 - client assert failure during cleanup after abort recovery (2701)
3713 - leak mdc device after failed mount (2712)
3714 - ptlrpc_check_set allows timedout requests to complete (2714)
3715 - wait for inflight reqs when ptlrpcd finishes (2710)
3716 - make sure unregistered services are removed from the srv_list
3717 - reset bulk XID's when resending them (caught by 1138 test)
3718 - unregister_bulk after timeout
3719 - fix lconf error (2694)
3720 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3721 - readahead locks pages, leaves pending causing memory pressure (2673)
3722 - increase OST request buffers to 4096 on large machines (2729)
3723 - fix up permission of existing directories in simple_mkdir (2661)
3724 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3725 - don't assign transno to errored transactions (2742)
3726 - don't delete objects on OST if given a bogus objid from MDS (2751)
3727 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3728 - if rq_no_resend, then timeout request after recovery (2432)
3729 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3730 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3731 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3732 - MDS recovery shouldn't delete objects that it has given out (2730)
3733 - if enqueue arrives after completion, don't clobber LVB (2819)
3734 - don't unlock pages twice when trigger_group_io returns error (2814)
3735 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3736 - don't write pages to disk if there was an error (1450)
3737 - don't ping imports that have recovery disabled (2676)
3738 - take buffered bytes into account when balancing socknal conn (2817)
3739 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3740 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3741 - remove little-endian swabbing of llog records (1987)
3742 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3743 - retry reposting request buffers if they fail (1191)
3744 - grow extent at grant time to avoid granting a revoked lock (2809)
3745 - lock revoke doesn't evict page if covered by a second lock (2765)
3746 - disable VM readahead to avoid reading outside lock extents (2805)
3748 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3749 - updated kernel patches for hp-2.4.20 kernel (2681)
3751 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3754 - fix truncated write corruption (2366)
3755 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3757 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3758 - ensure len(uuid) < 37 in lmc (1171)
3759 - fix ia64 OOPS in llog_test (2255)
3760 - zero end of page at obdfilter for partial page writes (2648)
3761 - don't leave stale dentries around after renames (bug 2428)
3762 - fix timeouts when evicting a client with a single lock held (2642)
3763 - set deadline for the initial HELLO message to drain (2634)
3764 - print out dotted-quad IP addresses in the socknal (2302)
3766 - additional debugging for MDS client eviction problem (2443)
3767 - fix mkfsoptions support for osts (2603, 2604)
3769 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3772 - add series for the vanilla 2.6.0 kernel
3773 - add series for the vanilla 2.4.24 kernel
3774 - add series for a cray x86/64 UL kernel drop
3775 - fix xattr patches for the vanilla 2.4.19 series
3777 - generate true UUIDs in lmc (1171)
3778 - have portals stack dumping break in UML (2466)
3779 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3780 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3781 - wake callback waiting threads on client eviction (2460)
3782 - Add --ptldebug and --subsystem to lmc (1719)
3783 - update assertion to allow safe interrupt allocation
3784 - set rq_no_resend for cancel requests (2432)
3785 - recalculate ptlrpcd timeout after resend (2494)
3786 - call vfs_rmdir when removing pending directories (2368)
3787 - fix renaming a file to itself (2429)
3788 - lmc creates a default one-stripe lov (2454)
3789 - expand procfs space to handle large clusters (2326)
3790 - increase UML stack to avoid overflow
3791 - update lconf's list of debug and subsystem masks
3792 - fix lfs find --obd (2510)
3793 - /proc tunable for disabling filter read caching (2591)
3794 - stop rpm packages from altering slapd.conf (2301)
3795 - disable nagle in the socknal under 0conf (2578)
3796 - choose mds inode size based on stripe count (2572)
3797 - fix kernel-source rpm problems (2516)
3799 - add --disable-doc to avoid pdf generation (2421)
3800 - update documentation, tests, type-os, comments
3801 - avoid format warnings on ia64
3802 - remove the TOE NAL
3803 - tiny code cleanups by removing unused fields
3805 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3808 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3809 - remove the most blatant lies from BUILDING file (2371)
3810 - change default debug level to reasonable production setting
3811 - reduce client side cache size to reduce cache flush time
3812 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3813 - make TCP zerocopy and pinger support enabled by default (2476)
3814 - sync writes completed after process exits caused crashes (2319)
3815 - maintain correct mount count on the MDS (2356)
3816 - backout 1557, because 2316 wasn't really fixed
3817 - better file I/O statistics gathering in /proc
3818 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3819 - another kernel patch to fix zero-copy TCP function export
3820 - don't take duplicate lock when processing re-sent getattr (2420)
3821 - lctl uses obd_self_export instead of creating new conn (2353)
3822 - MDS/OST recovery case which requires object creation asserted (2425)
3823 - move lfs from /usr/sbin to /usr/bin in packages
3824 - fix race between mds_client_add and mds_client_free (2417)
3825 - use kmalloc instead of slabs in portals (2430)
3826 - don't create duplicate records when a failover MDS is present (2442)
3827 - remove unnecessary mount age check (2332)
3828 - don't remove directory inodes from locks prematurely (2451)
3829 - don't break if MDS service name is the same as hostname (2103)
3830 - fix races in client write RPC generation when cache full (2482)
3832 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3835 - remove now-unused request->rq_obd (278)
3836 - if an allocation fails, print out how much memory we've used (1933)
3837 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3838 - add the "configurable stack size" patch to most series files (1256)
3839 - ability to write large log records, for 100+ OST configs (2306)
3840 - fix NULL deref when filter_prep fails (2314)
3841 - fix operator precedence error in filter_sync
3842 - dynamic allocation of socknal TX descriptors (2315)
3843 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3844 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3845 - ensure that CWARN messages in Portals always get to the syslog
3846 - __init/__exit are not for prototype decls (ldlm_init/exit)
3847 - x86-64 compile warning fixes
3848 - fix gateway LMC keyword conflict (2318)
3849 - fix MDS lock inversions in getattr/reint paths (1844)
3850 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3851 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3852 - DLM race condition prevented some lock evictions (2328)
3853 - ENOMEM detection and retry on socknal sends (2230)
3854 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3855 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3856 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3857 - better error messages when a client is rejected during recovery (1505)
3858 - avoid cancelling locks which were never granted, after failure (2330)
3859 - fix i_sem/journal inversion in mds_client_add (2333)
3860 - fix truncate/getattr lock cycle deadlock (2334)
3861 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3862 - fix two rare exit paths which could leak an l_lock() ref (2321)
3863 - fencepost error in MDS/OST orphan recovery (2226)
3864 - make log record alignment 8 bytes (1988)
3865 - lstripe now fails when requested offset > ost_count (2237)
3866 - ensure that all kernel series have a complete list.h (1607)
3867 - fix crashes in special-file operations (2316)
3868 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3870 - allow configurable automake binary, for testing new versions
3871 - small update to the lfs documentation
3873 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3875 * fix negative export reference count in fsfilt_sync (2312)
3877 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3878 * release candidate 0.9.1
3880 - orphans are moved into the PENDING directory for possible recovery
3881 - replayed opens now open by fid for orphan/rename safety (1042)
3882 - last close of an orphan inode generates a transno (683)
3883 - chdir() and mount() now pin the directory entry (1020)
3884 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3885 - discard very old requests without processing them (1502)
3886 - remove some common, well-understood CERRORs (1505)
3887 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3888 - clear "grant" flags in OST replies until OST grant code lands (1644)
3889 - fix read performance by not clobbering i_blksize on client (1598)
3890 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3891 - make LRU size a /proc tunable, clears locks when reduced (707)
3892 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3893 - server locks take references on exports now (1558)
3894 - build fixes for 2.4.20-rh trees (1663)
3895 - return an error from lov_create if all OSCs are inactive (1751)
3896 - fix import levels when a reconnect happens without a timeout (1597)
3897 - exit early from mds_open if we get a lookup error (1749)
3898 - partial page read at EOF wouldn't wait for disk before sending (1642)
3899 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3900 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3901 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3902 - write/truncate lock inversion (1639)
3903 - Don't auto-load obdclass, portals modules during cleanup (1495)
3904 - fix timestamps from jumping to "now" (1763)
3905 - extra journal assertions (1648)
3906 - add an extra multiunlink test (1771)
3907 - fix read_record/write_record API (1776)
3908 - fix leak of offset_extent, possible incorrect i_size later (1772)
3909 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3910 - Debugging for the kqswnal_get_idle_tx problems (1820)
3911 - Allow recovery to be attempted multiple times (1536)
3912 - Write out MDS last_rcvd file after it is first created (1600)
3913 - Fix tx_descriptor leak in failed transmit situations (1827)
3914 - ext3 journaling fixes for assertion failure after IO error (1871)
3915 - class_export_put() on freed export after completion AST error (1896)
3916 - Fix revalidate looping in VFS (1322)
3917 - Don't access a freed export during MDS_REINT timeout (1521)
3918 - Add open-unlink recovery support on the MDS (1673,1764)
3919 - Return an error if no MDS data was read from last_rcvd (1946)
3920 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3921 - Don't setup a disk device that doesn't match exported UUID (317)
3922 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3923 - avoid committing NULL handle in force close
3924 - local.sh is now a one-stripe LOV configuration
3925 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3926 - fix ext3 htree duplicate directory entry corruption (1516)
3927 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3928 - update to new LOV EA format (2097)
3929 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3931 2003-06-15 Phil Schwan <phil@clusterfs.com>
3934 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3935 - per-import/export recovery handling (958, 931, 959)
3936 - multiple last-rcvd slots, for serving multiple FSes (949)
3937 - connections are again shared between multiple imp/exports (963, 964)
3938 - "umount -f" would hang if any requests needed to be sent (393, 978)
3939 - avoid pinning large req buffer by copying for queued messages (989)
3940 - add "uuid" to "lctl device" command to help upcalls (991)
3941 - "open" RPCs with transnos would confuse recovery counters (1037)
3942 - do proper endian conversion of all wire messages (288, 340, 891)
3943 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3944 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3945 - fix ptlrpc_connection leak in target_handle_connect (1174)
3946 - fix import refcounting bug in OST and MDS cleanup (1134)
3947 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3948 - fix very unlikely obd_types race condition (501)
3949 - remove osc_open hack for echo_client (1187)
3950 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3951 - a failure in read_inode2 leads to deadlock (1139)
3952 - cancel ack-locks as soon as transaction is committed (1072)
3953 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3954 - make sure to commitrw after any preprw to avoid deadlock (1162)
3955 - failing to execute a file in a lustre FS would lock inode (1203)
3956 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3957 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3958 - obd_uuid2tgt would walk off the end of the list (1255)
3959 - on IA64 the osc would give portals incorrect bulk size (1258)
3960 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3961 - fix lock inversion caused by new llite matching code (1282)
3962 - limit the number of dirty pages on a client to 10MB (1286)
3963 - timed out locks were not being corrected cancelled (1289)
3964 - fix O_DIRECT above 4GB on IA-32 (1292)
3965 * major user-visible changes
3966 - fail out/fail over policy now controlled by the upcall (993)
3968 - add OBD_PING to check server availability and failure (954)
3969 - lustre messages are now sent in sending host order (288, 340, 891)
3970 - add eadatalen to MDS getattr reply (340)
3971 - OST read replies may contain second buffer, with per-page status (593)
3973 2003-03-11 Phil Schwan <phil@clusterfs.com>
3976 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3977 - failed open()s could cause deadlock; fixed (867, 869)
3978 - stop cancelling OST locks when files are closed (481)
3979 - overlapping XID spaces caused network corruption (851, 853)
3980 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3981 - setattr_raw added, to do single-RPC, server-side setattrs
3982 - lmc/lconf syntax change for OST UUIDs
3983 - fix crashy race condition between ptlrpc_free_req and osc_close
3984 - don't use request in mdc_enqueue if we hit a timeout (889)
3985 - don't set the inode i_size for regular files from the MDS (896)
3986 - handle out of order completion AST (842)
3987 - don't LBUG if a lock request times out after receiving AST (913)
3988 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3989 - if a bad lock AST arrives, send an error instead of dropping entirely
3990 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3991 - fix leak in bulk IO when only partially completed (899, 900, 926)
3992 - fix O_DIRECT for ia64 (55)
3993 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3994 - C-z after timeout could hang a process forever; fixed (977)
3996 - client-side I/O cache (678, 924, 929, 941, 970)
3998 - READPAGE and SETATTRs which don't take server-side locks get
4001 2003-02-11 Phil Schwan <phil@clusterfs.com>
4004 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4005 - Fix fsfilt_extN_readpage() to read a full page of directory
4006 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4007 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4008 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4009 - Most intent operations were removed, in favour of a new RPC mode
4010 that does a single RPC to the server and bypasses most of the VFS
4011 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4012 - Aggressively cancel local locks on DLM servers
4013 - mds_reint_unlink sends EA to the client if it's the last nlink.
4014 client uses that EA to unlink OST objects.
4015 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4016 - recursive symlinks were fixed (439)
4017 - fixed NULL deref in DEBUG_REQ
4018 - filter_update_lastobjid no longer calls sync, which annoyed extN
4019 - fixed multi-client small-writes to a single file problem (445)
4020 - fixed mtime updates during file writes (607)
4021 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4022 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4023 - fixed Linux OST size reporting problem (444, 656)
4024 - OST now updates object mtime with writes or setattr (607, 619)
4025 - client verifies file size before zeroing page past EOF (445)
4026 - OST now writes last allocated objid to disk with allocation (108)
4027 - LOV on echo now works (409)
4029 - mds_reint_unlink sends a new buffer, with the EA included. this
4030 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4031 set if a regular file was being unlinked, and it was the last link
4032 - use PtlGet from the target for bulk writes (315)
4033 - OST now updates object mtime with writes or setattr (607, 619)
4034 - LDLM now has a grant-time callback to revalidate locked items, if
4036 - Many MDS operations were reorganized to combat race conditions
4038 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4039 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4041 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4044 - Fully reactivate OST imports after reconnection (512, others)
4045 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4046 - More graceful error handling for truncating on dead OST (515)
4047 - Don't error out unless we're actually accessing dead stripes (474)
4048 - Fix garbage sizes when stripes are missing (410)
4049 - LRU counters were broken, causing constant lock purge (433, 432)
4050 - garbage on read from stripes with failed OSTs (441)
4051 - mark OSCs as active before reconnecting during recovery (438)
4052 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4053 - lfind did not preserve OST order in output (443)
4054 - symlinks cause hung clients, incorrect data (439)
4055 - stop dereferencing request after dropping refcount (457)
4056 - don't LASSERT(spin_is_locked) on non-SMP (455)
4057 - fixes for many rename() bugs
4058 - fstat didn't correctly synchronize attributes (399)
4059 - server must handle lock cancellation during blocking AST prep (487)
4060 - bulk descriptors were free()d too soon (511)
4061 - fix paths in lconf, which would load incorrect modules (451, 507)
4062 - fix confusing lconf 'host not found' error message (386)
4063 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4064 - fix race condition in mdc_blocking_ast() for inode access (526)
4065 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4066 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4067 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4068 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4069 - fix lstripe in several ways (kernel side) (527)
4070 - fix request leak in ldlm_cli_enqueue (262)
4071 - incorrect OSC was marked inactive after OST failure
4072 - call mds_fs_cleanup before unmounting filesystem (524)
4073 - fix races between taking ns_lock and ldlm_lock_change_resource
4074 - fix races updating LOV export open file list
4075 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4076 - fix recovery NULL deref in ldlm_cli_cancel_unused
4077 - fix some DLM races by using new hash table for lock handles (419)
4078 - permit the client to specify desired inodes, at replay
4079 - duplicate requests when we queue them for replay reintegration
4080 - fix last_rcvd offset calculation
4081 - sync after each recovered transaction, so we always make progress
4082 - never, not always, ERESTART requests without transnos
4083 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4085 - skip replay if the MDS says that the client is already connected
4086 - don't check for a recovery-enabled export to match lctl's UUID
4087 - don't INC_USE_COUNT for phantom exports
4088 - don't crash when cleaning up phantom exports (567)
4089 - don't double-finish or set replay data for errored mdc_open requests
4090 - abort requests when they time out, so we don't get old replies
4091 - send/receive replies for AST messages again
4092 - if the client says that it doesn't have the lock, cancel it on the
4094 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4095 mark it as destroyed, it will all work out when decref is called
4096 - fix module use counts (22, 581)
4098 - ASTs now expect a reply (server cancels lock on error reply)
4100 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4103 - fix many simultaneous client startup (392)
4104 - fix dentry->d_it clobbering
4105 - credentials weren't being shipped for readdir/getattr operations
4106 - remove invalid assertions triggered during some concurrent MD
4108 - proper Lustre versions added (336, 389)
4109 - fix memory leak for create error case (398)
4110 - fix LOV locking bug that would get cli/srv out of sync
4111 - fix echo client over LOV (409)
4112 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4113 - fix double-O_EXCL intent crash (424)
4114 - avoid sending multiple lock CANCELs (352)
4116 - MDS can do multi-client recovery (modulo bugs in new code)
4118 - many updates, edits, cleanups
4120 2002-11-18 Phil Schwan <phil@clusterfs.com>
4123 - fix null d_it dereference (346)
4124 - fix full OST/dbench hang (333)
4125 - fix permission problem with file removal (286)
4126 - fix removal of OSCs from LOV when they fail
4127 - fix NULL deref during bulk timeout (214)
4128 - fix problems related to multiple filesystems on one MDS (241)
4129 - fixed serious subtle metadata locking bugs
4130 - free locks on clients when inodes are removed due to memory
4132 - fix inode pointer in lock data (285)
4133 - partial support for multiple MDS on a single host (241)
4134 - data locks weren't cancelled at clear_inode time (290, 311)
4135 - intent locks could lead to unbounded lock growth (205)
4136 - added a maximum lock count, an LRU list, and a flusher
4137 - fix multiple rename (365)
4138 - properly abstracted the echo client
4139 - OSC locked 1 byte too many; fixed
4140 - rewrote brw callback code:
4141 - fixed recovery bugs related to LOVs (306)
4142 - fixed too-many-pages-in-one-write crash (191)
4143 - fixed (again) crash in sync_io_timeout (214)
4144 - probably fixed callback-related race (385)
4146 - Add capability to MDS protocol
4147 - LDLM cancellations and callbacks on different portals
4149 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4152 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4153 - defer open object destruction to close time (601981, 138)
4154 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4155 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4156 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4157 - don't call obd_setattr in ll_file_release for destroyed objects
4158 * protocol change to lustre_msg: move |version| and add |flags|
4159 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4160 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4161 * added replay of create, unlink, link and rename operations during
4162 MDS failover; recovery should be much more robust now
4163 * remove failed OSCs from LOVs (only lov_create uses this so far)
4164 * the lustre-HOWTO was brought (more) up to date (582544)
4166 2002-10-23 Phil Schwan <phil@clusterfs.com>
4169 - in-use dentries weren't being reused properly (617851)
4170 - prevent multiple LDLM setup (599178)
4171 - fix LOV size calculations for truncate (617853)
4172 - fix client handling of MDS intent errors (POSIX)
4173 - fix permission bug in lovstripe.c test (624321)
4174 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4175 - truncate past end of file could corrupt data
4176 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4177 - a race in recovery could return ETIMEDOUT to apps (623947)
4178 - building outside the source directory was fixed
4179 * the lustre-HOWTO was brought (more) up to date (582544)
4180 * major progress was made on recovery functionality
4182 2002-10-10 Phil Schwan <phil@clusterfs.com>
4185 - recovery deadlock fix
4186 - rm -rf causes LBUG fix (617817)
4187 - file open by multiple tasks fix (618962)
4188 - directory permissions bugs (602707 and 620007)
4189 - journal_stop fixed with locking (611313)
4190 - O_APPEND failures resolved (618273, perhaps 614459)
4191 - lconf PATH fix (619770)
4192 - IA64 build fix (621450)
4193 - RPC buffer sizes scale with amount of memory
4195 2002-10-01 Phil Schwan <phil@clusterfs.com>
4198 - locks would be cancelled without throwing away data pages,
4199 resulting in inconsistent data (605627)
4200 - inode attributes were not always being refreshed (605627, 612449)
4201 - lconf now continues to cleanup after lctl reports an error
4202 - MDS now enforces user permissions (602707)
4203 - lprocfs cleanup fixed, but not yet enabled (614157)
4204 - fixed infinite server hang, should a client not respond to an AST
4205 - avoid going into recovery if user calls readlink() with a buffer
4206 that's too small (613941)
4207 - AST RPCs no longer require replies (614867) -- this may be changed
4208 - don't crash server if client sends an IOV that's too big (611336)
4209 - fixed lock conversion deadlock (611892)
4210 - fixed the following of symlinks (614622)
4211 * recovery: the server can remove locks from a client that dies, other
4212 clients can make progress
4213 * more extN patch fixes
4214 * compile-time configurable ptlrpc buffer allocations
4216 - collaborative read cache document
4217 - Lustre Lite Performance CDR document-in-progress
4219 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4222 - fix typo in patch-2.4.18
4224 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4227 - clear ptlrpc request each time in handle_incoming_request()
4228 - unlink of files now destroys the object on the OST
4230 2002-09-19 Peter Braam <braam@clusterfs.com>
4232 * add hard link support
4233 * change obdfile creation method
4234 * kernel patch changed
4236 2002-09-19 Peter Braam <braam@clusterfs.com>
4239 - stack overflow bug in extN fixed
4241 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4243 * documentation updates
4244 - add man pages for config tools
4245 - update tests/README to describe testing with new config tools
4246 - finish metadata API descriptions
4247 * bug fixes and cleanups
4248 - statfs workaround for 16TB limit
4249 - LOV stripe allocation improved, can stripe on subset of OSTs
4250 - LOV file size/IO offset was wrong for files > 4GB in size
4251 - object EA data was being dropped, caused files to be unreadable
4252 - memory overflow with non-LOV OST caused memory corruption
4253 - fixed regression tests to work with new config tools, obdfilter
4254 - fixed bug when directory size became larger than 1 block
4255 - fixed bug (for single client case) when PWD was deleted
4256 - invalidate local directory pages when doing intent-based ops
4257 - avoid LDLM oops when lock callback contained bad data
4259 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4261 * documentation updates
4262 * bug fixes and cleanups
4263 - configuration tools
4266 - 64-bit compile warnings
4267 - 64-bit internal statfs data
4269 * test_brw on persistent OST devices
4271 * lprocfs (disabled)
4273 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4275 * documentation updates
4276 * bug fixes and cleanups
4277 * configuration tools
4279 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4282 * many small fixes to 0_5_4
4283 * io/network handling
4284 * thinkos in MDS operations
4286 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4289 * crucial basic fixes to 0.5.3
4290 * IOR, Iozone work over Elan
4293 2002-08-07 Phil Schwan <phil@clusterfs.com>
4294 * version 0_5_3, our first alpha
4295 * we use the new Portals iovs
4296 * documentation updates
4297 * bug fixes and cleanups
4298 * small changes in the DLM wire protocol
4300 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4301 * version 0_5_1 with some initial stability,
4302 * locking on MD and file I/O.
4303 * documentation updates
4304 * several bug fixes since 0.5.0
4305 * small changes in wire protocol
4307 2002-07-18 Phil Schwan <phil@clusterfs.com>
4309 * delivered as Lustre Light Alpha
4310 * fixed a crash after handling invalid MDS requests
4311 * fixed directory pages for architectures with non-4k pages sizes
4313 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4314 * release version v0_4_4
4315 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4317 * Fixes a number of interruption problems with OST operations.
4318 * Update documentation for portals header changes
4319 * Move all wire protocol structs/defines to lustre_idl.h
4320 * Fixes symlink length bug.
4321 * Add tcpdump to repository.
4323 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4324 * release version v0_4_3
4325 * Fixes statfs for inodes on extN.
4326 * Fixes bug in runtests which would delete /etc/hosts.
4327 * Use 64-bit object IDs wherever possible (not into VFS though)
4328 Remove ost_get_info, which is unused by lustre, and out of date.
4330 2002-07-03 Peter Braam <braam@clusterfs.com>
4331 * release version v0_4_2 Fixes a lookup error (type not passed)
4332 * move forward to head of Portals
4333 * move forward to latest Lustre kernel
4335 2002-06-25 Peter Braam <braam@clusterfs.com>
4336 * release version v0_4_1. Hopefully stable on single node use.