1 tbd Sun Microsystems, Inc.
4 2.6.9-67.0.1.EL (RHEL 4),
5 2.6.16.54-0.2.3 (SLES 10),
6 2.6.18-53.1.4.el5 (RHEL 5).
7 * Client support for unpatched kernels:
8 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
9 2.6.16 - 2.6.21 vanilla (kernel.org)
10 * Recommended e2fsprogs version: 1.40.4-cfs1
11 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
12 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
13 removed cwd "./" (refer to Bugzilla 14399).
17 Description: NULL lov_tgts causing MDS oops
18 Details : more safe checks for NULL lov_tgts for avoid oops.
20 Severity : enhancement
22 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
24 Severity : enhancement
26 Description: Add rhel5 support to HEAD.
30 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
31 Details : If an OST is down the MDS will hang indefinitely in
32 obd_statfs_rqset() waiting for the statfs data. While for
33 MDS QOS usage of statfs, it should not stuck in waiting.
35 Severity : enhancement
37 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
39 Severity : enhancement
41 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
45 Description: make mgs_setparam() handle fsname containing dash
46 Details : fsname containing a dash does not work with lctl conf_param
48 Severity : enhancement
50 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
52 Severity : enhancement
54 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
58 Frequency : rare, at shutdown
59 Description: access already free / zero obd_namespace.
60 Details : if client_disconnect_export was called without force flag set,
61 and exist connect request in flight, this can produce access to
62 NULL pointer (or already free pointer) when connect_interpret
63 store ocd flags in obd_namespace.
67 Frequency : only at startup
68 Description: not alloc memory with spinlock held.
69 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
72 Severity : enhancement
74 Description: make lustre randomly fail allocating memory
75 Details : Make lustre randomly failed allocating memory for testing purpose.
77 Severity : enhancement
79 Description: lost problems with lov objid file
80 Details : Fixes some scability and access to not inited memory problems
81 in work with lov objdid file.
86 Description: lfs find does not continue on file error
87 Details : Continue other files processing when a file/dir is absent.
91 Description: Inconsistent usage of lustre_pack_reply()
92 Details : Standardize the usage of lustre_pack_reply() such that it
93 always generate a CERROR on failure.
96 Frequency : occasional
98 Description: MDS hang or stay in waiting lock
99 Details : If client receive lock with CBPENDING flag ldlm need send lock
100 cancel as separate rpc, to avoid situation when cancel request
101 can't processed due all i/o threads stay in wait lock.
105 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
109 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
113 Description: Update to SLES9 kernel-2.6.5-7.287.3.
117 Description: Update to RHEL4 latest kernel.
119 Severity : enhancement
121 Description: Build SLES10 patchless client fails
122 Details : The configure was broken by run ./configure with
123 --with-linux-obj=.... argument for patchless client. When the
124 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
125 can't search header adequately. Use absolute path such as
126 -I($LINUX)/include instead.
130 Description: interrupt oig_wait produce painc on resend.
131 Details : brw_redo_request can be used for resend requests from ptlrpcd and
132 private set, and this produce situation when rq_ptlrpcd_data not
133 copyed to new allocated request and triggered LBUG on assert
134 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
135 ptlrpcd set if request is changed and can be safety changed to use
140 Description: LASSERT_{REQ,REP}SWAB macros are buggy
141 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
142 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
143 these macros and replace them with their difinitions instead.
147 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
148 Details : Update which_patch & target file for SLES10 latest kernel.
151 Description: Remove client patches from SLES 10 kernel.
152 Details : This causes SLES 10 clients to behave as patchless clients
153 even on a Lustre-patched (server) kernel.
155 Severity : enhancement
157 Description: self-adjustable client's lru lists
158 Details : use adaptive algorithm for managing client cached locks lru
159 lists according to current server load, other client's work
160 pattern, memory activities, etc. Both, server and client
161 side namespaces provide number of proc tunables for controlling
164 Severity : enhancement
166 Description: light-weight GSS support
167 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
168 and reduce performance overhead.
170 Severity : enhancement
172 Description: Linux keyring support
173 Details : Support using service of Linux keyring for Lustre GSS internal
174 context refresh/cache mechanism.
178 Description: Fix errors in lfs documentation
179 Details : Fixes man pages
183 Description: don't use GFP_* in generic Lustre code.
184 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
188 Description: obdclass is limited by single OBD_ALLOC(idarray)
189 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
193 Description: warnings with build patchless client with vanila 2.6.19 and up
194 Details : change old ctl_table style and replace ctl_table/ctl_table_header
195 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
199 Frequency : Only for SLES
200 Description: sanity_quota fail test_1
201 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
202 file, which makes TSTID[2] inunique.
206 Frequency : Only for SLES9
207 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
208 Details : Update target/ChangeLog/which_patch .
212 Frequency : Only for test_10 in sanity-quota.sh
213 Description: a bug in quota test code
214 Details : checking a test flag in a wrong place
218 Frequency : Only for test_13 in sanity-quota.sh
219 Description: a bug in quota test script
220 Details : shouldn't check overfull allocated quota
224 Description: jbd statistics
225 Details : Port older jbd statistics patch for sles10
227 Severity : enhancement
229 Description: Add printing inode info into message about error in writepage.
234 Description: reply_lock_interpret crash due to race with it and lock cancel.
235 Details : Do not replay locks that are being cancelled. Do not reference
236 locks by their address during replay, just by their handle.
240 Frequency : When flocks are used.
241 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
242 Details : Flock locks might destroy just granted lock if it could be merged
243 with another existing flock, this is done in completion handler,
244 so teach ldlm_cli_enquque_fini that this is a valid case for
250 Description: Oops in read and write path when failing to allocate lock.
251 Details : Check if lock allocation failed and return error back.
255 Description: lstripe command fails for valid OST index
256 Details : The stripe offset is compared to lov->desc.ld_tgt_count
257 instead of lov->desc.ld_active_tgt_count.
261 Description: sanity.sh failed test 103
262 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
267 Description: df doesn't work properly if diskfs blocksize != 4K
268 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
272 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
273 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
277 Description: obd_health_check_timeout too short
278 Details : set obd_health_check_timeout as 1.5x of obd_timeout
282 Description: llapi_file_create() does not allow some changes
283 Details : add llapi_file_open() that allows specifying the mode and
284 open flags, and also returns an open file handle.
288 Description: merge and cleanup kernel patches.
289 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
293 Description: Add journal checksum support.(Kernel part)
294 Details : The journal checksum feature adds two new flags i.e
295 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
296 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
297 indicates that the commit block contains the checksum for
298 the blocks described by the descriptor blocks. Now commit
299 record can be sent to disk without waiting for descriptor
300 blocks to be written to disk. This behavior is controlled
301 using JBD2_FEATURE_ASYNC_COMMIT flag.
305 Description: OSS needs mutliple precreate threads
306 Details : Add ability to start more than one create thread per OSS.
310 Description: Sanity.sh test_65k failed: setstripe should have succeeded
311 Details : execute lfs setstripe on client
315 Description: mds_obd_create error creating tmp object
316 Details : When the user sets quota on root, llog will be affected and can't
317 create files and write files.
320 Frequency : Always on ia64 patchless client, and possibly others.
322 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
323 Details : This allows the patchless client to be loaded on architectures
328 Description: RedHat Update kernel for RHEL5
329 Details : Add the kernel config file for RHEL5.
333 Description: RedHat Update kernel for RHEL5
334 Details : Modify the kernel config file more closer RHEL5.
338 Description: Build failure against Centos5 (RHEL5)
339 Details : Define PAGE_SIZE when it isn't present.
343 Description: test_7 on sanity-quota.sh can't be used on separate machines
344 Details : add facet to handle it
348 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
349 Details : with NFS, the anon dentry's parent was set to itself in
350 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
351 corresponding dentry other than use rec->ur_name.
353 Severity : enhancement
355 Description: lfs setstripe enhancement
356 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
360 Description: enable data checksumming by default
361 Details : enable checksum by default, allow --disable-checksum
362 configure option and "-o nochecksum" mount option. Checksums
363 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
367 Description: lustre support for RHEL5
368 Details : Add support for RHEL5.
372 Description: Client eviction due to failover config
373 Details : after a connection loss, the lustre client should attempt to
374 reconnect to the last active server first before trying the
375 other potential connections.
379 Description: when mds and osts use different quota unit(32bit and 64bit),
380 quota will be released repeatly.
381 Details : void sending multiple quota reqs to mds, which will keep the
382 status between the reqs.
386 Description: osts not allocated evenly to files
387 Details : change the condition to increase offset_idx
390 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
392 Description: Data corruption for OSTs that are formatted larger than 2TB
394 Details : When generating the bio request for lustre file writes the
395 sector number would overflow a temporary variable before being
396 used for the IO. The data reads correctly from Lustre (which
397 will overflow in a similar manner) but other file data or
398 filesystem metadata may be corrupted in some cases.
402 Description: Tune the kernel for good SCSI performance.
403 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
404 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
409 Description: rewrite ext2-derived code in obdclass/uuid.c
410 Details : rewrite inherited code (uuid parsing code from ext2 utils)
411 from scratch preserving functionality.
415 Description: "lfs find -obd UUID" prints directories
416 Details : "lfs find -obd UUID" will return all directory names instead
417 of just file names. It is incorrect because the directories
418 do not reside on the OSTs.
422 Description: use i_size_read and i_size_write in 2.6 port
423 Details : replace inode->i_size access with i_size_read/write()
426 Frequency : when removing large files
428 Description: scheduling issue during removal of large Lustre files
429 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
430 It causes scheduling issues when removing large files (17TB in the
434 Frequency : only with liblustre clients on XT3
436 Description: evictions taking too long
437 Details : allow llrd to evict clients directly on OSTs
440 Frequency : only on ppc
442 Description: /proc/fs/lustre/devices broken on ppc
443 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
444 We should make sure the type of 'index' is loff_t and then cast
445 explicitly as needed below. Do not assign an explicitly cast
449 Frequency : only for rhel5
451 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
452 Details : Modify the target file & which_kernel.
454 Severity : enhancement
456 Description: omit set fsid for export NFS
457 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
458 as FS_REQUIRES_DEV for avoid problems with generate fsid.
462 Frequency : Always, for kernels after 2.6.16
463 Description: Fix warning idr_remove called for id=.. which is not allocated.
464 Details : Last kernels save old s_dev before kill super and not allow
465 to restore from callback - restore it before call kill_anon_super.
469 Description: buffer overruns could theoretically occur
470 Details : llapi_semantic_traverse() modifies the "path" argument by
471 appending values to the end of the origin string, and a
472 overrun may occur. Adding buffer overrun check in liblustreapi.
476 Description: Fix error on 'ls .' at the top of the Lustre mount.
477 Details : Don't revalidate dentry if it is a root dentry.
481 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
482 Details : Modify vm-tunables-rhel4.patch.
486 Description: Kernel config for 2.6.18-vanilla.
487 Details : Modify targets/2.6-vanilla.target.in.
488 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
489 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
490 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
491 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
495 Description: improve handling recoverable errors
496 Details : if request processig with error which can be recoverable on server
497 request should be resend, otherwise page released from cache and
502 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
503 Details : Modify target file & which_patch.
504 A flaw was found in the IA32 system call emulation provided
505 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
506 value could be stored in the %RAX register, which could trigger an
507 out-of-bounds system call table access. An untrusted local user
508 could exploit this flaw to run code in the kernel
509 (ie a root privilege escalation). (CVE-2007-4573).
513 Description: O_DIRECT bypasses client statistics.
514 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
515 read_ahead_stats were updated. copied the stats section from
516 osc_send_oap_rpc() into async_internal().
520 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
524 Description: change order of libsysio includes
525 Details : '#include sysio.h' should always come before '#include xtio.h'
529 Description: handle "serious error: objid * already exists" more gracefully
530 Details : If LAST_ID value on disk is smaller than the objects existing in
531 the O/0/d* directories, it indicates disk corruption and causes an
532 LBUG(). If the object is 0-length, then we should use the existing
533 object. This will help to avoid a full fsck in most cases.
535 Severity : enhancement
537 Description: adapt the lustre_config script to support the upgrade case
538 Details : Add "-u" option for lustre_config script to support upgrading 1.4
539 server targets to 1.6 in parallel.
543 Description: To avoid grant space > avaible space when the disk is almost
544 full. Without this patch you might see the error "grant XXXX >
545 available" or some LBUG about grant, when the disk is almost
547 Details : In filter_check_grant, for non_grant cache write, we should
548 check the left space by if (*left > ungranted + bytes), instead
549 of (*left > ungranted), because only we are sure the left space
550 is enough for another "bytes", then the ungrant space should be
551 increase. In client, we should update cl_avail_grant only there
552 is OBD_MD_FLGRANT in the reply.
556 Description: Update RHEL 4 kernel to fix local root privilege escalation.
557 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
558 described in CVE-2007-4573. This problem could allow untrusted
559 local users to gain root access.
562 Frequency : when using O_DIRECT and quotas
564 Description: Incorrect file ownership on O_DIRECT output files
565 Details : block usage reported by 'lfs quota' does not take into account
566 files that have been written with O_DIRECT.
571 Description: touch file failed when fs is not full
572 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
573 otherwise we can get ENOSP while fs is not full.
577 Description: parallel lock callbacks
578 Details : Instead of sending blocking and completion callbacks as separated
579 requests, adding them to a set and sending in parallel.
582 Frequency : only for Cray XT3
583 Bugzilla : 12829/13455
584 Description: Changing primary group doesn't change the group lustre assigns to
586 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
587 overridden with the primary group provided by the group upcall,
588 whereas the supplied fsgid can be trusted if it is in the list of
589 supplementary groups returned by the group upcall.
591 Severity : enhancement
593 Description: Allow masking D_WARNING, D_ERROR messages from console
594 Details : Console messages can now be disabled via lnet.printk.
598 Description: User code with malformed file open parameter crashes client node
599 Details : Before packing join_file req, all the related reference should be
600 checked carefully in case some malformed flags cause fake
601 join_file req on client.
603 --------------------------------------------------------------------------------
605 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
607 * Support for kernels:
608 2.6.5-7.283 (SLES 9),
609 2.6.9-55.EL (RHEL 4),
610 2.6.16.46-0.14 (SLES 10),
611 2.6.18.8 vanilla (kernel.org)
612 * Client support for unpatched kernels:
613 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
614 2.6.16 - 2.6.22 vanilla (kernel.org)
615 * Due to recently discovered recovery problems, we do not recommend
616 using patchless RHEL 4 clients with this or any earlier release.
617 * Recommended e2fsprogs version: 1.39.cfs8
618 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
619 * Starting with this release, the ldiskfs backing filesystem required
620 by Lustre is now in its own package, lustre-ldiskfs. This package
621 should be installed. It is versioned separately from Lustre and
622 may be released separately in future.
626 Description: block reactivating mgc import until all deactivates complete
627 Details : Fix race when failing back MDT/MGS to itself (testing)
630 Frequency : at statup only
632 Description: mds_lov_synchronize race leads to various problems
633 Details : simultaneous MDT->OST connections at startup can cause the
634 sync to abort, leaving the OSC in a bad state.
636 Severity : enhancement
638 Description: add optional extra BUILD_VERSION info
639 Details : add a new environment variable (namely LUSTRE_VERS) which allows
640 to override the lustre version.
643 Frequency : 2.6.18 servers only
645 Description: ll_kern_mount() doesn't release the module reference
646 Details : The ldiskfs module reference count never drops down to 0
647 because ll_kern_mount() doesn't release the module reference.
652 Description: server LBUG when using old ost_num_threads parameter
653 Details : Accept the old ost_num_threads parameter but warn that it
654 is deprecated, and fix an off-by-one error that caused an LBUG.
659 Description: Transient SCSI error results in persistent IO issue
660 Details : iobuf->dr_error is not reinitialized to 0 between two
664 Frequency : sometimes when underlying device returns I/O errors
666 Description: OSTs not going read-only during write failures
667 Details : OSTs are not remounted read-only when the journal commit threads
668 get I/O errors because fsfilt_ext3 calls journal_start/stop()
669 instead of the ext3 wrappers.
673 Description: poor connect scaling with increasing client count
674 Details : Don't run filter_grant_sanity_check for more than 100 exports
675 to improve scaling for large numbers of clients.
678 Frequency : SLES10 only
680 Description: sanity-quota.sh quotacheck failed: rc = -22
681 Details : Quotas cannot be enabled on SLES10.
684 Frequency : liblustre clients only
686 Description: getdirentries does not give error when run on compute nodes
687 Details : getdirentries does not fail when the size specified as an argument
688 is too small to contain at least one entry
690 Severity : enhancement
692 Description: Add LNET router traceability for debug purposes
693 Details : If a checksum failure occurs with a router as part of the
694 IO path, the NID of the last router that forwarded the bulk data
695 is printed so it can be identified.
700 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
701 Details : Due to a race condition, liblustre clients were occasionally
704 Severity : enhancement
706 Description: lfs setstripe use optional parameters instead of postional
709 Severity : enhancement
711 Description: Nanosecond timestamp support for ldiskfs
712 Details : The on-disk ldiskfs filesystem has added support for nanosecond
713 resolution timestamps. There is not yet support for this at
714 the Lustre filesystem level.
717 Frequency : during server recovery
719 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
720 Details : request with rq_no_resend flag not awake l_wait_event if they get a
724 Frequency : nfs export on patchless client
726 Description: connectathon hang when test nfs export over patchless client
727 Details : Disconnected dentry cannot be found with lookup, so we do not need
728 to unhash it or make it invalid
731 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
732 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
733 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
738 Description: open req refcounting wrong on reconnect
739 Details : If reconnect happened between getting open reply from server and
740 call to mdc_set_replay_data in ll_file_open, we will schedule
741 replay for unreferenced request that we are about to free.
742 Subsequent close will crash in variety of ways.
743 Check that request is still eligible for replay in
744 mdc_set_replay_data().
749 Description: disable writes to filesystem when reading health_check file
750 Details : the default for reading the health_check proc file has changed
751 to NOT do a journal transaction and write to disk, because this
752 can cause reads of the /proc file to hang and block HA state
753 checking on a healthy but otherwise heavily loaded system. It
754 is possible to return to the previous behaviour during configure
755 with --enable-health-write.
757 Severity : enhancement
759 Description: 64-bit inode version
760 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
761 to the inode. This will be required for version-based recovery.
766 Description: MDS fails to start if a duplicate client export is detected
767 Details : in some rare cases it was possible for a client to connect to
768 an MDS multiple times. Upon recovery the MDS would detect this
769 and fail during startup. Handle this more gracefully.
771 Severity : enhancement
773 Description: Add -o localflock option to simulate old noflock
775 Details : This will achieve local-only flock/fcntl locks
781 Description: log_commit_thread vs filter_destroy race leads to crash
782 Details : Take import reference before releasing llog record semaphore
787 Description: Wrong request locking in request set processing
788 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
789 imp_delayed_list, in this place should be used imp_lock.
792 Frequency : when reconnection
794 Description: Grant Leak when osc reconnect to OST
795 Details : When osc reconnect ost, OST(filter) should check whether it
796 should grant more space to client by comparing fed_grant and
797 cl_avail_grant, and return the granted space to client instead
798 of "new granted" space, because client will call osc_init_grant
799 to update the client grant space info.
802 Frequency : when client reconnect to OST
804 Description: Grant Leak when osc do resend and replay bulk write
805 Details : When osc reconnect to OST, OST(filter)should clear grant info of
806 bulk write request, because the grant info while be sync between
807 OSC and OST when reconnect, and we should ignore the grant info
808 these of resend/replay write req.
813 Description: Grant space more than avaiable left space sometimes.
814 Details : When then OST is about to be full, if two bulk writing from
815 different clients came to OST. Accord the avaliable space of the
816 OST, the first req should be permitted, and the second one
817 should be denied by ENOSPC. But if the seconde arrived before
818 the first one is commited. The OST might wrongly permit second
819 writing, which will cause grant space > avaiable space.
822 Frequency : when client is evicted
824 Description: Grant might be wrongly erased when osc is evicted by OST
825 Details : when the import is evicted by server, it will fork another
826 thread ptlrpc_invalidate_import_thread to invalidate the
827 import, where the grant will be set to 0. While the original
828 thread will update the grant it got when connecting. So if
829 the former happened latter, the grant will be wrongly errased
830 because of this race.
835 Description: Checking Stale with correct fid
836 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
837 to check whether it is stale, because sometimes, we want the
838 enqueue happened anyway, and op_data.fid2 will not be initialized.
840 Severity : enhancement
842 Description: update patchless client
843 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
846 Frequency : only with 2.4 kernel
848 Description: random memory corruption
849 Details : size of struct ll_inode_info is to big for union inode.u and this
850 can be cause of random memory corruption.
855 Description: Memory leak in recovery
856 Details : Lov_mds_md was not free in an error handler in mds_create_object.
857 It should also check obd_fail before fsfilt_start, otherwise if
858 fsfilt_start return -EROFS,(failover mds during mds recovery).
859 then the req will return with repmsg->transno = 0 and rc = EROFS.
860 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
861 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
862 be freed no matter whether fsfilt_commit success or not.
865 Frequency : only with huge count clients
867 Description: Prevents from taking the superblock lock in llap_from_page for
869 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
870 for llap_from_page call in ll_removepage prevents from taking the
871 superblock lock for a soon died page.
876 Description: Not check open intent error before release open handle
877 Details : in some rare cases, the open intent error is not checked before
878 release open handle, which may cause
879 ASSERTION(open_req->rq_transno != 0), because it tries to release
880 the failed open handle.
885 Description: Set cat log bitmap only after create log success.
886 Details : in some rare cases, the cat log bitmap is set too early. and it
887 should be set only after create log success.
891 Description: Accessing a block bevice can re-enable I/O when Lustre is
892 tearing down a device.
893 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
898 Description: service threads may hog cpus when there are a lot of requests
899 Details : Insert cond_resched to give other threads a chance to use some CPU
904 Description: the cat log was not initialized in recovery
905 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
906 unlink log on mds will not be initialized until mds post
907 recovery. And also in mds post recovery, the unlink log will
908 initialization will be done asynchronausly, so there will be race
909 between add unlink log and unlink log initialization.
913 Description: brw_stats were being printed incorrectly
914 Details : brw_stats were being printed as log2 but all of them were not
915 recorded as log2. Also remove some code duplication arising from
916 filter_tally_{read,write}.
920 Frequency : rare, only in recovery.
921 Description: ASSERTION(req->rq_type != LI_POISON) failed
922 Details : imp_lock should be held while iterating over imp_sending_list for
923 prevent destroy request after get timeout in ptlrpc_queue_wait.
927 Description: replay-single.sh test 52 fails
928 Details : A lock's skiplist need to be cleanup when it being unlinked
929 from its resource list.
933 Description: Short directio read returns full requested size rather than
935 Details : Direct I/O operations should return actual amount of bytes
936 transferred rather than requested size.
938 Severity : enhancement
940 Description: metadata RPC reduction (e.g. for rm performance)
941 Details : decrease the amount of synchronous RPC between clients and servers
942 by canceling conflicing lock before the operation on the client
943 and packing thier handles into the main operation RPC to server.
945 Severity : enhancement
947 Description: Async OSC create to avoid the blocking unnecessarily.
948 Details : If a OST has no remain object, system will block on the creating
949 when need to create a new object on this OST. Now, ways use
950 pre-created objects when available, instead of blocking on an
951 empty osc while others are not empty. If we must block, we block
952 for the shortest possible period of time.
954 Severity : enhancement
956 Description: refine locking for avoid write wrong info into lov_objid file
957 Details : fix possible races with add new target and write/update data in
960 --------------------------------------------------------------------------------
962 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
967 Frequency : on some architectures
969 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
970 Details : Uninitialized flags sometimes cause configuration commands to
974 Frequency : patchless clients only
976 Description: missing __iget() symbol export
977 Details : The __iget() symbol export is missing. To avoid the need for
978 this on patchless clients the deathrow inode reaper is turned
979 off, and we depend on the VM to clean up old inodes. This
980 dependency was during via the fix for bug 12181.
982 --------------------------------------------------------------------------------
984 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
986 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
987 INTEROPERATE with older versions automatically. In many cases a
988 special upgrade step is needed. Please read the
989 user documentation before upgrading any part of a live system.
990 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
991 version will not interoperate with 1.6 betas before beta5 (1.5.95).
992 * WARNING: Lustre configuration and startup changes are required with
993 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
999 Severity : enhancement
1001 Description: MountConf
1002 Details : Lustre configuration is now managed via mkfs and mount
1003 commands instead of lmc and lconf. New obd types (MGS, MGC)
1004 are added for dynamic configuration management. See
1005 https://mail.clusterfs.com/wikis/lustre/MountConf for
1008 Severity : enhancement
1010 Description: dynamic OST addition
1011 Details : OSTs can now be added to a live filesystem
1013 Severity : enhancement
1015 Description: startup order invariance
1016 Details : MDTs and OSTs can be started in any order. Clients only
1017 require the MDT to complete startup.
1019 Severity : enhancement
1021 Description: parallel, asynchronous orphan cleanup
1022 Details : orphan cleanup is now performed in separate threads for each
1023 OST, allowing parallel non-blocking operation.
1025 Severity : enhancement
1027 Description: optimized stripe assignment
1028 Details : stripe assignments are now made based on ost space available,
1029 ost previous usage, and OSS previous usage, in order to try
1030 to optimize storage space and networking resources.
1032 Severity : enhancement
1034 Description: Permanently set tunables
1035 Details : All writable /proc/fs/lustre tunables can now be permanently
1036 set on a per-server basis, at mkfs time or on a live system.
1038 Severity : enhancement
1040 Description: Lustre message v2
1041 Details : Add lustre message format v2.
1043 Severity : enhancement
1045 Description: client OST exclusion list
1046 Details : Clients can be started with a list of OSTs that should be
1047 declared "inactive" for known non-responsive OSTs.
1051 Description: SPEC SFS validation failure on NFS v2 over lustre.
1052 Details : Changes the blocksize for regular files to be 2x RPC size,
1053 and not depend on stripe size.
1055 Severity : enhancement
1057 Description: Multiple MD RPCs in flight.
1058 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1059 To avoid overly-overloading MDS, introduce a limit on number of
1060 MDS RPCs in flight for a single client and add /proc controls
1061 to adjust this limit.
1063 Severity : enhancement
1065 Description: client read/write statistics
1066 Details : Add client read/write call usage stats for performance
1067 analysis of user processes.
1068 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1069 file access. extents_stats shows chunk size distribution.
1070 extents_stats_per_process show chunk size distribution per
1073 Severity : enhancement
1075 Description: mds statistics
1076 Details : Add detailed mds operations statistics in
1077 /proc/fs/lustre/mds/*/stats.
1081 Description: Failure of copying files with lustre special EAs.
1082 Details : Client side always return success for setxattr call for lustre
1083 special xattr (currently only "trusted.lov").
1088 Description: Refcount LNET uuids
1089 Details : The global LNET uuid list grew linearly with every startup;
1090 refcount repeated list entries instead of always adding to
1093 Severity : enhancement
1095 Description: Dynamic service threads
1096 Details : Within a small range, start extra service threads
1097 automatically when the request queue builds up.
1100 Frequency : mixed-endian client/server environments
1102 Description: mixed-endian crashes
1103 Details : The new msg_v2 system had some failures in mixed-endian
1106 Severity : enhancement
1108 Description: Easy OST removal
1109 Details : OSTs can be permanently deactivated with e.g. 'lctl
1110 conf_param lustre-OST0001.osc.active=0'
1112 Severity : enhancement
1114 Description: MGS proc entries
1115 Details : Added basic proc entries for the MGS showing what filesystems
1118 Severity : enhancement
1120 Description: provide MGS failover
1121 Details : Added config lock reacquisition after MGS server failover.
1123 Severity : enhancement
1125 Description: add Linux 2.4 support
1126 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1130 Description: a large application tries to do I/O to the same resource and dies
1131 in the middle of it.
1132 Details : Check the req->rq_arrival time after the call to
1133 ost_brw_lock_get(), but before we do anything about
1134 processing it & sending the BULK transfer request. This
1135 should help move old stale pending locks off the queue as
1136 quickly as obd_timeout.
1139 Frequency : when an incorrect nid is specified during startup
1141 Description: ptlrpc connect to non-existant node causes kernel crash
1142 Details : LNET can't be re-entered from an event callback, which
1143 happened when we expire a message after the export has been
1144 cleaned up. Instead, hand the zombie cleanup off to another
1147 Severity : enhancement
1149 Description: plain/inodebits lock performance improvement
1150 Details : Grouping plain/inodebits in granted list by their request modes
1151 and bits policy, thus improving the performance of search through
1155 Frequency : only if OST filesystem is corrupted
1157 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1158 Details : for a short time RPCs with bulk IO are in the replay list,
1159 but replay of bulk IOs is unimplemented. If the OST filesystem
1160 is corrupted due to disk cache incoherency and then replay is
1161 started it is possible to trip an assertion. Avoid putting
1162 committed RPCs into the replay list at all to avoid this issue.
1165 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1167 Description: System hang on startup
1168 Details : This bug allowed the liblustre (e.g. catamount) client to
1169 return to the app before handling all startup RPCs. This
1170 could leave the node unresponsive to lustre network traffic
1171 and manifested as a server ptllnd timeout.
1173 Severity : enhancement
1175 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1176 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1177 boolean to control whether to print peer debug info when a
1178 client's RPC times out.
1181 Frequency : only for kernels with patches from Lustre below 1.4.3
1183 Description: Remove old rdonly API
1184 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1187 Frequency : only for devices with external journals
1189 Description: Set external device read-only also
1190 Details : During a commanded failover stop, we set the disk device
1191 read-only while the server shuts down. We now also set any
1192 external journal device read-only at the same time.
1195 Frequency : when upgrading from 1.4 while trying to change parameters
1197 Description: The wrong (new) MDC name was used when setting parameters for
1198 upgraded MDT's. Also allows changing of OSC (and MDC)
1199 parameters if --writeconf is specified at tunefs upgrade time.
1202 Frequency : when setting specific ost indicies
1204 Description: QOS code breaks on skipped indicies
1205 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1206 created with --index need not be sequential.
1208 Severity : enhancement
1210 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1211 Details : The uninit_groups feature works in conjunction with the kernel
1212 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1213 up the pass1 processing of e2fsck. This is a read-only feature
1214 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1215 mount filesystems that have had this feature enabled.
1217 Severity : enhancement
1219 Description: Improve multi-block allocation algorithm to avoid fragmentation
1220 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1221 allocation locality and avoid filesystem fragmentation.
1223 ------------------------------------------------------------------------------
1225 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1227 * Support for kernels:
1228 2.4.21-47.0.1.EL (RHEL 3)
1229 2.6.5-7.283 (SLES 9)
1230 2.6.9-42.0.10.EL (RHEL 4)
1231 2.6.12.6 vanilla (kernel.org)
1232 2.6.16.27-0.9 (SLES 10)
1233 * Recommended e2fsprogs version: 1.39.cfs5
1235 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1239 Frequency : occasional, depends on client load and configuration
1240 Bugzilla : 12181, 12203
1241 Description: data loss for recently-modified files
1243 Details : In some cases it is possible that recently written or created
1244 files may not be written to disk in a timely manner (this should
1245 normally be within 30s unless client IO load is very high).
1246 The problem appears as zero-length files or files that are a
1247 multiple of 1MB in size after a client crash or client eviction
1248 that are missing data at the end of the file.
1250 This problem is more likely to be hit on clients where files are
1251 repeatedly created and unlinked in the same directory, clients
1252 have a large amount of RAM, have many CPUs, the filesystem has
1253 many OSTs, the clients are rebooted frequently, and/or the files
1254 are not accessed by other nodes after being written.
1256 The presence of the problem can be detected by looking at
1257 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1258 smaller than the second (nr_unused) then dirty files will not
1259 be flushed automatically to disk. "sync; sleep 10" should be
1260 run several times on the node before unmounting it to update
1261 Lustre (this is also safe to run on nodes without this problem).
1263 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1264 that can cause this same problem, so customers using that kernel
1265 also need to update the kernel in addition to Lustre. In order
1266 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1268 It is normal that files written just before a client crash (less
1269 than 30s) may not yet have been flushed to disk, even for local
1273 Frequency : frequent on thin XT3 nodes
1275 Description: UUID collision on thin XT3 Linux nodes
1276 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1277 randomly, since we relied on an insufficiently-seeded PRNG.
1282 Description: OSS hangs after "All ost request buffers busy"
1283 Details : A deadlock between quota and journal operations caused OSS
1284 hangs after printing "All ost request buffers busy."
1287 Frequency : always on liblustre builds
1289 Description: Cleanup compiler warnings on liblustre
1292 Frequency : always on liblustre builds on XT3
1294 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1298 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1299 > 32000 subdirectories
1300 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1301 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1302 subdirectory count crosses 32000. This will aid e2fsck to
1303 correctly handle more than 32000 subdirectories.
1306 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1308 Description: System hang on startup
1309 Details : This bug allowed the liblustre (e.g. catamount) client to
1310 return to the app before handling all startup RPCs. This
1311 could leave the node unresponsive to lustre network traffic
1312 and manifested as a server ptllnd timeout.
1314 Severity : enhancement
1316 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1317 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1318 boolean to control whether to print peer debug info when a
1319 client's RPC times out.
1324 Description: make O_SYNC working on 2.6 kernels
1325 Details : 2.6 kernels use different method for mark pages for write,
1326 so need add a code to lustre for O_SYNC work.
1331 Description: Failure to close file and release space on NFS
1332 Details : Put inode details into lock acquired in ll_intent_file_open.
1333 Use mdc_intent_lock in ll_intent_open to properly
1334 detect all kind of errors unhandled by mdc_enqueue.
1339 Description: proc file read during shutdown sometimes raced obd removal,
1341 Details : Add lock to prevent obd access after proc file removal.
1344 Frequency : Only for files larger than 4GB on 32-bit clients.
1346 Description: improperly doing page alignment of locks
1347 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1348 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1353 Description: RPCs being resent when they shouldn't be
1354 Details : Some RPCs that should not be resent are being resent. This
1355 can cause inconsistencies in the RPC state machine. Do not
1356 resend such requests.
1359 Frequency : rare, only with NFS export
1361 Description: Crash on NFS re-export node
1362 Details : under very unusual load conditions an assertion is hit in
1363 ll_intent_file_open()
1366 Frequency : only if OST filesystem is corrupted
1368 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1369 Details : for a short time RPCs with bulk IO are in the replay list,
1370 but replay of bulk IOs is unimplemented. If the OST filesystem
1371 is corrupted due to disk cache incoherency and then replay is
1372 started it is possible to trip an assertion. Avoid putting
1373 committed RPCs into the replay list at all to avoid this issue.
1378 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1379 Details : Large single O_DIRECT read and write calls can fail to allocate
1380 a sufficiently large buffer to process the request. In case of
1381 allocation failure the allocation is retried with a smaller
1382 buffer and broken into smaller requests.
1384 Severity : enhancement
1386 Description: Add -o localflock option to simulate old noflock behaviour.
1387 Details : This will achieve local-only flock/fcntl locks coherentness.
1392 Description: versioning check is incomplete
1393 Details : Checking the version difference of client vs. server, report
1394 error if the gap is too big.
1399 Description: add support PG_writeback bit
1400 Details : add support for PG_writeback bit for Lustre, for more carefull
1401 work with page cache in 2.6 kernel. This also fix some deadlocks
1402 and remove hack for work O_SYNC with 2.6 kernel.
1404 Severity : enhancement
1406 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1407 Details : The uninit_groups feature works in conjunction with the kernel
1408 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1409 up the pass1 processing of e2fsck. This is a read-only feature
1410 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1411 mount filesystems that have had this feature enabled.
1413 Severity : enhancement
1415 Description: Improve multi-block allocation algorithm to avoid fragmentation
1416 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1417 allocation locality and avoid filesystem fragmentation.
1419 ------------------------------------------------------------------------------
1421 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1423 * Support for kernels:
1424 2.6.9-42.0.3.EL (RHEL 4)
1425 2.6.5-7.276 (SLES 9)
1426 2.4.21-47.0.1.EL (RHEL 3)
1427 2.6.12.6 vanilla (kernel.org)
1428 2.6.16.21-0.8 (SLES10)
1429 * Recommended e2fsprogs version: 1.39.cfs2-0
1431 * The backwards-compatible /proc/sys/portals symlink has been removed
1432 in this release. Before upgrading, please ensure that you change
1433 any configuration scripts or /etc/sysctl.conf files that access
1434 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1435 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1436 in advance of the upgrade on any system running Lustre 1.4.6 or
1437 newer, since /proc/sys/lnet was added in that version.
1438 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1442 Frequency : only when quota is used
1444 Description: avoid scanning export list for quota master
1445 Details : Change the algorithms to avoid scanning export list in order
1446 to improve the efficiency.
1449 Frequency : MDS failover only, very rarely
1451 Description: "went back in time" messages on mds failover
1452 Details : The greatest transno may be lost when the current operation
1453 finishes with an error (transno==0) and the client's last_rcvd
1454 record is over-written. Save the greatest transno in the
1455 mds_last_transno for this case.
1458 Frequency : always for specific kernels and striping counts
1460 Description: client may get "Matching packet too big" without ACL support
1461 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1462 when trying to access files in certain configurations. The
1463 clients should in fact be denied when mounting because they do
1464 not understand ACLs.
1467 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1469 Description: many clients connecting with IO in progress causes connect timeouts
1470 Details : Avoid synchronous journal commits to avoid delays caused by many
1471 clients connecting/disconnecting when bulk IO is in progress.
1472 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1473 OST_IO_PORTAL to avoid delays behind potentially many pending
1477 Frequency : occasionally with multiple writers to a single file
1479 Description: shared writes to file may result in wrong size reported by stat()
1480 Details : Allow growing of kms when extent lock is cancelled
1483 Frequency : always with random mmap IO to multi-striped file
1485 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1486 Details : Only if the hole is at the end of OST object so that kms is too
1487 small. Fix is to increase kms accordingly in ll_nopage.
1490 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1492 Description: writes to a missing object would leak memory on the OST
1493 Details : If there is an inconsistency between the MDS and OST filesystems,
1494 such that the MDS references an object that doesn't exist, writes
1495 to that object will leak memory due to incorrect cleanup in the
1496 error handling path, eventually running out of memory on the OST.
1501 Description: Creating too long symlink causes lustre errors
1502 Details : Check symlink and name lengths before sending requests to MDS.
1505 Frequency : only if flock is enabled (not on by default)
1507 Description: posix locks not released on fd closure on 2.6.9+
1508 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1509 kernels, this caused such locks not to be released on fd close and
1510 then assertions on fs unmount about still used locks.
1513 Frequency : MDS failover only, very rarely
1515 Description: clients may get ASSERTION(granted_lock != NULL)
1516 Details : When request was taking a long time, and a client was resending
1517 a getattr by name lock request. The were multiple lock requests
1518 with the same client lock handle and
1519 mds_getattr_name->fixup_handle_for_resent_request found one of the
1520 lock handles but later failed with ASSERTION(granted_lock != NULL).
1525 Description: handle->h_buffer_credits > 0, assertion failure
1526 Details : h_buffer_credits is zero after truncate, causing assertion
1527 failure. This patch extends the transaction or creates a new
1531 Frequency : NFS re-export or patchless client
1532 Bugzilla : 11179, 10796
1533 Description: Crash on NFS re-export node (__d_move)
1534 Details : We do not want to hash the dentry if we don't have a lock.
1535 But if this dentry is later used in d_move, we'd hit uninitialised
1536 list head d_hash, so we just do this to init d_hash field but
1537 leave dentry unhashed.
1540 Frequency : NFS re-export or patchless client
1542 Description: NFS exports has problem with symbolic link
1543 Details : lustre client didn't properly install dentry when re-exported
1544 to NFS or running patchless client.
1547 Frequency : NFS re-export or patchless client
1549 Description: Various nfs/patchless fixes.
1550 Details : fixes reuse disconected alias for lookup process - this fixes
1551 warning "find_exported_dentry: npd != pd",
1552 fix permission error with open files at nfs.
1553 fix apply umask when do revalidate.
1556 Frequency : occasional
1558 Description: Crash on NFS re-export node
1559 Details : calling clear_page() on the wrong pointer triggered oops in
1560 generic_mapping_read().
1563 Frequency : rarely, using O_DIRECT IO
1565 Description: unaligned directio crashes client with LASSERT
1566 Details : check for unaligned buffers before trying any requests.
1569 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1571 Description: stale data returned from RAID cache
1572 Details : If only a small amount of IO is done to the RAID device before
1573 reading it again it is possible to get stale data from the RAID
1574 cache instead of reading it from disk.
1577 Frequency : always for sles10 kernel
1579 Description: sles10 support
1580 Details : ll_follow_link: compile fixes and using of nd_set_link
1581 under newer kernels.
1584 Frequency : depends on arch, kernel and compiler version, always on sles10
1587 Description: recursive or deep enough symlinks cause stack overflow
1588 Details : getting rid of large stack-allocated variable in
1592 Frequency : depends on hardware
1594 Description: lustre write performance loss in the SLES10 kernel
1595 Details : the performance loss is caused by using of write barriers in the
1596 ext3 code. The SLES10 kernel turns barrier support on by
1597 default. The fix is to undo that change for ldiskfs.
1599 ------------------------------------------------------------------------------
1601 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1603 * Support for kernels:
1604 2.6.9-42.0.3EL (RHEL 4)
1605 2.6.5-7.276 (SLES 9)
1606 2.4.21-47.0.1.EL (RHEL 3)
1607 2.6.12.6 vanilla (kernel.org)
1611 Frequency : quota enabled and large files being deleted
1613 Description: releasing more than 4GB of quota at once hangs OST
1614 Details : If a user deletes more than 4GB of files on a single OST it
1615 will cause the OST to spin in an infinite loop. Release
1616 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1621 Description: statfs data retrieved from /proc may be stale or zero
1622 Details : When reading per-device statfs data from /proc, in the
1623 {kbytes,files}_{total,free,avail} files, it may appear
1624 as zero or be out of date.
1627 Frequency : systems with MD RAID1 external journal devices
1629 Description: lconf's call to blkid is confused by RAID1 journal devices
1630 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1631 of returning all block devices that match the journal UUID.
1634 Frequency : always, for aggregate stripe size over 4GB
1636 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1637 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1638 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1641 Frequency : always if "lfs find" used on a local file/directory
1643 Description: "lfs find" segfaults if used on a local file/directory
1644 Details : The case where a directory component was not specified wasn't
1645 handled correctly. Handle this properly.
1648 Frequency : always on ppc64
1650 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1651 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1652 perform endianness conversion.
1655 Frequency : rarely (truncate to non-zero file size after write under load)
1656 Bugzilla : 10730, 10687
1657 Description: Files padded with zeros to next 4K multiple
1658 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1659 it is possible that files that are truncated to a non-zero size
1660 immediately after being written are filled with zero bytes beyond
1661 the truncated size. No file data is lost.
1663 Severity : enhancement
1665 Description: Allow recovery/failover for liblustre clients.
1666 Details : liblustre clients were unaware of failover configurations until
1669 Severity : enhancement
1671 Description: user file locks should fail when not mounting with flock option
1672 Details : Set up an error-returning stub in ll_file_operations.lock field
1673 to prevent incorrect behaviour when client is mounted without
1674 flock option. Also, set up properly f_op->flock field for
1678 Frequency : always on ia64
1680 Description: "lfs df" loops on printing out MDS statfs information
1681 Details : The obd_ioctl_data was not initialized and in some systems
1682 this caused a failure during the ioctl that did not return
1683 an error. Initialize the struct and return an error on failure.
1686 Frequency : SLES 9 only
1688 Description: Error of copying files with lustre special EAs as root
1689 Details : Client side always return success for setxattr call for lustre
1690 special xattr (currently only "trusted.lov").
1693 Frequency : rarely on clusters with both ia64+i386 clients
1695 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1696 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1697 clients are doing concurrent writes to the same file it is
1698 possible that the ia64 clients may LASSERT because the OST
1699 extent locks are not PAGE_SIZE aligned. Ensure that grown
1700 locks are always aligned on the request boundary.
1703 Frequency : specific use, occasional
1705 Description: Overwriting in use executable truncates on-disk binary image
1706 Details : If one node attempts to overwrite an executable in use by
1707 another node, we now correctly return ETXTBSY instead of
1708 truncating the file.
1710 Severity : enhancement
1712 Description: Async OSC create to avoid the blocking unnecessarily.
1713 Details : If a OST has no remain object, system will block on the creating
1714 when need to create a new object on this OST. Now, ways use
1715 pre-created objects when available, instead of blocking on an
1716 empty osc while others are not empty. If we must block, we block
1717 for the shortest possible period of time.
1722 Description: chmod on Lustre root is propagated to other clients
1723 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1724 invalid by the follow_mount time.
1727 Frequency : liblustre clients only
1729 Description: Race in 'instant cancel' lock handling could lead to such locks
1730 never to be granted in case of SMP MDS
1731 Details : Do not destroy not yet granted but cbpending locks in
1735 Frequency : replay/resend of open
1737 Description: non null lock assetion failure in mds_intent_policy
1738 Details : Trying to replay/resend lockless open requests resulted in
1739 mds_open() returning 0 with no lock. Now it sets a flag if
1740 a lock is going to be returned.
1742 Severity : enhancement
1744 Description: Checksum enhancements
1745 Details : New checksum enhancements allow for resending RPCs that failed
1748 Severity : enhancement
1750 Description: Tunables on number of dirty pages in cacche
1751 Details : Allow to set limit on number of dirty pages cached.
1756 Description: client crash on unmount - lock still has references
1757 Details : In some error handling cases it was possible to leak a lock
1758 reference on a client while accessing a file. This was not
1759 harmful to the client during operation, but would cause the
1760 client to crash when the filesystem is unmounted.
1763 Frequency : specific case, rare
1765 Description: ETXTBSY on mds though file not in use by client
1766 Details : ETXTBSY is no longer incorrectly returned when attempting to
1767 chmod or chown a directory that the user previously tried to
1768 execute or a currently-executing binary.
1771 Frequency : extremely rare except on liblustre-based clients
1773 Description: Lustre space not freed when files are deleted
1774 Details : Clean up open-unlinked files after client eviction. Previously
1775 the unlink was skipped and the files remained as orphans.
1780 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1781 Details : In some cases it was possible to send a blocking callback to
1782 a client doing a glimpse, even though that client didn't get
1783 a lock granted. When the glimpse lock is cancelled on the OST
1784 the freed lock is left on the waiting list and corrupted the list.
1787 Frequency : all core dumps
1789 Description: Broke core dumps to lustre
1790 Details : Negative dentry may be unhashed if parent does not have UPDATE
1791 lock, but some callers, e.g. do_coredump, expect dentry to be
1792 hashed after successful create, hash it in ll_create_it.
1794 ------------------------------------------------------------------------------
1796 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1798 * Support for kernels:
1799 2.6.9-42.0.2.EL (RHEL 4)
1800 2.6.5-7.276 (SLES 9)
1801 2.4.21-40.EL (RHEL 3)
1802 2.6.12.6 vanilla (kernel.org)
1806 Frequency : always on RHEL 3
1808 Description: Number of open files grows over time
1809 Details : The number of open files grows over time, whether or not
1810 Lustre is started. This was due to a filp leak introduced
1811 by one of our kernel patches.
1813 ------------------------------------------------------------------------------
1815 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1817 * Support for kernels:
1818 2.6.9-42.EL (RHEL 4)
1819 2.6.5-7.276 (SLES 9)
1820 2.4.21-40.EL (RHEL 3)
1821 2.6.12.6 vanilla (kernel.org)
1826 Bugzilla : 5719, 9635, 9792, 9684
1827 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1828 Details : If a server is under heavy load and cannot reply to new
1829 connection requests before the client resends the (re)connect,
1830 the connection handling code can behave badly if two service
1831 threads are concurrently handing separate (re)connections from
1832 the same client. Add better locking to the connection handling
1833 code, and ensure that only a single connection will be processed
1834 for a given client UUID, even if the lock is dropped.
1836 Severity : enhancement
1838 Description: add TCP zero-copy support to kernel
1839 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1840 sends if the hardware supports scatter-gather and checksumming.
1841 This allows socklnd to do client-write and server-read more
1842 efficiently and reduce CPU utilization from skbuf copying.
1845 Frequency : only if NFS exporting from client
1847 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1848 Details : If setting attributes on a file created under NFS that had
1849 never been opened it would be possible to oops the client
1850 if the file had no objects.
1853 Frequency : always for liblustre
1855 Description: liblustre client does MDS+OSTs setattr RPC for each write
1856 Details : When doing a write from a liblustre client, the client
1857 incorrectly issued an RPC to the MDS and each OST the file was
1858 striped over in order to update the timestamps. When writing
1859 with small chunks and many clients this could overwhelm the MDS
1860 with RPCs. In all cases it would slow down the write because
1861 these RPCs are unnecessary.
1863 Severity : enhancement
1865 Description: allow number of MDS service threads to be changed at module load
1866 Details : It is now possible to change the number of MDS service threads
1867 running. Adding "options mds mds_num_threads={N}" to the MDS's
1868 /etc/modprobe.conf will set the number of threads for the next
1869 time Lustre is restarted (assuming the "mds" module is also
1870 reloaded at that time). The default number of threads will
1871 stay the same, 32 for most systems.
1876 Description: OST crash if filesystem is unformatted or corrupt
1877 Details : If an OST is started on a device that has never been formatted
1878 or if the filesystem is corrupt and cannot even mount then the
1879 error handling cleanup routines would dereference a NULL pointer.
1884 Description: NULL pointer deref in llap_from_page.
1885 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1886 incorrect) mapping if the page was truncated/reclaimed while it was
1887 searched for. Check for this condition and skip such pages when
1888 doing readahead. Introduce extra check to llap_from_page() to
1889 verify page->mapping->host is non-NULL (so page is not anonymous).
1892 Frequency : Sometimes when using sys_sendfile
1894 Description: "page not covered by a lock" warnings from ll_readpage
1895 Details : sendfile called ll_readpage without right page locks present.
1896 Now we introduced ll_file_sendfile that does necessary locking
1897 around call to generic_file_sendfile() much like we do in
1901 Frequency : with certain MDS communication failures at client mount time
1903 Description: NULL pointer deref after failed client mount
1904 Details : a client connection request may delayed by the network layer
1905 and not be sent until after the PTLRPC layer has timed out the
1906 request. If the client fails the mount immediately it will try
1907 to clean up before the network times out the request. Add a
1908 reference from the request import to the obd device and delay
1909 the cleanup until the network drops the request.
1912 Frequency : occasionally during client (re)connect
1914 Description: assertion failure during client (re)connect
1915 Details : processing a client connection request may be delayed by the
1916 client or server longer than the client connect timeout. This
1917 causes the client to resend the connection request. If the
1918 original connection request is replied in this interval, the
1919 client may trip an assertion failure in ptlrpc_connect_interpret()
1920 which thought it would be the only running connect process.
1923 Frequency : only with obd_echo servers and clients that are rebooted
1925 Description: kernel BUG accessing uninitialized data structure
1926 Details : When running an obd_echo server it did not start the ping_evictor
1927 thread, and when a client was evicted an uninitialized data
1928 structure was accessed. Start the ping_evictor in the RPC
1929 service startup instead of the OBD startup.
1931 Severity : enhancement
1932 Bugzilla : 10193 (patchless)
1933 Description: Remove dependency on various unexported kernel interfaces.
1934 Details : No longer need reparent_to_init, exit_mm, exit_files,
1935 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1938 Frequency : rare (only users of deprecated and unsupported LDAP config)
1940 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1941 Details : LDAP apparently contains 'lustreName' attributes instead of
1942 'name'. A simple remapping of the name is sufficient.
1945 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1947 Description: waiting_locks_callback trips kernel BUG if client is evicted
1948 Details : Running with the dump_on_timeout debug flag turned on makes
1949 it possible that the waiting_locks_callback() can try to dump
1950 the Lustre kernel debug logs from an interrupt handler. Defer
1951 this log dumping to the expired_lock_main() thread.
1953 Severity : enhancement
1955 Description: Support NFS exporting on 2.6 kernels.
1956 Details : Implement non-rawops metadata methods for NFS server to use without
1957 changing NFS server code.
1960 Frequency : very rare (synthetic metadata workload only)
1962 Description: two racing renames might cause an MDS thread to deadlock
1963 Details : Running the "racer" program may cause one MDS thread to rename
1964 a file from being the source of a rename to being the target of
1965 a rename at exactly the same time that another thread is doing
1966 so, and the second thread has already enqueued these locks after
1967 doing a lookup of the target and is trying to relock them in
1968 order. Ensure that we don't try to re-lock the same resource.
1971 Frequency : only very large systems with liblustre clients
1973 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1974 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1975 all OSTs in parallel. This allows the eviction of stale liblustre
1976 clients to proceed much faster than if they were done in series,
1977 and also offers similar improvements for other set_info RPCs.
1982 Description: excessive CPU usage during initial read phase on client
1983 Details : During the initial read phase on a client, it would agressively
1984 retry readahead on the file, consuming too much CPU and impacting
1985 performance (since 1.4.5.8). Improve the readahead algorithm
1986 to avoid this, and also improve some other common cases (read
1987 of small files in particular, where "small" is files smaller than
1988 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1993 Description: MDS crash when receiving packet with unknown intent.
1994 Details : Do not LBUG in unknown intent case, just return -EFAULT
1996 Severity : enhancement
1997 Bugzilla : 9293, 9385
1998 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1999 Details : Do not serialize getattr (non-intent version) and statfs.
2002 Frequency : occasional, when OST network is overloaded/intermittent
2004 Description: client evicted by OST after bulk IO timeout
2005 Details : If a client sends a bulk IO request (read or write) the OST
2006 may evict the client if it is unresposive to its data GET/PUT
2007 request. This is incorrect if the network is overloaded (takes
2008 too long to transfer the RPC data) or dropped the OST GET/PUT
2009 request. There is no need to evict the client at all, since
2010 the pinger and/or lock callbacks will handle this, and the
2011 client can restart the bulk request.
2014 Frequency : Always when mmapping file with no objects
2016 Description: client crashes when mmapping file with no objects
2017 Details : Check that we actually have objects in a file before doing any
2018 operations on objects in ll_vm_open, ll_vm_close and
2024 Description: Request leak when working with deleted CWD
2025 Details : Introduce advanced request refcount tracking for requests
2026 referenced from lustre intent.
2028 Severity : Enhancement
2030 Description: Cache open file handles on client.
2031 Details : MDS now will return special lock along with openhandle, if
2032 requested and client is allowed to hold openhandle, even if unused,
2033 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2034 closing files for every read/write openration.
2036 Severity : Enhancement
2038 Description: Cache open negative dentries on client when possible.
2039 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2040 negative dentries on lock revocation.
2045 Description: Remounting a client read-only wasn't possible with a zconf mount
2046 Details : It wasn't possible to remount a client read-only with llmount.
2048 Severity : enhancement
2049 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2050 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2051 MPICH is included to add Lustre-specific ADIO interfaces.
2052 This is based closely on the UFS ADIO layer and only differs
2053 in file creation, in order to allow the OST striping to be set.
2054 This is user-contributed code and not supported by CFS.
2059 Description: extended inode attributes (immutable, append-only) work improperly
2060 when 2.4 and 2.6 kernels are used on client/server or vice versa
2061 Details : Introduce kernel-independent values for these flags.
2063 Severity : enhancement
2066 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2067 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2068 granularity. Now, Fractional MB granularity is be supported,
2069 this is very useful for low memory system.
2071 Severity : enhancement
2073 Description: Getattr by fid
2074 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2075 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2076 have locks and avoids getattr rpc requests for it on every stat.
2079 Frequency : Always, for filesystems larger than 2TB
2081 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2082 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2083 filesystems larger than 2TB to be mounted with mballoc enabled.
2086 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2088 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2089 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2090 filesystems larger than 2TB is mounted then the kernel will
2091 silently corrupt the start of the filesystem. CONFIG_LBD is
2092 enabled for all CFS-supported kernels, but the possibility of
2093 this happening with a modified kernel config exists.
2095 Severity : enhancement
2097 Description: add client O_DIRECT support for 2.6 kernels
2098 Details : It is now possible to do O_DIRECT reads and writes to files
2099 in the Lustre client mountpoint on 2.6 kernel clients.
2101 Severity : enhancement
2103 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2104 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2105 parallel, not waiting for response from every OST before sending
2106 a rpc to the next OST.
2111 Description: setattr vs write race when updating file timestamps
2112 Details : Client processes that update a file timestamp into the past
2113 right after writing to the file (e.g. tar) it is possible that
2114 the updated file modification time can be reset to the current
2115 time due to a race between processing the setattr and write RPC.
2117 Severity : enhancement
2119 Description: Bring 'lfs find' closer in line with regular Linux find.
2120 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2121 -print0 options and obtains all the needed info through the lustre
2124 Severity : enhancement
2126 Description: support up to 1024 configured devices on one node
2127 Details : change obd_dev array from statically allocated to dynamically
2128 allocated structs as they are first used to reduce memory usage
2133 Description: Flush dirty partially truncated pages during truncate
2134 Details : Immediatelly flush partially truncated pages in filter_setattr,
2135 this way we completely avoid having any pages in page cache on OST
2136 and can retire ugly workarounds during writes to flush such pages.
2141 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2142 Details : protect inode from truncation within vfs_unlink() context
2143 just take a reference before calling vfs_unlink() and release it
2144 when parent's i_sem is free.
2149 Description: last_id value checked outside lock on OST caused LASSERT failure
2150 Details : If there were multiple MDS->OST object precreate requests in
2151 flight, it was possible that the OST's last object id was checked
2152 outside a lock and incorrectly tripped an assertion. Move checks
2153 inside locks, and discard old precreate requests.
2156 Frequency : always, if extents are used on OSTs
2158 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2159 Details : OSTs using the extents format would not zero the high 16 bits of
2160 the index physical block number. This is not a problem for any
2161 OST filesystems smaller than 16TB, and no kernels support ext3
2162 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2163 new/modified files) and can be fixed for existing filesystems
2164 with e2fsprogs-1.39-cfs1.
2169 Description: import connection selection may be incorrect if timer wraps
2170 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2171 import connections to be ignored if the 32-bit jiffies counter
2172 wraps. Use a 64-bit jiffies counter.
2175 Frequency : very large clusters immediately after boot
2177 Description: LNET request buffers exhausted under heavy short-term load
2178 Details : If a large number of client requests are generated on a service
2179 that has previously never seen so many requests it is possible
2180 that the request buffer growth cannot keep up with the spike in
2181 demand. Instead of dropping incoming requests, they are held in
2182 the LND until the RPC service can accept more requests.
2185 Frequency : Sometimes during replay
2187 Description: Assertion failure in ll_local_open after replay.
2188 Details : If replay happened on an open request reply before we were able
2189 to set replay handler, reply will become not swabbed tripping the
2190 assertion in ll_local_open. Now we set the handler right after
2191 recognising of open request
2194 Frequency : very rare
2196 Description: kernel reports "badness in vsnprintf"
2197 Details : Reading from the "recovery_status" /proc file in small chunks
2198 may cause a negative length in lprocfs_obd_rd_recovery_status()
2199 call to vsnprintf() (which is otherwise harmless). Exit early
2200 if there is no more space in the output buffer.
2202 Severity : enhancement
2204 Description: clear OBD RPC statistics by writing to them
2205 Details : It is now possible to clear the OBD RPC statistics by writing
2206 to the "stats" file.
2211 Description: Client mtime is not the same on different clients after utimes
2212 Details : In some cases, the client was using the utimes() syscall on
2213 a file cached on another node. The clients now validate the
2214 ctime from the MDS + OSTs to determine which one is right.
2219 Description: Inability to activate failout mode
2220 Details : lconf script incorrectly assumed that in pythong string's numeric
2221 value is used in comparisons.
2224 Frequency : always with multiple stripes per file
2226 Description: Inefficient object allocation for mutli-stripe files
2227 Details : When selecting which OSTs to stripe files over, for files with
2228 a stripe count that divides evenly into the number of OSTs,
2229 the MDS is always picking the same starting OST for each file.
2230 Return the OST selection heuristic to the original design.
2235 Description: mount failures may take full timeout to return an error
2236 Details : Under some heavy load conditions it is possible that a
2237 failed mount can wait for the full obd_timeout interval,
2238 possibly several minutes, before reporting an error.
2239 Instead return an error as soon as the status is known.
2241 Frequency : quota enabled and large files being deleted
2243 Description: releasing more than 4GB of quota at once hangs OST
2244 Details : If a user deletes more than 4GB of files on a single OST it
2245 will cause the OST to spin in an infinite loop. Release
2246 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2251 Description: statfs data retrieved from /proc may be stale or zero
2252 Details : When reading per-device statfs data from /proc, in the
2253 {kbytes,files}_{total,free,avail} files, it may appear
2254 as zero or be out of date.
2257 Frequency : systems with MD RAID1 external journal devices
2259 Description: lconf's call to blkid is confused by RAID1 journal devices
2260 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2261 of returning all block devices that match the journal UUID.
2264 Frequency : always, for aggregate stripe size over 4GB
2266 Description: assertion fails when trying to use 4GB stripe size
2267 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2268 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2271 Frequency : always on ppc64
2273 Description: the first write on an ext3 filesystem with mballoc got stuck
2274 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2275 perform endianness conversion.
2277 ------------------------------------------------------------------------------
2279 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2281 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2282 INTEROPERATE with older versions automatically. Please read the
2283 user documentation before upgrading any part of a live system.
2284 * WARNING: Lustre networking configuration changes are required with
2285 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2288 * Support for kernels:
2289 2.6.9-22.0.2.EL (RHEL 4)
2290 2.6.5-7.244 (SLES 9)
2291 2.6.12.6 vanilla (kernel.org)
2294 Severity : enhancement
2295 Bugzilla : 7981/8208
2296 Description: Introduced Lustre Networking (LNET)
2297 Details : LNET is new networking infrastructure for Lustre, it includes
2298 a reorganized network configuration mode (see the user
2299 documentation for full details) as well as support for routing
2300 between different network fabrics. Lustre Networking Devices
2301 (LNDs) for the supported network fabrics have also been
2302 created for this new infrastructure.
2304 Severity : enhancement
2305 Description: Introduced Access control lists
2306 Details : clients can set ACLs on files and directories in order to have
2307 more fine-grained permissions than the standard Unix UGO+RWX.
2308 The MDS must be started with the "-o acl" mount option.
2310 Severity : enhancement
2311 Description: Introduced filesystem quotas
2312 Details : Administrators may now establish per-user quotas on the
2315 Severity : enhancement
2317 Description: Configuration change for the XT3
2318 The PTLLND is now used to run Lustre over Portals on the XT3
2319 The configure option(s) --with-cray-portals are no longer used.
2320 Rather --with-portals=<path-to-portals-includes> is used to
2321 enable building on the XT3. In addition to enable XT3 specific
2322 features the option --enable-cray-xt3 must be used.
2327 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2328 Details : A race between allocating a new llcd and re-getting the llcd_lock
2329 allowed another thread to grab newly-allocated llcd.
2331 Severity : enhancement
2333 Description: 2.6 OST async journal commit and locking fix to improve performance
2334 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2335 2.6 kernels are now async as they already were in 2.4 kernels so
2336 that they can commit concurrently with the network bulk transfer.
2337 For block-allocated files the filter allocation semaphore is held
2338 to avoid filesystem fragmentation during allocation. BKL lock
2339 removed for 2.6 xattr operations where it is no longer needed.
2344 Description: lconf incorrectly determined whether two IP networks could talk
2345 Details : In some more complicated routing and multiple-network
2346 configurations, lconf will avoid trying to make a network
2347 connection to a disjoint part of the IP space. It was doing the
2348 math incorrectly for one set of cases.
2353 Description: Fix for potential infinite loop processing records in an llog.
2354 Details : If an llog record is corrupted/zeroed, it is possible to loop
2355 forever in llog_process(). Validate the llog record length
2356 and skip the remainder of the block on error.
2359 Frequency : occasional (liblustre only)
2361 Description: liblustre could not open files whose last component is a symlink
2362 Details : sysio_path_walk() would incorrectly pass the open intent to
2363 intermediate path components.
2366 Frequency : rare (liblustre only with non-standard tuning)
2367 Bugzilla : 7201 (7350)
2368 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2369 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2370 after it has been released, only do this for referenced locks
2372 Severity : enhancement
2374 Description: specifying an (invalid) directory default stripe_size of -1
2375 would reset the directory default striping
2376 Details : stripe_size -1 was used internally to signal directory stripe
2377 removal, now use "all default" to signal dir stripe removal
2378 as a directory striping of "all default" is not useful
2381 Frequency : common for large clusters running liblustre clients
2383 Description: doing an ls when liblustre clients are running is slow
2384 Details : sending a glimpse AST to a liblustre client waits for every AST
2385 to time out, as liblustre clients will not respond. Since they
2386 cannot cache data we refresh the OST lock LVB from disk instead.
2388 Severity : enhancement
2390 Description: doing an ls at the same time as file IO can be slow
2391 Details : enqueue and other "small" requests can be blocked behind many
2392 large IO requests. Create a new OST IO portal for non-IO
2393 requests so they can be processed faster.
2396 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2398 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2399 created file with mode 000
2400 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2401 allow an ftruncate() as a non-root user to a file with mode 000.
2402 Lustre now respects this flag to disable mode checks when
2403 truncating a file owned by the user
2406 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2409 Description: Revoking locks from clients that went dead or catatonic might take
2411 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2412 cancellation of such locks instant on servers without waiting for
2413 any reply from clients. Clients drops these locks when cancel
2414 notification from server is received without replying.
2417 Frequency : liblustre-only, when liblustre client dies or becomes busy
2419 Description: Doing ls on Linux clients can take a long time with active
2421 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2422 granting such locks to it in the first place if possible. Locks
2423 are taken by proxy on the OST during the read or write and
2424 dropped immediately afterward. Add connect flags handling, do
2425 not grant locks to liblustre clients for glimpse ASTs.
2427 Severity : enhancement
2429 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2430 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2431 sequential reads and behaves badly with applications doing random
2432 reads. Improve it by reading ahead at least read region, and
2433 avoiding excessive large RPC for small reads.
2435 Severity : enhancement
2437 Description: Creating more than 1000 files for a single job may cause a load
2438 imbalance on the OSTs if there are also a large number of OSTs.
2439 Details : qos_prep_create() uses an OST index reseed value that is an
2440 even multiple of the number of available OSTs so that if the
2441 reseed happens in the middle of the object allocation it will
2442 still utilize the OSTs as uniformly as possible.
2447 Description: OST or MDS may oops in ping_evictor_main()
2448 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2449 but doesn't restart at beginning of obd_exports_timed list
2452 Severity : enhancement
2454 Description: improve by-nid export eviction on the MDS and OST
2455 Details : allow multiple exports with the same NID to be evicted at one
2456 time without re-searching the exports list.
2459 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2461 Description: MDS may oops in groups_free()
2462 Details : in rare race conditions a newly allocated group_info struct is
2463 freed again, and this can be NULL. The 2.4 compatibility code
2464 for groups_free() checked for a NULL pointer, but 2.6 did not.
2467 Frequency : common for liblustre clients doing little filesystem IO
2468 Bugzilla : 9352, 7313
2469 Description: server may evict liblustre clients accessing contended locks
2470 Details : if a client is granted a lock or receives a completion AST
2471 with a blocking AST already set it would not reply to the AST
2472 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2475 Frequency : lfs setstripe, only systems with more than 160 OSTs
2477 Description: unable to set striping with a starting offset beyond OST 160
2478 Details : llapi_create_file() incorrectly limited the starting stripe
2479 index to the maximum single-file stripe count.
2482 Frequency : LDAP users only
2484 Description: lconf did not handle in-kernel recovery with LDAP properly
2485 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2487 Severity : enhancement
2489 Description: bind OST threads to NUMA nodes to improve performance
2490 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2491 node and do their allocations there to localize memory access
2493 Severity : enhancement
2495 Description: llmount can determine client NID directly from Myrinet (GM)
2496 Details : the client NID code from gmnalnid was moved directly into
2497 llmount, removing the need to use this or specifying the
2498 client NID explicitly when mounting GM clients with zeroconf
2501 Frequency : if client is started with down MDS
2503 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2504 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2505 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2506 ptlrpc_import_delayed_req() because we didn't check if the
2507 request was interrupted, and we also didn't break out of the
2508 event loop if there was no timeout
2513 Description: data loss during non-page-aligned writes to a single file from
2514 both multiple nodes and multiple threads on one node at same time
2515 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2516 inconsistency led to false short-reads, that were cached and later
2517 used by ->prepare_write() to fill in partially written page,
2518 leading to data loss.
2521 Frequency : always, if lconf --abort_recovery used
2523 Description: lconf --abort_recovery fails with 'Operation not supported'
2524 Details : lconf was attempting to abort recovery on the MDT device and not
2527 Severity : enhancement
2529 Description: remove cleanup logs
2530 Details : replace lconf-generated cleanup logs with lustre internal
2531 cleanup routines. Eliminates the need for client-cleanup and
2534 Severity : enhancement
2536 Description: add support for EAs (user and system) on lustre filesystems
2537 Details : it is now possible to store extended attributes in the Lustre
2538 client filesystem, and with the user_xattr mount option it
2539 is possible to allow users to store EAs on their files also
2541 Severity : enhancement
2543 Description: Add possibility (config option) to show minimal available OST free
2545 Details : When compiled with --enable-mindf configure option, statfs(2)
2546 (and so, df) will return least minimal free space available from
2547 all OSTs as amount of free space on FS, instead of summary of
2548 free spaces of all OSTs.
2550 Severity : enhancement
2552 Description: do not expand extent locks acquired on OST-side
2553 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2554 by server: they are not cached anyway.
2557 Frequency : when mmap is used/binaries executed from Lustre
2559 Description: Unmmap pages before throwing them away from read cache.
2560 Details : llap_shrink cache now attempts to unmap pages before discarding
2561 them (if unmapping failed - do not discard). SLES9 kernel has
2562 extra checks that trigger if this unmapping is not done first.
2567 Description: lconf didn't resolve symlinks before checking to see whether a
2568 given mountpoint was already in use
2571 Frequency : when migrating failover services
2572 Bugzilla : 6395, 9514
2573 Description: When migrating a subset of services from a node (e.g. failback
2574 from a failover service node) the remaining services would
2575 time out and evict clients.
2576 Details : lconf --force (implied by --failover) sets the global obd_timeout
2577 to 5 seconds in order to quickly disconnect, but this caused
2578 other RPCs to time out too quickly. Do not change the global
2579 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2581 Severity : enhancement
2582 Frequency : if MDS is started with down OST
2583 Bugzilla : 9439,5706
2584 Description: Allow startup/shutdown of an MDS without depending on the
2585 availability of the OSTs.
2586 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2587 Add appropriate locking and lov-osc refcounts for safe
2588 cleaning. Add osc abort_inflight calls in case the
2589 synchronize never started.
2592 Frequency : occasional (Cray XT3 only)
2594 Description: root not authorized to access files in CRAY_PORTALS environment
2595 Details : The client process capabilities were not honoured on the MDS in
2596 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2597 been accessed by an authorized user then root was able to access
2598 the file on the local client also. The root user capabilities
2599 are now allowed on the MDS, as this environment has secure UID.
2602 Frequency : occasional
2604 Description: ldiskfs "too long searching" message happens too often
2605 Details : A debugging message (otherwise harmless) prints too often on
2606 the OST console. This has been reduced to only happen when
2607 there are fragmentation problems on the filesystem.
2612 Description: Division by zero in statfs when all OSCs are inactive
2613 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2614 lov_statfs divides by value returned by lov_get_stripecnt().
2618 Bugzilla : 9489, 3273
2619 Description: First write from each client to each OST was only 4kB in size,
2620 to initialize client writeback cache, which caused sub-optimal
2621 RPCs and poor layout on disk for the first writen file.
2622 Details : Clients now request an initial cache grant at (re)connect time
2623 and so that they can start streaming writes to the cache right
2624 away and always do full-sized RPCs if there is enough data.
2625 If the OST is rebooted the client also re-establishes its grant
2626 so that client cached writes will be honoured under the grant.
2631 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2632 Details : Now I/O RPCs go to different portal number and (presumably) fast
2633 lock requests (and glimses) and other RPCs get their own service
2634 threads pool that should be able to service those RPCs
2637 Severity : enhancement
2639 Description: Ability to exchange lustre version between client and servers and
2640 issue warnings at client side if client is too old. Also for
2641 liblustre clients there is ability to refuse connection of too old
2643 Details : New 'version' field is added to connect data structure that is
2644 filled with version info. That info is later checked by server and
2648 Frequency : rare, liblustre only.
2649 Bugzilla : 9296, 9581
2650 Description: Two simultaneous writes from liblustre at offset within same page
2651 might proceed at the same time overwriting eachother with stale
2653 Details : I/O lock withing llu_file_prwv was released too early, before data
2654 actually was hitting the wire. Extended lock-holding time until
2655 server acknowledges receiving data.
2658 Frequency : extremely rare. Never observed in practice.
2660 Description: avoid generating lustre_handle cookie of 0.
2661 Details : class_handle_hash() generates handle cookies by incrementing
2662 global counter, and can hit 0 occasionaly (this is unlikely, but
2663 not impossible, because initial value of cookie counter is
2664 selected randonly). Value of 0 is used as a sentinel meaning
2665 "unassigned handle" --- avoid it. Also coalesce two critical
2666 sections in this function into one.
2668 Severity : enhancement
2670 Description: allow liblustre clients to delegate truncate locking to OST
2671 Details : To avoid overhead of locking, liblustre client instructs OST to
2672 take extent lock in ost_punch() on client's behalf. New connection
2673 flag is added to handle backward compatibility.
2675 Severity : enhancement
2676 Bugzilla : 4928, 7341, 9758
2677 Description: allow number of OST service threads to be specified
2678 Details : a module parameter allows the number of OST service threads
2679 to be specified via "options ost ost_num_threads={N}" in the
2680 OSS's /etc/modules.conf or /etc/modprobe.conf.
2684 Bugzilla : 6146, 9635, 9895
2685 Description: servers crash with bad pointer in target_handle_connect()
2686 Details : In rare cases when a client is reconnecting it was possible that
2687 the connection request was the last reference for that export.
2688 We would temporarily drop the export reference and get a new
2689 one, but this may have been the last reference and the export
2690 was just destroyed. Get new reference before dropping old one.
2692 Severity : enhancement
2693 Frequency : if client is started with failover MDS
2695 Description: Allow multiple MDS hostnames in the mount command
2696 Details : Try to read the configuration from all specified MDS
2697 hostnames during a client mount in case the "primary"
2700 Severity : enhancement
2702 Description: Stop sending data to evicted clients as soon as possible.
2703 Details : Check if the client we are about to send or are sending data to
2704 was evicted already. (Check is done every second of waiting,
2705 for which l_wait_event interface was extended to allow checking
2706 of exit condition at specified intervals).
2709 Frequency : rare, normally only when NFS exporting is done from client
2711 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2713 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2716 Frequency : always, if srand() is called before liblustre initialization
2718 Description: Liblustre uses system PRNG disturbing its usage by user application
2719 Details : Introduce internal to lustre fast and high-quality PRNG for
2720 lustre usage and make liblustre and some other places in generic
2721 lustre code to use it.
2723 Severity : enhancement
2724 Bugzilla : 9477, 9557, 9870
2725 Description: Verify that the MDS configuration logs are updated when xml is
2726 Details : Check if the .xml configuration logs are newer than the config
2727 logs stored on the MDS and report an error if this is the case.
2728 Request --write-conf, or allow starting with --old_conf.
2730 Severity : enhancement
2732 Description: Handle symlinks in the path when checking if Lustre is mounted.
2733 Details : Resolve intermediate symlinks when checking if a client has
2734 mounted a filesystem to avoid duplicate client mounts.
2739 Description: lconf can hit an error exception but still return success.
2740 Details : The lconf command catches the Command error exception at the top
2741 level script context and will exit with the associated exit
2742 status, but doesn't ensure that this exit status is non-zero.
2747 Description: failure of ptlrpc thread startup can cause oops
2748 Details : Starting a ptlrpc service thread can fail if there are a large
2749 number of threads or the server memory is very fragmented.
2750 Handle this without oopsing.
2753 Frequency : always, only if liblustre and non-default acceptor port was used
2755 Description: liblustre cannot connect to servers with non-default acceptor port
2756 Details : tcpnal_set_default_params() was not called and was therefore
2757 ignoring the environment varaible TCPNAL_PORT, as well as other
2758 TCPNAL_ environment variables
2763 Description: two objects could be created on the same OST for a single file
2764 Details : If an OST is down, in some cases it was possible to create two
2765 objects on a single OST for a single file. No problems other
2766 than potential performance impact and spurious error messages.
2770 Bugzilla : 5681, 9562
2771 Description: Client may oops in ll_unhash_aliases
2772 Details : Client dcache may become inconsistent in race condition.
2773 In some cases "getcwd" can fail if the current directory is
2779 Description: Inode refcounting problems in NFS export code
2780 Details : link_raw functions used to call d_instantiate without obtaining
2781 extra inode reference first.
2785 Bugzilla : 9942, 9903
2786 Description: Referencing freed requests leading to crash, memleaks with NFS.
2787 Details : We used to require that call to ll_revalidate_it was always
2788 followed by ll_lookup_it. Also with revalidate_special() it is
2789 possible to call ll_revalidate_it() twice for the same dentry
2790 even if first occurence returned success. This fix changes semantic
2791 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2792 reference on a request referred from the intent.
2793 ll_intent_release() then releases such a request.
2796 Frequency : rare, normally benchmark loads only
2798 Description: unlinked inodes were kept in memory on the client
2799 Details : If a client is repeatedly creating and unlinking files it
2800 can accumulate a lot of stale inodes in the inode slab cache.
2801 If there is no other client load running this can cause the
2802 client node to run out of memory. Instead flush old inodes
2803 from client cache that have the same inode number as a new inode.
2806 Frequency : SLES9 2.6.5 kernel and long filenames only
2807 Bugzilla : 9969, 10379
2808 Description: utime reports stale NFS file handle
2809 Details : SLES9 uses out-of-dentry names in some cases, which confused
2810 the lustre dentry revalidation. Change it to always use the
2814 Frequency : rare, unless heavy write-truncate concurrency is continuous
2815 Bugzilla : 4180, 6984, 7171, 9963, 9331
2816 Description: OST becomes very slow and/or deadlocked during object unlink
2817 Details : filter_destroy() was holding onto the parent directory lock
2818 while truncating+unlinking objects. For very large objects this
2819 may block other threads for a long time and slow overall OST
2820 responsiveness. It may also be possible to get a lock ordering
2821 deadlock in this case, or run out of journal credits because of
2822 the combined truncate+unlink. Solution is to do object truncate
2823 first in one transaction without parent lock, and then do the
2824 final unlink in a new transaction with the parent lock. This
2825 reduces the lock hold time dramatically.
2828 Frequency : rare, 2.4 kernels only
2830 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2831 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2832 before unmounting their filesystem in order to avoid deadlock.
2833 The kernel_locked() function in 2.4 kernels only checks whether
2834 the kernel lock is held, not whether it is this process that is
2835 holding it as 2.6 kernels do.
2840 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2841 Details : The client ptlrpc code may be trying to reconnect to a down
2842 server before a previous connection attempt has timed out.
2843 Increase the reconnect interval to be longer than the connection
2844 timeout interval to avoid sending duplicate connections to
2848 Frequency : echo_client brw_test command
2850 Description: fix echo_client to work with OST preallocated code
2851 Details : OST preallocation code (5137) didn't take echo_client IO path
2852 into account: echo_client calls filter methods outside of any
2853 OST thread and, hence, there is no per-thread preallocated
2854 pages and buffers to use. Solution: hijack pga pages for IO. As
2855 a byproduct, this avoids unnecessary data copying.
2859 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2860 Description: Client can oops in mdc_commit_close() after open replay
2861 Details : It was possible for the MDS to return an open request with no
2862 transaction number in mds_finish_transno() if the client was
2863 evicted, but without actually returning an error. Clients
2864 would later try to replay that open and may trip an assertion
2865 Simplify the client close codepath, and always return an error
2866 from the MDS in case the open is not successful.
2869 Frequency : rare, 2.6 OSTs only
2871 Description: OST may deadlock under high load on fragmented files
2872 Details : If there was a heavy load and highly-fragmented OST filesystems
2873 it was possible to have all the OST threads deadlock waiting on
2874 allocation of biovecs, because the biovecs were not released
2875 until the entire RPC IO was completed. Instead, release biovecs
2876 as soon as they are complete to ensure forward IO progress.
2878 Severity : enhancement
2880 Description: Support for specifying external journal device at mount
2881 Details : If an OST or MDS device is formatted with an external journal
2882 device, this device major/minor is stored in the ext3 superblock
2883 and may not be valid for failover. Allow detecting and
2884 specifying the external journal at mount time.
2889 Description: Mounting an MDS with pending unlinked files may cause oops
2890 Details : target_finish_recovery() calls mds_postrecov() which returned
2891 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2892 considers this an error and immediately begins cleaning up the
2893 lov, just after starting the mds_lov process
2895 Severity : enhancement
2897 Description: Implement 'lfs df' to report actual free space on per-OST basis
2898 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2899 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2900 usage of objects; '-h' to report in human readable format.
2902 ------------------------------------------------------------------------------
2904 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2911 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2912 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2913 at startup during orphan unlinking. Instead, skip buddy update
2914 before setup, as it will be regenerated after recovery is complete.
2919 Description: If an OST is inactive, its locks might reference stale inodes.
2920 Details : lov_change_cbdata() must iterate over all namespaces, even if
2921 they are inactive to clear inode references from the lock.
2923 Severity : enhancement
2924 Frequency : occasional, if non-standard max_dirty_mb used
2926 Description: Client will block write RPCs if not enough grant
2927 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2928 then the client will block writes while waiting for another RPC
2929 to complete instead of consuming its dirty limit. With change
2930 we get improved performance when max_dirty_mb is small.
2932 Severity : enhancement
2933 Bugzilla : 3389, 6253
2934 Description: Add support for supplementary groups on the MDS.
2935 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2936 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2937 lookups for user supplementary groups into a cache.
2941 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2942 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2943 for liblustre O_EXCL is enforced by MDS.
2947 Description: readdir never returns NULL in liblustre.
2948 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2949 dentry in struct dirent.
2953 Frequency : liblustre only, depends on application IO pattern
2954 Description: liblustre clients evicted if not contacting servers
2955 Details : Don't put liblustre clients into the ping_evictor list, so
2956 they will not be evicted by the pinger ever.
2958 Severity : enhancement
2960 Description: Add ability to evict clients by NID from MDS.
2961 Details : By echoing "nid:$NID" string into
2962 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2963 $NID would be instantly evicted from this MDS and from all active
2964 OSTs connected to it.
2968 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2969 Details : lookup_it_finish() used to query file size from OSTs that was not
2974 Description: service threads change working directory to that of init
2975 Details : Starting lustre service threads may pin the working directory
2976 of the parent thread, making that filesystem busy. Threads
2977 now change to the working directory of init to avoid this.
2981 Frequency : during shutdown only
2982 Description: shutdown with a failed MDS or OST can cause unmount to hang
2983 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2988 Frequency : 2.6 only
2989 Description: chmod/chown may include an extra supplementary group
2990 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2991 supplementary group and if none is specified this is used.
2994 Bugzilla : 5479 (6816)
2996 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2997 Details : If lookup is in progress on a file that is unlinked we might try
2998 to revalidate the inode and fail in revalidate after lookup is
2999 complete and ll_file_open() enqueues the open again but
3000 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3003 Frequency : always, if lconf --abort_recovery used
3005 Description: lconf --abort_recovery fails with 'Operation not supported'
3006 Details : lconf was attempting to abort recovery on the MDT device and not
3009 ------------------------------------------------------------------------------
3011 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3016 Frequency : rare (only unsupported configurations with a node running as an
3018 Bugzilla : 6514, 5137
3019 Description: Mounting a Lustre file system on a node running as an OST could
3021 Details : OSTs now preallocates memory needed to write out data at
3022 startup, instead of when needed, to avoid having to
3023 allocate memory in possibly low memory situations.
3024 Specifically, if the file system is mounted on on OST,
3025 memory pressure could force it to try to write out data,
3026 which it needed to allocate memory to do. Due to the low
3027 memory, it would be unable to do so and the node would
3028 become unresponsive.
3030 Severity : enhancement
3032 Description: Addition of lconf --service command line option
3033 Details : lconf now accepts a '--service <arg>' option, which is
3034 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3036 Severity : enhancement
3038 Description: Failover mode is now the default for OSTs.
3039 Details : By default, OSTs will now run in failover mode. To return to
3040 the old behaviour, add '--failout' to the lmc line for OSTs.
3042 Severity : enhancement
3044 Description: Health checks are now provided for MDS and OSTs
3045 Details : Additional detailed health check information on MSD and OSTs
3046 is now provided through the procfs health_check value.
3049 Frequency : occasional, depends on IO load
3051 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3052 after numerous create/delete cycles
3053 Details : The ext3 inode allocation policy would not allocate new inodes
3054 very well on the OSTs because there are no new directories
3055 being created. Instead we look for groups with free space if
3056 the parent directories are nearly full.
3061 Description: Network or server problems during mount may cause partially
3062 mounted clients instead of returning an error.
3063 Details : The config llog parsing code may overwrite the error return
3064 code during mount error handling, returning success instead
3070 Description: MDS can fail to allocate large reply buffers
3071 Details : After long uptimes the MDS can fail to allocate large reply
3072 buffers (e.g. zconf client mount config records) due to memory
3073 fragmentation or consumption by the buffer cache. Preallocate
3074 some large reply buffers so that these replies can be sent even
3075 under memory pressure.
3079 Frequency : rare (liblustre)
3080 Description: fsx running with liblustre complained that using truncate() to
3081 extend the file doesn't work. This patch corrects that issue.
3082 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3083 ATTR_SIZE and lsm use in llu_setattr_raw.
3087 Frequency : rare, only 2.6 kernels
3088 Description: Unusual file access patterns on the MDS may result in inode
3089 data being lost in very rare circumstances.
3090 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3091 code in the 2.6 kernel caused the inode and/or EA data not to
3092 be read from disk, causing single-file corruption.
3096 Frequency : rare, only 2.6 filesystems using extents
3097 Description: Heavy concurrent write and delete load may cause data corruption.
3098 Details : It was possible under high-load situations to have an extent
3099 metadata block in the block device cache from a just-unlinked
3100 file overwrite a newly-allocated data block. We now unmap any
3101 metadata buffers that alias just-allocated data blocks.
3105 Frequency : filesystems with default stripe_count larger than 77
3106 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3107 Details : lconf specifies an inode size of 4096 bytes when the default
3108 stripe_count is larger than 77. This conflicts with the default
3109 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3110 this case to avoid pinning too much memory for large EAs.
3112 ------------------------------------------------------------------------------
3114 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3119 Frequency : rare (extremely heavy IO load with hundreds of clients)
3121 Description: Client is evicted, gets IO error writing to file
3122 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3123 caused clients to be evicted for AST timeouts. The fixes in
3124 bug 5192 mean we no longer need to have such short AST timeouts
3125 so ldlm_timeout has been increased.
3128 Frequency : occasional during --force or --failover shutdown under load
3129 Bugzilla : 5949, 4834
3130 Description: Server oops/LBUG if stopped with --force or --failover under load
3131 Details : a collection of import/export refcount and cleanup ordering
3132 issues fixed for safer force cleanup
3135 Frequency : only filesystems larger than 120 OSTs
3136 Bugzilla : 5990, 6223
3137 Description: lfs getstripe would oops on a very large filesystem
3138 Details : lov_getconfig used kfree on vmalloc'd memory
3141 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3142 Bugzilla : 6242, 6243
3143 Description: reading from files that had been truncated to a non-zero size
3144 but never opened returned no data
3145 Details : ll_file_read() reads zeros from no-object files to EOF
3150 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3151 Details : There's little harm in aborting MDS/OSS recovery and letting it
3152 try again, so I removed the LASSERT and return an error instead.
3154 Severity : enhancement
3156 Description: New debugging infrastructure for tracking down data corruption
3157 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3158 (b) cover more of the client-side code path, and (c) try to narrow
3159 down where problems occurred
3163 Bugzilla : 3819, 4364, 4397, 6313
3164 Description: Racing close and eviction MDS could cause assertion in mds_close
3165 Details : It was possible to get multiple mfd references during close and
3166 client eviction, leading to one thread referencing a freed mfd.
3168 Severity: : enhancement
3169 Bugzilla : 3262, 6359
3170 Description: Attempts to reconnect to servers are now more aggressive.
3171 Details : This builds on the enhanced upcall-less recovery that was added
3172 in 1.4.2. When trying to reconnect to servers, clients will
3173 now try each server in the failover group every 10 seconds. By
3174 default, clients would previously try one server every 25 seconds.
3179 Description: After recovery, certain operations trigger a failed
3180 assertion on a client.
3181 Details : Failing over an mds, using lconf -d --failover, while a
3182 client was doing a readdir() call would cause the client to
3183 LBUG after recovery completed and the readdir() was resent.
3185 Severity : enhancement
3187 Description: Default groups are now added by lconf
3188 Details : You can now run lconf --group <servicename> without having to
3189 manually add groups with lmc.
3192 Frequency : occasional
3194 Description: Nodes with an elan id of 0 trigger a failed assertion
3197 Frequency : always when accessing e.g. tty/console device nodes
3199 Description: tty and some other devices nodes cannot be used on lustre
3200 Details : file's private_data field is used by device data and lustre
3201 values in there got lost. New field was added to struct file to
3202 store fs-specific private data.
3205 Frequency : when exporting Lustre via NFS
3207 Description: NFSD failed occasionally when looking up a path component
3208 Details : NFSD is looking up ".." which was broken in ext3 directories
3209 that had grown large enough to become hashed.
3212 Frequency : Clusters with multiple interfaces not on the same subnet
3214 Description: Nodes will repeatedly try to reconnect to an interface which it
3215 cannot reach and report an error to the log.
3216 Details : Extra peer list entries will be created by lconf with some peers
3217 unreachable. lconf now validates the peer before adding it.
3220 Frequency : Only if a default stripe is set on the filesystem root.
3222 Description: Setting a default stripe on the filesystem root prevented the
3223 filesystem from being remounted.
3224 Details : The client was sending extra request flags in the root getattr
3225 request and did not allocate a reply buffer for the dir EA.
3228 Frequency : occasional, higher if lots of files are accessed by one client
3229 Bugzilla : 6159, 6097
3230 Description: Client trips assertion regarding lsm mismatch/magic
3231 Details : While revalidating inodes the VFS looks up inodes with ifind()
3232 and in rare cases can find an inode that is being freed.
3233 The ll_test_inode() code will free the lsm during ifind()
3234 when it finds an existing inode and then the VFS later attaches
3235 this free lsm to a new inode.
3239 Bugzilla : 6422, 7030
3240 Description: MDS deadlock between mkdir and client eviction
3241 Details : Creating a new file via mkdir or mknod (starting a transaction
3242 and getting the ns lock) can deadlock with client eviction
3243 (gets ns lock and trying to finish a synchronous transaction).
3246 Frequency : occasional
3247 Description: While starting a server, the fsfilt_ext3 module could not be
3249 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3250 kernels. Previously, lconf would still use the ext3 name
3251 when trying to load modules. Now, it will correctly use
3252 ext3 on 2.4 and ldiskfs on 2.6.
3254 Severity : enhancement
3255 Description: The default stripe count has been changed to 1
3256 Details : The interpretation of the default stripe count (0, to lfs
3257 or lmc) has been changed to mean striping across a single
3258 OST, rather than all available. For general usage we have
3259 found a stripe count of 1 or 2 works best.
3261 Severity : enhancement
3262 Description: Add support for compiling against Cray portals.
3263 Details : Conditional compiling for some areas that are different
3267 Frequency : occasional
3268 Bugzilla : 6409, 6834
3269 Description: Creating files with an explicit stripe count may lead to
3270 a failed assertion on the MDS
3271 Details : If some OSTs are full or unavailable, creating files may
3272 trigger a failed assertion on the MDS. Now, Lustre will
3273 try to use other servers or return an error to the
3277 Frequency : occasional
3279 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3280 caused lock timeout during readahead (since 1.4.2).
3281 Details : Processes doing readahead might match a lock that hasn't been
3282 granted yet if there are overlapping and conflicting lock
3283 requests. The readahead process waits on ungranted lock
3284 (original lock is CBPENDING), while OST waits for that process
3285 to cancel CBPENDING read lock and eventually evicts client.
3287 Severity : enhancement
3289 Description: Initial enabling of flock support for clients
3290 Details : Implements fcntl advisory locking and file status functions.
3291 This feature is provided as an optional mount flag (default
3292 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3293 locking are implemented yet, and those that are are not guaranteed
3294 to be completely correct in production environments.
3295 mount -t lustre -o [flock|noflock] ...
3298 Frequency : occasional
3300 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3301 assertion in the ext3 JBD (journaling) layer.
3302 Details : The b_committed_data struct is protected by the big kernel lock
3303 in 2.4 kernels, serializing journal_commit_transaction() and
3304 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3305 access to this struct. In 2.6 kernels there is finer grained
3306 locking to improve SMP performance of the JBD layer.
3310 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3315 Description: merge and cleanup kernel patches.
3318 -----------------------------------------------------------------------------
3320 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3322 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3323 versions of Lustre. Please update all servers and clients to
3324 version 1.4.2 or later at the same time. You must also run
3325 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3326 to update the configuration logs.
3328 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3329 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3330 - swab llog records of type '0' so we get proper header size/idx (5861)
3331 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3332 - fix rename of one directory over another leaking an inode (5953)
3333 - avoid SetPageDirty on 2.6 (5981)
3334 - don't re-add just-being-destroyed locks to the waiting list (5653)
3335 - when creating new directories, inherit the parent's custom
3336 striping settings if present parent (3048)
3337 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3338 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3339 - don't hold client locks on temporary worklist from l_lru (5666)
3340 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3341 - automatically evict dead clients (5921)
3342 - Update file size properly in create+truncate+fstat case (6196)
3343 - Do not unhash mountpoint dentries, do not allow removal of
3345 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3346 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3347 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3348 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3349 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3350 - Correctly handle reads of files with no objects (6243)
3351 - lctl recover will also mark a device active if deactivate used (5933)
3353 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3354 - allow --write-conf on an MDS with different nettype than client (5619)
3355 - don't write config llogs to MDS for mounts not from that MDS (5617)
3356 - lconf should create multiple TCP connections from a client (5201)
3357 - init scripts are now turned off by default; run chkconfig --on
3358 lustre and chkconfig --on lustrefs to use them
3359 - upcalls are no longer needed for clients to recover to failover
3361 - add --abort-recovery option to lconf to abort recovery on device
3363 - add support for an arbitrary number of OSTs (3026)
3364 - Quota support protocol changes.
3365 - forward compatibility changes to wire structs (6007)
3366 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3367 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3368 - improved "lustre status" script
3369 - initialize blocksize for non-regular files (6062)
3370 - added --disable-server and --disable-client configure options (5782)
3371 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3372 - Vanilla 2.4.29 support
3373 - increase maximum number of obd devices to 520 (6242)
3374 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3375 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3376 SLES 9 2.6.5 kernels
3377 - limit stripes per file to 160 (the maximum EA size) (6093)
3379 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3382 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3383 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3384 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3385 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3386 - small ext3 extents cleanups and fixes (5733)
3387 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3388 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3389 - avoid needless client->OST connect, fix handle mismatch (5317)
3390 - fix DLM error path that led to out-of-sync client, long delays (5779)
3391 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3392 - fix several locking issues related to i_size (5492,5624,5654,5672)
3393 - don't move pending lock onto export if it is already evicted (5683)
3394 - fix kernel oops when creating .foo in unlinked directory (5548)
3395 - fix deadlock in obdfilter statistics vs. object create (5811)
3396 - use time_{before,after} to avoid timer jiffies wrap (5882)
3397 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3398 - Do not leak request if server was not able to process it (5154)
3399 - If mds_open unable to find parent dir, make that negative lookup(5154)
3400 - don't create new directories with extent-mapping (5909, 5936)
3402 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3403 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3404 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3405 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3406 - support for RHEL4 kernel on i686 (5773)
3407 - provide error messages when incompatible logs are encountered (5898)
3409 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3410 * version 1.4.0.10 (1.4.1 release candidate 1)
3412 - don't keep a lock reference when lock is not granted (4238)
3413 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3414 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3415 - drop import inflight refcount on signal_completed_replay error (5255)
3416 - unlock page after async write error during send (3677)
3417 - handle missing objects in filter_preprw_read properly (5265)
3418 - no transno return for symlink open, don't save no-trasno open (3440)
3419 - don't try to complete elan receive that already failed (4012)
3420 - free RPC server reply state on error (5406)
3421 - clean up thread from ptlrpc_start_thread() on error (5160)
3422 - readahead could read extra page into cache that wasn't ejected (5388)
3423 - prevent races in class_attach/setup/cleanup/detach (5260)
3424 - don't dereference de->d_inode after l_dput of de (5458)
3425 - use "int" for stripe value returned from lock_to_stripe (5544)
3426 - mballoc allocation and error-checking fixes in 2.6 (5504)
3427 - block device patches to fix I/O request sizes in 2.6 (5482)
3428 - look up hostnames for IB nals (5602)
3429 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3430 - don't start multiple acceptors for the same port (5277)
3431 - fix incorrect LASSERT in mds_getattr_name (5635)
3432 - export a proc file for general "ping" checking (5628)
3433 - fix "lfs check" to not block when the MDS is down (5628)
3435 - service request history (4965)
3436 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3437 - create an "evict_client" /proc entry on OSTs, like the MDS has
3438 - fix mount usage message, return errors per mount(8) (5168)
3439 - change grep [] to grep "[]" in tests so they work in more UMLs
3440 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3441 - remove ancient LOV_MAGIC_V0 EA support (5047)
3442 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3443 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3444 - allow readahead allocations to fail when low on memory (5383)
3445 - mmap locking landed again, after considerable improvement (2828)
3446 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3448 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3451 - send OST transaction number in read/write reply to free req (4966)
3452 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3453 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3454 - the watchdog thread now runs as interruptible (5246)
3455 - flock/lockf fixes (but it's still disabled, pending 5135)
3456 - don't use EXT3 constants in llite code (5094)
3457 - memory shortage at startup could cause assertion (5176)
3459 - reorganization of lov code
3460 - single portals codebase
3462 - add extents/mballoc support (5025)
3463 - direct I/O reads in the obdfilter (4048)
3464 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3466 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3469 - send OST transaction number in read/write reply to free req (4966)
3470 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3471 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3472 - the watchdog thread now runs as interruptible (5246)
3473 - handle missing objects in filter_preprw_read properly (5265)
3474 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3475 - drop import inflight refcount on signal_completed_replay error (5255)
3476 - unlock page after async write error during send (3677)
3477 - return original error code on reconstructed replies (3761)
3478 - no transno return for symlink open, don't save no-trasno open (3440)
3480 - add pid to ldlm debugging output (4922)
3481 - bump the watchdog timeouts -- we can't handle 30sec yet
3482 - extra debugging for orphan dentry/inode bug (5259)
3484 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3487 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3488 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3489 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3490 - fix NULL dereference in /proc/sys/portals/routes (4827)
3491 - allow failed mdc_close() operations to be interrupted (4561)
3492 - stop precreate on OST before MDS would time out on it (4778)
3493 - don't send partial-page writes before EOF from client (4410)
3494 - discard client grant for sub-page writes on large-page clients (4520)
3495 - don't free dentries not owned by NFS code, check generation (4806)
3496 - fix lsm leak if mds_create_objects() fails (4801)
3497 - limit debug_daemon file size, always print CERROR messages (4789)
3498 - use transno after validating reply (3892)
3499 - process timed out requests if import state changes (3754)
3500 - update mtime on OST during writes, return in glimpse (4829)
3501 - add mkfsoptions to LDAP (4679)
3502 - use ->max_readahead method instead of zapping global ra (5039)
3503 - don't interrupt __l_wait_event() during strace
3505 - add software watchdogs to catch hung threads quickly (4941)
3506 - make lustrefs init script start after nfs is mounted
3507 - fix CWARN/ERROR duplication (4930)
3508 - return async write errors to application if possible (2248)
3509 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3510 - print NAL number in %x format (4645)
3511 - update barely-supported suse-2.4.21-171 series (4842)
3512 - support for sles 9 %post scripts
3513 - support for building 2.6 kernel-source packages
3514 - support for sles km_* packages
3516 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3519 - ignore -ENOENT errors in osc_destroy (3639)
3520 - notify osc create thread that OSC is being cleaned up (4600)
3521 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3522 - reconstruct ost_handle() like mds_handle() (4657)
3523 - create a new thread to do import eviction to avoid deadlock (3969)
3524 - let lconf resolve symlinked-to devices (4629)
3525 - don't unlink "objects" from directory with default EA (4554)
3526 - hold socknal file ref over connect in case target is down (4394)
3527 - allow more than 32000 subdirectories in a single directory (3244)
3528 - fix blocks count for O_DIRECT writes (3751)
3529 - OST returns ENOSPC from object create when no space left (4539)
3530 - don't send truncate RPC if file size isn't changing (4410)
3531 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3532 - bind to privileged port in socknal and tcpnal (3689)
3534 - rate limit CERROR/CWARN console message to avoid overload (4519)
3535 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3536 - basic mmap support (3918)
3537 - kernel patch series update from b1_4 (4711)
3539 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3542 - avoid crash during MDS cleanup with OST shut down (2775)
3543 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3544 - don't use bad inodes on the MDS (3744)
3545 - dynamic object preallocation to improve recovery speed (4236)
3546 - don't hold spinlock over lock dumping or change debug flags (4401)
3547 - don't zero obd_dev when it is force cleaned (3651)
3548 - print grants to console if they go negative (4431)
3549 - "lctl deactivate" will stop automatic recovery attempts (3406)
3550 - look for existing locks in ldlm_handle_enqueue() (3764)
3551 - don't resolve lock handle twice in recovery avoiding race (4401)
3552 - revalidate should check working dir is a directory (4134)
3554 - don't always mark "slow" obdfilter messages as errors (4418)
3556 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3559 - don't close LustreDB during write_conf until it is done (3860)
3560 - fix typo in lconf for_each_profile (3821)
3561 - allow dumping logs from multiple threads at one time (3820)
3562 - don't allow multiple threads in OSC recovery (3812)
3563 - fix debug_size parameters (3864)
3564 - fix mds_postrecov to initialize import for llog ctxt (3121)
3565 - replace config semaphore with spinlock (3306)
3566 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3567 - don't allow enqueue to complete on a destroyed export (3822)
3568 - down write_lock before checking llog header bitmap (3825)
3569 - recover from lock replay timeout (3764)
3570 - up llog sem before sending rpc (3652)
3571 - reduce ns lock hold times when setting kms (3267)
3572 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3573 - fix NULL deref and obd_dev leak on setup error (3312)
3574 - replace some LBUG about llog ops with error handling (3841)
3575 - don't match INVALID dentries from d_lookup and spin (3784)
3576 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3577 - fix invalid assertion in ptlrpc_set_wait (3880)
3579 - add libwrap support for the TCP acceptor (3996)
3580 - add /proc/sys/portals/routes for non-root route listing (3994)
3581 - allow setting MDS UUID in .xml (2580)
3582 - print the stack of a process that LBUGs (4228)
3584 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3587 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3588 - make sure to unset replay flag from failed open requests (3440)
3589 - if default stripe count is 0, use OST count for inode size (3636)
3590 - update parent mtime/ctime on client for create/unlink (2611)
3591 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3592 - free recovery state on server during a forced cleanup (3571)
3593 - unregister_reply for resent reqs (3063)
3594 - loop back devices mounting and status check on 2.6 (3563)
3595 - fix resource-creation race that can provoke i_size == 0 (3513)
3596 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3597 - more debugging for page-accounting assertion (3746)
3598 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3599 - avoid deadlock after precreation failure (3758)
3600 - fix race and lock order deadlock in orphan handling (3450, 3750)
3601 - add validity checks when grabbing inodes from l_ast_data (3599)
3603 - add /proc/.../recovery_status to obdfilter (3428)
3604 - lightweight CDEBUG infrastructure, debug daemon (3668)
3605 - change default OSC RPC parameters to be better on small clusters
3606 - turn off OST read cache for files smaller than 32MB
3607 - install man pages and include them in rpms (3100)
3608 - add new init script for (un)mounting lustre filesystems (2593)
3609 - run chkconfig in %post for init scripts (3701)
3610 - drop scimac NAL (unmaintained)
3612 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3615 - clean kiobufs before and after use (3485)
3616 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3617 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3618 - change init script to fail more clearly if not run as root (1528)
3619 - allow clients to reconnect during replay (1742)
3620 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3621 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3622 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3623 - fix cond_resched (3554)
3624 - only evict unfinished clients after recovery (3515)
3625 - allow bulk resend, prevent data loss (3570)
3626 - dynamic ptlrpc request buffer allocation (2102)
3627 - don't allow unlinking open directory if it isn't empty (2904)
3628 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3629 - remove extraneous obd dereference causing LASSERT failure (3334)
3630 - don't use get_cycles() when creating temp. files on the mds (3156)
3631 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3632 - handle EEXIST for set-stripe, set proper directory name (3336)
3634 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3635 - fix ksocknal_fmb_callback() error messages (2918)
3637 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3640 - don't copy lvb into (possibly NULL) reply on error (2983)
3641 - don't deref dentry after dput, don't free lvb on error (2922)
3642 - use the kms to determine writeback rpc length (2947)
3643 - increment oti_logcookies when osc is inactive (2948)
3644 - update client's i_blocks count via lvb messages (2543)
3645 - handle intent open/close of special files properly (1557)
3646 - mount MDS with errors=remount-ro, like obdfilter (2009)
3647 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3648 - don't use cancelling-locks' kms values (2947)
3649 - use highest lock extent for kms, not last one (2925)
3650 - don't dereference ERR_PTR() dentry in error handling path (3107)
3651 - fix thread race in portals_debug_dumplog() (3122)
3652 - create lprocfs device entries at setup instead of at attach (1519)
3653 - common AST error handler, don't evict client on completion race (3145)
3654 - zero nameidata in detach_mnt in 2.6 (3118)
3655 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3656 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3657 - initialize RPC timeout timer earlier for 2.6 (3219)
3658 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3659 - print nal/nid for unknown nid (3258)
3660 - additional checks for oscc recovery before doing precreate (3284)
3661 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3662 - don't crash in mdc_close for bad permissions on open (3285)
3663 - zero i_rdev for non-device files (3147)
3664 - clear page->private before handing to FS, better assertion (3119)
3665 - tune the read pipeline (3236)
3666 - fix incorrect decref of invalidated dentry (2350)
3667 - provide read-ahead stats and refine rpc in flight stats (3328)
3668 - don't hold journal transaction open across create RPC (3313)
3669 - update atime on MDS at close time (3265)
3670 - close LDAP connection when recovering to avoid server load (3315)
3671 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3672 - don't leak open file on MDS after open resend (3325)
3673 - serialize filter_precreate and filter_destroy_precreated (3329)
3674 - loop device shouldn't call sync_dev() for nul device (3092)
3675 - clear page cache after eviction (2766)
3676 - resynchronize MDS->OST in background (2824)
3677 - refuse to mount the same filesystem twice on same mountpoint (3394)
3678 - allow llmount to create routes for mounting behind routers (3320)
3679 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3680 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3681 - fix rare problem with rename on htree directories (3417)
3683 - allow default OST striping configuration per directory (1414)
3684 - fix compilation for qswnal for 2.6 kernels (3125)
3685 - increase maximum number of MDS request buffers for large systems
3686 - change liblustreapi to be useful for external progs like lfsck (3098)
3687 - increase local configuration timeout for slow disks (3353)
3688 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3690 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3693 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3694 - don't overwrite extent policy data in reply if lock was blocked (2901)
3695 - drop filter export grants atomically with removal from device (2663)
3696 - del obd_self_export from work_list in class_disconnect_exports (2908)
3697 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3698 - swab reply message in mdc_close, other PPC fixes (2464)
3699 - fix destroying of named logs (2325)
3700 - overwrite old logs when running lconf --write_conf (2264)
3701 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3702 - fix race in target_handle_connect (2898)
3703 - mds_reint_create() should take same inode create lock (2926)
3704 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3705 - don't close files for self_export to avoid uninitialized obd (2936)
3706 - allow MDS with the same name as client node (2939)
3707 - hold dentry reference for closed log files for unlink (2325)
3708 - reserve space for all logs during transactions (2059)
3709 - don't evict page beyond end of stripe extent (2925)
3710 - don't oops on a deleted current working directory (2399)
3711 - handle hard links to targets without a parent properly (2517)
3712 - don't dereference NULL lock when racing during eviction (2867)
3713 - don't grow lock extents when lots of conflicting locks (2919)
3715 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3718 - account for cache space usage on clients to avoid data loss (974)
3719 - lfsck support in lustre kernel code (2349)
3720 - reduce journal credits needed for BRW writes (2370)
3721 - orphan handling to avoid losing space on client/server crashes
3722 - ptlrpcd can be blocked, stopping ALL progress (2477)
3723 - use lock value blocks to assist in proper KMS, faster stat (1021)
3724 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3725 - recovery for initial connections (2355)
3726 - fixes for mds_cleanup_orphans (1934)
3727 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3728 - block all file creations until orphan recovery completes (1901)
3729 - client remove rq_connection from request struct (2423)
3730 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3731 - recovery timer race (2670)
3732 - mdc_close recovey bug (2532)
3733 - ptlrpc cleanup bug (2710)
3734 - mds timeout on local locks (2588)
3735 - namespace lock held during RPCs (2431)
3736 - handle interrupted sync write properly (2503)
3737 - don't try to handle a message that hasn't been replied to (2699)
3738 - client assert failure during cleanup after abort recovery (2701)
3739 - leak mdc device after failed mount (2712)
3740 - ptlrpc_check_set allows timedout requests to complete (2714)
3741 - wait for inflight reqs when ptlrpcd finishes (2710)
3742 - make sure unregistered services are removed from the srv_list
3743 - reset bulk XID's when resending them (caught by 1138 test)
3744 - unregister_bulk after timeout
3745 - fix lconf error (2694)
3746 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3747 - readahead locks pages, leaves pending causing memory pressure (2673)
3748 - increase OST request buffers to 4096 on large machines (2729)
3749 - fix up permission of existing directories in simple_mkdir (2661)
3750 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3751 - don't assign transno to errored transactions (2742)
3752 - don't delete objects on OST if given a bogus objid from MDS (2751)
3753 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3754 - if rq_no_resend, then timeout request after recovery (2432)
3755 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3756 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3757 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3758 - MDS recovery shouldn't delete objects that it has given out (2730)
3759 - if enqueue arrives after completion, don't clobber LVB (2819)
3760 - don't unlock pages twice when trigger_group_io returns error (2814)
3761 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3762 - don't write pages to disk if there was an error (1450)
3763 - don't ping imports that have recovery disabled (2676)
3764 - take buffered bytes into account when balancing socknal conn (2817)
3765 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3766 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3767 - remove little-endian swabbing of llog records (1987)
3768 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3769 - retry reposting request buffers if they fail (1191)
3770 - grow extent at grant time to avoid granting a revoked lock (2809)
3771 - lock revoke doesn't evict page if covered by a second lock (2765)
3772 - disable VM readahead to avoid reading outside lock extents (2805)
3774 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3775 - updated kernel patches for hp-2.4.20 kernel (2681)
3777 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3780 - fix truncated write corruption (2366)
3781 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3783 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3784 - ensure len(uuid) < 37 in lmc (1171)
3785 - fix ia64 OOPS in llog_test (2255)
3786 - zero end of page at obdfilter for partial page writes (2648)
3787 - don't leave stale dentries around after renames (bug 2428)
3788 - fix timeouts when evicting a client with a single lock held (2642)
3789 - set deadline for the initial HELLO message to drain (2634)
3790 - print out dotted-quad IP addresses in the socknal (2302)
3792 - additional debugging for MDS client eviction problem (2443)
3793 - fix mkfsoptions support for osts (2603, 2604)
3795 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3798 - add series for the vanilla 2.6.0 kernel
3799 - add series for the vanilla 2.4.24 kernel
3800 - add series for a cray x86/64 UL kernel drop
3801 - fix xattr patches for the vanilla 2.4.19 series
3803 - generate true UUIDs in lmc (1171)
3804 - have portals stack dumping break in UML (2466)
3805 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3806 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3807 - wake callback waiting threads on client eviction (2460)
3808 - Add --ptldebug and --subsystem to lmc (1719)
3809 - update assertion to allow safe interrupt allocation
3810 - set rq_no_resend for cancel requests (2432)
3811 - recalculate ptlrpcd timeout after resend (2494)
3812 - call vfs_rmdir when removing pending directories (2368)
3813 - fix renaming a file to itself (2429)
3814 - lmc creates a default one-stripe lov (2454)
3815 - expand procfs space to handle large clusters (2326)
3816 - increase UML stack to avoid overflow
3817 - update lconf's list of debug and subsystem masks
3818 - fix lfs find --obd (2510)
3819 - /proc tunable for disabling filter read caching (2591)
3820 - stop rpm packages from altering slapd.conf (2301)
3821 - disable nagle in the socknal under 0conf (2578)
3822 - choose mds inode size based on stripe count (2572)
3823 - fix kernel-source rpm problems (2516)
3825 - add --disable-doc to avoid pdf generation (2421)
3826 - update documentation, tests, type-os, comments
3827 - avoid format warnings on ia64
3828 - remove the TOE NAL
3829 - tiny code cleanups by removing unused fields
3831 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3834 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3835 - remove the most blatant lies from BUILDING file (2371)
3836 - change default debug level to reasonable production setting
3837 - reduce client side cache size to reduce cache flush time
3838 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3839 - make TCP zerocopy and pinger support enabled by default (2476)
3840 - sync writes completed after process exits caused crashes (2319)
3841 - maintain correct mount count on the MDS (2356)
3842 - backout 1557, because 2316 wasn't really fixed
3843 - better file I/O statistics gathering in /proc
3844 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3845 - another kernel patch to fix zero-copy TCP function export
3846 - don't take duplicate lock when processing re-sent getattr (2420)
3847 - lctl uses obd_self_export instead of creating new conn (2353)
3848 - MDS/OST recovery case which requires object creation asserted (2425)
3849 - move lfs from /usr/sbin to /usr/bin in packages
3850 - fix race between mds_client_add and mds_client_free (2417)
3851 - use kmalloc instead of slabs in portals (2430)
3852 - don't create duplicate records when a failover MDS is present (2442)
3853 - remove unnecessary mount age check (2332)
3854 - don't remove directory inodes from locks prematurely (2451)
3855 - don't break if MDS service name is the same as hostname (2103)
3856 - fix races in client write RPC generation when cache full (2482)
3858 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3861 - remove now-unused request->rq_obd (278)
3862 - if an allocation fails, print out how much memory we've used (1933)
3863 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3864 - add the "configurable stack size" patch to most series files (1256)
3865 - ability to write large log records, for 100+ OST configs (2306)
3866 - fix NULL deref when filter_prep fails (2314)
3867 - fix operator precedence error in filter_sync
3868 - dynamic allocation of socknal TX descriptors (2315)
3869 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3870 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3871 - ensure that CWARN messages in Portals always get to the syslog
3872 - __init/__exit are not for prototype decls (ldlm_init/exit)
3873 - x86-64 compile warning fixes
3874 - fix gateway LMC keyword conflict (2318)
3875 - fix MDS lock inversions in getattr/reint paths (1844)
3876 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3877 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3878 - DLM race condition prevented some lock evictions (2328)
3879 - ENOMEM detection and retry on socknal sends (2230)
3880 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3881 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3882 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3883 - better error messages when a client is rejected during recovery (1505)
3884 - avoid cancelling locks which were never granted, after failure (2330)
3885 - fix i_sem/journal inversion in mds_client_add (2333)
3886 - fix truncate/getattr lock cycle deadlock (2334)
3887 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3888 - fix two rare exit paths which could leak an l_lock() ref (2321)
3889 - fencepost error in MDS/OST orphan recovery (2226)
3890 - make log record alignment 8 bytes (1988)
3891 - lstripe now fails when requested offset > ost_count (2237)
3892 - ensure that all kernel series have a complete list.h (1607)
3893 - fix crashes in special-file operations (2316)
3894 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3896 - allow configurable automake binary, for testing new versions
3897 - small update to the lfs documentation
3899 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3901 * fix negative export reference count in fsfilt_sync (2312)
3903 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3904 * release candidate 0.9.1
3906 - orphans are moved into the PENDING directory for possible recovery
3907 - replayed opens now open by fid for orphan/rename safety (1042)
3908 - last close of an orphan inode generates a transno (683)
3909 - chdir() and mount() now pin the directory entry (1020)
3910 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3911 - discard very old requests without processing them (1502)
3912 - remove some common, well-understood CERRORs (1505)
3913 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3914 - clear "grant" flags in OST replies until OST grant code lands (1644)
3915 - fix read performance by not clobbering i_blksize on client (1598)
3916 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3917 - make LRU size a /proc tunable, clears locks when reduced (707)
3918 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3919 - server locks take references on exports now (1558)
3920 - build fixes for 2.4.20-rh trees (1663)
3921 - return an error from lov_create if all OSCs are inactive (1751)
3922 - fix import levels when a reconnect happens without a timeout (1597)
3923 - exit early from mds_open if we get a lookup error (1749)
3924 - partial page read at EOF wouldn't wait for disk before sending (1642)
3925 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3926 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3927 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3928 - write/truncate lock inversion (1639)
3929 - Don't auto-load obdclass, portals modules during cleanup (1495)
3930 - fix timestamps from jumping to "now" (1763)
3931 - extra journal assertions (1648)
3932 - add an extra multiunlink test (1771)
3933 - fix read_record/write_record API (1776)
3934 - fix leak of offset_extent, possible incorrect i_size later (1772)
3935 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3936 - Debugging for the kqswnal_get_idle_tx problems (1820)
3937 - Allow recovery to be attempted multiple times (1536)
3938 - Write out MDS last_rcvd file after it is first created (1600)
3939 - Fix tx_descriptor leak in failed transmit situations (1827)
3940 - ext3 journaling fixes for assertion failure after IO error (1871)
3941 - class_export_put() on freed export after completion AST error (1896)
3942 - Fix revalidate looping in VFS (1322)
3943 - Don't access a freed export during MDS_REINT timeout (1521)
3944 - Add open-unlink recovery support on the MDS (1673,1764)
3945 - Return an error if no MDS data was read from last_rcvd (1946)
3946 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3947 - Don't setup a disk device that doesn't match exported UUID (317)
3948 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3949 - avoid committing NULL handle in force close
3950 - local.sh is now a one-stripe LOV configuration
3951 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3952 - fix ext3 htree duplicate directory entry corruption (1516)
3953 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3954 - update to new LOV EA format (2097)
3955 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3957 2003-06-15 Phil Schwan <phil@clusterfs.com>
3960 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3961 - per-import/export recovery handling (958, 931, 959)
3962 - multiple last-rcvd slots, for serving multiple FSes (949)
3963 - connections are again shared between multiple imp/exports (963, 964)
3964 - "umount -f" would hang if any requests needed to be sent (393, 978)
3965 - avoid pinning large req buffer by copying for queued messages (989)
3966 - add "uuid" to "lctl device" command to help upcalls (991)
3967 - "open" RPCs with transnos would confuse recovery counters (1037)
3968 - do proper endian conversion of all wire messages (288, 340, 891)
3969 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3970 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3971 - fix ptlrpc_connection leak in target_handle_connect (1174)
3972 - fix import refcounting bug in OST and MDS cleanup (1134)
3973 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3974 - fix very unlikely obd_types race condition (501)
3975 - remove osc_open hack for echo_client (1187)
3976 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3977 - a failure in read_inode2 leads to deadlock (1139)
3978 - cancel ack-locks as soon as transaction is committed (1072)
3979 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3980 - make sure to commitrw after any preprw to avoid deadlock (1162)
3981 - failing to execute a file in a lustre FS would lock inode (1203)
3982 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3983 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3984 - obd_uuid2tgt would walk off the end of the list (1255)
3985 - on IA64 the osc would give portals incorrect bulk size (1258)
3986 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3987 - fix lock inversion caused by new llite matching code (1282)
3988 - limit the number of dirty pages on a client to 10MB (1286)
3989 - timed out locks were not being corrected cancelled (1289)
3990 - fix O_DIRECT above 4GB on IA-32 (1292)
3991 * major user-visible changes
3992 - fail out/fail over policy now controlled by the upcall (993)
3994 - add OBD_PING to check server availability and failure (954)
3995 - lustre messages are now sent in sending host order (288, 340, 891)
3996 - add eadatalen to MDS getattr reply (340)
3997 - OST read replies may contain second buffer, with per-page status (593)
3999 2003-03-11 Phil Schwan <phil@clusterfs.com>
4002 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4003 - failed open()s could cause deadlock; fixed (867, 869)
4004 - stop cancelling OST locks when files are closed (481)
4005 - overlapping XID spaces caused network corruption (851, 853)
4006 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4007 - setattr_raw added, to do single-RPC, server-side setattrs
4008 - lmc/lconf syntax change for OST UUIDs
4009 - fix crashy race condition between ptlrpc_free_req and osc_close
4010 - don't use request in mdc_enqueue if we hit a timeout (889)
4011 - don't set the inode i_size for regular files from the MDS (896)
4012 - handle out of order completion AST (842)
4013 - don't LBUG if a lock request times out after receiving AST (913)
4014 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4015 - if a bad lock AST arrives, send an error instead of dropping entirely
4016 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4017 - fix leak in bulk IO when only partially completed (899, 900, 926)
4018 - fix O_DIRECT for ia64 (55)
4019 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4020 - C-z after timeout could hang a process forever; fixed (977)
4022 - client-side I/O cache (678, 924, 929, 941, 970)
4024 - READPAGE and SETATTRs which don't take server-side locks get
4027 2003-02-11 Phil Schwan <phil@clusterfs.com>
4030 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4031 - Fix fsfilt_extN_readpage() to read a full page of directory
4032 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4033 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4034 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4035 - Most intent operations were removed, in favour of a new RPC mode
4036 that does a single RPC to the server and bypasses most of the VFS
4037 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4038 - Aggressively cancel local locks on DLM servers
4039 - mds_reint_unlink sends EA to the client if it's the last nlink.
4040 client uses that EA to unlink OST objects.
4041 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4042 - recursive symlinks were fixed (439)
4043 - fixed NULL deref in DEBUG_REQ
4044 - filter_update_lastobjid no longer calls sync, which annoyed extN
4045 - fixed multi-client small-writes to a single file problem (445)
4046 - fixed mtime updates during file writes (607)
4047 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4048 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4049 - fixed Linux OST size reporting problem (444, 656)
4050 - OST now updates object mtime with writes or setattr (607, 619)
4051 - client verifies file size before zeroing page past EOF (445)
4052 - OST now writes last allocated objid to disk with allocation (108)
4053 - LOV on echo now works (409)
4055 - mds_reint_unlink sends a new buffer, with the EA included. this
4056 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4057 set if a regular file was being unlinked, and it was the last link
4058 - use PtlGet from the target for bulk writes (315)
4059 - OST now updates object mtime with writes or setattr (607, 619)
4060 - LDLM now has a grant-time callback to revalidate locked items, if
4062 - Many MDS operations were reorganized to combat race conditions
4064 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4065 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4067 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4070 - Fully reactivate OST imports after reconnection (512, others)
4071 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4072 - More graceful error handling for truncating on dead OST (515)
4073 - Don't error out unless we're actually accessing dead stripes (474)
4074 - Fix garbage sizes when stripes are missing (410)
4075 - LRU counters were broken, causing constant lock purge (433, 432)
4076 - garbage on read from stripes with failed OSTs (441)
4077 - mark OSCs as active before reconnecting during recovery (438)
4078 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4079 - lfind did not preserve OST order in output (443)
4080 - symlinks cause hung clients, incorrect data (439)
4081 - stop dereferencing request after dropping refcount (457)
4082 - don't LASSERT(spin_is_locked) on non-SMP (455)
4083 - fixes for many rename() bugs
4084 - fstat didn't correctly synchronize attributes (399)
4085 - server must handle lock cancellation during blocking AST prep (487)
4086 - bulk descriptors were free()d too soon (511)
4087 - fix paths in lconf, which would load incorrect modules (451, 507)
4088 - fix confusing lconf 'host not found' error message (386)
4089 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4090 - fix race condition in mdc_blocking_ast() for inode access (526)
4091 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4092 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4093 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4094 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4095 - fix lstripe in several ways (kernel side) (527)
4096 - fix request leak in ldlm_cli_enqueue (262)
4097 - incorrect OSC was marked inactive after OST failure
4098 - call mds_fs_cleanup before unmounting filesystem (524)
4099 - fix races between taking ns_lock and ldlm_lock_change_resource
4100 - fix races updating LOV export open file list
4101 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4102 - fix recovery NULL deref in ldlm_cli_cancel_unused
4103 - fix some DLM races by using new hash table for lock handles (419)
4104 - permit the client to specify desired inodes, at replay
4105 - duplicate requests when we queue them for replay reintegration
4106 - fix last_rcvd offset calculation
4107 - sync after each recovered transaction, so we always make progress
4108 - never, not always, ERESTART requests without transnos
4109 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4111 - skip replay if the MDS says that the client is already connected
4112 - don't check for a recovery-enabled export to match lctl's UUID
4113 - don't INC_USE_COUNT for phantom exports
4114 - don't crash when cleaning up phantom exports (567)
4115 - don't double-finish or set replay data for errored mdc_open requests
4116 - abort requests when they time out, so we don't get old replies
4117 - send/receive replies for AST messages again
4118 - if the client says that it doesn't have the lock, cancel it on the
4120 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4121 mark it as destroyed, it will all work out when decref is called
4122 - fix module use counts (22, 581)
4124 - ASTs now expect a reply (server cancels lock on error reply)
4126 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4129 - fix many simultaneous client startup (392)
4130 - fix dentry->d_it clobbering
4131 - credentials weren't being shipped for readdir/getattr operations
4132 - remove invalid assertions triggered during some concurrent MD
4134 - proper Lustre versions added (336, 389)
4135 - fix memory leak for create error case (398)
4136 - fix LOV locking bug that would get cli/srv out of sync
4137 - fix echo client over LOV (409)
4138 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4139 - fix double-O_EXCL intent crash (424)
4140 - avoid sending multiple lock CANCELs (352)
4142 - MDS can do multi-client recovery (modulo bugs in new code)
4144 - many updates, edits, cleanups
4146 2002-11-18 Phil Schwan <phil@clusterfs.com>
4149 - fix null d_it dereference (346)
4150 - fix full OST/dbench hang (333)
4151 - fix permission problem with file removal (286)
4152 - fix removal of OSCs from LOV when they fail
4153 - fix NULL deref during bulk timeout (214)
4154 - fix problems related to multiple filesystems on one MDS (241)
4155 - fixed serious subtle metadata locking bugs
4156 - free locks on clients when inodes are removed due to memory
4158 - fix inode pointer in lock data (285)
4159 - partial support for multiple MDS on a single host (241)
4160 - data locks weren't cancelled at clear_inode time (290, 311)
4161 - intent locks could lead to unbounded lock growth (205)
4162 - added a maximum lock count, an LRU list, and a flusher
4163 - fix multiple rename (365)
4164 - properly abstracted the echo client
4165 - OSC locked 1 byte too many; fixed
4166 - rewrote brw callback code:
4167 - fixed recovery bugs related to LOVs (306)
4168 - fixed too-many-pages-in-one-write crash (191)
4169 - fixed (again) crash in sync_io_timeout (214)
4170 - probably fixed callback-related race (385)
4172 - Add capability to MDS protocol
4173 - LDLM cancellations and callbacks on different portals
4175 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4178 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4179 - defer open object destruction to close time (601981, 138)
4180 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4181 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4182 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4183 - don't call obd_setattr in ll_file_release for destroyed objects
4184 * protocol change to lustre_msg: move |version| and add |flags|
4185 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4186 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4187 * added replay of create, unlink, link and rename operations during
4188 MDS failover; recovery should be much more robust now
4189 * remove failed OSCs from LOVs (only lov_create uses this so far)
4190 * the lustre-HOWTO was brought (more) up to date (582544)
4192 2002-10-23 Phil Schwan <phil@clusterfs.com>
4195 - in-use dentries weren't being reused properly (617851)
4196 - prevent multiple LDLM setup (599178)
4197 - fix LOV size calculations for truncate (617853)
4198 - fix client handling of MDS intent errors (POSIX)
4199 - fix permission bug in lovstripe.c test (624321)
4200 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4201 - truncate past end of file could corrupt data
4202 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4203 - a race in recovery could return ETIMEDOUT to apps (623947)
4204 - building outside the source directory was fixed
4205 * the lustre-HOWTO was brought (more) up to date (582544)
4206 * major progress was made on recovery functionality
4208 2002-10-10 Phil Schwan <phil@clusterfs.com>
4211 - recovery deadlock fix
4212 - rm -rf causes LBUG fix (617817)
4213 - file open by multiple tasks fix (618962)
4214 - directory permissions bugs (602707 and 620007)
4215 - journal_stop fixed with locking (611313)
4216 - O_APPEND failures resolved (618273, perhaps 614459)
4217 - lconf PATH fix (619770)
4218 - IA64 build fix (621450)
4219 - RPC buffer sizes scale with amount of memory
4221 2002-10-01 Phil Schwan <phil@clusterfs.com>
4224 - locks would be cancelled without throwing away data pages,
4225 resulting in inconsistent data (605627)
4226 - inode attributes were not always being refreshed (605627, 612449)
4227 - lconf now continues to cleanup after lctl reports an error
4228 - MDS now enforces user permissions (602707)
4229 - lprocfs cleanup fixed, but not yet enabled (614157)
4230 - fixed infinite server hang, should a client not respond to an AST
4231 - avoid going into recovery if user calls readlink() with a buffer
4232 that's too small (613941)
4233 - AST RPCs no longer require replies (614867) -- this may be changed
4234 - don't crash server if client sends an IOV that's too big (611336)
4235 - fixed lock conversion deadlock (611892)
4236 - fixed the following of symlinks (614622)
4237 * recovery: the server can remove locks from a client that dies, other
4238 clients can make progress
4239 * more extN patch fixes
4240 * compile-time configurable ptlrpc buffer allocations
4242 - collaborative read cache document
4243 - Lustre Lite Performance CDR document-in-progress
4245 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4248 - fix typo in patch-2.4.18
4250 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4253 - clear ptlrpc request each time in handle_incoming_request()
4254 - unlink of files now destroys the object on the OST
4256 2002-09-19 Peter Braam <braam@clusterfs.com>
4258 * add hard link support
4259 * change obdfile creation method
4260 * kernel patch changed
4262 2002-09-19 Peter Braam <braam@clusterfs.com>
4265 - stack overflow bug in extN fixed
4267 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4269 * documentation updates
4270 - add man pages for config tools
4271 - update tests/README to describe testing with new config tools
4272 - finish metadata API descriptions
4273 * bug fixes and cleanups
4274 - statfs workaround for 16TB limit
4275 - LOV stripe allocation improved, can stripe on subset of OSTs
4276 - LOV file size/IO offset was wrong for files > 4GB in size
4277 - object EA data was being dropped, caused files to be unreadable
4278 - memory overflow with non-LOV OST caused memory corruption
4279 - fixed regression tests to work with new config tools, obdfilter
4280 - fixed bug when directory size became larger than 1 block
4281 - fixed bug (for single client case) when PWD was deleted
4282 - invalidate local directory pages when doing intent-based ops
4283 - avoid LDLM oops when lock callback contained bad data
4285 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4287 * documentation updates
4288 * bug fixes and cleanups
4289 - configuration tools
4292 - 64-bit compile warnings
4293 - 64-bit internal statfs data
4295 * test_brw on persistent OST devices
4297 * lprocfs (disabled)
4299 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4301 * documentation updates
4302 * bug fixes and cleanups
4303 * configuration tools
4305 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4308 * many small fixes to 0_5_4
4309 * io/network handling
4310 * thinkos in MDS operations
4312 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4315 * crucial basic fixes to 0.5.3
4316 * IOR, Iozone work over Elan
4319 2002-08-07 Phil Schwan <phil@clusterfs.com>
4320 * version 0_5_3, our first alpha
4321 * we use the new Portals iovs
4322 * documentation updates
4323 * bug fixes and cleanups
4324 * small changes in the DLM wire protocol
4326 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4327 * version 0_5_1 with some initial stability,
4328 * locking on MD and file I/O.
4329 * documentation updates
4330 * several bug fixes since 0.5.0
4331 * small changes in wire protocol
4333 2002-07-18 Phil Schwan <phil@clusterfs.com>
4335 * delivered as Lustre Light Alpha
4336 * fixed a crash after handling invalid MDS requests
4337 * fixed directory pages for architectures with non-4k pages sizes
4339 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4340 * release version v0_4_4
4341 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4343 * Fixes a number of interruption problems with OST operations.
4344 * Update documentation for portals header changes
4345 * Move all wire protocol structs/defines to lustre_idl.h
4346 * Fixes symlink length bug.
4347 * Add tcpdump to repository.
4349 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4350 * release version v0_4_3
4351 * Fixes statfs for inodes on extN.
4352 * Fixes bug in runtests which would delete /etc/hosts.
4353 * Use 64-bit object IDs wherever possible (not into VFS though)
4354 Remove ost_get_info, which is unused by lustre, and out of date.
4356 2002-07-03 Peter Braam <braam@clusterfs.com>
4357 * release version v0_4_2 Fixes a lookup error (type not passed)
4358 * move forward to head of Portals
4359 * move forward to latest Lustre kernel
4361 2002-06-25 Peter Braam <braam@clusterfs.com>
4362 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob