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).
15 Severity : enhancement
17 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
19 Severity : enhancement
21 Description: Add rhel5 support to HEAD.
25 Descriptoin: make lov_create() will not stuck in obd_statfs_rqset()
26 Details : If an OST is down the MDS will hang indefinitely in
27 obd_statfs_rqset() waiting for the statfs data. While for
28 MDS QOS usage of statfs, it should not stuck in waiting.
30 Severity : enhancement
32 Description: Update to RHEL4 U6 kernel-2.6.9-67.EL.
34 Severity : enhancement
36 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
40 Description: make mgs_setparam() handle fsname containing dash
41 Details : fsname containing a dash does not work with lctl conf_param
43 Severity : enhancement
45 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
47 Severity : enhancement
49 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
53 Frequency : rare, at shutdown
54 Description: access already free / zero obd_namespace.
55 Details : if client_disconnect_export was called without force flag set,
56 and exist connect request in flight, this can produce access to
57 NULL pointer (or already free pointer) when connect_interpret
58 store ocd flags in obd_namespace.
62 Frequency : only at startup
63 Description: not alloc memory with spinlock held.
64 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
67 Severity : enhancement
69 Description: make lustre randomly fail allocating memory
70 Details : Make lustre randomly failed allocating memory for testing purpose.
72 Severity : enhancement
74 Description: lost problems with lov objid file
75 Details : Fixes some scability and access to not inited memory problems
76 in work with lov objdid file.
81 Description: lfs find does not continue on file error
82 Details : Continue other files processing when a file/dir is absent.
86 Description: Inconsistent usage of lustre_pack_reply()
87 Details : Standardize the usage of lustre_pack_reply() such that it
88 always generate a CERROR on failure.
91 Frequency : occasional
93 Description: MDS hang or stay in waiting lock
94 Details : If client receive lock with CBPENDING flag ldlm need send lock
95 cancel as separate rpc, to avoid situation when cancel request
96 can't processed due all i/o threads stay in wait lock.
100 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
104 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
108 Description: Update to SLES9 kernel-2.6.5-7.287.3.
112 Description: Update to RHEL4 latest kernel.
114 Severity : enhancement
116 Description: Build SLES10 patchless client fails
117 Details : The configure was broken by run ./configure with
118 --with-linux-obj=.... argument for patchless client. When the
119 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
120 can't search header adequately. Use absolute path such as
121 -I($LINUX)/include instead.
125 Description: interrupt oig_wait produce painc on resend.
126 Details : brw_redo_request can be used for resend requests from ptlrpcd and
127 private set, and this produce situation when rq_ptlrpcd_data not
128 copyed to new allocated request and triggered LBUG on assert
129 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
130 ptlrpcd set if request is changed and can be safety changed to use
135 Description: LASSERT_{REQ,REP}SWAB macros are buggy
136 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
137 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
138 these macros and replace them with their difinitions instead.
142 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
143 Details : Update which_patch & target file for SLES10 latest kernel.
146 Description: Remove client patches from SLES 10 kernel.
147 Details : This causes SLES 10 clients to behave as patchless clients
148 even on a Lustre-patched (server) kernel.
150 Severity : enhancement
152 Description: self-adjustable client's lru lists
153 Details : use adaptive algorithm for managing client cached locks lru
154 lists according to current server load, other client's work
155 pattern, memory activities, etc. Both, server and client
156 side namespaces provide number of proc tunables for controlling
159 Severity : enhancement
161 Description: light-weight GSS support
162 Details : Support krb5n and krb5a mode, which keep Kerberos 5 authentication
163 and reduce performance overhead.
165 Severity : enhancement
167 Description: Linux keyring support
168 Details : Support using service of Linux keyring for Lustre GSS internal
169 context refresh/cache mechanism.
173 Description: Fix errors in lfs documentation
174 Details : Fixes man pages
178 Description: don't use GFP_* in generic Lustre code.
179 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
183 Description: obdclass is limited by single OBD_ALLOC(idarray)
184 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
188 Description: warnings with build patchless client with vanila 2.6.19 and up
189 Details : change old ctl_table style and replace ctl_table/ctl_table_header
190 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
194 Frequency : Only for SLES
195 Description: sanity_quota fail test_1
196 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
197 file, which makes TSTID[2] inunique.
201 Frequency : Only for SLES9
202 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
203 Details : Update target/ChangeLog/which_patch .
207 Frequency : Only for test_10 in sanity-quota.sh
208 Description: a bug in quota test code
209 Details : checking a test flag in a wrong place
213 Frequency : Only for test_13 in sanity-quota.sh
214 Description: a bug in quota test script
215 Details : shouldn't check overfull allocated quota
219 Description: jbd statistics
220 Details : Port older jbd statistics patch for sles10
222 Severity : enhancement
224 Description: Add printing inode info into message about error in writepage.
229 Description: reply_lock_interpret crash due to race with it and lock cancel.
230 Details : Do not replay locks that are being cancelled. Do not reference
231 locks by their address during replay, just by their handle.
235 Frequency : When flocks are used.
236 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
237 Details : Flock locks might destroy just granted lock if it could be merged
238 with another existing flock, this is done in completion handler,
239 so teach ldlm_cli_enquque_fini that this is a valid case for
245 Description: Oops in read and write path when failing to allocate lock.
246 Details : Check if lock allocation failed and return error back.
250 Description: lstripe command fails for valid OST index
251 Details : The stripe offset is compared to lov->desc.ld_tgt_count
252 instead of lov->desc.ld_active_tgt_count.
256 Description: sanity.sh failed test 103
257 Details : RHEL mis-interpret setfacl "-X" param, so we won't test setfacl
262 Description: df doesn't work properly if diskfs blocksize != 4K
263 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
267 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
268 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
272 Description: obd_health_check_timeout too short
273 Details : set obd_health_check_timeout as 1.5x of obd_timeout
277 Description: llapi_file_create() does not allow some changes
278 Details : add llapi_file_open() that allows specifying the mode and
279 open flags, and also returns an open file handle.
283 Description: merge and cleanup kernel patches.
284 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
288 Description: Add journal checksum support.(Kernel part)
289 Details : The journal checksum feature adds two new flags i.e
290 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
291 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
292 indicates that the commit block contains the checksum for
293 the blocks described by the descriptor blocks. Now commit
294 record can be sent to disk without waiting for descriptor
295 blocks to be written to disk. This behavior is controlled
296 using JBD2_FEATURE_ASYNC_COMMIT flag.
300 Description: OSS needs mutliple precreate threads
301 Details : Add ability to start more than one create thread per OSS.
305 Description: Sanity.sh test_65k failed: setstripe should have succeeded
306 Details : execute lfs setstripe on client
310 Description: mds_obd_create error creating tmp object
311 Details : When the user sets quota on root, llog will be affected and can't
312 create files and write files.
315 Frequency : Always on ia64 patchless client, and possibly others.
317 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
318 Details : This allows the patchless client to be loaded on architectures
323 Description: RedHat Update kernel for RHEL5
324 Details : Add the kernel config file for RHEL5.
328 Description: RedHat Update kernel for RHEL5
329 Details : Modify the kernel config file more closer RHEL5.
333 Description: Build failure against Centos5 (RHEL5)
334 Details : Define PAGE_SIZE when it isn't present.
338 Description: test_7 on sanity-quota.sh can't be used on separate machines
339 Details : add facet to handle it
343 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
344 Details : with NFS, the anon dentry's parent was set to itself in
345 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
346 corresponding dentry other than use rec->ur_name.
348 Severity : enhancement
350 Description: lfs setstripe enhancement
351 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
355 Description: enable data checksumming by default
356 Details : enable checksum by default, allow --disable-checksum
357 configure option and "-o nochecksum" mount option. Checksums
358 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
362 Description: lustre support for RHEL5
363 Details : Add support for RHEL5.
367 Description: Client eviction due to failover config
368 Details : after a connection loss, the lustre client should attempt to
369 reconnect to the last active server first before trying the
370 other potential connections.
374 Description: when mds and osts use different quota unit(32bit and 64bit),
375 quota will be released repeatly.
376 Details : void sending multiple quota reqs to mds, which will keep the
377 status between the reqs.
381 Description: osts not allocated evenly to files
382 Details : change the condition to increase offset_idx
385 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
387 Description: Data corruption for OSTs that are formatted larger than 2TB
389 Details : When generating the bio request for lustre file writes the
390 sector number would overflow a temporary variable before being
391 used for the IO. The data reads correctly from Lustre (which
392 will overflow in a similar manner) but other file data or
393 filesystem metadata may be corrupted in some cases.
397 Description: Tune the kernel for good SCSI performance.
398 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
399 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
404 Description: rewrite ext2-derived code in obdclass/uuid.c
405 Details : rewrite inherited code (uuid parsing code from ext2 utils)
406 from scratch preserving functionality.
410 Description: "lfs find -obd UUID" prints directories
411 Details : "lfs find -obd UUID" will return all directory names instead
412 of just file names. It is incorrect because the directories
413 do not reside on the OSTs.
417 Description: use i_size_read and i_size_write in 2.6 port
418 Details : replace inode->i_size access with i_size_read/write()
421 Frequency : when removing large files
423 Description: scheduling issue during removal of large Lustre files
424 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
425 It causes scheduling issues when removing large files (17TB in the
429 Frequency : only with liblustre clients on XT3
431 Description: evictions taking too long
432 Details : allow llrd to evict clients directly on OSTs
435 Frequency : only on ppc
437 Description: /proc/fs/lustre/devices broken on ppc
438 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
439 We should make sure the type of 'index' is loff_t and then cast
440 explicitly as needed below. Do not assign an explicitly cast
444 Frequency : only for rhel5
446 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
447 Details : Modify the target file & which_kernel.
449 Severity : enhancement
451 Description: omit set fsid for export NFS
452 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
453 as FS_REQUIRES_DEV for avoid problems with generate fsid.
457 Frequency : Always, for kernels after 2.6.16
458 Description: Fix warning idr_remove called for id=.. which is not allocated.
459 Details : Last kernels save old s_dev before kill super and not allow
460 to restore from callback - restore it before call kill_anon_super.
464 Description: buffer overruns could theoretically occur
465 Details : llapi_semantic_traverse() modifies the "path" argument by
466 appending values to the end of the origin string, and a
467 overrun may occur. Adding buffer overrun check in liblustreapi.
471 Description: Fix error on 'ls .' at the top of the Lustre mount.
472 Details : Don't revalidate dentry if it is a root dentry.
476 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
477 Details : Modify vm-tunables-rhel4.patch.
481 Description: Kernel config for 2.6.18-vanilla.
482 Details : Modify targets/2.6-vanilla.target.in.
483 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
484 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
485 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
486 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
490 Description: improve handling recoverable errors
491 Details : if request processig with error which can be recoverable on server
492 request should be resend, otherwise page released from cache and
497 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
498 Details : Modify target file & which_patch.
499 A flaw was found in the IA32 system call emulation provided
500 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
501 value could be stored in the %RAX register, which could trigger an
502 out-of-bounds system call table access. An untrusted local user
503 could exploit this flaw to run code in the kernel
504 (ie a root privilege escalation). (CVE-2007-4573).
508 Description: O_DIRECT bypasses client statistics.
509 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
510 read_ahead_stats were updated. copied the stats section from
511 osc_send_oap_rpc() into async_internal().
515 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla
519 Description: change order of libsysio includes
520 Details : '#include sysio.h' should always come before '#include xtio.h'
524 Description: handle "serious error: objid * already exists" more gracefully
525 Details : If LAST_ID value on disk is smaller than the objects existing in
526 the O/0/d* directories, it indicates disk corruption and causes an
527 LBUG(). If the object is 0-length, then we should use the existing
528 object. This will help to avoid a full fsck in most cases.
530 Severity : enhancement
532 Description: adapt the lustre_config script to support the upgrade case
533 Details : Add "-u" option for lustre_config script to support upgrading 1.4
534 server targets to 1.6 in parallel.
538 Description: To avoid grant space > avaible space when the disk is almost
539 full. Without this patch you might see the error "grant XXXX >
540 available" or some LBUG about grant, when the disk is almost
542 Details : In filter_check_grant, for non_grant cache write, we should
543 check the left space by if (*left > ungranted + bytes), instead
544 of (*left > ungranted), because only we are sure the left space
545 is enough for another "bytes", then the ungrant space should be
546 increase. In client, we should update cl_avail_grant only there
547 is OBD_MD_FLGRANT in the reply.
551 Description: Update RHEL 4 kernel to fix local root privilege escalation.
552 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
553 described in CVE-2007-4573. This problem could allow untrusted
554 local users to gain root access.
557 Frequency : when using O_DIRECT and quotas
559 Description: Incorrect file ownership on O_DIRECT output files
560 Details : block usage reported by 'lfs quota' does not take into account
561 files that have been written with O_DIRECT.
566 Description: touch file failed when fs is not full
567 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
568 otherwise we can get ENOSP while fs is not full.
572 Description: parallel lock callbacks
573 Details : Instead of sending blocking and completion callbacks as separated
574 requests, adding them to a set and sending in parallel.
577 Frequency : only for Cray XT3
578 Bugzilla : 12829/13455
579 Description: Changing primary group doesn't change the group lustre assigns to
581 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
582 overridden with the primary group provided by the group upcall,
583 whereas the supplied fsgid can be trusted if it is in the list of
584 supplementary groups returned by the group upcall.
586 Severity : enhancement
588 Description: Allow masking D_WARNING, D_ERROR messages from console
589 Details : Console messages can now be disabled via lnet.printk.
593 Description: User code with malformed file open parameter crashes client node
594 Details : Before packing join_file req, all the related reference should be
595 checked carefully in case some malformed flags cause fake
596 join_file req on client.
598 --------------------------------------------------------------------------------
600 2007-08-10 Cluster File Systems, Inc. <info@clusterfs.com>
602 * Support for kernels:
603 2.6.5-7.283 (SLES 9),
604 2.6.9-55.EL (RHEL 4),
605 2.6.16.46-0.14 (SLES 10),
606 2.6.18.8 vanilla (kernel.org)
607 * Client support for unpatched kernels:
608 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
609 2.6.16 - 2.6.22 vanilla (kernel.org)
610 * Due to recently discovered recovery problems, we do not recommend
611 using patchless RHEL 4 clients with this or any earlier release.
612 * Recommended e2fsprogs version: 1.39.cfs8
613 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
614 * Starting with this release, the ldiskfs backing filesystem required
615 by Lustre is now in its own package, lustre-ldiskfs. This package
616 should be installed. It is versioned separately from Lustre and
617 may be released separately in future.
621 Description: block reactivating mgc import until all deactivates complete
622 Details : Fix race when failing back MDT/MGS to itself (testing)
625 Frequency : at statup only
627 Description: mds_lov_synchronize race leads to various problems
628 Details : simultaneous MDT->OST connections at startup can cause the
629 sync to abort, leaving the OSC in a bad state.
631 Severity : enhancement
633 Description: add optional extra BUILD_VERSION info
634 Details : add a new environment variable (namely LUSTRE_VERS) which allows
635 to override the lustre version.
638 Frequency : 2.6.18 servers only
640 Description: ll_kern_mount() doesn't release the module reference
641 Details : The ldiskfs module reference count never drops down to 0
642 because ll_kern_mount() doesn't release the module reference.
647 Description: server LBUG when using old ost_num_threads parameter
648 Details : Accept the old ost_num_threads parameter but warn that it
649 is deprecated, and fix an off-by-one error that caused an LBUG.
654 Description: Transient SCSI error results in persistent IO issue
655 Details : iobuf->dr_error is not reinitialized to 0 between two
659 Frequency : sometimes when underlying device returns I/O errors
661 Description: OSTs not going read-only during write failures
662 Details : OSTs are not remounted read-only when the journal commit threads
663 get I/O errors because fsfilt_ext3 calls journal_start/stop()
664 instead of the ext3 wrappers.
668 Description: poor connect scaling with increasing client count
669 Details : Don't run filter_grant_sanity_check for more than 100 exports
670 to improve scaling for large numbers of clients.
673 Frequency : SLES10 only
675 Description: sanity-quota.sh quotacheck failed: rc = -22
676 Details : Quotas cannot be enabled on SLES10.
679 Frequency : liblustre clients only
681 Description: getdirentries does not give error when run on compute nodes
682 Details : getdirentries does not fail when the size specified as an argument
683 is too small to contain at least one entry
685 Severity : enhancement
687 Description: Add LNET router traceability for debug purposes
688 Details : If a checksum failure occurs with a router as part of the
689 IO path, the NID of the last router that forwarded the bulk data
690 is printed so it can be identified.
695 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
696 Details : Due to a race condition, liblustre clients were occasionally
699 Severity : enhancement
701 Description: lfs setstripe use optional parameters instead of postional
704 Severity : enhancement
706 Description: Nanosecond timestamp support for ldiskfs
707 Details : The on-disk ldiskfs filesystem has added support for nanosecond
708 resolution timestamps. There is not yet support for this at
709 the Lustre filesystem level.
712 Frequency : during server recovery
714 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
715 Details : request with rq_no_resend flag not awake l_wait_event if they get a
719 Frequency : nfs export on patchless client
721 Description: connectathon hang when test nfs export over patchless client
722 Details : Disconnected dentry cannot be found with lookup, so we do not need
723 to unhash it or make it invalid
726 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
727 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
728 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
733 Description: open req refcounting wrong on reconnect
734 Details : If reconnect happened between getting open reply from server and
735 call to mdc_set_replay_data in ll_file_open, we will schedule
736 replay for unreferenced request that we are about to free.
737 Subsequent close will crash in variety of ways.
738 Check that request is still eligible for replay in
739 mdc_set_replay_data().
744 Description: disable writes to filesystem when reading health_check file
745 Details : the default for reading the health_check proc file has changed
746 to NOT do a journal transaction and write to disk, because this
747 can cause reads of the /proc file to hang and block HA state
748 checking on a healthy but otherwise heavily loaded system. It
749 is possible to return to the previous behaviour during configure
750 with --enable-health-write.
752 Severity : enhancement
754 Description: 64-bit inode version
755 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
756 to the inode. This will be required for version-based recovery.
761 Description: MDS fails to start if a duplicate client export is detected
762 Details : in some rare cases it was possible for a client to connect to
763 an MDS multiple times. Upon recovery the MDS would detect this
764 and fail during startup. Handle this more gracefully.
766 Severity : enhancement
768 Description: Add -o localflock option to simulate old noflock
770 Details : This will achieve local-only flock/fcntl locks
776 Description: log_commit_thread vs filter_destroy race leads to crash
777 Details : Take import reference before releasing llog record semaphore
782 Description: Wrong request locking in request set processing
783 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
784 imp_delayed_list, in this place should be used imp_lock.
787 Frequency : when reconnection
789 Description: Grant Leak when osc reconnect to OST
790 Details : When osc reconnect ost, OST(filter) should check whether it
791 should grant more space to client by comparing fed_grant and
792 cl_avail_grant, and return the granted space to client instead
793 of "new granted" space, because client will call osc_init_grant
794 to update the client grant space info.
797 Frequency : when client reconnect to OST
799 Description: Grant Leak when osc do resend and replay bulk write
800 Details : When osc reconnect to OST, OST(filter)should clear grant info of
801 bulk write request, because the grant info while be sync between
802 OSC and OST when reconnect, and we should ignore the grant info
803 these of resend/replay write req.
808 Description: Grant space more than avaiable left space sometimes.
809 Details : When then OST is about to be full, if two bulk writing from
810 different clients came to OST. Accord the avaliable space of the
811 OST, the first req should be permitted, and the second one
812 should be denied by ENOSPC. But if the seconde arrived before
813 the first one is commited. The OST might wrongly permit second
814 writing, which will cause grant space > avaiable space.
817 Frequency : when client is evicted
819 Description: Grant might be wrongly erased when osc is evicted by OST
820 Details : when the import is evicted by server, it will fork another
821 thread ptlrpc_invalidate_import_thread to invalidate the
822 import, where the grant will be set to 0. While the original
823 thread will update the grant it got when connecting. So if
824 the former happened latter, the grant will be wrongly errased
825 because of this race.
830 Description: Checking Stale with correct fid
831 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
832 to check whether it is stale, because sometimes, we want the
833 enqueue happened anyway, and op_data.fid2 will not be initialized.
835 Severity : enhancement
837 Description: update patchless client
838 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
841 Frequency : only with 2.4 kernel
843 Description: random memory corruption
844 Details : size of struct ll_inode_info is to big for union inode.u and this
845 can be cause of random memory corruption.
850 Description: Memory leak in recovery
851 Details : Lov_mds_md was not free in an error handler in mds_create_object.
852 It should also check obd_fail before fsfilt_start, otherwise if
853 fsfilt_start return -EROFS,(failover mds during mds recovery).
854 then the req will return with repmsg->transno = 0 and rc = EROFS.
855 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
856 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
857 be freed no matter whether fsfilt_commit success or not.
860 Frequency : only with huge count clients
862 Description: Prevents from taking the superblock lock in llap_from_page for
864 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
865 for llap_from_page call in ll_removepage prevents from taking the
866 superblock lock for a soon died page.
871 Description: Not check open intent error before release open handle
872 Details : in some rare cases, the open intent error is not checked before
873 release open handle, which may cause
874 ASSERTION(open_req->rq_transno != 0), because it tries to release
875 the failed open handle.
880 Description: Set cat log bitmap only after create log success.
881 Details : in some rare cases, the cat log bitmap is set too early. and it
882 should be set only after create log success.
886 Description: Accessing a block bevice can re-enable I/O when Lustre is
887 tearing down a device.
888 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
893 Description: service threads may hog cpus when there are a lot of requests
894 Details : Insert cond_resched to give other threads a chance to use some CPU
899 Description: the cat log was not initialized in recovery
900 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
901 unlink log on mds will not be initialized until mds post
902 recovery. And also in mds post recovery, the unlink log will
903 initialization will be done asynchronausly, so there will be race
904 between add unlink log and unlink log initialization.
908 Description: brw_stats were being printed incorrectly
909 Details : brw_stats were being printed as log2 but all of them were not
910 recorded as log2. Also remove some code duplication arising from
911 filter_tally_{read,write}.
915 Frequency : rare, only in recovery.
916 Description: ASSERTION(req->rq_type != LI_POISON) failed
917 Details : imp_lock should be held while iterating over imp_sending_list for
918 prevent destroy request after get timeout in ptlrpc_queue_wait.
922 Description: replay-single.sh test 52 fails
923 Details : A lock's skiplist need to be cleanup when it being unlinked
924 from its resource list.
928 Description: Short directio read returns full requested size rather than
930 Details : Direct I/O operations should return actual amount of bytes
931 transferred rather than requested size.
933 Severity : enhancement
935 Description: metadata RPC reduction (e.g. for rm performance)
936 Details : decrease the amount of synchronous RPC between clients and servers
937 by canceling conflicing lock before the operation on the client
938 and packing thier handles into the main operation RPC to server.
940 Severity : enhancement
942 Description: Async OSC create to avoid the blocking unnecessarily.
943 Details : If a OST has no remain object, system will block on the creating
944 when need to create a new object on this OST. Now, ways use
945 pre-created objects when available, instead of blocking on an
946 empty osc while others are not empty. If we must block, we block
947 for the shortest possible period of time.
949 Severity : enhancement
951 Description: refine locking for avoid write wrong info into lov_objid file
952 Details : fix possible races with add new target and write/update data in
955 --------------------------------------------------------------------------------
957 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
962 Frequency : on some architectures
964 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
965 Details : Uninitialized flags sometimes cause configuration commands to
969 Frequency : patchless clients only
971 Description: missing __iget() symbol export
972 Details : The __iget() symbol export is missing. To avoid the need for
973 this on patchless clients the deathrow inode reaper is turned
974 off, and we depend on the VM to clean up old inodes. This
975 dependency was during via the fix for bug 12181.
977 --------------------------------------------------------------------------------
979 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
981 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
982 INTEROPERATE with older versions automatically. In many cases a
983 special upgrade step is needed. Please read the
984 user documentation before upgrading any part of a live system.
985 * WIRE PROTOCOL CHANGE from previous 1.6 beta versions. This
986 version will not interoperate with 1.6 betas before beta5 (1.5.95).
987 * WARNING: Lustre configuration and startup changes are required with
988 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
994 Severity : enhancement
996 Description: MountConf
997 Details : Lustre configuration is now managed via mkfs and mount
998 commands instead of lmc and lconf. New obd types (MGS, MGC)
999 are added for dynamic configuration management. See
1000 https://mail.clusterfs.com/wikis/lustre/MountConf for
1003 Severity : enhancement
1005 Description: dynamic OST addition
1006 Details : OSTs can now be added to a live filesystem
1008 Severity : enhancement
1010 Description: startup order invariance
1011 Details : MDTs and OSTs can be started in any order. Clients only
1012 require the MDT to complete startup.
1014 Severity : enhancement
1016 Description: parallel, asynchronous orphan cleanup
1017 Details : orphan cleanup is now performed in separate threads for each
1018 OST, allowing parallel non-blocking operation.
1020 Severity : enhancement
1022 Description: optimized stripe assignment
1023 Details : stripe assignments are now made based on ost space available,
1024 ost previous usage, and OSS previous usage, in order to try
1025 to optimize storage space and networking resources.
1027 Severity : enhancement
1029 Description: Permanently set tunables
1030 Details : All writable /proc/fs/lustre tunables can now be permanently
1031 set on a per-server basis, at mkfs time or on a live system.
1033 Severity : enhancement
1035 Description: Lustre message v2
1036 Details : Add lustre message format v2.
1038 Severity : enhancement
1040 Description: client OST exclusion list
1041 Details : Clients can be started with a list of OSTs that should be
1042 declared "inactive" for known non-responsive OSTs.
1046 Description: SPEC SFS validation failure on NFS v2 over lustre.
1047 Details : Changes the blocksize for regular files to be 2x RPC size,
1048 and not depend on stripe size.
1050 Severity : enhancement
1052 Description: Multiple MD RPCs in flight.
1053 Details : Further unserialise some read-only MDS RPCs - learn about intents.
1054 To avoid overly-overloading MDS, introduce a limit on number of
1055 MDS RPCs in flight for a single client and add /proc controls
1056 to adjust this limit.
1058 Severity : enhancement
1060 Description: client read/write statistics
1061 Details : Add client read/write call usage stats for performance
1062 analysis of user processes.
1063 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1064 file access. extents_stats shows chunk size distribution.
1065 extents_stats_per_process show chunk size distribution per
1068 Severity : enhancement
1070 Description: mds statistics
1071 Details : Add detailed mds operations statistics in
1072 /proc/fs/lustre/mds/*/stats.
1076 Description: Failure of copying files with lustre special EAs.
1077 Details : Client side always return success for setxattr call for lustre
1078 special xattr (currently only "trusted.lov").
1083 Description: Refcount LNET uuids
1084 Details : The global LNET uuid list grew linearly with every startup;
1085 refcount repeated list entries instead of always adding to
1088 Severity : enhancement
1090 Description: Dynamic service threads
1091 Details : Within a small range, start extra service threads
1092 automatically when the request queue builds up.
1095 Frequency : mixed-endian client/server environments
1097 Description: mixed-endian crashes
1098 Details : The new msg_v2 system had some failures in mixed-endian
1101 Severity : enhancement
1103 Description: Easy OST removal
1104 Details : OSTs can be permanently deactivated with e.g. 'lctl
1105 conf_param lustre-OST0001.osc.active=0'
1107 Severity : enhancement
1109 Description: MGS proc entries
1110 Details : Added basic proc entries for the MGS showing what filesystems
1113 Severity : enhancement
1115 Description: provide MGS failover
1116 Details : Added config lock reacquisition after MGS server failover.
1118 Severity : enhancement
1120 Description: add Linux 2.4 support
1121 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1125 Description: a large application tries to do I/O to the same resource and dies
1126 in the middle of it.
1127 Details : Check the req->rq_arrival time after the call to
1128 ost_brw_lock_get(), but before we do anything about
1129 processing it & sending the BULK transfer request. This
1130 should help move old stale pending locks off the queue as
1131 quickly as obd_timeout.
1134 Frequency : when an incorrect nid is specified during startup
1136 Description: ptlrpc connect to non-existant node causes kernel crash
1137 Details : LNET can't be re-entered from an event callback, which
1138 happened when we expire a message after the export has been
1139 cleaned up. Instead, hand the zombie cleanup off to another
1142 Severity : enhancement
1144 Description: plain/inodebits lock performance improvement
1145 Details : Grouping plain/inodebits in granted list by their request modes
1146 and bits policy, thus improving the performance of search through
1150 Frequency : only if OST filesystem is corrupted
1152 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1153 Details : for a short time RPCs with bulk IO are in the replay list,
1154 but replay of bulk IOs is unimplemented. If the OST filesystem
1155 is corrupted due to disk cache incoherency and then replay is
1156 started it is possible to trip an assertion. Avoid putting
1157 committed RPCs into the replay list at all to avoid this issue.
1160 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1162 Description: System hang on startup
1163 Details : This bug allowed the liblustre (e.g. catamount) client to
1164 return to the app before handling all startup RPCs. This
1165 could leave the node unresponsive to lustre network traffic
1166 and manifested as a server ptllnd timeout.
1168 Severity : enhancement
1170 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1171 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1172 boolean to control whether to print peer debug info when a
1173 client's RPC times out.
1176 Frequency : only for kernels with patches from Lustre below 1.4.3
1178 Description: Remove old rdonly API
1179 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1182 Frequency : only for devices with external journals
1184 Description: Set external device read-only also
1185 Details : During a commanded failover stop, we set the disk device
1186 read-only while the server shuts down. We now also set any
1187 external journal device read-only at the same time.
1190 Frequency : when upgrading from 1.4 while trying to change parameters
1192 Description: The wrong (new) MDC name was used when setting parameters for
1193 upgraded MDT's. Also allows changing of OSC (and MDC)
1194 parameters if --writeconf is specified at tunefs upgrade time.
1197 Frequency : when setting specific ost indicies
1199 Description: QOS code breaks on skipped indicies
1200 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1201 created with --index need not be sequential.
1203 Severity : enhancement
1205 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1206 Details : The uninit_groups feature works in conjunction with the kernel
1207 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1208 up the pass1 processing of e2fsck. This is a read-only feature
1209 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1210 mount filesystems that have had this feature enabled.
1212 Severity : enhancement
1214 Description: Improve multi-block allocation algorithm to avoid fragmentation
1215 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1216 allocation locality and avoid filesystem fragmentation.
1218 ------------------------------------------------------------------------------
1220 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1222 * Support for kernels:
1223 2.4.21-47.0.1.EL (RHEL 3)
1224 2.6.5-7.283 (SLES 9)
1225 2.6.9-42.0.10.EL (RHEL 4)
1226 2.6.12.6 vanilla (kernel.org)
1227 2.6.16.27-0.9 (SLES 10)
1228 * Recommended e2fsprogs version: 1.39.cfs5
1230 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1234 Frequency : occasional, depends on client load and configuration
1235 Bugzilla : 12181, 12203
1236 Description: data loss for recently-modified files
1238 Details : In some cases it is possible that recently written or created
1239 files may not be written to disk in a timely manner (this should
1240 normally be within 30s unless client IO load is very high).
1241 The problem appears as zero-length files or files that are a
1242 multiple of 1MB in size after a client crash or client eviction
1243 that are missing data at the end of the file.
1245 This problem is more likely to be hit on clients where files are
1246 repeatedly created and unlinked in the same directory, clients
1247 have a large amount of RAM, have many CPUs, the filesystem has
1248 many OSTs, the clients are rebooted frequently, and/or the files
1249 are not accessed by other nodes after being written.
1251 The presence of the problem can be detected by looking at
1252 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1253 smaller than the second (nr_unused) then dirty files will not
1254 be flushed automatically to disk. "sync; sleep 10" should be
1255 run several times on the node before unmounting it to update
1256 Lustre (this is also safe to run on nodes without this problem).
1258 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1259 that can cause this same problem, so customers using that kernel
1260 also need to update the kernel in addition to Lustre. In order
1261 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1263 It is normal that files written just before a client crash (less
1264 than 30s) may not yet have been flushed to disk, even for local
1268 Frequency : frequent on thin XT3 nodes
1270 Description: UUID collision on thin XT3 Linux nodes
1271 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1272 randomly, since we relied on an insufficiently-seeded PRNG.
1277 Description: OSS hangs after "All ost request buffers busy"
1278 Details : A deadlock between quota and journal operations caused OSS
1279 hangs after printing "All ost request buffers busy."
1282 Frequency : always on liblustre builds
1284 Description: Cleanup compiler warnings on liblustre
1287 Frequency : always on liblustre builds on XT3
1289 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1293 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1294 > 32000 subdirectories
1295 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1296 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1297 subdirectory count crosses 32000. This will aid e2fsck to
1298 correctly handle more than 32000 subdirectories.
1301 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1303 Description: System hang on startup
1304 Details : This bug allowed the liblustre (e.g. catamount) client to
1305 return to the app before handling all startup RPCs. This
1306 could leave the node unresponsive to lustre network traffic
1307 and manifested as a server ptllnd timeout.
1309 Severity : enhancement
1311 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1312 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1313 boolean to control whether to print peer debug info when a
1314 client's RPC times out.
1319 Description: make O_SYNC working on 2.6 kernels
1320 Details : 2.6 kernels use different method for mark pages for write,
1321 so need add a code to lustre for O_SYNC work.
1326 Description: Failure to close file and release space on NFS
1327 Details : Put inode details into lock acquired in ll_intent_file_open.
1328 Use mdc_intent_lock in ll_intent_open to properly
1329 detect all kind of errors unhandled by mdc_enqueue.
1334 Description: proc file read during shutdown sometimes raced obd removal,
1336 Details : Add lock to prevent obd access after proc file removal.
1339 Frequency : Only for files larger than 4GB on 32-bit clients.
1341 Description: improperly doing page alignment of locks
1342 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1343 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1348 Description: RPCs being resent when they shouldn't be
1349 Details : Some RPCs that should not be resent are being resent. This
1350 can cause inconsistencies in the RPC state machine. Do not
1351 resend such requests.
1354 Frequency : rare, only with NFS export
1356 Description: Crash on NFS re-export node
1357 Details : under very unusual load conditions an assertion is hit in
1358 ll_intent_file_open()
1361 Frequency : only if OST filesystem is corrupted
1363 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1364 Details : for a short time RPCs with bulk IO are in the replay list,
1365 but replay of bulk IOs is unimplemented. If the OST filesystem
1366 is corrupted due to disk cache incoherency and then replay is
1367 started it is possible to trip an assertion. Avoid putting
1368 committed RPCs into the replay list at all to avoid this issue.
1373 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1374 Details : Large single O_DIRECT read and write calls can fail to allocate
1375 a sufficiently large buffer to process the request. In case of
1376 allocation failure the allocation is retried with a smaller
1377 buffer and broken into smaller requests.
1379 Severity : enhancement
1381 Description: Add -o localflock option to simulate old noflock behaviour.
1382 Details : This will achieve local-only flock/fcntl locks coherentness.
1387 Description: versioning check is incomplete
1388 Details : Checking the version difference of client vs. server, report
1389 error if the gap is too big.
1394 Description: add support PG_writeback bit
1395 Details : add support for PG_writeback bit for Lustre, for more carefull
1396 work with page cache in 2.6 kernel. This also fix some deadlocks
1397 and remove hack for work O_SYNC with 2.6 kernel.
1399 Severity : enhancement
1401 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1402 Details : The uninit_groups feature works in conjunction with the kernel
1403 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1404 up the pass1 processing of e2fsck. This is a read-only feature
1405 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1406 mount filesystems that have had this feature enabled.
1408 Severity : enhancement
1410 Description: Improve multi-block allocation algorithm to avoid fragmentation
1411 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1412 allocation locality and avoid filesystem fragmentation.
1414 ------------------------------------------------------------------------------
1416 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1418 * Support for kernels:
1419 2.6.9-42.0.3.EL (RHEL 4)
1420 2.6.5-7.276 (SLES 9)
1421 2.4.21-47.0.1.EL (RHEL 3)
1422 2.6.12.6 vanilla (kernel.org)
1423 2.6.16.21-0.8 (SLES10)
1424 * Recommended e2fsprogs version: 1.39.cfs2-0
1426 * The backwards-compatible /proc/sys/portals symlink has been removed
1427 in this release. Before upgrading, please ensure that you change
1428 any configuration scripts or /etc/sysctl.conf files that access
1429 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1430 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1431 in advance of the upgrade on any system running Lustre 1.4.6 or
1432 newer, since /proc/sys/lnet was added in that version.
1433 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1437 Frequency : only when quota is used
1439 Description: avoid scanning export list for quota master
1440 Details : Change the algorithms to avoid scanning export list in order
1441 to improve the efficiency.
1444 Frequency : MDS failover only, very rarely
1446 Description: "went back in time" messages on mds failover
1447 Details : The greatest transno may be lost when the current operation
1448 finishes with an error (transno==0) and the client's last_rcvd
1449 record is over-written. Save the greatest transno in the
1450 mds_last_transno for this case.
1453 Frequency : always for specific kernels and striping counts
1455 Description: client may get "Matching packet too big" without ACL support
1456 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1457 when trying to access files in certain configurations. The
1458 clients should in fact be denied when mounting because they do
1459 not understand ACLs.
1462 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1464 Description: many clients connecting with IO in progress causes connect timeouts
1465 Details : Avoid synchronous journal commits to avoid delays caused by many
1466 clients connecting/disconnecting when bulk IO is in progress.
1467 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1468 OST_IO_PORTAL to avoid delays behind potentially many pending
1472 Frequency : occasionally with multiple writers to a single file
1474 Description: shared writes to file may result in wrong size reported by stat()
1475 Details : Allow growing of kms when extent lock is cancelled
1478 Frequency : always with random mmap IO to multi-striped file
1480 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1481 Details : Only if the hole is at the end of OST object so that kms is too
1482 small. Fix is to increase kms accordingly in ll_nopage.
1485 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1487 Description: writes to a missing object would leak memory on the OST
1488 Details : If there is an inconsistency between the MDS and OST filesystems,
1489 such that the MDS references an object that doesn't exist, writes
1490 to that object will leak memory due to incorrect cleanup in the
1491 error handling path, eventually running out of memory on the OST.
1496 Description: Creating too long symlink causes lustre errors
1497 Details : Check symlink and name lengths before sending requests to MDS.
1500 Frequency : only if flock is enabled (not on by default)
1502 Description: posix locks not released on fd closure on 2.6.9+
1503 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1504 kernels, this caused such locks not to be released on fd close and
1505 then assertions on fs unmount about still used locks.
1508 Frequency : MDS failover only, very rarely
1510 Description: clients may get ASSERTION(granted_lock != NULL)
1511 Details : When request was taking a long time, and a client was resending
1512 a getattr by name lock request. The were multiple lock requests
1513 with the same client lock handle and
1514 mds_getattr_name->fixup_handle_for_resent_request found one of the
1515 lock handles but later failed with ASSERTION(granted_lock != NULL).
1520 Description: handle->h_buffer_credits > 0, assertion failure
1521 Details : h_buffer_credits is zero after truncate, causing assertion
1522 failure. This patch extends the transaction or creates a new
1526 Frequency : NFS re-export or patchless client
1527 Bugzilla : 11179, 10796
1528 Description: Crash on NFS re-export node (__d_move)
1529 Details : We do not want to hash the dentry if we don't have a lock.
1530 But if this dentry is later used in d_move, we'd hit uninitialised
1531 list head d_hash, so we just do this to init d_hash field but
1532 leave dentry unhashed.
1535 Frequency : NFS re-export or patchless client
1537 Description: NFS exports has problem with symbolic link
1538 Details : lustre client didn't properly install dentry when re-exported
1539 to NFS or running patchless client.
1542 Frequency : NFS re-export or patchless client
1544 Description: Various nfs/patchless fixes.
1545 Details : fixes reuse disconected alias for lookup process - this fixes
1546 warning "find_exported_dentry: npd != pd",
1547 fix permission error with open files at nfs.
1548 fix apply umask when do revalidate.
1551 Frequency : occasional
1553 Description: Crash on NFS re-export node
1554 Details : calling clear_page() on the wrong pointer triggered oops in
1555 generic_mapping_read().
1558 Frequency : rarely, using O_DIRECT IO
1560 Description: unaligned directio crashes client with LASSERT
1561 Details : check for unaligned buffers before trying any requests.
1564 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1566 Description: stale data returned from RAID cache
1567 Details : If only a small amount of IO is done to the RAID device before
1568 reading it again it is possible to get stale data from the RAID
1569 cache instead of reading it from disk.
1572 Frequency : always for sles10 kernel
1574 Description: sles10 support
1575 Details : ll_follow_link: compile fixes and using of nd_set_link
1576 under newer kernels.
1579 Frequency : depends on arch, kernel and compiler version, always on sles10
1582 Description: recursive or deep enough symlinks cause stack overflow
1583 Details : getting rid of large stack-allocated variable in
1587 Frequency : depends on hardware
1589 Description: lustre write performance loss in the SLES10 kernel
1590 Details : the performance loss is caused by using of write barriers in the
1591 ext3 code. The SLES10 kernel turns barrier support on by
1592 default. The fix is to undo that change for ldiskfs.
1594 ------------------------------------------------------------------------------
1596 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1598 * Support for kernels:
1599 2.6.9-42.0.3EL (RHEL 4)
1600 2.6.5-7.276 (SLES 9)
1601 2.4.21-47.0.1.EL (RHEL 3)
1602 2.6.12.6 vanilla (kernel.org)
1606 Frequency : quota enabled and large files being deleted
1608 Description: releasing more than 4GB of quota at once hangs OST
1609 Details : If a user deletes more than 4GB of files on a single OST it
1610 will cause the OST to spin in an infinite loop. Release
1611 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1616 Description: statfs data retrieved from /proc may be stale or zero
1617 Details : When reading per-device statfs data from /proc, in the
1618 {kbytes,files}_{total,free,avail} files, it may appear
1619 as zero or be out of date.
1622 Frequency : systems with MD RAID1 external journal devices
1624 Description: lconf's call to blkid is confused by RAID1 journal devices
1625 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1626 of returning all block devices that match the journal UUID.
1629 Frequency : always, for aggregate stripe size over 4GB
1631 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1632 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1633 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1636 Frequency : always if "lfs find" used on a local file/directory
1638 Description: "lfs find" segfaults if used on a local file/directory
1639 Details : The case where a directory component was not specified wasn't
1640 handled correctly. Handle this properly.
1643 Frequency : always on ppc64
1645 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1646 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1647 perform endianness conversion.
1650 Frequency : rarely (truncate to non-zero file size after write under load)
1651 Bugzilla : 10730, 10687
1652 Description: Files padded with zeros to next 4K multiple
1653 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1654 it is possible that files that are truncated to a non-zero size
1655 immediately after being written are filled with zero bytes beyond
1656 the truncated size. No file data is lost.
1658 Severity : enhancement
1660 Description: Allow recovery/failover for liblustre clients.
1661 Details : liblustre clients were unaware of failover configurations until
1664 Severity : enhancement
1666 Description: user file locks should fail when not mounting with flock option
1667 Details : Set up an error-returning stub in ll_file_operations.lock field
1668 to prevent incorrect behaviour when client is mounted without
1669 flock option. Also, set up properly f_op->flock field for
1673 Frequency : always on ia64
1675 Description: "lfs df" loops on printing out MDS statfs information
1676 Details : The obd_ioctl_data was not initialized and in some systems
1677 this caused a failure during the ioctl that did not return
1678 an error. Initialize the struct and return an error on failure.
1681 Frequency : SLES 9 only
1683 Description: Error of copying files with lustre special EAs as root
1684 Details : Client side always return success for setxattr call for lustre
1685 special xattr (currently only "trusted.lov").
1688 Frequency : rarely on clusters with both ia64+i386 clients
1690 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1691 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1692 clients are doing concurrent writes to the same file it is
1693 possible that the ia64 clients may LASSERT because the OST
1694 extent locks are not PAGE_SIZE aligned. Ensure that grown
1695 locks are always aligned on the request boundary.
1698 Frequency : specific use, occasional
1700 Description: Overwriting in use executable truncates on-disk binary image
1701 Details : If one node attempts to overwrite an executable in use by
1702 another node, we now correctly return ETXTBSY instead of
1703 truncating the file.
1705 Severity : enhancement
1707 Description: Async OSC create to avoid the blocking unnecessarily.
1708 Details : If a OST has no remain object, system will block on the creating
1709 when need to create a new object on this OST. Now, ways use
1710 pre-created objects when available, instead of blocking on an
1711 empty osc while others are not empty. If we must block, we block
1712 for the shortest possible period of time.
1717 Description: chmod on Lustre root is propagated to other clients
1718 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1719 invalid by the follow_mount time.
1722 Frequency : liblustre clients only
1724 Description: Race in 'instant cancel' lock handling could lead to such locks
1725 never to be granted in case of SMP MDS
1726 Details : Do not destroy not yet granted but cbpending locks in
1730 Frequency : replay/resend of open
1732 Description: non null lock assetion failure in mds_intent_policy
1733 Details : Trying to replay/resend lockless open requests resulted in
1734 mds_open() returning 0 with no lock. Now it sets a flag if
1735 a lock is going to be returned.
1737 Severity : enhancement
1739 Description: Checksum enhancements
1740 Details : New checksum enhancements allow for resending RPCs that failed
1743 Severity : enhancement
1745 Description: Tunables on number of dirty pages in cacche
1746 Details : Allow to set limit on number of dirty pages cached.
1751 Description: client crash on unmount - lock still has references
1752 Details : In some error handling cases it was possible to leak a lock
1753 reference on a client while accessing a file. This was not
1754 harmful to the client during operation, but would cause the
1755 client to crash when the filesystem is unmounted.
1758 Frequency : specific case, rare
1760 Description: ETXTBSY on mds though file not in use by client
1761 Details : ETXTBSY is no longer incorrectly returned when attempting to
1762 chmod or chown a directory that the user previously tried to
1763 execute or a currently-executing binary.
1766 Frequency : extremely rare except on liblustre-based clients
1768 Description: Lustre space not freed when files are deleted
1769 Details : Clean up open-unlinked files after client eviction. Previously
1770 the unlink was skipped and the files remained as orphans.
1775 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1776 Details : In some cases it was possible to send a blocking callback to
1777 a client doing a glimpse, even though that client didn't get
1778 a lock granted. When the glimpse lock is cancelled on the OST
1779 the freed lock is left on the waiting list and corrupted the list.
1782 Frequency : all core dumps
1784 Description: Broke core dumps to lustre
1785 Details : Negative dentry may be unhashed if parent does not have UPDATE
1786 lock, but some callers, e.g. do_coredump, expect dentry to be
1787 hashed after successful create, hash it in ll_create_it.
1789 ------------------------------------------------------------------------------
1791 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1793 * Support for kernels:
1794 2.6.9-42.0.2.EL (RHEL 4)
1795 2.6.5-7.276 (SLES 9)
1796 2.4.21-40.EL (RHEL 3)
1797 2.6.12.6 vanilla (kernel.org)
1801 Frequency : always on RHEL 3
1803 Description: Number of open files grows over time
1804 Details : The number of open files grows over time, whether or not
1805 Lustre is started. This was due to a filp leak introduced
1806 by one of our kernel patches.
1808 ------------------------------------------------------------------------------
1810 08-20-2006 Cluster File Systems, Inc. <info@clusterfs.com>
1812 * Support for kernels:
1813 2.6.9-42.EL (RHEL 4)
1814 2.6.5-7.276 (SLES 9)
1815 2.4.21-40.EL (RHEL 3)
1816 2.6.12.6 vanilla (kernel.org)
1821 Bugzilla : 5719, 9635, 9792, 9684
1822 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1823 Details : If a server is under heavy load and cannot reply to new
1824 connection requests before the client resends the (re)connect,
1825 the connection handling code can behave badly if two service
1826 threads are concurrently handing separate (re)connections from
1827 the same client. Add better locking to the connection handling
1828 code, and ensure that only a single connection will be processed
1829 for a given client UUID, even if the lock is dropped.
1831 Severity : enhancement
1833 Description: add TCP zero-copy support to kernel
1834 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1835 sends if the hardware supports scatter-gather and checksumming.
1836 This allows socklnd to do client-write and server-read more
1837 efficiently and reduce CPU utilization from skbuf copying.
1840 Frequency : only if NFS exporting from client
1842 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1843 Details : If setting attributes on a file created under NFS that had
1844 never been opened it would be possible to oops the client
1845 if the file had no objects.
1848 Frequency : always for liblustre
1850 Description: liblustre client does MDS+OSTs setattr RPC for each write
1851 Details : When doing a write from a liblustre client, the client
1852 incorrectly issued an RPC to the MDS and each OST the file was
1853 striped over in order to update the timestamps. When writing
1854 with small chunks and many clients this could overwhelm the MDS
1855 with RPCs. In all cases it would slow down the write because
1856 these RPCs are unnecessary.
1858 Severity : enhancement
1860 Description: allow number of MDS service threads to be changed at module load
1861 Details : It is now possible to change the number of MDS service threads
1862 running. Adding "options mds mds_num_threads={N}" to the MDS's
1863 /etc/modprobe.conf will set the number of threads for the next
1864 time Lustre is restarted (assuming the "mds" module is also
1865 reloaded at that time). The default number of threads will
1866 stay the same, 32 for most systems.
1871 Description: OST crash if filesystem is unformatted or corrupt
1872 Details : If an OST is started on a device that has never been formatted
1873 or if the filesystem is corrupt and cannot even mount then the
1874 error handling cleanup routines would dereference a NULL pointer.
1879 Description: NULL pointer deref in llap_from_page.
1880 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1881 incorrect) mapping if the page was truncated/reclaimed while it was
1882 searched for. Check for this condition and skip such pages when
1883 doing readahead. Introduce extra check to llap_from_page() to
1884 verify page->mapping->host is non-NULL (so page is not anonymous).
1887 Frequency : Sometimes when using sys_sendfile
1889 Description: "page not covered by a lock" warnings from ll_readpage
1890 Details : sendfile called ll_readpage without right page locks present.
1891 Now we introduced ll_file_sendfile that does necessary locking
1892 around call to generic_file_sendfile() much like we do in
1896 Frequency : with certain MDS communication failures at client mount time
1898 Description: NULL pointer deref after failed client mount
1899 Details : a client connection request may delayed by the network layer
1900 and not be sent until after the PTLRPC layer has timed out the
1901 request. If the client fails the mount immediately it will try
1902 to clean up before the network times out the request. Add a
1903 reference from the request import to the obd device and delay
1904 the cleanup until the network drops the request.
1907 Frequency : occasionally during client (re)connect
1909 Description: assertion failure during client (re)connect
1910 Details : processing a client connection request may be delayed by the
1911 client or server longer than the client connect timeout. This
1912 causes the client to resend the connection request. If the
1913 original connection request is replied in this interval, the
1914 client may trip an assertion failure in ptlrpc_connect_interpret()
1915 which thought it would be the only running connect process.
1918 Frequency : only with obd_echo servers and clients that are rebooted
1920 Description: kernel BUG accessing uninitialized data structure
1921 Details : When running an obd_echo server it did not start the ping_evictor
1922 thread, and when a client was evicted an uninitialized data
1923 structure was accessed. Start the ping_evictor in the RPC
1924 service startup instead of the OBD startup.
1926 Severity : enhancement
1927 Bugzilla : 10193 (patchless)
1928 Description: Remove dependency on various unexported kernel interfaces.
1929 Details : No longer need reparent_to_init, exit_mm, exit_files,
1930 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1933 Frequency : rare (only users of deprecated and unsupported LDAP config)
1935 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1936 Details : LDAP apparently contains 'lustreName' attributes instead of
1937 'name'. A simple remapping of the name is sufficient.
1940 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1942 Description: waiting_locks_callback trips kernel BUG if client is evicted
1943 Details : Running with the dump_on_timeout debug flag turned on makes
1944 it possible that the waiting_locks_callback() can try to dump
1945 the Lustre kernel debug logs from an interrupt handler. Defer
1946 this log dumping to the expired_lock_main() thread.
1948 Severity : enhancement
1950 Description: Support NFS exporting on 2.6 kernels.
1951 Details : Implement non-rawops metadata methods for NFS server to use without
1952 changing NFS server code.
1955 Frequency : very rare (synthetic metadata workload only)
1957 Description: two racing renames might cause an MDS thread to deadlock
1958 Details : Running the "racer" program may cause one MDS thread to rename
1959 a file from being the source of a rename to being the target of
1960 a rename at exactly the same time that another thread is doing
1961 so, and the second thread has already enqueued these locks after
1962 doing a lookup of the target and is trying to relock them in
1963 order. Ensure that we don't try to re-lock the same resource.
1966 Frequency : only very large systems with liblustre clients
1968 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1969 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1970 all OSTs in parallel. This allows the eviction of stale liblustre
1971 clients to proceed much faster than if they were done in series,
1972 and also offers similar improvements for other set_info RPCs.
1977 Description: excessive CPU usage during initial read phase on client
1978 Details : During the initial read phase on a client, it would agressively
1979 retry readahead on the file, consuming too much CPU and impacting
1980 performance (since 1.4.5.8). Improve the readahead algorithm
1981 to avoid this, and also improve some other common cases (read
1982 of small files in particular, where "small" is files smaller than
1983 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1988 Description: MDS crash when receiving packet with unknown intent.
1989 Details : Do not LBUG in unknown intent case, just return -EFAULT
1991 Severity : enhancement
1992 Bugzilla : 9293, 9385
1993 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1994 Details : Do not serialize getattr (non-intent version) and statfs.
1997 Frequency : occasional, when OST network is overloaded/intermittent
1999 Description: client evicted by OST after bulk IO timeout
2000 Details : If a client sends a bulk IO request (read or write) the OST
2001 may evict the client if it is unresposive to its data GET/PUT
2002 request. This is incorrect if the network is overloaded (takes
2003 too long to transfer the RPC data) or dropped the OST GET/PUT
2004 request. There is no need to evict the client at all, since
2005 the pinger and/or lock callbacks will handle this, and the
2006 client can restart the bulk request.
2009 Frequency : Always when mmapping file with no objects
2011 Description: client crashes when mmapping file with no objects
2012 Details : Check that we actually have objects in a file before doing any
2013 operations on objects in ll_vm_open, ll_vm_close and
2019 Description: Request leak when working with deleted CWD
2020 Details : Introduce advanced request refcount tracking for requests
2021 referenced from lustre intent.
2023 Severity : Enhancement
2025 Description: Cache open file handles on client.
2026 Details : MDS now will return special lock along with openhandle, if
2027 requested and client is allowed to hold openhandle, even if unused,
2028 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2029 closing files for every read/write openration.
2031 Severity : Enhancement
2033 Description: Cache open negative dentries on client when possible.
2034 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2035 negative dentries on lock revocation.
2040 Description: Remounting a client read-only wasn't possible with a zconf mount
2041 Details : It wasn't possible to remount a client read-only with llmount.
2043 Severity : enhancement
2044 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2045 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2046 MPICH is included to add Lustre-specific ADIO interfaces.
2047 This is based closely on the UFS ADIO layer and only differs
2048 in file creation, in order to allow the OST striping to be set.
2049 This is user-contributed code and not supported by CFS.
2054 Description: extended inode attributes (immutable, append-only) work improperly
2055 when 2.4 and 2.6 kernels are used on client/server or vice versa
2056 Details : Introduce kernel-independent values for these flags.
2058 Severity : enhancement
2061 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2062 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2063 granularity. Now, Fractional MB granularity is be supported,
2064 this is very useful for low memory system.
2066 Severity : enhancement
2068 Description: Getattr by fid
2069 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2070 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2071 have locks and avoids getattr rpc requests for it on every stat.
2074 Frequency : Always, for filesystems larger than 2TB
2076 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2077 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2078 filesystems larger than 2TB to be mounted with mballoc enabled.
2081 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2083 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2084 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2085 filesystems larger than 2TB is mounted then the kernel will
2086 silently corrupt the start of the filesystem. CONFIG_LBD is
2087 enabled for all CFS-supported kernels, but the possibility of
2088 this happening with a modified kernel config exists.
2090 Severity : enhancement
2092 Description: add client O_DIRECT support for 2.6 kernels
2093 Details : It is now possible to do O_DIRECT reads and writes to files
2094 in the Lustre client mountpoint on 2.6 kernel clients.
2096 Severity : enhancement
2098 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2099 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2100 parallel, not waiting for response from every OST before sending
2101 a rpc to the next OST.
2106 Description: setattr vs write race when updating file timestamps
2107 Details : Client processes that update a file timestamp into the past
2108 right after writing to the file (e.g. tar) it is possible that
2109 the updated file modification time can be reset to the current
2110 time due to a race between processing the setattr and write RPC.
2112 Severity : enhancement
2114 Description: Bring 'lfs find' closer in line with regular Linux find.
2115 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2116 -print0 options and obtains all the needed info through the lustre
2119 Severity : enhancement
2121 Description: support up to 1024 configured devices on one node
2122 Details : change obd_dev array from statically allocated to dynamically
2123 allocated structs as they are first used to reduce memory usage
2128 Description: Flush dirty partially truncated pages during truncate
2129 Details : Immediatelly flush partially truncated pages in filter_setattr,
2130 this way we completely avoid having any pages in page cache on OST
2131 and can retire ugly workarounds during writes to flush such pages.
2136 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2137 Details : protect inode from truncation within vfs_unlink() context
2138 just take a reference before calling vfs_unlink() and release it
2139 when parent's i_sem is free.
2144 Description: last_id value checked outside lock on OST caused LASSERT failure
2145 Details : If there were multiple MDS->OST object precreate requests in
2146 flight, it was possible that the OST's last object id was checked
2147 outside a lock and incorrectly tripped an assertion. Move checks
2148 inside locks, and discard old precreate requests.
2151 Frequency : always, if extents are used on OSTs
2153 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2154 Details : OSTs using the extents format would not zero the high 16 bits of
2155 the index physical block number. This is not a problem for any
2156 OST filesystems smaller than 16TB, and no kernels support ext3
2157 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2158 new/modified files) and can be fixed for existing filesystems
2159 with e2fsprogs-1.39-cfs1.
2164 Description: import connection selection may be incorrect if timer wraps
2165 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2166 import connections to be ignored if the 32-bit jiffies counter
2167 wraps. Use a 64-bit jiffies counter.
2170 Frequency : very large clusters immediately after boot
2172 Description: LNET request buffers exhausted under heavy short-term load
2173 Details : If a large number of client requests are generated on a service
2174 that has previously never seen so many requests it is possible
2175 that the request buffer growth cannot keep up with the spike in
2176 demand. Instead of dropping incoming requests, they are held in
2177 the LND until the RPC service can accept more requests.
2180 Frequency : Sometimes during replay
2182 Description: Assertion failure in ll_local_open after replay.
2183 Details : If replay happened on an open request reply before we were able
2184 to set replay handler, reply will become not swabbed tripping the
2185 assertion in ll_local_open. Now we set the handler right after
2186 recognising of open request
2189 Frequency : very rare
2191 Description: kernel reports "badness in vsnprintf"
2192 Details : Reading from the "recovery_status" /proc file in small chunks
2193 may cause a negative length in lprocfs_obd_rd_recovery_status()
2194 call to vsnprintf() (which is otherwise harmless). Exit early
2195 if there is no more space in the output buffer.
2197 Severity : enhancement
2199 Description: clear OBD RPC statistics by writing to them
2200 Details : It is now possible to clear the OBD RPC statistics by writing
2201 to the "stats" file.
2206 Description: Client mtime is not the same on different clients after utimes
2207 Details : In some cases, the client was using the utimes() syscall on
2208 a file cached on another node. The clients now validate the
2209 ctime from the MDS + OSTs to determine which one is right.
2214 Description: Inability to activate failout mode
2215 Details : lconf script incorrectly assumed that in pythong string's numeric
2216 value is used in comparisons.
2219 Frequency : always with multiple stripes per file
2221 Description: Inefficient object allocation for mutli-stripe files
2222 Details : When selecting which OSTs to stripe files over, for files with
2223 a stripe count that divides evenly into the number of OSTs,
2224 the MDS is always picking the same starting OST for each file.
2225 Return the OST selection heuristic to the original design.
2230 Description: mount failures may take full timeout to return an error
2231 Details : Under some heavy load conditions it is possible that a
2232 failed mount can wait for the full obd_timeout interval,
2233 possibly several minutes, before reporting an error.
2234 Instead return an error as soon as the status is known.
2236 Frequency : quota enabled and large files being deleted
2238 Description: releasing more than 4GB of quota at once hangs OST
2239 Details : If a user deletes more than 4GB of files on a single OST it
2240 will cause the OST to spin in an infinite loop. Release
2241 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2246 Description: statfs data retrieved from /proc may be stale or zero
2247 Details : When reading per-device statfs data from /proc, in the
2248 {kbytes,files}_{total,free,avail} files, it may appear
2249 as zero or be out of date.
2252 Frequency : systems with MD RAID1 external journal devices
2254 Description: lconf's call to blkid is confused by RAID1 journal devices
2255 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2256 of returning all block devices that match the journal UUID.
2259 Frequency : always, for aggregate stripe size over 4GB
2261 Description: assertion fails when trying to use 4GB stripe size
2262 Details : Use "setstripe" to set stripe size over 4GB will fail the kernel,
2263 complaining "ASSERTION(lsm->lsm_xfersize != 0)"
2266 Frequency : always on ppc64
2268 Description: the first write on an ext3 filesystem with mballoc got stuck
2269 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2270 perform endianness conversion.
2272 ------------------------------------------------------------------------------
2274 02-14-2006 Cluster File Systems, Inc. <info@clusterfs.com>
2276 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2277 INTEROPERATE with older versions automatically. Please read the
2278 user documentation before upgrading any part of a live system.
2279 * WARNING: Lustre networking configuration changes are required with
2280 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2283 * Support for kernels:
2284 2.6.9-22.0.2.EL (RHEL 4)
2285 2.6.5-7.244 (SLES 9)
2286 2.6.12.6 vanilla (kernel.org)
2289 Severity : enhancement
2290 Bugzilla : 7981/8208
2291 Description: Introduced Lustre Networking (LNET)
2292 Details : LNET is new networking infrastructure for Lustre, it includes
2293 a reorganized network configuration mode (see the user
2294 documentation for full details) as well as support for routing
2295 between different network fabrics. Lustre Networking Devices
2296 (LNDs) for the supported network fabrics have also been
2297 created for this new infrastructure.
2299 Severity : enhancement
2300 Description: Introduced Access control lists
2301 Details : clients can set ACLs on files and directories in order to have
2302 more fine-grained permissions than the standard Unix UGO+RWX.
2303 The MDS must be started with the "-o acl" mount option.
2305 Severity : enhancement
2306 Description: Introduced filesystem quotas
2307 Details : Administrators may now establish per-user quotas on the
2310 Severity : enhancement
2312 Description: Configuration change for the XT3
2313 The PTLLND is now used to run Lustre over Portals on the XT3
2314 The configure option(s) --with-cray-portals are no longer used.
2315 Rather --with-portals=<path-to-portals-includes> is used to
2316 enable building on the XT3. In addition to enable XT3 specific
2317 features the option --enable-cray-xt3 must be used.
2322 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2323 Details : A race between allocating a new llcd and re-getting the llcd_lock
2324 allowed another thread to grab newly-allocated llcd.
2326 Severity : enhancement
2328 Description: 2.6 OST async journal commit and locking fix to improve performance
2329 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2330 2.6 kernels are now async as they already were in 2.4 kernels so
2331 that they can commit concurrently with the network bulk transfer.
2332 For block-allocated files the filter allocation semaphore is held
2333 to avoid filesystem fragmentation during allocation. BKL lock
2334 removed for 2.6 xattr operations where it is no longer needed.
2339 Description: lconf incorrectly determined whether two IP networks could talk
2340 Details : In some more complicated routing and multiple-network
2341 configurations, lconf will avoid trying to make a network
2342 connection to a disjoint part of the IP space. It was doing the
2343 math incorrectly for one set of cases.
2348 Description: Fix for potential infinite loop processing records in an llog.
2349 Details : If an llog record is corrupted/zeroed, it is possible to loop
2350 forever in llog_process(). Validate the llog record length
2351 and skip the remainder of the block on error.
2354 Frequency : occasional (liblustre only)
2356 Description: liblustre could not open files whose last component is a symlink
2357 Details : sysio_path_walk() would incorrectly pass the open intent to
2358 intermediate path components.
2361 Frequency : rare (liblustre only with non-standard tuning)
2362 Bugzilla : 7201 (7350)
2363 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2364 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2365 after it has been released, only do this for referenced locks
2367 Severity : enhancement
2369 Description: specifying an (invalid) directory default stripe_size of -1
2370 would reset the directory default striping
2371 Details : stripe_size -1 was used internally to signal directory stripe
2372 removal, now use "all default" to signal dir stripe removal
2373 as a directory striping of "all default" is not useful
2376 Frequency : common for large clusters running liblustre clients
2378 Description: doing an ls when liblustre clients are running is slow
2379 Details : sending a glimpse AST to a liblustre client waits for every AST
2380 to time out, as liblustre clients will not respond. Since they
2381 cannot cache data we refresh the OST lock LVB from disk instead.
2383 Severity : enhancement
2385 Description: doing an ls at the same time as file IO can be slow
2386 Details : enqueue and other "small" requests can be blocked behind many
2387 large IO requests. Create a new OST IO portal for non-IO
2388 requests so they can be processed faster.
2391 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2393 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2394 created file with mode 000
2395 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2396 allow an ftruncate() as a non-root user to a file with mode 000.
2397 Lustre now respects this flag to disable mode checks when
2398 truncating a file owned by the user
2401 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2404 Description: Revoking locks from clients that went dead or catatonic might take
2406 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2407 cancellation of such locks instant on servers without waiting for
2408 any reply from clients. Clients drops these locks when cancel
2409 notification from server is received without replying.
2412 Frequency : liblustre-only, when liblustre client dies or becomes busy
2414 Description: Doing ls on Linux clients can take a long time with active
2416 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2417 granting such locks to it in the first place if possible. Locks
2418 are taken by proxy on the OST during the read or write and
2419 dropped immediately afterward. Add connect flags handling, do
2420 not grant locks to liblustre clients for glimpse ASTs.
2422 Severity : enhancement
2424 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2425 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2426 sequential reads and behaves badly with applications doing random
2427 reads. Improve it by reading ahead at least read region, and
2428 avoiding excessive large RPC for small reads.
2430 Severity : enhancement
2432 Description: Creating more than 1000 files for a single job may cause a load
2433 imbalance on the OSTs if there are also a large number of OSTs.
2434 Details : qos_prep_create() uses an OST index reseed value that is an
2435 even multiple of the number of available OSTs so that if the
2436 reseed happens in the middle of the object allocation it will
2437 still utilize the OSTs as uniformly as possible.
2442 Description: OST or MDS may oops in ping_evictor_main()
2443 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2444 but doesn't restart at beginning of obd_exports_timed list
2447 Severity : enhancement
2449 Description: improve by-nid export eviction on the MDS and OST
2450 Details : allow multiple exports with the same NID to be evicted at one
2451 time without re-searching the exports list.
2454 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2456 Description: MDS may oops in groups_free()
2457 Details : in rare race conditions a newly allocated group_info struct is
2458 freed again, and this can be NULL. The 2.4 compatibility code
2459 for groups_free() checked for a NULL pointer, but 2.6 did not.
2462 Frequency : common for liblustre clients doing little filesystem IO
2463 Bugzilla : 9352, 7313
2464 Description: server may evict liblustre clients accessing contended locks
2465 Details : if a client is granted a lock or receives a completion AST
2466 with a blocking AST already set it would not reply to the AST
2467 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2470 Frequency : lfs setstripe, only systems with more than 160 OSTs
2472 Description: unable to set striping with a starting offset beyond OST 160
2473 Details : llapi_create_file() incorrectly limited the starting stripe
2474 index to the maximum single-file stripe count.
2477 Frequency : LDAP users only
2479 Description: lconf did not handle in-kernel recovery with LDAP properly
2480 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2482 Severity : enhancement
2484 Description: bind OST threads to NUMA nodes to improve performance
2485 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2486 node and do their allocations there to localize memory access
2488 Severity : enhancement
2490 Description: llmount can determine client NID directly from Myrinet (GM)
2491 Details : the client NID code from gmnalnid was moved directly into
2492 llmount, removing the need to use this or specifying the
2493 client NID explicitly when mounting GM clients with zeroconf
2496 Frequency : if client is started with down MDS
2498 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2499 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2500 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2501 ptlrpc_import_delayed_req() because we didn't check if the
2502 request was interrupted, and we also didn't break out of the
2503 event loop if there was no timeout
2508 Description: data loss during non-page-aligned writes to a single file from
2509 both multiple nodes and multiple threads on one node at same time
2510 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2511 inconsistency led to false short-reads, that were cached and later
2512 used by ->prepare_write() to fill in partially written page,
2513 leading to data loss.
2516 Frequency : always, if lconf --abort_recovery used
2518 Description: lconf --abort_recovery fails with 'Operation not supported'
2519 Details : lconf was attempting to abort recovery on the MDT device and not
2522 Severity : enhancement
2524 Description: remove cleanup logs
2525 Details : replace lconf-generated cleanup logs with lustre internal
2526 cleanup routines. Eliminates the need for client-cleanup and
2529 Severity : enhancement
2531 Description: add support for EAs (user and system) on lustre filesystems
2532 Details : it is now possible to store extended attributes in the Lustre
2533 client filesystem, and with the user_xattr mount option it
2534 is possible to allow users to store EAs on their files also
2536 Severity : enhancement
2538 Description: Add possibility (config option) to show minimal available OST free
2540 Details : When compiled with --enable-mindf configure option, statfs(2)
2541 (and so, df) will return least minimal free space available from
2542 all OSTs as amount of free space on FS, instead of summary of
2543 free spaces of all OSTs.
2545 Severity : enhancement
2547 Description: do not expand extent locks acquired on OST-side
2548 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2549 by server: they are not cached anyway.
2552 Frequency : when mmap is used/binaries executed from Lustre
2554 Description: Unmmap pages before throwing them away from read cache.
2555 Details : llap_shrink cache now attempts to unmap pages before discarding
2556 them (if unmapping failed - do not discard). SLES9 kernel has
2557 extra checks that trigger if this unmapping is not done first.
2562 Description: lconf didn't resolve symlinks before checking to see whether a
2563 given mountpoint was already in use
2566 Frequency : when migrating failover services
2567 Bugzilla : 6395, 9514
2568 Description: When migrating a subset of services from a node (e.g. failback
2569 from a failover service node) the remaining services would
2570 time out and evict clients.
2571 Details : lconf --force (implied by --failover) sets the global obd_timeout
2572 to 5 seconds in order to quickly disconnect, but this caused
2573 other RPCs to time out too quickly. Do not change the global
2574 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2576 Severity : enhancement
2577 Frequency : if MDS is started with down OST
2578 Bugzilla : 9439,5706
2579 Description: Allow startup/shutdown of an MDS without depending on the
2580 availability of the OSTs.
2581 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2582 Add appropriate locking and lov-osc refcounts for safe
2583 cleaning. Add osc abort_inflight calls in case the
2584 synchronize never started.
2587 Frequency : occasional (Cray XT3 only)
2589 Description: root not authorized to access files in CRAY_PORTALS environment
2590 Details : The client process capabilities were not honoured on the MDS in
2591 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2592 been accessed by an authorized user then root was able to access
2593 the file on the local client also. The root user capabilities
2594 are now allowed on the MDS, as this environment has secure UID.
2597 Frequency : occasional
2599 Description: ldiskfs "too long searching" message happens too often
2600 Details : A debugging message (otherwise harmless) prints too often on
2601 the OST console. This has been reduced to only happen when
2602 there are fragmentation problems on the filesystem.
2607 Description: Division by zero in statfs when all OSCs are inactive
2608 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2609 lov_statfs divides by value returned by lov_get_stripecnt().
2613 Bugzilla : 9489, 3273
2614 Description: First write from each client to each OST was only 4kB in size,
2615 to initialize client writeback cache, which caused sub-optimal
2616 RPCs and poor layout on disk for the first writen file.
2617 Details : Clients now request an initial cache grant at (re)connect time
2618 and so that they can start streaming writes to the cache right
2619 away and always do full-sized RPCs if there is enough data.
2620 If the OST is rebooted the client also re-establishes its grant
2621 so that client cached writes will be honoured under the grant.
2626 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2627 Details : Now I/O RPCs go to different portal number and (presumably) fast
2628 lock requests (and glimses) and other RPCs get their own service
2629 threads pool that should be able to service those RPCs
2632 Severity : enhancement
2634 Description: Ability to exchange lustre version between client and servers and
2635 issue warnings at client side if client is too old. Also for
2636 liblustre clients there is ability to refuse connection of too old
2638 Details : New 'version' field is added to connect data structure that is
2639 filled with version info. That info is later checked by server and
2643 Frequency : rare, liblustre only.
2644 Bugzilla : 9296, 9581
2645 Description: Two simultaneous writes from liblustre at offset within same page
2646 might proceed at the same time overwriting eachother with stale
2648 Details : I/O lock withing llu_file_prwv was released too early, before data
2649 actually was hitting the wire. Extended lock-holding time until
2650 server acknowledges receiving data.
2653 Frequency : extremely rare. Never observed in practice.
2655 Description: avoid generating lustre_handle cookie of 0.
2656 Details : class_handle_hash() generates handle cookies by incrementing
2657 global counter, and can hit 0 occasionaly (this is unlikely, but
2658 not impossible, because initial value of cookie counter is
2659 selected randonly). Value of 0 is used as a sentinel meaning
2660 "unassigned handle" --- avoid it. Also coalesce two critical
2661 sections in this function into one.
2663 Severity : enhancement
2665 Description: allow liblustre clients to delegate truncate locking to OST
2666 Details : To avoid overhead of locking, liblustre client instructs OST to
2667 take extent lock in ost_punch() on client's behalf. New connection
2668 flag is added to handle backward compatibility.
2670 Severity : enhancement
2671 Bugzilla : 4928, 7341, 9758
2672 Description: allow number of OST service threads to be specified
2673 Details : a module parameter allows the number of OST service threads
2674 to be specified via "options ost ost_num_threads={N}" in the
2675 OSS's /etc/modules.conf or /etc/modprobe.conf.
2679 Bugzilla : 6146, 9635, 9895
2680 Description: servers crash with bad pointer in target_handle_connect()
2681 Details : In rare cases when a client is reconnecting it was possible that
2682 the connection request was the last reference for that export.
2683 We would temporarily drop the export reference and get a new
2684 one, but this may have been the last reference and the export
2685 was just destroyed. Get new reference before dropping old one.
2687 Severity : enhancement
2688 Frequency : if client is started with failover MDS
2690 Description: Allow multiple MDS hostnames in the mount command
2691 Details : Try to read the configuration from all specified MDS
2692 hostnames during a client mount in case the "primary"
2695 Severity : enhancement
2697 Description: Stop sending data to evicted clients as soon as possible.
2698 Details : Check if the client we are about to send or are sending data to
2699 was evicted already. (Check is done every second of waiting,
2700 for which l_wait_event interface was extended to allow checking
2701 of exit condition at specified intervals).
2704 Frequency : rare, normally only when NFS exporting is done from client
2706 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2708 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2711 Frequency : always, if srand() is called before liblustre initialization
2713 Description: Liblustre uses system PRNG disturbing its usage by user application
2714 Details : Introduce internal to lustre fast and high-quality PRNG for
2715 lustre usage and make liblustre and some other places in generic
2716 lustre code to use it.
2718 Severity : enhancement
2719 Bugzilla : 9477, 9557, 9870
2720 Description: Verify that the MDS configuration logs are updated when xml is
2721 Details : Check if the .xml configuration logs are newer than the config
2722 logs stored on the MDS and report an error if this is the case.
2723 Request --write-conf, or allow starting with --old_conf.
2725 Severity : enhancement
2727 Description: Handle symlinks in the path when checking if Lustre is mounted.
2728 Details : Resolve intermediate symlinks when checking if a client has
2729 mounted a filesystem to avoid duplicate client mounts.
2734 Description: lconf can hit an error exception but still return success.
2735 Details : The lconf command catches the Command error exception at the top
2736 level script context and will exit with the associated exit
2737 status, but doesn't ensure that this exit status is non-zero.
2742 Description: failure of ptlrpc thread startup can cause oops
2743 Details : Starting a ptlrpc service thread can fail if there are a large
2744 number of threads or the server memory is very fragmented.
2745 Handle this without oopsing.
2748 Frequency : always, only if liblustre and non-default acceptor port was used
2750 Description: liblustre cannot connect to servers with non-default acceptor port
2751 Details : tcpnal_set_default_params() was not called and was therefore
2752 ignoring the environment varaible TCPNAL_PORT, as well as other
2753 TCPNAL_ environment variables
2758 Description: two objects could be created on the same OST for a single file
2759 Details : If an OST is down, in some cases it was possible to create two
2760 objects on a single OST for a single file. No problems other
2761 than potential performance impact and spurious error messages.
2765 Bugzilla : 5681, 9562
2766 Description: Client may oops in ll_unhash_aliases
2767 Details : Client dcache may become inconsistent in race condition.
2768 In some cases "getcwd" can fail if the current directory is
2774 Description: Inode refcounting problems in NFS export code
2775 Details : link_raw functions used to call d_instantiate without obtaining
2776 extra inode reference first.
2780 Bugzilla : 9942, 9903
2781 Description: Referencing freed requests leading to crash, memleaks with NFS.
2782 Details : We used to require that call to ll_revalidate_it was always
2783 followed by ll_lookup_it. Also with revalidate_special() it is
2784 possible to call ll_revalidate_it() twice for the same dentry
2785 even if first occurence returned success. This fix changes semantic
2786 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2787 reference on a request referred from the intent.
2788 ll_intent_release() then releases such a request.
2791 Frequency : rare, normally benchmark loads only
2793 Description: unlinked inodes were kept in memory on the client
2794 Details : If a client is repeatedly creating and unlinking files it
2795 can accumulate a lot of stale inodes in the inode slab cache.
2796 If there is no other client load running this can cause the
2797 client node to run out of memory. Instead flush old inodes
2798 from client cache that have the same inode number as a new inode.
2801 Frequency : SLES9 2.6.5 kernel and long filenames only
2802 Bugzilla : 9969, 10379
2803 Description: utime reports stale NFS file handle
2804 Details : SLES9 uses out-of-dentry names in some cases, which confused
2805 the lustre dentry revalidation. Change it to always use the
2809 Frequency : rare, unless heavy write-truncate concurrency is continuous
2810 Bugzilla : 4180, 6984, 7171, 9963, 9331
2811 Description: OST becomes very slow and/or deadlocked during object unlink
2812 Details : filter_destroy() was holding onto the parent directory lock
2813 while truncating+unlinking objects. For very large objects this
2814 may block other threads for a long time and slow overall OST
2815 responsiveness. It may also be possible to get a lock ordering
2816 deadlock in this case, or run out of journal credits because of
2817 the combined truncate+unlink. Solution is to do object truncate
2818 first in one transaction without parent lock, and then do the
2819 final unlink in a new transaction with the parent lock. This
2820 reduces the lock hold time dramatically.
2823 Frequency : rare, 2.4 kernels only
2825 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2826 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2827 before unmounting their filesystem in order to avoid deadlock.
2828 The kernel_locked() function in 2.4 kernels only checks whether
2829 the kernel lock is held, not whether it is this process that is
2830 holding it as 2.6 kernels do.
2835 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2836 Details : The client ptlrpc code may be trying to reconnect to a down
2837 server before a previous connection attempt has timed out.
2838 Increase the reconnect interval to be longer than the connection
2839 timeout interval to avoid sending duplicate connections to
2843 Frequency : echo_client brw_test command
2845 Description: fix echo_client to work with OST preallocated code
2846 Details : OST preallocation code (5137) didn't take echo_client IO path
2847 into account: echo_client calls filter methods outside of any
2848 OST thread and, hence, there is no per-thread preallocated
2849 pages and buffers to use. Solution: hijack pga pages for IO. As
2850 a byproduct, this avoids unnecessary data copying.
2854 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2855 Description: Client can oops in mdc_commit_close() after open replay
2856 Details : It was possible for the MDS to return an open request with no
2857 transaction number in mds_finish_transno() if the client was
2858 evicted, but without actually returning an error. Clients
2859 would later try to replay that open and may trip an assertion
2860 Simplify the client close codepath, and always return an error
2861 from the MDS in case the open is not successful.
2864 Frequency : rare, 2.6 OSTs only
2866 Description: OST may deadlock under high load on fragmented files
2867 Details : If there was a heavy load and highly-fragmented OST filesystems
2868 it was possible to have all the OST threads deadlock waiting on
2869 allocation of biovecs, because the biovecs were not released
2870 until the entire RPC IO was completed. Instead, release biovecs
2871 as soon as they are complete to ensure forward IO progress.
2873 Severity : enhancement
2875 Description: Support for specifying external journal device at mount
2876 Details : If an OST or MDS device is formatted with an external journal
2877 device, this device major/minor is stored in the ext3 superblock
2878 and may not be valid for failover. Allow detecting and
2879 specifying the external journal at mount time.
2884 Description: Mounting an MDS with pending unlinked files may cause oops
2885 Details : target_finish_recovery() calls mds_postrecov() which returned
2886 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2887 considers this an error and immediately begins cleaning up the
2888 lov, just after starting the mds_lov process
2890 Severity : enhancement
2892 Description: Implement 'lfs df' to report actual free space on per-OST basis
2893 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2894 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2895 usage of objects; '-h' to report in human readable format.
2897 ------------------------------------------------------------------------------
2899 08-26-2005 Cluster File Systems, Inc. <info@clusterfs.com>
2906 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2907 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2908 at startup during orphan unlinking. Instead, skip buddy update
2909 before setup, as it will be regenerated after recovery is complete.
2914 Description: If an OST is inactive, its locks might reference stale inodes.
2915 Details : lov_change_cbdata() must iterate over all namespaces, even if
2916 they are inactive to clear inode references from the lock.
2918 Severity : enhancement
2919 Frequency : occasional, if non-standard max_dirty_mb used
2921 Description: Client will block write RPCs if not enough grant
2922 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2923 then the client will block writes while waiting for another RPC
2924 to complete instead of consuming its dirty limit. With change
2925 we get improved performance when max_dirty_mb is small.
2927 Severity : enhancement
2928 Bugzilla : 3389, 6253
2929 Description: Add support for supplementary groups on the MDS.
2930 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2931 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2932 lookups for user supplementary groups into a cache.
2936 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2937 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2938 for liblustre O_EXCL is enforced by MDS.
2942 Description: readdir never returns NULL in liblustre.
2943 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2944 dentry in struct dirent.
2948 Frequency : liblustre only, depends on application IO pattern
2949 Description: liblustre clients evicted if not contacting servers
2950 Details : Don't put liblustre clients into the ping_evictor list, so
2951 they will not be evicted by the pinger ever.
2953 Severity : enhancement
2955 Description: Add ability to evict clients by NID from MDS.
2956 Details : By echoing "nid:$NID" string into
2957 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2958 $NID would be instantly evicted from this MDS and from all active
2959 OSTs connected to it.
2963 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2964 Details : lookup_it_finish() used to query file size from OSTs that was not
2969 Description: service threads change working directory to that of init
2970 Details : Starting lustre service threads may pin the working directory
2971 of the parent thread, making that filesystem busy. Threads
2972 now change to the working directory of init to avoid this.
2976 Frequency : during shutdown only
2977 Description: shutdown with a failed MDS or OST can cause unmount to hang
2978 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2983 Frequency : 2.6 only
2984 Description: chmod/chown may include an extra supplementary group
2985 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2986 supplementary group and if none is specified this is used.
2989 Bugzilla : 5479 (6816)
2991 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2992 Details : If lookup is in progress on a file that is unlinked we might try
2993 to revalidate the inode and fail in revalidate after lookup is
2994 complete and ll_file_open() enqueues the open again but
2995 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2998 Frequency : always, if lconf --abort_recovery used
3000 Description: lconf --abort_recovery fails with 'Operation not supported'
3001 Details : lconf was attempting to abort recovery on the MDT device and not
3004 ------------------------------------------------------------------------------
3006 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3011 Frequency : rare (only unsupported configurations with a node running as an
3013 Bugzilla : 6514, 5137
3014 Description: Mounting a Lustre file system on a node running as an OST could
3016 Details : OSTs now preallocates memory needed to write out data at
3017 startup, instead of when needed, to avoid having to
3018 allocate memory in possibly low memory situations.
3019 Specifically, if the file system is mounted on on OST,
3020 memory pressure could force it to try to write out data,
3021 which it needed to allocate memory to do. Due to the low
3022 memory, it would be unable to do so and the node would
3023 become unresponsive.
3025 Severity : enhancement
3027 Description: Addition of lconf --service command line option
3028 Details : lconf now accepts a '--service <arg>' option, which is
3029 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3031 Severity : enhancement
3033 Description: Failover mode is now the default for OSTs.
3034 Details : By default, OSTs will now run in failover mode. To return to
3035 the old behaviour, add '--failout' to the lmc line for OSTs.
3037 Severity : enhancement
3039 Description: Health checks are now provided for MDS and OSTs
3040 Details : Additional detailed health check information on MSD and OSTs
3041 is now provided through the procfs health_check value.
3044 Frequency : occasional, depends on IO load
3046 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3047 after numerous create/delete cycles
3048 Details : The ext3 inode allocation policy would not allocate new inodes
3049 very well on the OSTs because there are no new directories
3050 being created. Instead we look for groups with free space if
3051 the parent directories are nearly full.
3056 Description: Network or server problems during mount may cause partially
3057 mounted clients instead of returning an error.
3058 Details : The config llog parsing code may overwrite the error return
3059 code during mount error handling, returning success instead
3065 Description: MDS can fail to allocate large reply buffers
3066 Details : After long uptimes the MDS can fail to allocate large reply
3067 buffers (e.g. zconf client mount config records) due to memory
3068 fragmentation or consumption by the buffer cache. Preallocate
3069 some large reply buffers so that these replies can be sent even
3070 under memory pressure.
3074 Frequency : rare (liblustre)
3075 Description: fsx running with liblustre complained that using truncate() to
3076 extend the file doesn't work. This patch corrects that issue.
3077 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3078 ATTR_SIZE and lsm use in llu_setattr_raw.
3082 Frequency : rare, only 2.6 kernels
3083 Description: Unusual file access patterns on the MDS may result in inode
3084 data being lost in very rare circumstances.
3085 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3086 code in the 2.6 kernel caused the inode and/or EA data not to
3087 be read from disk, causing single-file corruption.
3091 Frequency : rare, only 2.6 filesystems using extents
3092 Description: Heavy concurrent write and delete load may cause data corruption.
3093 Details : It was possible under high-load situations to have an extent
3094 metadata block in the block device cache from a just-unlinked
3095 file overwrite a newly-allocated data block. We now unmap any
3096 metadata buffers that alias just-allocated data blocks.
3100 Frequency : filesystems with default stripe_count larger than 77
3101 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3102 Details : lconf specifies an inode size of 4096 bytes when the default
3103 stripe_count is larger than 77. This conflicts with the default
3104 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3105 this case to avoid pinning too much memory for large EAs.
3107 ------------------------------------------------------------------------------
3109 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3114 Frequency : rare (extremely heavy IO load with hundreds of clients)
3116 Description: Client is evicted, gets IO error writing to file
3117 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3118 caused clients to be evicted for AST timeouts. The fixes in
3119 bug 5192 mean we no longer need to have such short AST timeouts
3120 so ldlm_timeout has been increased.
3123 Frequency : occasional during --force or --failover shutdown under load
3124 Bugzilla : 5949, 4834
3125 Description: Server oops/LBUG if stopped with --force or --failover under load
3126 Details : a collection of import/export refcount and cleanup ordering
3127 issues fixed for safer force cleanup
3130 Frequency : only filesystems larger than 120 OSTs
3131 Bugzilla : 5990, 6223
3132 Description: lfs getstripe would oops on a very large filesystem
3133 Details : lov_getconfig used kfree on vmalloc'd memory
3136 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3137 Bugzilla : 6242, 6243
3138 Description: reading from files that had been truncated to a non-zero size
3139 but never opened returned no data
3140 Details : ll_file_read() reads zeros from no-object files to EOF
3145 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3146 Details : There's little harm in aborting MDS/OSS recovery and letting it
3147 try again, so I removed the LASSERT and return an error instead.
3149 Severity : enhancement
3151 Description: New debugging infrastructure for tracking down data corruption
3152 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3153 (b) cover more of the client-side code path, and (c) try to narrow
3154 down where problems occurred
3158 Bugzilla : 3819, 4364, 4397, 6313
3159 Description: Racing close and eviction MDS could cause assertion in mds_close
3160 Details : It was possible to get multiple mfd references during close and
3161 client eviction, leading to one thread referencing a freed mfd.
3163 Severity: : enhancement
3164 Bugzilla : 3262, 6359
3165 Description: Attempts to reconnect to servers are now more aggressive.
3166 Details : This builds on the enhanced upcall-less recovery that was added
3167 in 1.4.2. When trying to reconnect to servers, clients will
3168 now try each server in the failover group every 10 seconds. By
3169 default, clients would previously try one server every 25 seconds.
3174 Description: After recovery, certain operations trigger a failed
3175 assertion on a client.
3176 Details : Failing over an mds, using lconf -d --failover, while a
3177 client was doing a readdir() call would cause the client to
3178 LBUG after recovery completed and the readdir() was resent.
3180 Severity : enhancement
3182 Description: Default groups are now added by lconf
3183 Details : You can now run lconf --group <servicename> without having to
3184 manually add groups with lmc.
3187 Frequency : occasional
3189 Description: Nodes with an elan id of 0 trigger a failed assertion
3192 Frequency : always when accessing e.g. tty/console device nodes
3194 Description: tty and some other devices nodes cannot be used on lustre
3195 Details : file's private_data field is used by device data and lustre
3196 values in there got lost. New field was added to struct file to
3197 store fs-specific private data.
3200 Frequency : when exporting Lustre via NFS
3202 Description: NFSD failed occasionally when looking up a path component
3203 Details : NFSD is looking up ".." which was broken in ext3 directories
3204 that had grown large enough to become hashed.
3207 Frequency : Clusters with multiple interfaces not on the same subnet
3209 Description: Nodes will repeatedly try to reconnect to an interface which it
3210 cannot reach and report an error to the log.
3211 Details : Extra peer list entries will be created by lconf with some peers
3212 unreachable. lconf now validates the peer before adding it.
3215 Frequency : Only if a default stripe is set on the filesystem root.
3217 Description: Setting a default stripe on the filesystem root prevented the
3218 filesystem from being remounted.
3219 Details : The client was sending extra request flags in the root getattr
3220 request and did not allocate a reply buffer for the dir EA.
3223 Frequency : occasional, higher if lots of files are accessed by one client
3224 Bugzilla : 6159, 6097
3225 Description: Client trips assertion regarding lsm mismatch/magic
3226 Details : While revalidating inodes the VFS looks up inodes with ifind()
3227 and in rare cases can find an inode that is being freed.
3228 The ll_test_inode() code will free the lsm during ifind()
3229 when it finds an existing inode and then the VFS later attaches
3230 this free lsm to a new inode.
3234 Bugzilla : 6422, 7030
3235 Description: MDS deadlock between mkdir and client eviction
3236 Details : Creating a new file via mkdir or mknod (starting a transaction
3237 and getting the ns lock) can deadlock with client eviction
3238 (gets ns lock and trying to finish a synchronous transaction).
3241 Frequency : occasional
3242 Description: While starting a server, the fsfilt_ext3 module could not be
3244 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3245 kernels. Previously, lconf would still use the ext3 name
3246 when trying to load modules. Now, it will correctly use
3247 ext3 on 2.4 and ldiskfs on 2.6.
3249 Severity : enhancement
3250 Description: The default stripe count has been changed to 1
3251 Details : The interpretation of the default stripe count (0, to lfs
3252 or lmc) has been changed to mean striping across a single
3253 OST, rather than all available. For general usage we have
3254 found a stripe count of 1 or 2 works best.
3256 Severity : enhancement
3257 Description: Add support for compiling against Cray portals.
3258 Details : Conditional compiling for some areas that are different
3262 Frequency : occasional
3263 Bugzilla : 6409, 6834
3264 Description: Creating files with an explicit stripe count may lead to
3265 a failed assertion on the MDS
3266 Details : If some OSTs are full or unavailable, creating files may
3267 trigger a failed assertion on the MDS. Now, Lustre will
3268 try to use other servers or return an error to the
3272 Frequency : occasional
3274 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3275 caused lock timeout during readahead (since 1.4.2).
3276 Details : Processes doing readahead might match a lock that hasn't been
3277 granted yet if there are overlapping and conflicting lock
3278 requests. The readahead process waits on ungranted lock
3279 (original lock is CBPENDING), while OST waits for that process
3280 to cancel CBPENDING read lock and eventually evicts client.
3282 Severity : enhancement
3284 Description: Initial enabling of flock support for clients
3285 Details : Implements fcntl advisory locking and file status functions.
3286 This feature is provided as an optional mount flag (default
3287 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3288 locking are implemented yet, and those that are are not guaranteed
3289 to be completely correct in production environments.
3290 mount -t lustre -o [flock|noflock] ...
3293 Frequency : occasional
3295 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3296 assertion in the ext3 JBD (journaling) layer.
3297 Details : The b_committed_data struct is protected by the big kernel lock
3298 in 2.4 kernels, serializing journal_commit_transaction() and
3299 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3300 access to this struct. In 2.6 kernels there is finer grained
3301 locking to improve SMP performance of the JBD layer.
3305 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3310 Description: merge and cleanup kernel patches.
3313 -----------------------------------------------------------------------------
3315 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3317 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3318 versions of Lustre. Please update all servers and clients to
3319 version 1.4.2 or later at the same time. You must also run
3320 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3321 to update the configuration logs.
3323 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3324 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3325 - swab llog records of type '0' so we get proper header size/idx (5861)
3326 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3327 - fix rename of one directory over another leaking an inode (5953)
3328 - avoid SetPageDirty on 2.6 (5981)
3329 - don't re-add just-being-destroyed locks to the waiting list (5653)
3330 - when creating new directories, inherit the parent's custom
3331 striping settings if present parent (3048)
3332 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3333 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3334 - don't hold client locks on temporary worklist from l_lru (5666)
3335 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3336 - automatically evict dead clients (5921)
3337 - Update file size properly in create+truncate+fstat case (6196)
3338 - Do not unhash mountpoint dentries, do not allow removal of
3340 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3341 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3342 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3343 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3344 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3345 - Correctly handle reads of files with no objects (6243)
3346 - lctl recover will also mark a device active if deactivate used (5933)
3348 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3349 - allow --write-conf on an MDS with different nettype than client (5619)
3350 - don't write config llogs to MDS for mounts not from that MDS (5617)
3351 - lconf should create multiple TCP connections from a client (5201)
3352 - init scripts are now turned off by default; run chkconfig --on
3353 lustre and chkconfig --on lustrefs to use them
3354 - upcalls are no longer needed for clients to recover to failover
3356 - add --abort-recovery option to lconf to abort recovery on device
3358 - add support for an arbitrary number of OSTs (3026)
3359 - Quota support protocol changes.
3360 - forward compatibility changes to wire structs (6007)
3361 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3362 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3363 - improved "lustre status" script
3364 - initialize blocksize for non-regular files (6062)
3365 - added --disable-server and --disable-client configure options (5782)
3366 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3367 - Vanilla 2.4.29 support
3368 - increase maximum number of obd devices to 520 (6242)
3369 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3370 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3371 SLES 9 2.6.5 kernels
3372 - limit stripes per file to 160 (the maximum EA size) (6093)
3374 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3377 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3378 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3379 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3380 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3381 - small ext3 extents cleanups and fixes (5733)
3382 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3383 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3384 - avoid needless client->OST connect, fix handle mismatch (5317)
3385 - fix DLM error path that led to out-of-sync client, long delays (5779)
3386 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3387 - fix several locking issues related to i_size (5492,5624,5654,5672)
3388 - don't move pending lock onto export if it is already evicted (5683)
3389 - fix kernel oops when creating .foo in unlinked directory (5548)
3390 - fix deadlock in obdfilter statistics vs. object create (5811)
3391 - use time_{before,after} to avoid timer jiffies wrap (5882)
3392 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3393 - Do not leak request if server was not able to process it (5154)
3394 - If mds_open unable to find parent dir, make that negative lookup(5154)
3395 - don't create new directories with extent-mapping (5909, 5936)
3397 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3398 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3399 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3400 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3401 - support for RHEL4 kernel on i686 (5773)
3402 - provide error messages when incompatible logs are encountered (5898)
3404 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3405 * version 1.4.0.10 (1.4.1 release candidate 1)
3407 - don't keep a lock reference when lock is not granted (4238)
3408 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3409 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3410 - drop import inflight refcount on signal_completed_replay error (5255)
3411 - unlock page after async write error during send (3677)
3412 - handle missing objects in filter_preprw_read properly (5265)
3413 - no transno return for symlink open, don't save no-trasno open (3440)
3414 - don't try to complete elan receive that already failed (4012)
3415 - free RPC server reply state on error (5406)
3416 - clean up thread from ptlrpc_start_thread() on error (5160)
3417 - readahead could read extra page into cache that wasn't ejected (5388)
3418 - prevent races in class_attach/setup/cleanup/detach (5260)
3419 - don't dereference de->d_inode after l_dput of de (5458)
3420 - use "int" for stripe value returned from lock_to_stripe (5544)
3421 - mballoc allocation and error-checking fixes in 2.6 (5504)
3422 - block device patches to fix I/O request sizes in 2.6 (5482)
3423 - look up hostnames for IB nals (5602)
3424 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3425 - don't start multiple acceptors for the same port (5277)
3426 - fix incorrect LASSERT in mds_getattr_name (5635)
3427 - export a proc file for general "ping" checking (5628)
3428 - fix "lfs check" to not block when the MDS is down (5628)
3430 - service request history (4965)
3431 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3432 - create an "evict_client" /proc entry on OSTs, like the MDS has
3433 - fix mount usage message, return errors per mount(8) (5168)
3434 - change grep [] to grep "[]" in tests so they work in more UMLs
3435 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3436 - remove ancient LOV_MAGIC_V0 EA support (5047)
3437 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3438 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3439 - allow readahead allocations to fail when low on memory (5383)
3440 - mmap locking landed again, after considerable improvement (2828)
3441 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3443 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3446 - send OST transaction number in read/write reply to free req (4966)
3447 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3448 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3449 - the watchdog thread now runs as interruptible (5246)
3450 - flock/lockf fixes (but it's still disabled, pending 5135)
3451 - don't use EXT3 constants in llite code (5094)
3452 - memory shortage at startup could cause assertion (5176)
3454 - reorganization of lov code
3455 - single portals codebase
3457 - add extents/mballoc support (5025)
3458 - direct I/O reads in the obdfilter (4048)
3459 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3461 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3464 - send OST transaction number in read/write reply to free req (4966)
3465 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3466 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3467 - the watchdog thread now runs as interruptible (5246)
3468 - handle missing objects in filter_preprw_read properly (5265)
3469 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3470 - drop import inflight refcount on signal_completed_replay error (5255)
3471 - unlock page after async write error during send (3677)
3472 - return original error code on reconstructed replies (3761)
3473 - no transno return for symlink open, don't save no-trasno open (3440)
3475 - add pid to ldlm debugging output (4922)
3476 - bump the watchdog timeouts -- we can't handle 30sec yet
3477 - extra debugging for orphan dentry/inode bug (5259)
3479 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3482 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3483 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3484 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3485 - fix NULL dereference in /proc/sys/portals/routes (4827)
3486 - allow failed mdc_close() operations to be interrupted (4561)
3487 - stop precreate on OST before MDS would time out on it (4778)
3488 - don't send partial-page writes before EOF from client (4410)
3489 - discard client grant for sub-page writes on large-page clients (4520)
3490 - don't free dentries not owned by NFS code, check generation (4806)
3491 - fix lsm leak if mds_create_objects() fails (4801)
3492 - limit debug_daemon file size, always print CERROR messages (4789)
3493 - use transno after validating reply (3892)
3494 - process timed out requests if import state changes (3754)
3495 - update mtime on OST during writes, return in glimpse (4829)
3496 - add mkfsoptions to LDAP (4679)
3497 - use ->max_readahead method instead of zapping global ra (5039)
3498 - don't interrupt __l_wait_event() during strace
3500 - add software watchdogs to catch hung threads quickly (4941)
3501 - make lustrefs init script start after nfs is mounted
3502 - fix CWARN/ERROR duplication (4930)
3503 - return async write errors to application if possible (2248)
3504 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3505 - print NAL number in %x format (4645)
3506 - update barely-supported suse-2.4.21-171 series (4842)
3507 - support for sles 9 %post scripts
3508 - support for building 2.6 kernel-source packages
3509 - support for sles km_* packages
3511 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3514 - ignore -ENOENT errors in osc_destroy (3639)
3515 - notify osc create thread that OSC is being cleaned up (4600)
3516 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3517 - reconstruct ost_handle() like mds_handle() (4657)
3518 - create a new thread to do import eviction to avoid deadlock (3969)
3519 - let lconf resolve symlinked-to devices (4629)
3520 - don't unlink "objects" from directory with default EA (4554)
3521 - hold socknal file ref over connect in case target is down (4394)
3522 - allow more than 32000 subdirectories in a single directory (3244)
3523 - fix blocks count for O_DIRECT writes (3751)
3524 - OST returns ENOSPC from object create when no space left (4539)
3525 - don't send truncate RPC if file size isn't changing (4410)
3526 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3527 - bind to privileged port in socknal and tcpnal (3689)
3529 - rate limit CERROR/CWARN console message to avoid overload (4519)
3530 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3531 - basic mmap support (3918)
3532 - kernel patch series update from b1_4 (4711)
3534 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3537 - avoid crash during MDS cleanup with OST shut down (2775)
3538 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3539 - don't use bad inodes on the MDS (3744)
3540 - dynamic object preallocation to improve recovery speed (4236)
3541 - don't hold spinlock over lock dumping or change debug flags (4401)
3542 - don't zero obd_dev when it is force cleaned (3651)
3543 - print grants to console if they go negative (4431)
3544 - "lctl deactivate" will stop automatic recovery attempts (3406)
3545 - look for existing locks in ldlm_handle_enqueue() (3764)
3546 - don't resolve lock handle twice in recovery avoiding race (4401)
3547 - revalidate should check working dir is a directory (4134)
3549 - don't always mark "slow" obdfilter messages as errors (4418)
3551 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3554 - don't close LustreDB during write_conf until it is done (3860)
3555 - fix typo in lconf for_each_profile (3821)
3556 - allow dumping logs from multiple threads at one time (3820)
3557 - don't allow multiple threads in OSC recovery (3812)
3558 - fix debug_size parameters (3864)
3559 - fix mds_postrecov to initialize import for llog ctxt (3121)
3560 - replace config semaphore with spinlock (3306)
3561 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3562 - don't allow enqueue to complete on a destroyed export (3822)
3563 - down write_lock before checking llog header bitmap (3825)
3564 - recover from lock replay timeout (3764)
3565 - up llog sem before sending rpc (3652)
3566 - reduce ns lock hold times when setting kms (3267)
3567 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3568 - fix NULL deref and obd_dev leak on setup error (3312)
3569 - replace some LBUG about llog ops with error handling (3841)
3570 - don't match INVALID dentries from d_lookup and spin (3784)
3571 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3572 - fix invalid assertion in ptlrpc_set_wait (3880)
3574 - add libwrap support for the TCP acceptor (3996)
3575 - add /proc/sys/portals/routes for non-root route listing (3994)
3576 - allow setting MDS UUID in .xml (2580)
3577 - print the stack of a process that LBUGs (4228)
3579 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3582 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3583 - make sure to unset replay flag from failed open requests (3440)
3584 - if default stripe count is 0, use OST count for inode size (3636)
3585 - update parent mtime/ctime on client for create/unlink (2611)
3586 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3587 - free recovery state on server during a forced cleanup (3571)
3588 - unregister_reply for resent reqs (3063)
3589 - loop back devices mounting and status check on 2.6 (3563)
3590 - fix resource-creation race that can provoke i_size == 0 (3513)
3591 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3592 - more debugging for page-accounting assertion (3746)
3593 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3594 - avoid deadlock after precreation failure (3758)
3595 - fix race and lock order deadlock in orphan handling (3450, 3750)
3596 - add validity checks when grabbing inodes from l_ast_data (3599)
3598 - add /proc/.../recovery_status to obdfilter (3428)
3599 - lightweight CDEBUG infrastructure, debug daemon (3668)
3600 - change default OSC RPC parameters to be better on small clusters
3601 - turn off OST read cache for files smaller than 32MB
3602 - install man pages and include them in rpms (3100)
3603 - add new init script for (un)mounting lustre filesystems (2593)
3604 - run chkconfig in %post for init scripts (3701)
3605 - drop scimac NAL (unmaintained)
3607 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3610 - clean kiobufs before and after use (3485)
3611 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3612 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3613 - change init script to fail more clearly if not run as root (1528)
3614 - allow clients to reconnect during replay (1742)
3615 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3616 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3617 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3618 - fix cond_resched (3554)
3619 - only evict unfinished clients after recovery (3515)
3620 - allow bulk resend, prevent data loss (3570)
3621 - dynamic ptlrpc request buffer allocation (2102)
3622 - don't allow unlinking open directory if it isn't empty (2904)
3623 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3624 - remove extraneous obd dereference causing LASSERT failure (3334)
3625 - don't use get_cycles() when creating temp. files on the mds (3156)
3626 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3627 - handle EEXIST for set-stripe, set proper directory name (3336)
3629 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3630 - fix ksocknal_fmb_callback() error messages (2918)
3632 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3635 - don't copy lvb into (possibly NULL) reply on error (2983)
3636 - don't deref dentry after dput, don't free lvb on error (2922)
3637 - use the kms to determine writeback rpc length (2947)
3638 - increment oti_logcookies when osc is inactive (2948)
3639 - update client's i_blocks count via lvb messages (2543)
3640 - handle intent open/close of special files properly (1557)
3641 - mount MDS with errors=remount-ro, like obdfilter (2009)
3642 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3643 - don't use cancelling-locks' kms values (2947)
3644 - use highest lock extent for kms, not last one (2925)
3645 - don't dereference ERR_PTR() dentry in error handling path (3107)
3646 - fix thread race in portals_debug_dumplog() (3122)
3647 - create lprocfs device entries at setup instead of at attach (1519)
3648 - common AST error handler, don't evict client on completion race (3145)
3649 - zero nameidata in detach_mnt in 2.6 (3118)
3650 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3651 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3652 - initialize RPC timeout timer earlier for 2.6 (3219)
3653 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3654 - print nal/nid for unknown nid (3258)
3655 - additional checks for oscc recovery before doing precreate (3284)
3656 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3657 - don't crash in mdc_close for bad permissions on open (3285)
3658 - zero i_rdev for non-device files (3147)
3659 - clear page->private before handing to FS, better assertion (3119)
3660 - tune the read pipeline (3236)
3661 - fix incorrect decref of invalidated dentry (2350)
3662 - provide read-ahead stats and refine rpc in flight stats (3328)
3663 - don't hold journal transaction open across create RPC (3313)
3664 - update atime on MDS at close time (3265)
3665 - close LDAP connection when recovering to avoid server load (3315)
3666 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3667 - don't leak open file on MDS after open resend (3325)
3668 - serialize filter_precreate and filter_destroy_precreated (3329)
3669 - loop device shouldn't call sync_dev() for nul device (3092)
3670 - clear page cache after eviction (2766)
3671 - resynchronize MDS->OST in background (2824)
3672 - refuse to mount the same filesystem twice on same mountpoint (3394)
3673 - allow llmount to create routes for mounting behind routers (3320)
3674 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3675 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3676 - fix rare problem with rename on htree directories (3417)
3678 - allow default OST striping configuration per directory (1414)
3679 - fix compilation for qswnal for 2.6 kernels (3125)
3680 - increase maximum number of MDS request buffers for large systems
3681 - change liblustreapi to be useful for external progs like lfsck (3098)
3682 - increase local configuration timeout for slow disks (3353)
3683 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3685 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3688 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3689 - don't overwrite extent policy data in reply if lock was blocked (2901)
3690 - drop filter export grants atomically with removal from device (2663)
3691 - del obd_self_export from work_list in class_disconnect_exports (2908)
3692 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3693 - swab reply message in mdc_close, other PPC fixes (2464)
3694 - fix destroying of named logs (2325)
3695 - overwrite old logs when running lconf --write_conf (2264)
3696 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3697 - fix race in target_handle_connect (2898)
3698 - mds_reint_create() should take same inode create lock (2926)
3699 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3700 - don't close files for self_export to avoid uninitialized obd (2936)
3701 - allow MDS with the same name as client node (2939)
3702 - hold dentry reference for closed log files for unlink (2325)
3703 - reserve space for all logs during transactions (2059)
3704 - don't evict page beyond end of stripe extent (2925)
3705 - don't oops on a deleted current working directory (2399)
3706 - handle hard links to targets without a parent properly (2517)
3707 - don't dereference NULL lock when racing during eviction (2867)
3708 - don't grow lock extents when lots of conflicting locks (2919)
3710 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3713 - account for cache space usage on clients to avoid data loss (974)
3714 - lfsck support in lustre kernel code (2349)
3715 - reduce journal credits needed for BRW writes (2370)
3716 - orphan handling to avoid losing space on client/server crashes
3717 - ptlrpcd can be blocked, stopping ALL progress (2477)
3718 - use lock value blocks to assist in proper KMS, faster stat (1021)
3719 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3720 - recovery for initial connections (2355)
3721 - fixes for mds_cleanup_orphans (1934)
3722 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3723 - block all file creations until orphan recovery completes (1901)
3724 - client remove rq_connection from request struct (2423)
3725 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3726 - recovery timer race (2670)
3727 - mdc_close recovey bug (2532)
3728 - ptlrpc cleanup bug (2710)
3729 - mds timeout on local locks (2588)
3730 - namespace lock held during RPCs (2431)
3731 - handle interrupted sync write properly (2503)
3732 - don't try to handle a message that hasn't been replied to (2699)
3733 - client assert failure during cleanup after abort recovery (2701)
3734 - leak mdc device after failed mount (2712)
3735 - ptlrpc_check_set allows timedout requests to complete (2714)
3736 - wait for inflight reqs when ptlrpcd finishes (2710)
3737 - make sure unregistered services are removed from the srv_list
3738 - reset bulk XID's when resending them (caught by 1138 test)
3739 - unregister_bulk after timeout
3740 - fix lconf error (2694)
3741 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3742 - readahead locks pages, leaves pending causing memory pressure (2673)
3743 - increase OST request buffers to 4096 on large machines (2729)
3744 - fix up permission of existing directories in simple_mkdir (2661)
3745 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3746 - don't assign transno to errored transactions (2742)
3747 - don't delete objects on OST if given a bogus objid from MDS (2751)
3748 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3749 - if rq_no_resend, then timeout request after recovery (2432)
3750 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3751 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3752 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3753 - MDS recovery shouldn't delete objects that it has given out (2730)
3754 - if enqueue arrives after completion, don't clobber LVB (2819)
3755 - don't unlock pages twice when trigger_group_io returns error (2814)
3756 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3757 - don't write pages to disk if there was an error (1450)
3758 - don't ping imports that have recovery disabled (2676)
3759 - take buffered bytes into account when balancing socknal conn (2817)
3760 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3761 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3762 - remove little-endian swabbing of llog records (1987)
3763 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3764 - retry reposting request buffers if they fail (1191)
3765 - grow extent at grant time to avoid granting a revoked lock (2809)
3766 - lock revoke doesn't evict page if covered by a second lock (2765)
3767 - disable VM readahead to avoid reading outside lock extents (2805)
3769 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3770 - updated kernel patches for hp-2.4.20 kernel (2681)
3772 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3775 - fix truncated write corruption (2366)
3776 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3778 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3779 - ensure len(uuid) < 37 in lmc (1171)
3780 - fix ia64 OOPS in llog_test (2255)
3781 - zero end of page at obdfilter for partial page writes (2648)
3782 - don't leave stale dentries around after renames (bug 2428)
3783 - fix timeouts when evicting a client with a single lock held (2642)
3784 - set deadline for the initial HELLO message to drain (2634)
3785 - print out dotted-quad IP addresses in the socknal (2302)
3787 - additional debugging for MDS client eviction problem (2443)
3788 - fix mkfsoptions support for osts (2603, 2604)
3790 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3793 - add series for the vanilla 2.6.0 kernel
3794 - add series for the vanilla 2.4.24 kernel
3795 - add series for a cray x86/64 UL kernel drop
3796 - fix xattr patches for the vanilla 2.4.19 series
3798 - generate true UUIDs in lmc (1171)
3799 - have portals stack dumping break in UML (2466)
3800 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3801 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3802 - wake callback waiting threads on client eviction (2460)
3803 - Add --ptldebug and --subsystem to lmc (1719)
3804 - update assertion to allow safe interrupt allocation
3805 - set rq_no_resend for cancel requests (2432)
3806 - recalculate ptlrpcd timeout after resend (2494)
3807 - call vfs_rmdir when removing pending directories (2368)
3808 - fix renaming a file to itself (2429)
3809 - lmc creates a default one-stripe lov (2454)
3810 - expand procfs space to handle large clusters (2326)
3811 - increase UML stack to avoid overflow
3812 - update lconf's list of debug and subsystem masks
3813 - fix lfs find --obd (2510)
3814 - /proc tunable for disabling filter read caching (2591)
3815 - stop rpm packages from altering slapd.conf (2301)
3816 - disable nagle in the socknal under 0conf (2578)
3817 - choose mds inode size based on stripe count (2572)
3818 - fix kernel-source rpm problems (2516)
3820 - add --disable-doc to avoid pdf generation (2421)
3821 - update documentation, tests, type-os, comments
3822 - avoid format warnings on ia64
3823 - remove the TOE NAL
3824 - tiny code cleanups by removing unused fields
3826 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3829 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3830 - remove the most blatant lies from BUILDING file (2371)
3831 - change default debug level to reasonable production setting
3832 - reduce client side cache size to reduce cache flush time
3833 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3834 - make TCP zerocopy and pinger support enabled by default (2476)
3835 - sync writes completed after process exits caused crashes (2319)
3836 - maintain correct mount count on the MDS (2356)
3837 - backout 1557, because 2316 wasn't really fixed
3838 - better file I/O statistics gathering in /proc
3839 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3840 - another kernel patch to fix zero-copy TCP function export
3841 - don't take duplicate lock when processing re-sent getattr (2420)
3842 - lctl uses obd_self_export instead of creating new conn (2353)
3843 - MDS/OST recovery case which requires object creation asserted (2425)
3844 - move lfs from /usr/sbin to /usr/bin in packages
3845 - fix race between mds_client_add and mds_client_free (2417)
3846 - use kmalloc instead of slabs in portals (2430)
3847 - don't create duplicate records when a failover MDS is present (2442)
3848 - remove unnecessary mount age check (2332)
3849 - don't remove directory inodes from locks prematurely (2451)
3850 - don't break if MDS service name is the same as hostname (2103)
3851 - fix races in client write RPC generation when cache full (2482)
3853 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3856 - remove now-unused request->rq_obd (278)
3857 - if an allocation fails, print out how much memory we've used (1933)
3858 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3859 - add the "configurable stack size" patch to most series files (1256)
3860 - ability to write large log records, for 100+ OST configs (2306)
3861 - fix NULL deref when filter_prep fails (2314)
3862 - fix operator precedence error in filter_sync
3863 - dynamic allocation of socknal TX descriptors (2315)
3864 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3865 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3866 - ensure that CWARN messages in Portals always get to the syslog
3867 - __init/__exit are not for prototype decls (ldlm_init/exit)
3868 - x86-64 compile warning fixes
3869 - fix gateway LMC keyword conflict (2318)
3870 - fix MDS lock inversions in getattr/reint paths (1844)
3871 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3872 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3873 - DLM race condition prevented some lock evictions (2328)
3874 - ENOMEM detection and retry on socknal sends (2230)
3875 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3876 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3877 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3878 - better error messages when a client is rejected during recovery (1505)
3879 - avoid cancelling locks which were never granted, after failure (2330)
3880 - fix i_sem/journal inversion in mds_client_add (2333)
3881 - fix truncate/getattr lock cycle deadlock (2334)
3882 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3883 - fix two rare exit paths which could leak an l_lock() ref (2321)
3884 - fencepost error in MDS/OST orphan recovery (2226)
3885 - make log record alignment 8 bytes (1988)
3886 - lstripe now fails when requested offset > ost_count (2237)
3887 - ensure that all kernel series have a complete list.h (1607)
3888 - fix crashes in special-file operations (2316)
3889 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3891 - allow configurable automake binary, for testing new versions
3892 - small update to the lfs documentation
3894 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3896 * fix negative export reference count in fsfilt_sync (2312)
3898 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3899 * release candidate 0.9.1
3901 - orphans are moved into the PENDING directory for possible recovery
3902 - replayed opens now open by fid for orphan/rename safety (1042)
3903 - last close of an orphan inode generates a transno (683)
3904 - chdir() and mount() now pin the directory entry (1020)
3905 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3906 - discard very old requests without processing them (1502)
3907 - remove some common, well-understood CERRORs (1505)
3908 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3909 - clear "grant" flags in OST replies until OST grant code lands (1644)
3910 - fix read performance by not clobbering i_blksize on client (1598)
3911 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3912 - make LRU size a /proc tunable, clears locks when reduced (707)
3913 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3914 - server locks take references on exports now (1558)
3915 - build fixes for 2.4.20-rh trees (1663)
3916 - return an error from lov_create if all OSCs are inactive (1751)
3917 - fix import levels when a reconnect happens without a timeout (1597)
3918 - exit early from mds_open if we get a lookup error (1749)
3919 - partial page read at EOF wouldn't wait for disk before sending (1642)
3920 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3921 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3922 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3923 - write/truncate lock inversion (1639)
3924 - Don't auto-load obdclass, portals modules during cleanup (1495)
3925 - fix timestamps from jumping to "now" (1763)
3926 - extra journal assertions (1648)
3927 - add an extra multiunlink test (1771)
3928 - fix read_record/write_record API (1776)
3929 - fix leak of offset_extent, possible incorrect i_size later (1772)
3930 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3931 - Debugging for the kqswnal_get_idle_tx problems (1820)
3932 - Allow recovery to be attempted multiple times (1536)
3933 - Write out MDS last_rcvd file after it is first created (1600)
3934 - Fix tx_descriptor leak in failed transmit situations (1827)
3935 - ext3 journaling fixes for assertion failure after IO error (1871)
3936 - class_export_put() on freed export after completion AST error (1896)
3937 - Fix revalidate looping in VFS (1322)
3938 - Don't access a freed export during MDS_REINT timeout (1521)
3939 - Add open-unlink recovery support on the MDS (1673,1764)
3940 - Return an error if no MDS data was read from last_rcvd (1946)
3941 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3942 - Don't setup a disk device that doesn't match exported UUID (317)
3943 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3944 - avoid committing NULL handle in force close
3945 - local.sh is now a one-stripe LOV configuration
3946 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3947 - fix ext3 htree duplicate directory entry corruption (1516)
3948 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3949 - update to new LOV EA format (2097)
3950 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3952 2003-06-15 Phil Schwan <phil@clusterfs.com>
3955 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3956 - per-import/export recovery handling (958, 931, 959)
3957 - multiple last-rcvd slots, for serving multiple FSes (949)
3958 - connections are again shared between multiple imp/exports (963, 964)
3959 - "umount -f" would hang if any requests needed to be sent (393, 978)
3960 - avoid pinning large req buffer by copying for queued messages (989)
3961 - add "uuid" to "lctl device" command to help upcalls (991)
3962 - "open" RPCs with transnos would confuse recovery counters (1037)
3963 - do proper endian conversion of all wire messages (288, 340, 891)
3964 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3965 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3966 - fix ptlrpc_connection leak in target_handle_connect (1174)
3967 - fix import refcounting bug in OST and MDS cleanup (1134)
3968 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3969 - fix very unlikely obd_types race condition (501)
3970 - remove osc_open hack for echo_client (1187)
3971 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3972 - a failure in read_inode2 leads to deadlock (1139)
3973 - cancel ack-locks as soon as transaction is committed (1072)
3974 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3975 - make sure to commitrw after any preprw to avoid deadlock (1162)
3976 - failing to execute a file in a lustre FS would lock inode (1203)
3977 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3978 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3979 - obd_uuid2tgt would walk off the end of the list (1255)
3980 - on IA64 the osc would give portals incorrect bulk size (1258)
3981 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3982 - fix lock inversion caused by new llite matching code (1282)
3983 - limit the number of dirty pages on a client to 10MB (1286)
3984 - timed out locks were not being corrected cancelled (1289)
3985 - fix O_DIRECT above 4GB on IA-32 (1292)
3986 * major user-visible changes
3987 - fail out/fail over policy now controlled by the upcall (993)
3989 - add OBD_PING to check server availability and failure (954)
3990 - lustre messages are now sent in sending host order (288, 340, 891)
3991 - add eadatalen to MDS getattr reply (340)
3992 - OST read replies may contain second buffer, with per-page status (593)
3994 2003-03-11 Phil Schwan <phil@clusterfs.com>
3997 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3998 - failed open()s could cause deadlock; fixed (867, 869)
3999 - stop cancelling OST locks when files are closed (481)
4000 - overlapping XID spaces caused network corruption (851, 853)
4001 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4002 - setattr_raw added, to do single-RPC, server-side setattrs
4003 - lmc/lconf syntax change for OST UUIDs
4004 - fix crashy race condition between ptlrpc_free_req and osc_close
4005 - don't use request in mdc_enqueue if we hit a timeout (889)
4006 - don't set the inode i_size for regular files from the MDS (896)
4007 - handle out of order completion AST (842)
4008 - don't LBUG if a lock request times out after receiving AST (913)
4009 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4010 - if a bad lock AST arrives, send an error instead of dropping entirely
4011 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4012 - fix leak in bulk IO when only partially completed (899, 900, 926)
4013 - fix O_DIRECT for ia64 (55)
4014 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4015 - C-z after timeout could hang a process forever; fixed (977)
4017 - client-side I/O cache (678, 924, 929, 941, 970)
4019 - READPAGE and SETATTRs which don't take server-side locks get
4022 2003-02-11 Phil Schwan <phil@clusterfs.com>
4025 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4026 - Fix fsfilt_extN_readpage() to read a full page of directory
4027 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4028 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4029 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4030 - Most intent operations were removed, in favour of a new RPC mode
4031 that does a single RPC to the server and bypasses most of the VFS
4032 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4033 - Aggressively cancel local locks on DLM servers
4034 - mds_reint_unlink sends EA to the client if it's the last nlink.
4035 client uses that EA to unlink OST objects.
4036 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4037 - recursive symlinks were fixed (439)
4038 - fixed NULL deref in DEBUG_REQ
4039 - filter_update_lastobjid no longer calls sync, which annoyed extN
4040 - fixed multi-client small-writes to a single file problem (445)
4041 - fixed mtime updates during file writes (607)
4042 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4043 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4044 - fixed Linux OST size reporting problem (444, 656)
4045 - OST now updates object mtime with writes or setattr (607, 619)
4046 - client verifies file size before zeroing page past EOF (445)
4047 - OST now writes last allocated objid to disk with allocation (108)
4048 - LOV on echo now works (409)
4050 - mds_reint_unlink sends a new buffer, with the EA included. this
4051 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4052 set if a regular file was being unlinked, and it was the last link
4053 - use PtlGet from the target for bulk writes (315)
4054 - OST now updates object mtime with writes or setattr (607, 619)
4055 - LDLM now has a grant-time callback to revalidate locked items, if
4057 - Many MDS operations were reorganized to combat race conditions
4059 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4060 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4062 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4065 - Fully reactivate OST imports after reconnection (512, others)
4066 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4067 - More graceful error handling for truncating on dead OST (515)
4068 - Don't error out unless we're actually accessing dead stripes (474)
4069 - Fix garbage sizes when stripes are missing (410)
4070 - LRU counters were broken, causing constant lock purge (433, 432)
4071 - garbage on read from stripes with failed OSTs (441)
4072 - mark OSCs as active before reconnecting during recovery (438)
4073 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4074 - lfind did not preserve OST order in output (443)
4075 - symlinks cause hung clients, incorrect data (439)
4076 - stop dereferencing request after dropping refcount (457)
4077 - don't LASSERT(spin_is_locked) on non-SMP (455)
4078 - fixes for many rename() bugs
4079 - fstat didn't correctly synchronize attributes (399)
4080 - server must handle lock cancellation during blocking AST prep (487)
4081 - bulk descriptors were free()d too soon (511)
4082 - fix paths in lconf, which would load incorrect modules (451, 507)
4083 - fix confusing lconf 'host not found' error message (386)
4084 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4085 - fix race condition in mdc_blocking_ast() for inode access (526)
4086 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4087 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4088 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4089 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4090 - fix lstripe in several ways (kernel side) (527)
4091 - fix request leak in ldlm_cli_enqueue (262)
4092 - incorrect OSC was marked inactive after OST failure
4093 - call mds_fs_cleanup before unmounting filesystem (524)
4094 - fix races between taking ns_lock and ldlm_lock_change_resource
4095 - fix races updating LOV export open file list
4096 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4097 - fix recovery NULL deref in ldlm_cli_cancel_unused
4098 - fix some DLM races by using new hash table for lock handles (419)
4099 - permit the client to specify desired inodes, at replay
4100 - duplicate requests when we queue them for replay reintegration
4101 - fix last_rcvd offset calculation
4102 - sync after each recovered transaction, so we always make progress
4103 - never, not always, ERESTART requests without transnos
4104 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4106 - skip replay if the MDS says that the client is already connected
4107 - don't check for a recovery-enabled export to match lctl's UUID
4108 - don't INC_USE_COUNT for phantom exports
4109 - don't crash when cleaning up phantom exports (567)
4110 - don't double-finish or set replay data for errored mdc_open requests
4111 - abort requests when they time out, so we don't get old replies
4112 - send/receive replies for AST messages again
4113 - if the client says that it doesn't have the lock, cancel it on the
4115 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4116 mark it as destroyed, it will all work out when decref is called
4117 - fix module use counts (22, 581)
4119 - ASTs now expect a reply (server cancels lock on error reply)
4121 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4124 - fix many simultaneous client startup (392)
4125 - fix dentry->d_it clobbering
4126 - credentials weren't being shipped for readdir/getattr operations
4127 - remove invalid assertions triggered during some concurrent MD
4129 - proper Lustre versions added (336, 389)
4130 - fix memory leak for create error case (398)
4131 - fix LOV locking bug that would get cli/srv out of sync
4132 - fix echo client over LOV (409)
4133 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4134 - fix double-O_EXCL intent crash (424)
4135 - avoid sending multiple lock CANCELs (352)
4137 - MDS can do multi-client recovery (modulo bugs in new code)
4139 - many updates, edits, cleanups
4141 2002-11-18 Phil Schwan <phil@clusterfs.com>
4144 - fix null d_it dereference (346)
4145 - fix full OST/dbench hang (333)
4146 - fix permission problem with file removal (286)
4147 - fix removal of OSCs from LOV when they fail
4148 - fix NULL deref during bulk timeout (214)
4149 - fix problems related to multiple filesystems on one MDS (241)
4150 - fixed serious subtle metadata locking bugs
4151 - free locks on clients when inodes are removed due to memory
4153 - fix inode pointer in lock data (285)
4154 - partial support for multiple MDS on a single host (241)
4155 - data locks weren't cancelled at clear_inode time (290, 311)
4156 - intent locks could lead to unbounded lock growth (205)
4157 - added a maximum lock count, an LRU list, and a flusher
4158 - fix multiple rename (365)
4159 - properly abstracted the echo client
4160 - OSC locked 1 byte too many; fixed
4161 - rewrote brw callback code:
4162 - fixed recovery bugs related to LOVs (306)
4163 - fixed too-many-pages-in-one-write crash (191)
4164 - fixed (again) crash in sync_io_timeout (214)
4165 - probably fixed callback-related race (385)
4167 - Add capability to MDS protocol
4168 - LDLM cancellations and callbacks on different portals
4170 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4173 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4174 - defer open object destruction to close time (601981, 138)
4175 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4176 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4177 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4178 - don't call obd_setattr in ll_file_release for destroyed objects
4179 * protocol change to lustre_msg: move |version| and add |flags|
4180 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4181 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4182 * added replay of create, unlink, link and rename operations during
4183 MDS failover; recovery should be much more robust now
4184 * remove failed OSCs from LOVs (only lov_create uses this so far)
4185 * the lustre-HOWTO was brought (more) up to date (582544)
4187 2002-10-23 Phil Schwan <phil@clusterfs.com>
4190 - in-use dentries weren't being reused properly (617851)
4191 - prevent multiple LDLM setup (599178)
4192 - fix LOV size calculations for truncate (617853)
4193 - fix client handling of MDS intent errors (POSIX)
4194 - fix permission bug in lovstripe.c test (624321)
4195 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4196 - truncate past end of file could corrupt data
4197 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4198 - a race in recovery could return ETIMEDOUT to apps (623947)
4199 - building outside the source directory was fixed
4200 * the lustre-HOWTO was brought (more) up to date (582544)
4201 * major progress was made on recovery functionality
4203 2002-10-10 Phil Schwan <phil@clusterfs.com>
4206 - recovery deadlock fix
4207 - rm -rf causes LBUG fix (617817)
4208 - file open by multiple tasks fix (618962)
4209 - directory permissions bugs (602707 and 620007)
4210 - journal_stop fixed with locking (611313)
4211 - O_APPEND failures resolved (618273, perhaps 614459)
4212 - lconf PATH fix (619770)
4213 - IA64 build fix (621450)
4214 - RPC buffer sizes scale with amount of memory
4216 2002-10-01 Phil Schwan <phil@clusterfs.com>
4219 - locks would be cancelled without throwing away data pages,
4220 resulting in inconsistent data (605627)
4221 - inode attributes were not always being refreshed (605627, 612449)
4222 - lconf now continues to cleanup after lctl reports an error
4223 - MDS now enforces user permissions (602707)
4224 - lprocfs cleanup fixed, but not yet enabled (614157)
4225 - fixed infinite server hang, should a client not respond to an AST
4226 - avoid going into recovery if user calls readlink() with a buffer
4227 that's too small (613941)
4228 - AST RPCs no longer require replies (614867) -- this may be changed
4229 - don't crash server if client sends an IOV that's too big (611336)
4230 - fixed lock conversion deadlock (611892)
4231 - fixed the following of symlinks (614622)
4232 * recovery: the server can remove locks from a client that dies, other
4233 clients can make progress
4234 * more extN patch fixes
4235 * compile-time configurable ptlrpc buffer allocations
4237 - collaborative read cache document
4238 - Lustre Lite Performance CDR document-in-progress
4240 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4243 - fix typo in patch-2.4.18
4245 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4248 - clear ptlrpc request each time in handle_incoming_request()
4249 - unlink of files now destroys the object on the OST
4251 2002-09-19 Peter Braam <braam@clusterfs.com>
4253 * add hard link support
4254 * change obdfile creation method
4255 * kernel patch changed
4257 2002-09-19 Peter Braam <braam@clusterfs.com>
4260 - stack overflow bug in extN fixed
4262 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4264 * documentation updates
4265 - add man pages for config tools
4266 - update tests/README to describe testing with new config tools
4267 - finish metadata API descriptions
4268 * bug fixes and cleanups
4269 - statfs workaround for 16TB limit
4270 - LOV stripe allocation improved, can stripe on subset of OSTs
4271 - LOV file size/IO offset was wrong for files > 4GB in size
4272 - object EA data was being dropped, caused files to be unreadable
4273 - memory overflow with non-LOV OST caused memory corruption
4274 - fixed regression tests to work with new config tools, obdfilter
4275 - fixed bug when directory size became larger than 1 block
4276 - fixed bug (for single client case) when PWD was deleted
4277 - invalidate local directory pages when doing intent-based ops
4278 - avoid LDLM oops when lock callback contained bad data
4280 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4282 * documentation updates
4283 * bug fixes and cleanups
4284 - configuration tools
4287 - 64-bit compile warnings
4288 - 64-bit internal statfs data
4290 * test_brw on persistent OST devices
4292 * lprocfs (disabled)
4294 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4296 * documentation updates
4297 * bug fixes and cleanups
4298 * configuration tools
4300 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4303 * many small fixes to 0_5_4
4304 * io/network handling
4305 * thinkos in MDS operations
4307 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4310 * crucial basic fixes to 0.5.3
4311 * IOR, Iozone work over Elan
4314 2002-08-07 Phil Schwan <phil@clusterfs.com>
4315 * version 0_5_3, our first alpha
4316 * we use the new Portals iovs
4317 * documentation updates
4318 * bug fixes and cleanups
4319 * small changes in the DLM wire protocol
4321 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4322 * version 0_5_1 with some initial stability,
4323 * locking on MD and file I/O.
4324 * documentation updates
4325 * several bug fixes since 0.5.0
4326 * small changes in wire protocol
4328 2002-07-18 Phil Schwan <phil@clusterfs.com>
4330 * delivered as Lustre Light Alpha
4331 * fixed a crash after handling invalid MDS requests
4332 * fixed directory pages for architectures with non-4k pages sizes
4334 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4335 * release version v0_4_4
4336 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4338 * Fixes a number of interruption problems with OST operations.
4339 * Update documentation for portals header changes
4340 * Move all wire protocol structs/defines to lustre_idl.h
4341 * Fixes symlink length bug.
4342 * Add tcpdump to repository.
4344 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4345 * release version v0_4_3
4346 * Fixes statfs for inodes on extN.
4347 * Fixes bug in runtests which would delete /etc/hosts.
4348 * Use 64-bit object IDs wherever possible (not into VFS though)
4349 Remove ost_get_info, which is unused by lustre, and out of date.
4351 2002-07-03 Peter Braam <braam@clusterfs.com>
4352 * release version v0_4_2 Fixes a lookup error (type not passed)
4353 * move forward to head of Portals
4354 * move forward to latest Lustre kernel
4356 2002-06-25 Peter Braam <braam@clusterfs.com>
4357 * release version v0_4_1. Hopefully stable on single node use.
git://git.whamcloud.com / fs / lustre-release.git / blob