1 tbd Sun Microsystems, Inc.
5 2.6.9-67.0.1.EL (RHEL 4),
6 2.6.16.54-0.2.3 (SLES 10),
7 2.6.18-53.1.4.el5 (RHEL 5),
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.22 vanilla (kernel.org)
12 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
13 we do not recommend using patchless RHEL4 clients with kernels
14 prior to 2.6.9-55EL (RHEL4U5).
15 * Recommended e2fsprogs version: 1.40.4-cfs1
16 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
18 removed cwd "./" (refer to Bugzilla 14399).
20 Severity : enhancement
22 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
25 Frequency : rare on shutdown ost
27 Description: If llog cancel was not send before clean_exports phase, this can
28 produce deadlock in llog code.
29 Details : If llog thread has last reference to obd and call class_import_put
30 this produce deadlock because llog_cleanup_commit_master wait when
31 last llog_commit_thread exited, but this never success because was
32 called from llog_commit_thread.
36 Description: NULL lov_tgts causing MDS oops
37 Details : more safe checks for NULL lov_tgts for avoid oops.
39 Severity : enhancement
41 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
43 Severity : enhancement
45 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
49 Description: make mgs_setparam() handle fsname containing dash
50 Details : fsname containing a dash does not work with lctl conf_param
52 Severity : enhancement
54 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
58 Frequency : rare, in recovery and (or) destroy lovobjid file.
59 Description: rewrite lov objid code.
60 Details : Cleanup for lov objid code, remove scability problems and wrong
61 locking. Fix sending last_id into ost.
63 Severity : enhancement
65 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
67 Severity : enhancement
69 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
70 Details : Use d_move_locked instead of __d_move.
74 Frequency : rare, at shutdown
75 Description: access already free / zero obd_namespace.
76 Details : if client_disconnect_export was called without force flag set,
77 and exist connect request in flight, this can produce access to
78 NULL pointer (or already free pointer) when connect_interpret
79 store ocd flags in obd_namespace.
83 Frequency : only at startup
84 Description: not alloc memory with spinlock held.
85 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
91 Description: lfs find does not continue on file error
92 Details : Continue other files processing when a file/dir is absent.
96 Description: Inconsistent usage of lustre_pack_reply()
97 Details : Standardize the usage of lustre_pack_reply() such that it
98 always generate a CERROR on failure.
101 Frequency : very rare
103 Description: Fix replay if there is an un-replied request and open
104 Details : In some cases, older replay request will revert the
105 mcd->mcd_last_xid on MDS which is used to record the client's
108 Severity : enhancement
110 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
112 Severity : enhancement
114 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
116 Severity : enhancement
118 Description: Update to SLES9 kernel-2.6.5-7.287.3.
120 Severity : enhancement
122 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
124 Severity : enhancement
126 Description: Build SLES10 patchless client fails
127 Details : The configure was broken by run ./configure with
128 --with-linux-obj=.... argument for patchless client. When the
129 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
130 can't search header adequately. Use absolute path such as
131 -I($LINUX)/include instead.
133 Severity : enhancement
135 Description: Lustre Page Accounting
136 Details : New macros for page alloc and free which enable accounting
137 of page allocation of Lustre. Use percpu counters to store memory
141 Frequency : only if debugging is disabled
143 Description: LASSERT_{REQ,REP}SWAB macros are buggy
144 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
145 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
146 these macros and replace them with their difinitions instead.
151 Description: interrupt oig_wait produce painc on resend.
152 Details : brw_redo_request can be used for resend requests from ptlrpcd and
153 private set, and this produce situation when rq_ptlrpcd_data not
154 copyed to new allocated request and triggered LBUG on assert
155 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
156 ptlrpcd set if request is changed and can be safety changed to use
159 Severity : enhancement
161 Description: organize the server-side client stats on per-nid basis
162 Details : Change the structure of stats under obdfilter and mds to
170 The "uuid"s file would list the uuids of _active_ exports.
171 And the clear entry is to clear all stats and stale nids.
176 Description: Processes looping in ll_readdir() on Lustre clients finally causing
177 a full node pseudo-hang
178 Details : Concurrent access to the same directory from multiple clients with
179 intensive file creation/removal can cause a client node to spin in
180 ll_readdir(). i_version must be increased every time the lock is
181 cancelled to ensure a revalidate is done.
186 Description: touch file failed when fs is not full
187 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
188 otherwise we can get ENOSP while fs is not full.
191 Frequency : only for Cray XT3
192 Bugzilla : 12829/13455
193 Description: Changing primary group doesn't change the group lustre assigns to
195 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
196 overridden with the primary group provided by the group upcall,
197 whereas the supplied fsgid can be trusted if it is in the list of
198 supplementary groups returned by the group upcall.
202 Description: Root Squash Functionality
203 Details : Implementation of NFS-like root squash capability. Specifically,
204 don't allow someone with root access on a client node to be able
205 to manipulate files owned by root on a server node.
209 Description: Slow trucate/writes to huge files at high offsets.
210 Details : Directly associate cached pages to lock that protect those pages,
211 this allows us to quickly find what pages to write and remove
212 once lock callback is received.
216 Description: Too many locks accumulating on client during NFS usage
217 Details : mds_open improperly used accmode to find out access mode to a
218 file. Also mdc_intent_lock logic to find out if we already have
219 lock similar to just received was flawed since introduction of
220 skiplists - locks are now added to the front of the granted
225 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
227 Details : In obd setup failure handler, obd_stopping will not necessarily to
228 be 1, and obd_set_up should also be checked to make sure whether
229 obd is completely setup.
231 Severity : enhancement
233 Description: Allow masking D_WARNING, D_ERROR messages from console
234 Details : Console messages can now be disabled via lnet.printk.
238 Description: User code with malformed file open parameter crashes client node
239 Details : Before packing join_file req, all the related reference should be
240 checked carefully in case some malformed flags cause fake
241 join_file req on client.
246 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
247 grain of quota doing harm to quota's accuracy
248 Details : qunit size will be changed when quota limitation is too low/high;
249 record the pending quota write in order to get more accureate
250 quota; delete the patch for bug12588, which is unnecessary when
251 this patch is landed. This bug also contains 14526, 14299, 14601
252 and 13794, which are found and landed during v1_4_12.
256 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
257 Details : ldlm_completion_ast() assumes that a lock is granted when the req
258 mode is equal to the granted mode. However, it should also check
259 that LDLM_FL_CP_REQD is not set.
263 Description: Heavy nfs access might result in deadlocks
264 Details : After ELC code landed, it is now improper to enqueue any mds
265 locks under och_sem, because enqueue might want to decide to
266 cancel open locks for same inode we are holding och_sem for.
268 --------------------------------------------------------------------------------
270 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
272 * Support for kernels:
273 2.6.5-7.286 (SLES 9),
274 2.6.9-55.0.9.EL (RHEL 4),
275 2.6.16.53-0.8 (SLES 10),
276 2.6.18-8.1.14.el5 (RHEL 5),
277 2.6.18.8 vanilla (kernel.org)
278 * Client support for unpatched kernels:
279 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
280 2.6.16 - 2.6.22 vanilla (kernel.org)
281 * Due to recently discovered recovery problems, we do not recommend
282 using patchless RHEL 4 clients with this or any earlier release.
283 * Recommended e2fsprogs version: 1.40.2-cfs1
284 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
287 Frequency : occasional
289 Description: MDS hang or stay in waiting lock
290 Details : If client receive lock with CBPENDING flag ldlm need send lock
291 cancel as separate rpc, to avoid situation when cancel request
292 can't processed due all i/o threads stay in wait lock.
295 Frequency : occasional
297 Description: Do not fail import if osc_interpret_create gets -EAGAIN
298 Details : If osc_interpret_create got -EAGAIN it immediately exits and
299 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
300 oscc_has_objects and see OSC has no objests and call
301 oscc_internal_create to resend create request.
303 Severity : enhancement
305 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
306 Details : Update which_patch & target file for SLES10 latest kernel.
308 Severity : enhancement
310 Description: add --type and --size parameters to lfs find
311 Details : Enhance lfs find by adding filetype and filesize parameters. Also
312 multiple OBDs can now be specified for the --obd option.
314 Severity : enhancement
316 Description: eliminate client locks in face of contention
317 Details : file contention detection and lockless i/o implementation
320 Severity : enhancement
322 Description: Remove client patches from SLES 10 kernel.
323 Details : This causes SLES 10 clients to behave as patchless clients
324 even on a Lustre-patched (server) kernel.
326 Severity : enhancement
328 Description: use i_size_read and i_size_write in 2.6 port
329 Details : replace inode->i_size access with i_size_read/write()
332 Frequency : when removing large files
334 Description: scheduling issue during removal of large Lustre files
335 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
336 It causes scheduling issues when removing large files (17TB in the
342 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
343 Details : Impossible (EPROTO is returned) to access a directory that has a
344 non-default striping and ACLs.
347 Frequency : only on ppc
349 Description: /proc/fs/lustre/devices broken on ppc
350 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
351 We should make sure the type of 'index' is loff_t and then cast
352 explicitly as needed below. Do not assign an explicitly cast
356 Frequency : only for rhel5
358 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
359 Details : Modify the target file & which_kernel.
362 Frequency : if the uninit_groups feature is enabled on ldiskfs
364 Description: e2fsck reports "invalid unused inodes count"
365 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
366 feature and only a single inode is created in a group then the
367 "bg_unused_inodes" count is incorrectly updated. Creating a
368 second inode in that group would update it correctly.
373 Description: buffer overruns could theoretically occur
374 Details : llapi_semantic_traverse() modifies the "path" argument by
375 appending values to the end of the origin string, and a buffer
376 overrun may occur. Adding buffer overrun check in liblustreapi.
378 Severity : enhancement
380 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
383 Frequency : only if filesystem is inconsistent
385 Description: handle "serious error: objid * already exists" more gracefully
386 Details : If LAST_ID value on disk is smaller than the objects existing in
387 the O/0/d* directories, it indicates disk corruption and causes an
388 LBUG(). If the object is 0-length, then we should use the existing
389 object. This will help to avoid a full fsck in most cases.
391 Severity : enhancement
393 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
394 Details : Modify vm-tunables-rhel4.patch.
396 Severity : enhancement
398 Description: Kernel config for 2.6.18-vanilla.
399 Details : Modify targets/2.6-vanilla.target.in.
400 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
401 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
402 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
403 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
406 Frequency : occasional
408 Description: improve handling recoverable errors
409 Details : If request processed with error which can be recoverable on server
410 request should be resend, otherwise page released from cache and
416 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
417 Details : Modify target file & which_patch.
418 A flaw was found in the IA32 system call emulation provided
419 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
420 value could be stored in the %RAX register, which could trigger an
421 out-of-bounds system call table access. An untrusted local user
422 could exploit this flaw to run code in the kernel
423 (ie a root privilege escalation). (CVE-2007-4573).
427 Description: change order of libsysio includes
428 Details : '#include sysio.h' should always come before '#include xtio.h'
430 Severity : enhancement
432 Description: adapt the lustre_config script to support the upgrade case
433 Details : Add "-u" option for lustre_config script to support upgrading 1.4
434 server targets to 1.6 in parallel.
439 Description: To avoid grant space > avaible space when the disk is almost
440 full. Without this patch you might see the error "grant XXXX >
441 available" or some LBUG about grant, when the disk is almost
443 Details : In filter_check_grant, for non_grant cache write, we should
444 check the left space by if (*left > ungranted + bytes), instead
445 of (*left > ungranted), because only we are sure the left space
446 is enough for another "bytes", then the ungrant space should be
447 increase. In client, we should update cl_avail_grant only there is
448 OBD_MD_FLGRANT in the reply.
453 Description: Update RHEL 4 kernel to fix local root privilege escalation.
454 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
455 described in CVE-2007-4573. This problem could allow untrusted
456 local users to gain root access.
459 Frequency : when using O_DIRECT and quotas
461 Description: Incorrect file ownership on O_DIRECT output files
462 Details : block usage reported by 'lfs quota' does not take into account
463 files that have been written with O_DIRECT.
465 --------------------------------------------------------------------------------
467 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
469 * Support for kernels:
470 2.6.5-7.286 (SLES 9),
471 2.6.9-55.0.2.EL (RHEL 4),
472 2.6.16.46-0.14 (SLES 10),
473 2.6.18-8.1.8.el5 (RHEL 5),
474 2.6.18.8 vanilla (kernel.org)
475 * Client support for unpatched kernels:
476 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
477 2.6.16 - 2.6.21 vanilla (kernel.org)
478 * Due to recently discovered recovery problems, we do not recommend
479 using patchless RHEL 4 clients with this or any earlier release.
480 * Recommended e2fsprogs version: 1.40.2-cfs1
481 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
485 Description: Fix errors in lfs documentation
486 Details : Fixes man pages
488 Severity : enhancement
490 Description: Adaptive timeouts
491 Details : RPC timeouts adapt to changing server load and network
492 conditions to reduce resend attempts and improve recovery time.
494 Severity : enhancement
496 Description: llapi_file_create() does not allow some changes
497 Details : add llapi_file_open() that allows specifying the file creation
498 mode and open flags, and also returns an open file handle.
500 Severity : enhancement
502 Description: df doesn't work properly if diskfs blocksize != 4K
503 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
506 Frequency : always on directories with default striping set
508 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
509 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
511 Severity : enhancement
513 Description: merge and cleanup kernel patches.
514 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
516 Severity : enhancement
518 Description: RedHat Update kernel for RHEL5
519 Details : Kernel config file for RHEL5.
521 Severity : enhancement
523 Description: OSS needs mutliple precreate threads
524 Details : Add ability to start more than one create thread per OSS.
527 Frequency : only with quota on the root user
529 Description: mds_obd_create error creating tmp object
530 Details : When the user sets quota on root, llog will be affected and can't
531 create files and write files.
534 Frequency : Always on ia64 patchless client, and possibly others.
536 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
537 Details : This allows the patchless client to be loaded on architectures
543 Description: disorder of journal start and llog_add cause deadlock.
544 Details : in llog_origin_connect, journal start should happen before
545 llog_add keep the same order as other functions to avoid
548 Severity : enhancement
550 Description: RedHat Update kernel for RHEL5
551 Details : Modify the kernel config file more closer RHEL5.
553 Severity : enhancement
555 Description: Build failure against Centos5 (RHEL5)
556 Details : Define PAGE_SIZE when it isn't present.
559 Frequency : occasionally when using NFS
561 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
562 Details : with NFS, the anon dentry's parent was set to itself in
563 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
564 corresponding dentry other than use rec->ur_name.
566 Severity : enhancement
568 Description: enable data checksumming by default
569 Details : enable checksum by default, allow --disable-checksum
570 configure option and "-o nochecksum" mount option. Checksums
571 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
574 Frequency : Occasionally with failover
576 Description: Client eviction due to failover config
577 Details : after a connection loss, the lustre client should attempt to
578 reconnect to the last active server first before trying the
579 other potential connections.
581 Severity : enhancement
583 Description: client-side metadata stat-ahead during readdir(directory readahead)
584 Details : perform client-side metadata stat-ahead when the client detects
585 readdir and sequential stat of dir entries therein
590 Description: when mds and osts use different quota unit(32bit and 64bit),
591 quota will be released repeatly.
592 Details : void sending multiple quota reqs to mds, which will keep the
593 status between the reqs.
596 Frequency : only with liblustre clients on XT3
598 Description: evictions taking too long
599 Details : allow llrd to evict clients directly on OSTs
603 Description: osts not allocated evenly to files
604 Details : change the condition to increase offset_idx
606 Severity : enhancement
608 Description: self-adjustable client's lru lists
609 Details : use adaptive algorithm for managing client cached locks lru
610 lists according to current server load, other client's work
611 pattern, memory activities, etc. Both, server and client
612 side namespaces provide number of proc tunables for controlling
617 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
618 Details : rewrite inherited code (uuid parsing code from ext2 utils and
619 readdir code from ext3) from scratch preserving functionality.
623 Description: Only those disconnect error should be returned by rq_status.
624 Details : In open/enqueue processs, Some errors, which will cause client
625 disconnected, should be returned by rq_status, while other
626 errors should still be returned by intent, then mdc or llite will
629 Severity : enhancement
631 Description: Tune the kernel for good SCSI performance.
632 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
633 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
637 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
638 Bugzilla : 13547, 13627
639 Description: Data corruption for OSTs that are formatted larger than 2TB
641 Details : When generating the bio request for lustre file writes the
642 sector number would overflow a temporary variable before being
643 used for the IO. The data reads correctly from Lustre (which
644 will overflow in a similar manner) but other file data or
645 filesystem metadata may be corrupted in some cases.
649 Description: "lfs find -obd UUID" prints directories
650 Details : "lfs find -obd UUID" will return all directory names instead
651 of just file names. It is incorrect because the directories
652 do not reside on the OSTs.
655 Bugzilla : 13596, 13614
656 Description: MDS hang after unclean shutdown of lots of clients
657 Details : Never resend AST requests.
661 Frequency : Always, for kernels after 2.6.16
662 Description: Fix warning idr_remove called for id=.. which is not allocated.
663 Details : Last kernels save old s_dev before kill super and not allow
664 to restore from callback - restore it before call kill_anon_super.
668 Description: Fix error on 'ls .' at the top of the Lustre mount.
669 Details : Don't revalidate dentry if it is a root dentry.
671 --------------------------------------------------------------------------------
673 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
675 * Support for kernels:
676 2.6.5-7.286 (SLES 9),
677 2.6.9-55.0.2.EL (RHEL 4),
678 2.6.16.46-0.14 (SLES 10),
679 2.6.18-8.1.8.el5 (RHEL 5),
680 2.6.18.8 vanilla (kernel.org)
681 * Client support for unpatched kernels:
682 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
683 2.6.16 - 2.6.22 vanilla (kernel.org)
684 * Due to recently discovered recovery problems, we do not recommend
685 using patchless RHEL 4 clients with this or any earlier release.
686 * Recommended e2fsprogs version: 1.39.cfs8
687 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
692 Description: block reactivating mgc import until all deactivates complete
693 Details : Fix race when failing back MDT/MGS to itself (testing)
695 Severity : enhancement
697 Description: lfs setstripe enhancement
698 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
701 Frequency : mds/oss recovery
703 Description: llog ctxt is refrenced after it has been freed.
704 Details : llog ctxt refcount was added to avoide the race between ctxt free
705 and llog recovery process. Each llog user must hold ctxt refcount
706 before it access the llog. And the llog ctxt can only be freed
707 when its refcount is zero.
709 Severity : enhancement
711 Description: randomly memory allocation failure util
712 Details : Make lustre randomly failed allocating memory for testing purpose.
715 Frequency : only for SLES10
717 Description: Update kernel patch for SLES10 SP1
718 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
720 Severity : enhancement
722 Description: omit set fsid for export NFS
723 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
724 as FS_REQUIRES_DEV for avoid problems with generate fsid.
727 Frequency : after network failures
729 Description: Add sync option to mount_lustre.c
730 Details : Client loses data written to lustre after a network interruption.
732 Severity : enhancement
734 Description: Error message improvement.
735 Details : Merging of two LCONSOLE_ERROR_MSG into one.
737 Severity : enhancement
739 Description: don't use GFP_* in generic Lustre code.
740 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
742 Severity : enhancement
744 Description: obdclass is limited by single OBD_ALLOC(idarray)
745 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
747 Severity : enhancement
749 Description: updated patchess for new RHEL4 kernel
750 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
752 Severity : enhancement
754 Description: warnings with build patchless client with vanila 2.6.19 and up
755 Details : change old ctl_table style and replace ctl_table/ctl_table_header
756 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
758 Severity : enhancement
760 Description: O_DIRECT bypasses client statistics.
761 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
762 read_ahead_stats were updated. Copied stats section from
763 osc_send_oap_rpc() into async_internal().
766 Frequency : only for Cray XT3
768 Description: peer credits not enough on many OST per OSS systems.
769 Details : Use new lnet way to add credits as we need those for pings and ASTs
772 Frequency : only with liblustre
774 Description: Liblustre is not releasing flock locks on file close.
775 Details : Release flock locks on file close.
778 Frequency : only for RHEL4
780 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
781 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
782 Update target file and kernel config.
786 Description: lustre support for RHEL5
787 Details : Add support for RHEL5.
792 Description: ASSERTION(export != NULL) failed in target_handle_connect
793 Details : Assetion hit is result of rare race between disconnect and connet
794 to same nid. target_handle_connect found old connect cockie and
795 tried to reconnect, but can't find export for this cockie.
800 Description: umount blocks forever on error
801 Details : In result of wrong using obd_no_recov and obd_force flags client
802 can hand if cancel or some other requests is lost.
805 Frequency : Only for SLES
807 Description: sanity_quota fail test_1
808 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
809 file, which makes TSTID[2] inunique.
811 Severity : enhancement
813 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
814 Details : Update target/ChangeLog/which_patch .
816 Severity : enhancement
818 Description: jbd statistics
819 Details : Port older jbd statistics patch for sles10
824 Description: lvbo_init failed for resource with missing objects.
825 Details : Fix returning error if we do stat for file with missing/corrupted
826 objects and i_size set to all sum of size of all avaible objects.
827 if we to truncate/write to missing object - it is recreated.
832 Description: Oops in read and write path when failing to allocate lock.
833 Details : Check if lock allocation failed and return error back.
836 Frequency : When flocks are used.
838 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
839 Details : Flock locks might destroy just granted lock if it could be merged
840 with another existing flock, this is done in completion handler,
841 so teach ldlm_cli_enquque_fini that this is a valid case for
847 Description: reply_lock_interpret crash due to race with it and lock cancel.
848 Details : Do not replay locks that are being cancelled. Do not reference
849 locks by their address during replay, just by their handle.
852 Frequency : only with deactivated OSTs
854 Description: lstripe command fails for valid OST index
855 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
856 instead of lov->desc.ld_active_tgt_count.
858 Severity : enhancement
860 Description: Build failure against Centos5 (RHEL5)
861 Details : Use getpagesize() instead of PAGE_SIZE.
863 --------------------------------------------------------------------------------
865 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
867 * Support for kernels:
868 2.6.5-7.283 (SLES 9),
869 2.6.9-55.EL (RHEL 4),
870 2.6.16.46-0.14 (SLES 10),
871 2.6.18.8 vanilla (kernel.org)
872 * Client support for unpatched kernels:
873 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
874 2.6.16 - 2.6.22 vanilla (kernel.org)
875 * Due to recently discovered recovery problems, we do not recommend
876 using patchless RHEL 4 clients with this or any earlier release.
877 * Recommended e2fsprogs version: 1.39.cfs8
878 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
879 * Starting with this release, the ldiskfs backing filesystem required
880 by Lustre is now in its own package, lustre-ldiskfs. This package
881 should be installed. It is versioned separately from Lustre and
882 may be released separately in future.
887 Description: server LBUG when shutting down
888 Details : Block umount forever until the mount refcount is zero rather
889 than giving up after an arbitrary timeout.
891 Severity : enhancement
893 Description: add optional extra BUILD_VERSION info
894 Details : add a new environment variable (namely LUSTRE_VERS) which allows
895 to override the lustre version.
898 Frequency : 2.6.18 servers only
900 Description: ll_kern_mount() doesn't release the module reference
901 Details : The ldiskfs module reference count never drops down to 0
902 because ll_kern_mount() doesn't release the module reference.
907 Description: server LBUG when using old ost_num_threads parameter
908 Details : Accept the old ost_num_threads parameter but warn that it
909 is deprecated, and fix an off-by-one error that caused an LBUG.
914 Description: Transient SCSI error results in persistent IO issue
915 Details : iobuf->dr_error is not reinitialized to 0 between two
919 Frequency : sometimes when underlying device returns I/O errors
921 Description: OSTs not going read-only during write failures
922 Details : OSTs are not remounted read-only when the journal commit threads
923 get I/O errors because fsfilt_ext3 calls journal_start/stop()
924 instead of the ext3 wrappers.
927 Frequency : only with 10000 clients or more
929 Description: poor connect scaling with increasing client count
930 Details : Don't run filter_grant_sanity_check for more than 100 exports
931 to improve scaling for large numbers of clients.
934 Frequency : SLES10 only
936 Description: sanity-quota.sh quotacheck failed: rc = -22
937 Details : Quotas cannot be enabled on SLES10.
940 Frequency : liblustre clients only
942 Description: getdirentries does not give error when run on compute nodes
943 Details : getdirentries does not fail when the size specified as an argument
944 is too small to contain at least one entry
946 Severity : enhancement
948 Description: Add LNET router traceability for debug purposes
949 Details : If a checksum failure occurs with a router as part of the
950 IO path, the NID of the last router that forwarded the bulk data
951 is printed so it can be identified.
956 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
957 Details : Due to a race condition, liblustre clients were occasionally
960 Severity : enhancement
962 Description: lfs setstripe use optional parameters instead of postional
965 Severity : enhancement
967 Description: Nanosecond timestamp support for ldiskfs
968 Details : The on-disk ldiskfs filesystem has added support for nanosecond
969 resolution timestamps. There is not yet support for this at
970 the Lustre filesystem level.
973 Frequency : during server recovery
975 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
976 Details : request with rq_no_resend flag not awake l_wait_event if they get a
980 Frequency : nfs export on patchless client
982 Description: connectathon hang when test nfs export over patchless client
983 Details : Disconnected dentry cannot be found with lookup, so we do not need
984 to unhash it or make it invalid
987 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
988 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
989 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
994 Description: open req refcounting wrong on reconnect
995 Details : If reconnect happened between getting open reply from server and
996 call to mdc_set_replay_data in ll_file_open, we will schedule
997 replay for unreferenced request that we are about to free.
998 Subsequent close will crash in variety of ways.
999 Check that request is still eligible for replay in
1000 mdc_set_replay_data().
1005 Description: disable writes to filesystem when reading health_check file
1006 Details : the default for reading the health_check proc file has changed
1007 to NOT do a journal transaction and write to disk, because this
1008 can cause reads of the /proc file to hang and block HA state
1009 checking on a healthy but otherwise heavily loaded system. It
1010 is possible to return to the previous behaviour during configure
1011 with --enable-health-write.
1013 Severity : enhancement
1015 Description: 64-bit inode version
1016 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1017 to the inode. This will be required for version-based recovery.
1022 Description: MDS fails to start if a duplicate client export is detected
1023 Details : in some rare cases it was possible for a client to connect to
1024 an MDS multiple times. Upon recovery the MDS would detect this
1025 and fail during startup. Handle this more gracefully.
1027 Severity : enhancement
1029 Description: Add -o localflock option to simulate old noflock behaviour.
1030 Details : This will achieve local-only flock/fcntl locks coherentness.
1035 Description: log_commit_thread vs filter_destroy race leads to crash
1036 Details : Take import reference before releasing llog record semaphore
1041 Description: Wrong request locking in request set processing
1042 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1043 imp_delayed_list, in this place should be used imp_lock.
1046 Frequency : when reconnecting
1048 Description: Grant leak when OSC reconnect to OST
1049 Details : When osc reconnect ost, OST(filter) should check whether it
1050 should grant more space to client by comparing fed_grant and
1051 cl_avail_grant, and return the granted space to client instead
1052 of "new granted" space, because client will call osc_init_grant
1053 to update the client grant space info.
1056 Frequency : when client reconnects to OST
1058 Description: Grant leak when OSC does a resend and replays bulk write
1059 Details : When osc reconnect to OST, OST(filter) should clear grant info of
1060 bulk write request, because the grant info while be sync between
1061 OSC and OST when reconnect, and we should ignore the grant info
1062 these of resend/replay write req.
1067 Description: Grant space more than avaiable space sometimes.
1068 Details : When then OST is about to be full, if two bulk writing from
1069 different clients came to OST. Accord the avaliable space of the
1070 OST, the first req should be permitted, and the second one
1071 should be denied by ENOSPC. But if the seconde arrived before
1072 the first one is commited. The OST might wrongly permit second
1073 writing, which will cause grant space > avaiable space.
1076 Frequency : when client is evicted
1078 Description: Grant might be wrongly erased when osc is evicted by OST
1079 Details : when the import is evicted by server, it will fork another
1080 thread ptlrpc_invalidate_import_thread to invalidate the
1081 import, where the grant will be set to 0. While the original
1082 thread will update the grant it got when connecting. So if
1083 the former happened latter, the grant will be wrongly errased
1084 because of this race.
1089 Description: Checking Stale with correct fid
1090 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1091 to check whether it is stale, because sometimes, we want the
1092 enqueue happened anyway, and op_data.fid2 will not be initialized.
1094 Severity : enhancement
1096 Description: update patchless client
1097 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1100 Frequency : only with 2.4 kernel
1102 Description: random memory corruption
1103 Details : size of struct ll_inode_info is to big for union inode.u and this
1104 can be cause of random memory corruption.
1109 Description: Memory leak in recovery
1110 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1111 It should also check obd_fail before fsfilt_start, otherwise if
1112 fsfilt_start return -EROFS,(failover mds during mds recovery).
1113 then the req will return with repmsg->transno = 0 and rc = EROFS.
1114 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1115 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1116 be freed no matter whether fsfilt_commit success or not.
1119 Frequency : only with huge numbers of clients
1121 Description: Prevents from taking the superblock lock in llap_from_page for
1123 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1124 for llap_from_page call in ll_removepage() prevents from taking
1125 the superblock lock for a soon died page.
1130 Description: Not check open intent error before release open handle
1131 Details : in some rare cases, the open intent error is not checked before
1132 release open handle, which may cause
1133 ASSERTION(open_req->rq_transno != 0), because it tries to release
1134 the failed open handle.
1139 Description: Set cat log bitmap only after create log success.
1140 Details : in some rare cases, the cat log bitmap is set too early. and it
1141 should be set only after create log success.
1145 Description: Accessing a block bevice can re-enable I/O when Lustre is
1146 tearing down a device.
1147 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1152 Description: service threads may hog cpus when there are a lot of requests
1153 Details : Insert cond_resched to give other threads a chance to use some CPU
1158 Description: the cat log was not initialized in recovery
1159 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1160 unlink log on mds will not be initialized until mds post
1161 recovery. And also in mds post recovery, the unlink log will
1162 initialization will be done asynchronausly, so there will be race
1163 between add unlink log and unlink log initialization.
1167 Description: brw_stats were being printed incorrectly
1168 Details : brw_stats were being printed as log2 but all of them were not
1169 recorded as log2. Also remove some code duplication arising from
1170 filter_tally_{read,write}.
1174 Frequency : rare, only in recovery.
1175 Description: ASSERTION(req->rq_type != LI_POISON) failed
1176 Details : imp_lock should be held while iterating over imp_sending_list for
1177 prevent destroy request after get timeout in ptlrpc_queue_wait.
1181 Description: replay-single.sh test 52 fails
1182 Details : A lock's skiplist need to be cleanup when it being unlinked
1183 from its resource list.
1188 Description: Short directio read returns full requested size rather than
1190 Details : Direct I/O operations should return actual amount of bytes
1191 transferred rather than requested size.
1193 Severity : enhancement
1195 Description: metadata RPC reduction (e.g. for rm performance)
1196 Details : decrease the amount of synchronous RPC between clients and servers
1197 by canceling conflicing lock before the operation on the client
1198 and packing thier handles into the main operation RPC to server.
1200 Severity : enhancement
1202 Description: add #ifdef HAVE_KERNEL_CONFIG_H
1203 Details : kernels from 2.6.19 not need include linux/config.h, but add
1204 include linux/autoconf.h in commpiler command line.
1206 Severity : enhancement
1208 Description: patchless client support for 2.6.22 kernel
1209 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
1210 removed. In this case we need drop using os depended interface to
1211 kmem_cache and use cfs_mem_cache API.
1216 Description: fix mal-formatted messages
1217 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
1221 Frequency : always in liblustre
1222 Description: wrong IS_ERR implementation in liblustre.h
1223 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
1228 Description: Correct condition for output debug message.
1229 Details : inode i_nlink equal zero is not enough for output message about
1230 disk corruption, i_ctime and i_mode should be also checked.
1234 Frequency : always in patchless client
1235 Description: add configure check for truncate_complete_page
1236 Details : improve checks for exported symbols. This allow run check without
1237 sources, but with Module.symvers shipped with kernel distribution.
1238 add check for truncate_complete_page used by patchless client.
1242 Description: sanity.sh test_77h fails with "test_77h file compare failed"
1243 Details : test_77h uses a file which was messed by other test case.
1247 Description: Not Check whether lov_tgts is NULL in some lov functions
1248 Details : Checking whether lov_tgts is NULL in some functions.
1252 Description: replace obdo_alloc() with OBDO_ALLOC macro
1253 Details : nothing special is done in obdo_alloc() function, and for
1254 debugging purpose, it needs to be replaced with macros.
1258 Description: bad return value and errno from fcntl call
1259 Details : In liblustre API, errno should be a negative value if error
1264 Description: ptlrpc_check_set() LBUG
1265 Details : In case of positive reply from server and failed client bulk
1266 callback after bulk transfer shouldn't LBUG, but process this
1267 request as erroneous.
1269 Severity : enhancement
1271 Description: VFS operations stats tool.
1272 Details : Tool which collects stats by tracking value written in pid,
1273 ppid, gid and uses llstat to generate output to plot graph using
1275 Updated lustre/utils/Makefile.am
1276 Added lustre/utils/ltrack_stats.c
1278 Severity : enhancement
1280 Description: 2.6.18 server support (lustre 1.6.1)
1281 Details : Support for 2.6.18 kernels on the server side.
1286 Description: ASSERTION(imp->imp_conn_current) failed
1287 Details : an assertion failure is hit if a client node boots and attempts to
1288 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
1291 Frequency : only for i686
1293 Description: 1.4.11 RC1 build fails for RHEL 4, i686
1294 Details : Fixed config variable for build.
1299 Description: Updated patchess for new RHEL4 kernel
1300 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
1301 Updated patch jbd-stats-2.6.9.patch
1302 Updated patch qsnet-rhel4-2.6.patch
1303 Updated patch quota-deadlock-on-pagelock-core.patch
1304 Updated patch vfs_intent-2.6-rhel4.patch
1305 Updated patch vfs_races-2.6-rhel4.patch
1306 Updated series file 2.6-rhel4-titech.series
1307 Updated series file 2.6-rhel4.series
1308 Updated kernel config files
1313 Description: lquota slave complains LBUG when reconnecting with mds
1315 Details : quota slave depends qctxt->lqc_import to send its quota request.
1316 This pointer will be invalid if mds did failover or broke its
1317 connect to osts, which leads to LBUG.
1320 Frequency : when qunit size is too small(less than 20M)
1322 Description: write is stopped by improper -EDQUOT
1323 Details : If the master is busy and qunit size is small enough(let's say 1M),
1324 the slave can not get quota from master on time, which will lead
1325 slave to trigger a -EQUOTA to client.
1330 Description: Deadlock during metadata tests
1331 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
1332 called with the per-dentry lock held.
1335 Frequency : SLES9 only
1337 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
1338 Details : set CONFIG_NR_CPUS to 128 instead of 8.
1340 Severity : enhancement
1342 Description: remove fs_prep_san_write operation and related patches
1343 Details : remove the ext3-san-jdike patches which are no longer useful.
1348 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
1349 Details : a disk corruption can cause the mballoc code to assert on a
1350 double free or other extent corruptions. Handle these with
1351 ext3_error() instead of assertions.
1354 Frequency : only with mballoc3 code and deep extent trees
1356 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
1357 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
1358 can cause the filesystem to be remounted read-only.
1363 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
1364 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
1365 few external users of this ioctl number is lfsck's call to
1366 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
1367 Add the old ioctl number to the handler so both old and new
1372 Description: parallel lock callbacks
1373 Details : Instead of sending blocking and completion callbacks as separated
1374 requests, adding them to a set and sending in parallel.
1378 Description: Disable most debugging by default
1379 Details : To improve performance, disable most logging (for debug purposes)
1380 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
1381 default, and HA includes fewer messages.
1385 Frequency : only run on patchless client.
1386 Description: use do_facet on sanity.sh for test handling recoverables errors
1387 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1391 Description: hash tables for lists of nids, connections and uuids
1392 Details : Hash tables noticeably help when a lot of clients connect to a
1393 server, to faster identify duplicate connections or reconnects,
1394 also to faster find export to evict in manual eviction case.
1398 Description: Sometimes, when the server evict a client, and the client will
1399 not be evicted as soon as possible.
1400 Details : In enqueue req, the error was returned by intent, instead of
1401 rq_status which make ptlrpc layer not detect this error, and
1402 does not evict the client. So enqueue error should be returned
1406 Frequency : only at startup
1408 Description: Delay client connections to MDT until fist MDT->OST connect
1409 Details : If a client tried to create a new file before the MDT had
1410 connected to any OSTs, the create would return EIO. Now
1411 the client will simply block until the MDT connects to the
1412 first OST and the create can succeed.
1415 Frequency : only for SLES9
1417 Description: Routinely utilize latest Quadrics drivers in CFS releases
1418 Details : Update patch qsnet-suse-2.6.patch.
1421 Frequency : only for sles10
1423 Description: Update patches for SLES 10 SP1 kernel.
1424 Details : Update the patch vfs_intent-2.6-sles10.patch.
1429 Description: Routinely utilize latest Quadrics drivers in CFS releases
1430 Details : Update patch qsnet-rhel4-2.6.patch.
1433 Frequency : at statup only
1435 Description: mds_lov_synchronize race leads to various problems
1436 Details : simultaneous MDT->OST connections at startup can cause the
1437 sync to abort, leaving the OSC in a bad state.
1442 Description: Using wrong pointer in osc_brw_prep_request
1443 Details : Access to array[-1] can produce panic if kernel compiled with
1444 CONFIG_PAGE_ALLOC enabled
1446 Severity : enhancement
1448 Description: Async OSC create to avoid the blocking unnecessarily.
1449 Details : If a OST has no remain object, system will block on the creating
1450 when need to create a new object on this OST. Now, ways use
1451 pre-created objects when available, instead of blocking on an
1452 empty osc while others are not empty. If we must block, we block
1453 for the shortest possible period of time.
1457 Frequency : only in recovery
1458 Description: Mark OST as early accessible if his start SYNC.
1459 Details : osc_precreate return flag early accessible if oscc marked as
1460 OSCC_FLAG_SYNC_IN_PROGRESS.
1465 Description: Sometimes precreate code can triger create object on wrong ost
1466 Details : Wrong protected or not not restored variables aftre precreate loop
1467 can produce creation object on wrong ost.
1470 Frequency : oss recovery
1472 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1474 Details : llog_commit_thread_count should be synced between llog_commit
1475 start and cleanup, so new llog_commit thread should not be started
1476 when llog_commit threads being stopped to avoid accessing some
1479 Severity : enhancement
1481 Description: Add printing inode info into message about error in writepage.
1483 --------------------------------------------------------------------------------
1485 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1490 Frequency : on some architectures
1492 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1493 Details : Uninitialized flags sometimes cause configuration commands to
1497 Frequency : patchless clients only
1499 Description: missing __iget() symbol export
1500 Details : The __iget() symbol export is missing. To avoid the need for
1501 this on patchless clients the deathrow inode reaper is turned
1502 off, and we depend on the VM to clean up old inodes. This
1503 dependency was during via the fix for bug 12181.
1508 Description: sanity.sh fail: test_52b
1509 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1510 incorrect inode's flags to client.
1512 --------------------------------------------------------------------------------
1514 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1516 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1517 INTEROPERATE with older versions automatically. In many cases a
1518 special upgrade step is needed. Please read the
1519 user documentation before upgrading any part of a 1.4.x system.
1520 * WARNING: Lustre configuration and startup changes are required with
1521 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1523 * Support for kernels:
1524 2.4.21-47.0.1.EL (RHEL 3),
1525 2.6.5-7.283 (SLES 9),
1526 2.6.9-42.0.10.EL (RHEL 4),
1527 2.6.12.6 vanilla (kernel.org),
1528 2.6.16.27-0.9 (SLES10)
1529 * Client support for unpatched kernels:
1530 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1531 2.6.16 - 2.6.19 vanilla (kernel.org),
1532 2.6.9-42.0.8EL (RHEL 4)
1533 * Recommended e2fsprogs version: 1.39.cfs6
1534 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1537 Severity : enhancement
1539 Description: Async OSC create to avoid the blocking unnecessarily.
1540 Details : If a OST has no remain object, system will block on the creating
1541 when need to create a new object on this OST. Now, ways use
1542 pre-created objects when available, instead of blocking on an
1543 empty osc while others are not empty. If we must block, we block
1544 for the shortest possible period of time.
1546 Severity : enhancement
1548 Description: MountConf
1549 Details : Lustre configuration is now managed via mkfs and mount
1550 commands instead of lmc and lconf. New obd types (MGS, MGC)
1551 are added for dynamic configuration management. See
1552 https://mail.clusterfs.com/wikis/lustre/MountConf for
1555 Severity : enhancement
1557 Description: dynamic OST addition
1558 Details : OSTs can now be added to a live filesystem
1560 Severity : enhancement
1562 Description: startup order invariance
1563 Details : MDTs and OSTs can be started in any order. Clients only
1564 require the MDT to complete startup.
1566 Severity : enhancement
1568 Description: parallel, asynchronous orphan cleanup
1569 Details : orphan cleanup is now performed in separate threads for each
1570 OST, allowing parallel non-blocking operation.
1572 Severity : enhancement
1574 Description: optimized stripe assignment
1575 Details : stripe assignments are now made based on ost space available,
1576 ost previous usage, and OSS previous usage, in order to try
1577 to optimize storage space and networking resources.
1579 Severity : enhancement
1581 Description: Permanently set tunables
1582 Details : All writable /proc/fs/lustre tunables can now be permanently
1583 set on a per-server basis, at mkfs time or on a live
1586 Severity : enhancement
1588 Description: Lustre message v2
1589 Details : Add lustre message format v2.
1591 Severity : enhancement
1593 Description: client OST exclusion list
1594 Details : Clients can be started with a list of OSTs that should be
1595 declared "inactive" for known non-responsive OSTs.
1599 Description: ENOENT returned for valid filehandle during dbench.
1600 Details : Check if a directory has children when invalidating dentries
1601 associated with an inode during lock cancellation. This fixes
1602 an incorrect ENOENT sometimes seen for valid filehandles during
1603 testing with dbench.
1606 Frequency : SFS test only (otherwise harmless)
1608 Description: SPEC SFS validation failure on NFS v2 over lustre.
1609 Details : Changes the blocksize for regular files to be 2x RPC size,
1610 and not depend on stripe size.
1612 Severity : enhancement
1614 Description: fine-grained SMP locking inside DLM
1615 Details : Improve DLM performance on SMP systems by removing the single
1616 per-namespace lock and replace it with per-resource locks.
1618 Severity : enhancement
1620 Description: don't hold multiple extent locks at one time
1621 Details : To avoid client eviction during large writes, locks are not
1622 held on multiple stripes at one time or for very large writes.
1623 Otherwise, clients can block waiting for a lock on a failed OST
1624 while holding locks on other OSTs and be evicted.
1626 Severity : enhancement
1628 Description: Multiple MD RPCs in flight.
1629 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1630 To avoid overly-overloading MDT, introduce a limit on number of
1631 MDT RPCs in flight for a single client and add /proc controls
1632 to adjust this limit.
1634 Severity : enhancement
1636 Description: client read/write statistics
1637 Details : Add client read/write call usage stats for performance
1638 analysis of user processes.
1639 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1640 file access. extents_stats shows chunk size distribution.
1641 extents_stats_per_process show chunk size distribution per
1644 Severity : enhancement
1646 Description: per-client statistics on server
1647 Details : Add ldlm and operations statistics for each client in
1648 /proc/fs/lustre/mds|obdfilter/*/exports/
1650 Severity : enhancement
1652 Description: improved MDT statistics
1653 Details : Add detailed MDT operations statistics in
1654 /proc/fs/lustre/mds/*/stats
1656 Severity : enhancement
1658 Description: VFS operations stats
1659 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1660 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1665 Description: Fix client-side osc byte counters
1666 Details : The osc read/write byte counters in
1667 /proc/fs/lustre/osc/*/stats are now working
1670 Frequency : always as root on SLES
1672 Description: Failure of copying files with lustre special EAs.
1673 Details : Client side always return success for setxattr call for lustre
1674 special xattr (currently only "trusted.lov").
1679 Description: Refcount LNET uuids
1680 Details : The global LNET uuid list grew linearly with every startup;
1681 refcount repeated list entries instead of always adding to
1684 Severity : enhancement
1686 Description: Dynamic service threads
1687 Details : Within a small range, start extra service threads
1688 automatically when the request queue builds up.
1691 Frequency : mixed-endian client/server environments
1693 Description: mixed-endian crashes
1694 Details : The new msg_v2 system had some failures in mixed-endian
1697 Severity : enhancement
1699 Description: Easy OST removal
1700 Details : OSTs can be permanently deactivated with e.g. 'lctl
1701 conf_param lustre-OST0001.osc.active=0'
1703 Severity : enhancement
1705 Description: MGS proc entries
1706 Details : Added basic proc entries for the MGS showing what filesystems
1709 Severity : enhancement
1711 Description: provide MGS failover
1712 Details : Added config lock reacquisition after MGS server failover.
1714 Severity : enhancement
1716 Description: add Linux 2.4 support
1717 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1721 Description: a large application tries to do I/O to the same resource and dies
1722 in the middle of it.
1723 Details : Check the req->rq_arrival time after the call to
1724 ost_brw_lock_get(), but before we do anything about
1725 processing it & sending the BULK transfer request. This
1726 should help move old stale pending locks off the queue as
1727 quickly as obd_timeout.
1730 Frequency : when an incorrect nid is specified during startup
1732 Description: ptlrpc connect to non-existant node causes kernel crash
1733 Details : LNET can't be re-entered from an event callback, which
1734 happened when we expire a message after the export has been
1735 cleaned up. Instead, hand the zombie cleanup off to another
1738 Severity : enhancement
1740 Description: plain/inodebits lock performance improvement
1741 Details : Grouping plain/inodebits in granted list by their request modes
1742 and bits policy, thus improving the performance of search through
1746 Frequency : only if OST filesystem is corrupted
1748 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1749 Details : for a short time RPCs with bulk IO are in the replay list,
1750 but replay of bulk IOs is unimplemented. If the OST filesystem
1751 is corrupted due to disk cache incoherency and then replay is
1752 started it is possible to trip an assertion. Avoid putting
1753 committed RPCs into the replay list at all to avoid this issue.
1756 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1758 Description: System hang on startup
1759 Details : This bug allowed the liblustre (e.g. catamount) client to
1760 return to the app before handling all startup RPCs. This
1761 could leave the node unresponsive to lustre network traffic
1762 and manifested as a server ptllnd timeout.
1764 Severity : enhancement
1766 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1767 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1768 boolean to control whether to print peer debug info when a
1769 client's RPC times out.
1772 Frequency : only for kernels with patches from Lustre below 1.4.3
1774 Description: Remove old rdonly API
1775 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1778 Frequency : only for devices with external journals
1780 Description: Set external device read-only also
1781 Details : During a commanded failover stop, we set the disk device
1782 read-only while the server shuts down. We now also set any
1783 external journal device read-only at the same time.
1786 Frequency : when upgrading from 1.4 while trying to change parameters
1788 Description: The wrong (new) MDC name was used when setting parameters for
1789 upgraded MDT's. Also allows changing of OSC (and MDC)
1790 parameters if --writeconf is specified at tunefs upgrade time.
1793 Frequency : when setting specific ost indicies
1795 Description: QOS code breaks on skipped indicies
1796 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1797 created with --index need not be sequential.
1799 Severity : enhancement
1801 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1802 Details : The uninit_groups feature works in conjunction with the kernel
1803 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1804 up the pass1 processing of e2fsck. This is a read-only feature
1805 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1806 mount filesystems that have had this feature enabled.
1808 Severity : enhancement
1810 Description: Improve multi-block allocation algorithm to avoid fragmentation
1811 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1812 allocation locality and avoid filesystem fragmentation.
1817 Description: lfs quota fails with deactivated OSTS
1818 Details : With this patch, three improvements are included:
1819 1. detete the softlimit in mds and osts when use "lfs quota".
1820 2. display the inaccurate data in the output of "lfs quota".
1821 3. try to get quota info when "lfs quota" is executed.
1823 ------------------------------------------------------------------------------
1825 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1827 * Support for kernels:
1828 2.4.21-47.0.1.EL (RHEL 3)
1829 2.6.5-7.283 (SLES 9)
1830 2.6.9-42.0.10.EL (RHEL 4)
1831 2.6.12.6 vanilla (kernel.org)
1832 2.6.16.27-0.9 (SLES 10)
1833 * Recommended e2fsprogs version: 1.39.cfs5
1835 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1839 Frequency : occasional, depends on client load and configuration
1840 Bugzilla : 12181, 12203
1841 Description: data loss for recently-modified files
1843 Details : In some cases it is possible that recently written or created
1844 files may not be written to disk in a timely manner (this should
1845 normally be within 30s unless client IO load is very high).
1846 The problem appears as zero-length files or files that are a
1847 multiple of 1MB in size after a client crash or client eviction
1848 that are missing data at the end of the file.
1850 This problem is more likely to be hit on clients where files are
1851 repeatedly created and unlinked in the same directory, clients
1852 have a large amount of RAM, have many CPUs, the filesystem has
1853 many OSTs, the clients are rebooted frequently, and/or the files
1854 are not accessed by other nodes after being written.
1856 The presence of the problem can be detected by looking at
1857 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1858 smaller than the second (nr_unused) then dirty files will not
1859 be flushed automatically to disk. "sync; sleep 10" should be
1860 run several times on the node before unmounting it to update
1861 Lustre (this is also safe to run on nodes without this problem).
1863 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1864 that can cause this same problem, so customers using that kernel
1865 also need to update the kernel in addition to Lustre. In order
1866 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1868 It is normal that files written just before a client crash (less
1869 than 30s) may not yet have been flushed to disk, even for local
1873 Frequency : frequent on thin XT3 nodes
1875 Description: UUID collision on thin XT3 Linux nodes
1876 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1877 randomly, since we relied on an insufficiently-seeded PRNG.
1882 Description: OSS hangs after "All ost request buffers busy"
1883 Details : A deadlock between quota and journal operations caused OSS
1884 hangs after printing "All ost request buffers busy."
1887 Frequency : always on liblustre builds
1889 Description: Cleanup compiler warnings on liblustre
1892 Frequency : always on liblustre builds on XT3
1894 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1898 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1899 > 32000 subdirectories
1900 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1901 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1902 subdirectory count crosses 32000. This will aid e2fsck to
1903 correctly handle more than 32000 subdirectories.
1906 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1908 Description: System hang on startup
1909 Details : This bug allowed the liblustre (e.g. catamount) client to
1910 return to the app before handling all startup RPCs. This
1911 could leave the node unresponsive to lustre network traffic
1912 and manifested as a server ptllnd timeout.
1914 Severity : enhancement
1916 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1917 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1918 boolean to control whether to print peer debug info when a
1919 client's RPC times out.
1924 Description: make O_SYNC working on 2.6 kernels
1925 Details : 2.6 kernels use different method for mark pages for write,
1926 so need add a code to lustre for O_SYNC work.
1931 Description: Failure to close file and release space on NFS
1932 Details : Put inode details into lock acquired in ll_intent_file_open.
1933 Use mdc_intent_lock in ll_intent_open to properly
1934 detect all kind of errors unhandled by mdc_enqueue.
1939 Description: proc file read during shutdown sometimes raced obd removal,
1941 Details : Add lock to prevent obd access after proc file removal.
1944 Frequency : Only for files larger than 4GB on 32-bit clients.
1946 Description: improperly doing page alignment of locks
1947 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1948 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1953 Description: RPCs being resent when they shouldn't be
1954 Details : Some RPCs that should not be resent are being resent. This
1955 can cause inconsistencies in the RPC state machine. Do not
1956 resend such requests.
1959 Frequency : rare, only with NFS export
1961 Description: Crash on NFS re-export node
1962 Details : under very unusual load conditions an assertion is hit in
1963 ll_intent_file_open()
1966 Frequency : only if OST filesystem is corrupted
1968 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1969 Details : for a short time RPCs with bulk IO are in the replay list,
1970 but replay of bulk IOs is unimplemented. If the OST filesystem
1971 is corrupted due to disk cache incoherency and then replay is
1972 started it is possible to trip an assertion. Avoid putting
1973 committed RPCs into the replay list at all to avoid this issue.
1978 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1979 Details : Large single O_DIRECT read and write calls can fail to allocate
1980 a sufficiently large buffer to process the request. In case of
1981 allocation failure the allocation is retried with a smaller
1982 buffer and broken into smaller requests.
1984 Severity : enhancement
1986 Description: Add -o localflock option to simulate old noflock behaviour.
1987 Details : This will achieve local-only flock/fcntl locks coherentness.
1992 Description: versioning check is incomplete
1993 Details : Checking the version difference of client vs. server, report
1994 error if the gap is too big.
1999 Description: add support PG_writeback bit
2000 Details : add support for PG_writeback bit for Lustre, for more carefull
2001 work with page cache in 2.6 kernel. This also fix some deadlocks
2002 and remove hack for work O_SYNC with 2.6 kernel.
2004 Severity : enhancement
2006 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2007 Details : The uninit_groups feature works in conjunction with the kernel
2008 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2009 up the pass1 processing of e2fsck. This is a read-only feature
2010 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2011 mount filesystems that have had this feature enabled.
2013 Severity : enhancement
2015 Description: Improve multi-block allocation algorithm to avoid fragmentation
2016 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2017 allocation locality and avoid filesystem fragmentation.
2019 ------------------------------------------------------------------------------
2021 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2023 * Support for kernels:
2024 2.6.9-42.0.3.EL (RHEL 4)
2025 2.6.5-7.276 (SLES 9)
2026 2.4.21-47.0.1.EL (RHEL 3)
2027 2.6.12.6 vanilla (kernel.org)
2028 2.6.16.21-0.8 (SLES10)
2029 * Recommended e2fsprogs version: 1.39.cfs2-0
2031 * The backwards-compatible /proc/sys/portals symlink has been removed
2032 in this release. Before upgrading, please ensure that you change
2033 any configuration scripts or /etc/sysctl.conf files that access
2034 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2035 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2036 in advance of the upgrade on any system running Lustre 1.4.6 or
2037 newer, since /proc/sys/lnet was added in that version.
2038 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2042 Frequency : only when quota is used
2044 Description: avoid scanning export list for quota master
2045 Details : Change the algorithms to avoid scanning export list in order
2046 to improve the efficiency.
2049 Frequency : MDS failover only, very rarely
2051 Description: "went back in time" messages on mds failover
2052 Details : The greatest transno may be lost when the current operation
2053 finishes with an error (transno==0) and the client's last_rcvd
2054 record is over-written. Save the greatest transno in the
2055 mds_last_transno for this case.
2058 Frequency : always for specific kernels and striping counts
2060 Description: client may get "Matching packet too big" without ACL support
2061 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2062 when trying to access files in certain configurations. The
2063 clients should in fact be denied when mounting because they do
2064 not understand ACLs.
2067 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2069 Description: many clients connecting with IO in progress causes connect timeouts
2070 Details : Avoid synchronous journal commits to avoid delays caused by many
2071 clients connecting/disconnecting when bulk IO is in progress.
2072 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2073 OST_IO_PORTAL to avoid delays behind potentially many pending
2077 Frequency : occasionally with multiple writers to a single file
2079 Description: shared writes to file may result in wrong size reported by stat()
2080 Details : Allow growing of kms when extent lock is cancelled
2083 Frequency : always with random mmap IO to multi-striped file
2085 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2086 Details : Only if the hole is at the end of OST object so that kms is too
2087 small. Fix is to increase kms accordingly in ll_nopage.
2090 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2092 Description: writes to a missing object would leak memory on the OST
2093 Details : If there is an inconsistency between the MDS and OST filesystems,
2094 such that the MDS references an object that doesn't exist, writes
2095 to that object will leak memory due to incorrect cleanup in the
2096 error handling path, eventually running out of memory on the OST.
2101 Description: Creating too long symlink causes lustre errors
2102 Details : Check symlink and name lengths before sending requests to MDS.
2105 Frequency : only if flock is enabled (not on by default)
2107 Description: posix locks not released on fd closure on 2.6.9+
2108 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2109 kernels, this caused such locks not to be released on fd close and
2110 then assertions on fs unmount about still used locks.
2113 Frequency : MDS failover only, very rarely
2115 Description: clients may get ASSERTION(granted_lock != NULL)
2116 Details : When request was taking a long time, and a client was resending
2117 a getattr by name lock request. The were multiple lock requests
2118 with the same client lock handle and
2119 mds_getattr_name->fixup_handle_for_resent_request found one of the
2120 lock handles but later failed with ASSERTION(granted_lock != NULL).
2125 Description: handle->h_buffer_credits > 0, assertion failure
2126 Details : h_buffer_credits is zero after truncate, causing assertion
2127 failure. This patch extends the transaction or creates a new
2131 Frequency : NFS re-export or patchless client
2132 Bugzilla : 11179, 10796
2133 Description: Crash on NFS re-export node (__d_move)
2134 Details : We do not want to hash the dentry if we don't have a lock.
2135 But if this dentry is later used in d_move, we'd hit uninitialised
2136 list head d_hash, so we just do this to init d_hash field but
2137 leave dentry unhashed.
2140 Frequency : NFS re-export or patchless client
2142 Description: NFS exports has problem with symbolic link
2143 Details : lustre client didn't properly install dentry when re-exported
2144 to NFS or running patchless client.
2147 Frequency : NFS re-export or patchless client
2149 Description: Various nfs/patchless fixes.
2150 Details : fixes reuse disconected alias for lookup process - this fixes
2151 warning "find_exported_dentry: npd != pd",
2152 fix permission error with open files at nfs.
2153 fix apply umask when do revalidate.
2156 Frequency : occasional
2158 Description: Crash on NFS re-export node
2159 Details : calling clear_page() on the wrong pointer triggered oops in
2160 generic_mapping_read().
2163 Frequency : rarely, using O_DIRECT IO
2165 Description: unaligned directio crashes client with LASSERT
2166 Details : check for unaligned buffers before trying any requests.
2169 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2171 Description: stale data returned from RAID cache
2172 Details : If only a small amount of IO is done to the RAID device before
2173 reading it again it is possible to get stale data from the RAID
2174 cache instead of reading it from disk.
2177 Frequency : always for sles10 kernel
2179 Description: sles10 support
2180 Details : ll_follow_link: compile fixes and using of nd_set_link
2181 under newer kernels.
2184 Frequency : depends on arch, kernel and compiler version, always on sles10
2187 Description: recursive or deep enough symlinks cause stack overflow
2188 Details : getting rid of large stack-allocated variable in
2192 Frequency : depends on hardware
2194 Description: lustre write performance loss in the SLES10 kernel
2195 Details : the performance loss is caused by using of write barriers in the
2196 ext3 code. The SLES10 kernel turns barrier support on by
2197 default. The fix is to undo that change for ldiskfs.
2199 ------------------------------------------------------------------------------
2201 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2203 * Support for kernels:
2204 2.6.9-42.0.3EL (RHEL 4)
2205 2.6.5-7.276 (SLES 9)
2206 2.4.21-47.0.1.EL (RHEL 3)
2207 2.6.12.6 vanilla (kernel.org)
2211 Frequency : quota enabled and large files being deleted
2213 Description: releasing more than 4GB of quota at once hangs OST
2214 Details : If a user deletes more than 4GB of files on a single OST it
2215 will cause the OST to spin in an infinite loop. Release
2216 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2221 Description: statfs data retrieved from /proc may be stale or zero
2222 Details : When reading per-device statfs data from /proc, in the
2223 {kbytes,files}_{total,free,avail} files, it may appear
2224 as zero or be out of date.
2227 Frequency : systems with MD RAID1 external journal devices
2229 Description: lconf's call to blkid is confused by RAID1 journal devices
2230 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2231 of returning all block devices that match the journal UUID.
2234 Frequency : always, for aggregate stripe size over 4GB
2236 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2237 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2238 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2241 Frequency : always if "lfs find" used on a local file/directory
2243 Description: "lfs find" segfaults if used on a local file/directory
2244 Details : The case where a directory component was not specified wasn't
2245 handled correctly. Handle this properly.
2248 Frequency : always on ppc64
2250 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2251 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2252 perform endianness conversion.
2255 Frequency : rarely (truncate to non-zero file size after write under load)
2256 Bugzilla : 10730, 10687
2257 Description: Files padded with zeros to next 4K multiple
2258 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2259 it is possible that files that are truncated to a non-zero size
2260 immediately after being written are filled with zero bytes beyond
2261 the truncated size. No file data is lost.
2263 Severity : enhancement
2265 Description: Allow recovery/failover for liblustre clients.
2266 Details : liblustre clients were unaware of failover configurations until
2269 Severity : enhancement
2271 Description: user file locks should fail when not mounting with flock option
2272 Details : Set up an error-returning stub in ll_file_operations.lock field
2273 to prevent incorrect behaviour when client is mounted without
2274 flock option. Also, set up properly f_op->flock field for
2278 Frequency : always on ia64
2280 Description: "lfs df" loops on printing out MDS statfs information
2281 Details : The obd_ioctl_data was not initialized and in some systems
2282 this caused a failure during the ioctl that did not return
2283 an error. Initialize the struct and return an error on failure.
2286 Frequency : SLES 9 only
2288 Description: Error of copying files with lustre special EAs as root
2289 Details : Client side always return success for setxattr call for lustre
2290 special xattr (currently only "trusted.lov").
2293 Frequency : rarely on clusters with both ia64+i386 clients
2295 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2296 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2297 clients are doing concurrent writes to the same file it is
2298 possible that the ia64 clients may LASSERT because the OST
2299 extent locks are not PAGE_SIZE aligned. Ensure that grown
2300 locks are always aligned on the request boundary.
2303 Frequency : specific use, occasional
2305 Description: Overwriting in use executable truncates on-disk binary image
2306 Details : If one node attempts to overwrite an executable in use by
2307 another node, we now correctly return ETXTBSY instead of
2308 truncating the file.
2310 Severity : enhancement
2312 Description: Async OSC create to avoid the blocking unnecessarily.
2313 Details : If a OST has no remain object, system will block on the creating
2314 when need to create a new object on this OST. Now, ways use
2315 pre-created objects when available, instead of blocking on an
2316 empty osc while others are not empty. If we must block, we block
2317 for the shortest possible period of time.
2322 Description: chmod on Lustre root is propagated to other clients
2323 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2324 invalid by the follow_mount time.
2327 Frequency : liblustre clients only
2329 Description: Race in 'instant cancel' lock handling could lead to such locks
2330 never to be granted in case of SMP MDS
2331 Details : Do not destroy not yet granted but cbpending locks in
2335 Frequency : replay/resend of open
2337 Description: non null lock assetion failure in mds_intent_policy
2338 Details : Trying to replay/resend lockless open requests resulted in
2339 mds_open() returning 0 with no lock. Now it sets a flag if
2340 a lock is going to be returned.
2342 Severity : enhancement
2344 Description: Checksum enhancements
2345 Details : New checksum enhancements allow for resending RPCs that failed
2348 Severity : enhancement
2350 Description: Tunables on number of dirty pages in cacche
2351 Details : Allow to set limit on number of dirty pages cached.
2356 Description: client crash on unmount - lock still has references
2357 Details : In some error handling cases it was possible to leak a lock
2358 reference on a client while accessing a file. This was not
2359 harmful to the client during operation, but would cause the
2360 client to crash when the filesystem is unmounted.
2363 Frequency : specific case, rare
2365 Description: ETXTBSY on mds though file not in use by client
2366 Details : ETXTBSY is no longer incorrectly returned when attempting to
2367 chmod or chown a directory that the user previously tried to
2368 execute or a currently-executing binary.
2371 Frequency : extremely rare except on liblustre-based clients
2373 Description: Lustre space not freed when files are deleted
2374 Details : Clean up open-unlinked files after client eviction. Previously
2375 the unlink was skipped and the files remained as orphans.
2380 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2381 Details : In some cases it was possible to send a blocking callback to
2382 a client doing a glimpse, even though that client didn't get
2383 a lock granted. When the glimpse lock is cancelled on the OST
2384 the freed lock is left on the waiting list and corrupted the list.
2387 Frequency : all core dumps
2389 Description: Broke core dumps to lustre
2390 Details : Negative dentry may be unhashed if parent does not have UPDATE
2391 lock, but some callers, e.g. do_coredump, expect dentry to be
2392 hashed after successful create, hash it in ll_create_it.
2394 ------------------------------------------------------------------------------
2396 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2398 * Support for kernels:
2399 2.6.9-42.0.2.EL (RHEL 4)
2400 2.6.5-7.276 (SLES 9)
2401 2.4.21-40.EL (RHEL 3)
2402 2.6.12.6 vanilla (kernel.org)
2406 Frequency : always on RHEL 3
2408 Description: Number of open files grows over time
2409 Details : The number of open files grows over time, whether or not
2410 Lustre is started. This was due to a filp leak introduced
2411 by one of our kernel patches.
2413 ------------------------------------------------------------------------------
2415 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2417 * Support for kernels:
2418 2.6.9-42.EL (RHEL 4)
2419 2.6.5-7.267 (SLES 9)
2420 2.4.21-40.EL (RHEL 3)
2421 2.6.12.6 vanilla (kernel.org)
2426 Bugzilla : 5719, 9635, 9792, 9684
2427 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2428 Details : If a server is under heavy load and cannot reply to new
2429 connection requests before the client resends the (re)connect,
2430 the connection handling code can behave badly if two service
2431 threads are concurrently handing separate (re)connections from
2432 the same client. Add better locking to the connection handling
2433 code, and ensure that only a single connection will be processed
2434 for a given client UUID, even if the lock is dropped.
2436 Severity : enhancement
2438 Description: add TCP zero-copy support to kernel
2439 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2440 sends if the hardware supports scatter-gather and checksumming.
2441 This allows socklnd to do client-write and server-read more
2442 efficiently and reduce CPU utilization from skbuf copying.
2445 Frequency : only if NFS exporting from client
2447 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2448 Details : If setting attributes on a file created under NFS that had
2449 never been opened it would be possible to oops the client
2450 if the file had no objects.
2454 Bugzilla : 9326, 10402, 10897
2455 Description: client crash in ptlrpcd_wake() thread when sending async RPC
2456 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2457 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2458 before ptlrpcd_wake() was called and the request was freed.
2461 Frequency : always for liblustre
2463 Description: liblustre client does MDS+OSTs setattr RPC for each write
2464 Details : When doing a write from a liblustre client, the client
2465 incorrectly issued an RPC to the MDS and each OST the file was
2466 striped over in order to update the timestamps. When writing
2467 with small chunks and many clients this could overwhelm the MDS
2468 with RPCs. In all cases it would slow down the write because
2469 these RPCs are unnecessary.
2471 Severity : enhancement
2473 Description: allow number of MDS service threads to be changed at module load
2474 Details : It is now possible to change the number of MDS service threads
2475 running. Adding "options mds mds_num_threads={N}" to the MDS's
2476 /etc/modprobe.conf will set the number of threads for the next
2477 time Lustre is restarted (assuming the "mds" module is also
2478 reloaded at that time). The default number of threads will
2479 stay the same, 32 for most systems.
2484 Description: OST crash if filesystem is unformatted or corrupt
2485 Details : If an OST is started on a device that has never been formatted
2486 or if the filesystem is corrupt and cannot even mount then the
2487 error handling cleanup routines would dereference a NULL pointer.
2492 Description: NULL pointer deref in llap_from_page.
2493 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2494 incorrect) mapping if the page was truncated/reclaimed while it was
2495 searched for. Check for this condition and skip such pages when
2496 doing readahead. Introduce extra check to llap_from_page() to
2497 verify page->mapping->host is non-NULL (so page is not anonymous).
2500 Frequency : Sometimes when using sys_sendfile
2502 Description: "page not covered by a lock" warnings from ll_readpage
2503 Details : sendfile called ll_readpage without right page locks present.
2504 Now we introduced ll_file_sendfile that does necessary locking
2505 around call to generic_file_sendfile() much like we do in
2509 Frequency : with certain MDS communication failures at client mount time
2511 Description: NULL pointer deref after failed client mount
2512 Details : a client connection request may delayed by the network layer
2513 and not be sent until after the PTLRPC layer has timed out the
2514 request. If the client fails the mount immediately it will try
2515 to clean up before the network times out the request. Add a
2516 reference from the request import to the obd device and delay
2517 the cleanup until the network drops the request.
2520 Frequency : occasionally during client (re)connect
2522 Description: assertion failure during client (re)connect
2523 Details : processing a client connection request may be delayed by the
2524 client or server longer than the client connect timeout. This
2525 causes the client to resend the connection request. If the
2526 original connection request is replied in this interval, the
2527 client may trip an assertion failure in ptlrpc_connect_interpret()
2528 which thought it would be the only running connect process.
2531 Frequency : only with obd_echo servers and clients that are rebooted
2533 Description: kernel BUG accessing uninitialized data structure
2534 Details : When running an obd_echo server it did not start the ping_evictor
2535 thread, and when a client was evicted an uninitialized data
2536 structure was accessed. Start the ping_evictor in the RPC
2537 service startup instead of the OBD startup.
2539 Severity : enhancement
2540 Bugzilla : 10193 (patchless)
2541 Description: Remove dependency on various unexported kernel interfaces.
2542 Details : No longer need reparent_to_init, exit_mm, exit_files,
2543 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2546 Frequency : rare (only users of deprecated and unsupported LDAP config)
2548 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2549 Details : LDAP apparently contains 'lustreName' attributes instead of
2550 'name'. A simple remapping of the name is sufficient.
2553 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2555 Description: waiting_locks_callback trips kernel BUG if client is evicted
2556 Details : Running with the dump_on_timeout debug flag turned on makes
2557 it possible that the waiting_locks_callback() can try to dump
2558 the Lustre kernel debug logs from an interrupt handler. Defer
2559 this log dumping to the expired_lock_main() thread.
2561 Severity : enhancement
2563 Description: Support NFS exporting on 2.6 kernels.
2564 Details : Implement non-rawops metadata methods for NFS server to use without
2565 changing NFS server code.
2568 Frequency : very rare (synthetic metadata workload only)
2570 Description: two racing renames might cause an MDS thread to deadlock
2571 Details : Running the "racer" program may cause one MDS thread to rename
2572 a file from being the source of a rename to being the target of
2573 a rename at exactly the same time that another thread is doing
2574 so, and the second thread has already enqueued these locks after
2575 doing a lookup of the target and is trying to relock them in
2576 order. Ensure that we don't try to re-lock the same resource.
2579 Frequency : only very large systems with liblustre clients
2581 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2582 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2583 all OSTs in parallel. This allows the eviction of stale liblustre
2584 clients to proceed much faster than if they were done in series,
2585 and also offers similar improvements for other set_info RPCs.
2590 Description: excessive CPU usage during initial read phase on client
2591 Details : During the initial read phase on a client, it would agressively
2592 retry readahead on the file, consuming too much CPU and impacting
2593 performance (since 1.4.5.8). Improve the readahead algorithm
2594 to avoid this, and also improve some other common cases (read
2595 of small files in particular, where "small" is files smaller than
2596 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2601 Description: MDS crash when receiving packet with unknown intent.
2602 Details : Do not LBUG in unknown intent case, just return -EFAULT
2604 Severity : enhancement
2605 Bugzilla : 9293, 9385
2606 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2607 Details : Do not serialize getattr (non-intent version) and statfs.
2610 Frequency : occasional, when OST network is overloaded/intermittent
2612 Description: client evicted by OST after bulk IO timeout
2613 Details : If a client sends a bulk IO request (read or write) the OST
2614 may evict the client if it is unresposive to its data GET/PUT
2615 request. This is incorrect if the network is overloaded (takes
2616 too long to transfer the RPC data) or dropped the OST GET/PUT
2617 request. There is no need to evict the client at all, since
2618 the pinger and/or lock callbacks will handle this, and the
2619 client can restart the bulk request.
2622 Frequency : Always when mmapping file with no objects
2624 Description: client crashes when mmapping file with no objects
2625 Details : Check that we actually have objects in a file before doing any
2626 operations on objects in ll_vm_open, ll_vm_close and
2632 Description: Request leak when working with deleted CWD
2633 Details : Introduce advanced request refcount tracking for requests
2634 referenced from lustre intent.
2636 Severity : Enhancement
2638 Description: Cache open file handles on client.
2639 Details : MDS now will return special lock along with openhandle, if
2640 requested and client is allowed to hold openhandle, even if unused,
2641 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2642 closing files for every read/write openration.
2644 Severity : Enhancement
2646 Description: Cache open negative dentries on client when possible.
2647 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2648 negative dentries on lock revocation.
2653 Description: Remounting a client read-only wasn't possible with a zconf mount
2654 Details : It wasn't possible to remount a client read-only with llmount.
2656 Severity : enhancement
2657 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2658 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2659 MPICH is included to add Lustre-specific ADIO interfaces.
2660 This is based closely on the UFS ADIO layer and only differs
2661 in file creation, in order to allow the OST striping to be set.
2662 This is user-contributed code and not supported by CFS.
2667 Description: extended inode attributes (immutable, append-only) work improperly
2668 when 2.4 and 2.6 kernels are used on client/server or vice versa
2669 Details : Introduce kernel-independent values for these flags.
2671 Severity : enhancement
2674 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2675 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2676 granularity. Now, Fractional MB granularity is be supported,
2677 this is very useful for low memory system.
2679 Severity : enhancement
2681 Description: Getattr by fid
2682 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2683 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2684 have locks and avoids getattr rpc requests for it on every stat.
2687 Frequency : Always, for filesystems larger than 2TB
2689 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2690 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2691 filesystems larger than 2TB to be mounted with mballoc enabled.
2694 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2696 Description: filesystem corruption for non-standard kernels and very large OSTs
2697 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2698 filesystems larger than 2TB is mounted then the kernel will
2699 silently corrupt the start of the filesystem. CONFIG_LBD is
2700 enabled for all CFS-supported kernels, but the possibility of
2701 this happening with a modified kernel config exists.
2703 Severity : enhancement
2705 Description: add client O_DIRECT support for 2.6 kernels
2706 Details : It is now possible to do O_DIRECT reads and writes to files
2707 in the Lustre client mountpoint on 2.6 kernel clients.
2709 Severity : enhancement
2711 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2712 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2713 parallel, not waiting for response from every OST before sending
2714 a rpc to the next OST.
2719 Description: setattr vs write race when updating file timestamps
2720 Details : Client processes that update a file timestamp into the past
2721 right after writing to the file (e.g. tar) it is possible that
2722 the updated file modification time can be reset to the current
2723 time due to a race between processing the setattr and write RPC.
2725 Severity : enhancement
2727 Description: Bring 'lfs find' closer in line with regular Linux find.
2728 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2729 -print0 options and obtains all the needed info through the lustre
2732 Severity : enhancement
2734 Description: support up to 1024 configured devices on one node
2735 Details : change obd_dev array from statically allocated to dynamically
2736 allocated structs as they are first used to reduce memory usage
2741 Description: Flush dirty partially truncated pages during truncate
2742 Details : Immediatelly flush partially truncated pages in filter_setattr,
2743 this way we completely avoid having any pages in page cache on OST
2744 and can retire ugly workarounds during writes to flush such pages.
2749 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2750 Details : protect inode from truncation within vfs_unlink() context
2751 just take a reference before calling vfs_unlink() and release it
2752 when parent's i_sem is free.
2755 Frequency : always, if extents are used on OSTs
2757 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2758 Details : OSTs using the extents format would not zero the high 16 bits of
2759 the index physical block number. This is not a problem for any
2760 OST filesystems smaller than 16TB, and no kernels support ext3
2761 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2762 new/modified files) and can be fixed for existing filesystems
2763 with e2fsprogs-1.39-cfs1.
2768 Description: import connection selection may be incorrect if timer wraps
2769 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2770 import connections to be ignored if the 32-bit jiffies counter
2771 wraps. Use a 64-bit jiffies counter.
2774 Frequency : during server recovery
2776 Description: crash after server is denying duplicate export
2777 Details : If clients are resending connect requests to the server, the
2778 server refuses to allow a client to connect multiple times.
2779 Fixed a bug in the handling of this case.
2782 Frequency : very large clusters immediately after boot
2784 Description: LNET request buffers exhausted under heavy short-term load
2785 Details : If a large number of client requests are generated on a service
2786 that has previously never seen so many requests it is possible
2787 that the request buffer growth cannot keep up with the spike in
2788 demand. Instead of dropping incoming requests, they are held in
2789 the LND until the RPC service can accept more requests.
2792 Frequency : Sometimes during replay
2794 Description: Assertion failure in ll_local_open after replay.
2795 Details : If replay happened on an open request reply before we were able
2796 to set replay handler, reply will become not swabbed tripping the
2797 assertion in ll_local_open. Now we set the handler right after
2798 recognising of open request
2801 Frequency : very rare
2803 Description: kernel reports "badness in vsnprintf"
2804 Details : Reading from the "recovery_status" /proc file in small chunks
2805 may cause a negative length in lprocfs_obd_rd_recovery_status()
2806 call to vsnprintf() (which is otherwise harmless). Exit early
2807 if there is no more space in the output buffer.
2809 Severity : enhancement
2811 Description: clear OBD RPC statistics by writing to them
2812 Details : It is now possible to clear the OBD RPC statistics by writing
2813 to the "stats" file.
2818 Description: Client mtime is not the same on different clients after utimes
2819 Details : In some cases, the client was using the utimes() syscall on
2820 a file cached on another node. The clients now validate the
2821 ctime from the MDS + OSTs to determine which one is right.
2826 Description: Inability to activate failout mode
2827 Details : lconf script incorrectly assumed that in python string's numeric
2828 value is used in comparisons.
2831 Frequency : always with multiple stripes per file
2833 Description: Inefficient object allocation for mutli-stripe files
2834 Details : When selecting which OSTs to stripe files over, for files with
2835 a stripe count that divides evenly into the number of OSTs,
2836 the MDS is always picking the same starting OST for each file.
2837 Return the OST selection heuristic to the original design.
2842 Description: mount failures may take full timeout to return an error
2843 Details : Under some heavy load conditions it is possible that a
2844 failed mount can wait for the full obd_timeout interval,
2845 possibly several minutes, before reporting an error.
2846 Instead return an error as soon as the status is known.
2848 ------------------------------------------------------------------------------
2850 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2852 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2853 INTEROPERATE with older versions automatically. Please read the
2854 user documentation before upgrading any part of a live system.
2855 * WARNING: Lustre networking configuration changes are required with
2856 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2859 * Support for kernels:
2860 2.6.9-22.0.2.EL (RHEL 4)
2861 2.6.5-7.244 (SLES 9)
2862 2.6.12.6 vanilla (kernel.org)
2865 Severity : enhancement
2866 Bugzilla : 7981/8208
2867 Description: Introduced Lustre Networking (LNET)
2868 Details : LNET is new networking infrastructure for Lustre, it includes
2869 a reorganized network configuration mode (see the user
2870 documentation for full details) as well as support for routing
2871 between different network fabrics. Lustre Networking Devices
2872 (LNDs) for the supported network fabrics have also been
2873 created for this new infrastructure.
2875 Severity : enhancement
2876 Description: Introduced Access control lists
2877 Details : clients can set ACLs on files and directories in order to have
2878 more fine-grained permissions than the standard Unix UGO+RWX.
2879 The MDS must be started with the "-o acl" mount option.
2881 Severity : enhancement
2882 Description: Introduced filesystem quotas
2883 Details : Administrators may now establish per-user quotas on the
2886 Severity : enhancement
2888 Description: Configuration change for the XT3
2889 The PTLLND is now used to run Lustre over Portals on the XT3
2890 The configure option(s) --with-cray-portals are no longer used.
2891 Rather --with-portals=<path-to-portals-includes> is used to
2892 enable building on the XT3. In addition to enable XT3 specific
2893 features the option --enable-cray-xt3 must be used.
2898 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2899 Details : A race between allocating a new llcd and re-getting the llcd_lock
2900 allowed another thread to grab newly-allocated llcd.
2902 Severity : enhancement
2904 Description: 2.6 OST async journal commit and locking fix to improve performance
2905 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2906 2.6 kernels are now async as they already were in 2.4 kernels so
2907 that they can commit concurrently with the network bulk transfer.
2908 For block-allocated files the filter allocation semaphore is held
2909 to avoid filesystem fragmentation during allocation. BKL lock
2910 removed for 2.6 xattr operations where it is no longer needed.
2915 Description: lconf incorrectly determined whether two IP networks could talk
2916 Details : In some more complicated routing and multiple-network
2917 configurations, lconf will avoid trying to make a network
2918 connection to a disjoint part of the IP space. It was doing the
2919 math incorrectly for one set of cases.
2924 Description: Fix for potential infinite loop processing records in an llog.
2925 Details : If an llog record is corrupted/zeroed, it is possible to loop
2926 forever in llog_process(). Validate the llog record length
2927 and skip the remainder of the block on error.
2930 Frequency : occasional (liblustre only)
2932 Description: liblustre could not open files whose last component is a symlink
2933 Details : sysio_path_walk() would incorrectly pass the open intent to
2934 intermediate path components.
2937 Frequency : rare (liblustre only with non-standard tuning)
2938 Bugzilla : 7201 (7350)
2939 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2940 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2941 after it has been released, only do this for referenced locks
2943 Severity : enhancement
2945 Description: specifying an (invalid) directory default stripe_size of -1
2946 would reset the directory default striping
2947 Details : stripe_size -1 was used internally to signal directory stripe
2948 removal, now use "all default" to signal dir stripe removal
2949 as a directory striping of "all default" is not useful
2952 Frequency : common for large clusters running liblustre clients
2954 Description: doing an ls when liblustre clients are running is slow
2955 Details : sending a glimpse AST to a liblustre client waits for every AST
2956 to time out, as liblustre clients will not respond. Since they
2957 cannot cache data we refresh the OST lock LVB from disk instead.
2959 Severity : enhancement
2961 Description: doing an ls at the same time as file IO can be slow
2962 Details : enqueue and other "small" requests can be blocked behind many
2963 large IO requests. Create a new OST IO portal for non-IO
2964 requests so they can be processed faster.
2967 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2969 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2970 created file with mode 000
2971 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2972 allow an ftruncate() as a non-root user to a file with mode 000.
2973 Lustre now respects this flag to disable mode checks when
2974 truncating a file owned by the user
2977 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2980 Description: Revoking locks from clients that went dead or catatonic might take
2982 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2983 cancellation of such locks instant on servers without waiting for
2984 any reply from clients. Clients drops these locks when cancel
2985 notification from server is received without replying.
2988 Frequency : liblustre-only, when liblustre client dies or becomes busy
2990 Description: Doing ls on Linux clients can take a long time with active
2992 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2993 granting such locks to it in the first place if possible. Locks
2994 are taken by proxy on the OST during the read or write and
2995 dropped immediately afterward. Add connect flags handling, do
2996 not grant locks to liblustre clients for glimpse ASTs.
2998 Severity : enhancement
3000 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3001 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3002 sequential reads and behaves badly with applications doing random
3003 reads. Improve it by reading ahead at least read region, and
3004 avoiding excessive large RPC for small reads.
3006 Severity : enhancement
3008 Description: Creating more than 1000 files for a single job may cause a load
3009 imbalance on the OSTs if there are also a large number of OSTs.
3010 Details : qos_prep_create() uses an OST index reseed value that is an
3011 even multiple of the number of available OSTs so that if the
3012 reseed happens in the middle of the object allocation it will
3013 still utilize the OSTs as uniformly as possible.
3018 Description: OST or MDS may oops in ping_evictor_main()
3019 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3020 but doesn't restart at beginning of obd_exports_timed list
3023 Severity : enhancement
3025 Description: improve by-nid export eviction on the MDS and OST
3026 Details : allow multiple exports with the same NID to be evicted at one
3027 time without re-searching the exports list.
3030 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3032 Description: MDS may oops in groups_free()
3033 Details : in rare race conditions a newly allocated group_info struct is
3034 freed again, and this can be NULL. The 2.4 compatibility code
3035 for groups_free() checked for a NULL pointer, but 2.6 did not.
3038 Frequency : common for liblustre clients doing little filesystem IO
3039 Bugzilla : 9352, 7313
3040 Description: server may evict liblustre clients accessing contended locks
3041 Details : if a client is granted a lock or receives a completion AST
3042 with a blocking AST already set it would not reply to the AST
3043 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3046 Frequency : lfs setstripe, only systems with more than 160 OSTs
3048 Description: unable to set striping with a starting offset beyond OST 160
3049 Details : llapi_create_file() incorrectly limited the starting stripe
3050 index to the maximum single-file stripe count.
3053 Frequency : LDAP users only
3055 Description: lconf did not handle in-kernel recovery with LDAP properly
3056 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3058 Severity : enhancement
3060 Description: bind OST threads to NUMA nodes to improve performance
3061 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3062 node and do their allocations there to localize memory access
3064 Severity : enhancement
3066 Description: llmount can determine client NID directly from Myrinet (GM)
3067 Details : the client NID code from gmnalnid was moved directly into
3068 llmount, removing the need to use this or specifying the
3069 client NID explicitly when mounting GM clients with zeroconf
3072 Frequency : if client is started with down MDS
3074 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3075 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3076 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3077 ptlrpc_import_delayed_req() because we didn't check if the
3078 request was interrupted, and we also didn't break out of the
3079 event loop if there was no timeout
3084 Description: data loss during non-page-aligned writes to a single file from
3085 both multiple nodes and multiple threads on one node at same time
3086 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3087 inconsistency led to false short-reads, that were cached and later
3088 used by ->prepare_write() to fill in partially written page,
3089 leading to data loss.
3092 Frequency : always, if lconf --abort_recovery used
3094 Description: lconf --abort_recovery fails with 'Operation not supported'
3095 Details : lconf was attempting to abort recovery on the MDT device and not
3098 Severity : enhancement
3100 Description: remove cleanup logs
3101 Details : replace lconf-generated cleanup logs with lustre internal
3102 cleanup routines. Eliminates the need for client-cleanup and
3105 Severity : enhancement
3107 Description: add support for EAs (user and system) on lustre filesystems
3108 Details : it is now possible to store extended attributes in the Lustre
3109 client filesystem, and with the user_xattr mount option it
3110 is possible to allow users to store EAs on their files also
3112 Severity : enhancement
3114 Description: Add possibility (config option) to show minimal available OST free
3116 Details : When compiled with --enable-mindf configure option, statfs(2)
3117 (and so, df) will return least minimal free space available from
3118 all OSTs as amount of free space on FS, instead of summary of
3119 free spaces of all OSTs.
3121 Severity : enhancement
3123 Description: do not expand extent locks acquired on OST-side
3124 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3125 by server: they are not cached anyway.
3128 Frequency : when mmap is used/binaries executed from Lustre
3130 Description: Unmmap pages before throwing them away from read cache.
3131 Details : llap_shrink cache now attempts to unmap pages before discarding
3132 them (if unmapping failed - do not discard). SLES9 kernel has
3133 extra checks that trigger if this unmapping is not done first.
3138 Description: lconf didn't resolve symlinks before checking to see whether a
3139 given mountpoint was already in use
3142 Frequency : when migrating failover services
3143 Bugzilla : 6395, 9514
3144 Description: When migrating a subset of services from a node (e.g. failback
3145 from a failover service node) the remaining services would
3146 time out and evict clients.
3147 Details : lconf --force (implied by --failover) sets the global obd_timeout
3148 to 5 seconds in order to quickly disconnect, but this caused
3149 other RPCs to time out too quickly. Do not change the global
3150 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3152 Severity : enhancement
3153 Frequency : if MDS is started with down OST
3154 Bugzilla : 9439,5706
3155 Description: Allow startup/shutdown of an MDS without depending on the
3156 availability of the OSTs.
3157 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3158 Add appropriate locking and lov-osc refcounts for safe
3159 cleaning. Add osc abort_inflight calls in case the
3160 synchronize never started.
3163 Frequency : occasional (Cray XT3 only)
3165 Description: root not authorized to access files in CRAY_PORTALS environment
3166 Details : The client process capabilities were not honoured on the MDS in
3167 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3168 been accessed by an authorized user then root was able to access
3169 the file on the local client also. The root user capabilities
3170 are now allowed on the MDS, as this environment has secure UID.
3173 Frequency : occasional
3175 Description: ldiskfs "too long searching" message happens too often
3176 Details : A debugging message (otherwise harmless) prints too often on
3177 the OST console. This has been reduced to only happen when
3178 there are fragmentation problems on the filesystem.
3183 Description: Division by zero in statfs when all OSCs are inactive
3184 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3185 lov_statfs divides by value returned by lov_get_stripecnt().
3189 Bugzilla : 9489, 3273
3190 Description: First write from each client to each OST was only 4kB in size,
3191 to initialize client writeback cache, which caused sub-optimal
3192 RPCs and poor layout on disk for the first writen file.
3193 Details : Clients now request an initial cache grant at (re)connect time
3194 and so that they can start streaming writes to the cache right
3195 away and always do full-sized RPCs if there is enough data.
3196 If the OST is rebooted the client also re-establishes its grant
3197 so that client cached writes will be honoured under the grant.
3202 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3203 Details : Now I/O RPCs go to different portal number and (presumably) fast
3204 lock requests (and glimses) and other RPCs get their own service
3205 threads pool that should be able to service those RPCs
3208 Severity : enhancement
3210 Description: Ability to exchange lustre version between client and servers and
3211 issue warnings at client side if client is too old. Also for
3212 liblustre clients there is ability to refuse connection of too old
3214 Details : New 'version' field is added to connect data structure that is
3215 filled with version info. That info is later checked by server and
3219 Frequency : rare, liblustre only.
3220 Bugzilla : 9296, 9581
3221 Description: Two simultaneous writes from liblustre at offset within same page
3222 might proceed at the same time overwriting eachother with stale
3224 Details : I/O lock withing llu_file_prwv was released too early, before data
3225 actually was hitting the wire. Extended lock-holding time until
3226 server acknowledges receiving data.
3229 Frequency : extremely rare. Never observed in practice.
3231 Description: avoid generating lustre_handle cookie of 0.
3232 Details : class_handle_hash() generates handle cookies by incrementing
3233 global counter, and can hit 0 occasionaly (this is unlikely, but
3234 not impossible, because initial value of cookie counter is
3235 selected randonly). Value of 0 is used as a sentinel meaning
3236 "unassigned handle" --- avoid it. Also coalesce two critical
3237 sections in this function into one.
3239 Severity : enhancement
3241 Description: allow liblustre clients to delegate truncate locking to OST
3242 Details : To avoid overhead of locking, liblustre client instructs OST to
3243 take extent lock in ost_punch() on client's behalf. New connection
3244 flag is added to handle backward compatibility.
3246 Severity : enhancement
3247 Bugzilla : 4928, 7341, 9758
3248 Description: allow number of OST service threads to be specified
3249 Details : a module parameter allows the number of OST service threads
3250 to be specified via "options ost ost_num_threads={N}" in the
3251 OSS's /etc/modules.conf or /etc/modprobe.conf.
3255 Bugzilla : 6146, 9635, 9895
3256 Description: servers crash with bad pointer in target_handle_connect()
3257 Details : In rare cases when a client is reconnecting it was possible that
3258 the connection request was the last reference for that export.
3259 We would temporarily drop the export reference and get a new
3260 one, but this may have been the last reference and the export
3261 was just destroyed. Get new reference before dropping old one.
3263 Severity : enhancement
3264 Frequency : if client is started with failover MDS
3266 Description: Allow multiple MDS hostnames in the mount command
3267 Details : Try to read the configuration from all specified MDS
3268 hostnames during a client mount in case the "primary"
3271 Severity : enhancement
3273 Description: Stop sending data to evicted clients as soon as possible.
3274 Details : Check if the client we are about to send or are sending data to
3275 was evicted already. (Check is done every second of waiting,
3276 for which l_wait_event interface was extended to allow checking
3277 of exit condition at specified intervals).
3280 Frequency : rare, normally only when NFS exporting is done from client
3282 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3284 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3287 Frequency : always, if srand() is called before liblustre initialization
3289 Description: Liblustre uses system PRNG disturbing its usage by user application
3290 Details : Introduce internal to lustre fast and high-quality PRNG for
3291 lustre usage and make liblustre and some other places in generic
3292 lustre code to use it.
3294 Severity : enhancement
3295 Bugzilla : 9477, 9557, 9870
3296 Description: Verify that the MDS configuration logs are updated when xml is
3297 Details : Check if the .xml configuration logs are newer than the config
3298 logs stored on the MDS and report an error if this is the case.
3299 Request --write-conf, or allow starting with --old_conf.
3301 Severity : enhancement
3303 Description: Handle symlinks in the path when checking if Lustre is mounted.
3304 Details : Resolve intermediate symlinks when checking if a client has
3305 mounted a filesystem to avoid duplicate client mounts.
3310 Description: lconf can hit an error exception but still return success.
3311 Details : The lconf command catches the Command error exception at the top
3312 level script context and will exit with the associated exit
3313 status, but doesn't ensure that this exit status is non-zero.
3318 Description: failure of ptlrpc thread startup can cause oops
3319 Details : Starting a ptlrpc service thread can fail if there are a large
3320 number of threads or the server memory is very fragmented.
3321 Handle this without oopsing.
3324 Frequency : always, only if liblustre and non-default acceptor port was used
3326 Description: liblustre cannot connect to servers with non-default acceptor port
3327 Details : tcpnal_set_default_params() was not called and was therefore
3328 ignoring the environment varaible TCPNAL_PORT, as well as other
3329 TCPNAL_ environment variables
3334 Description: two objects could be created on the same OST for a single file
3335 Details : If an OST is down, in some cases it was possible to create two
3336 objects on a single OST for a single file. No problems other
3337 than potential performance impact and spurious error messages.
3341 Bugzilla : 5681, 9562
3342 Description: Client may oops in ll_unhash_aliases
3343 Details : Client dcache may become inconsistent in race condition.
3344 In some cases "getcwd" can fail if the current directory is
3350 Description: Inode refcounting problems in NFS export code
3351 Details : link_raw functions used to call d_instantiate without obtaining
3352 extra inode reference first.
3356 Bugzilla : 9942, 9903
3357 Description: Referencing freed requests leading to crash, memleaks with NFS.
3358 Details : We used to require that call to ll_revalidate_it was always
3359 followed by ll_lookup_it. Also with revalidate_special() it is
3360 possible to call ll_revalidate_it() twice for the same dentry
3361 even if first occurence returned success. This fix changes semantic
3362 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3363 reference on a request referred from the intent.
3364 ll_intent_release() then releases such a request.
3367 Frequency : rare, normally benchmark loads only
3369 Description: unlinked inodes were kept in memory on the client
3370 Details : If a client is repeatedly creating and unlinking files it
3371 can accumulate a lot of stale inodes in the inode slab cache.
3372 If there is no other client load running this can cause the
3373 client node to run out of memory. Instead flush old inodes
3374 from client cache that have the same inode number as a new inode.
3377 Frequency : SLES9 2.6.5 kernel and long filenames only
3378 Bugzilla : 9969, 10379
3379 Description: utime reports stale NFS file handle
3380 Details : SLES9 uses out-of-dentry names in some cases, which confused
3381 the lustre dentry revalidation. Change it to always use the
3385 Frequency : rare, unless heavy write-truncate concurrency is continuous
3386 Bugzilla : 4180, 6984, 7171, 9963, 9331
3387 Description: OST becomes very slow and/or deadlocked during object unlink
3388 Details : filter_destroy() was holding onto the parent directory lock
3389 while truncating+unlinking objects. For very large objects this
3390 may block other threads for a long time and slow overall OST
3391 responsiveness. It may also be possible to get a lock ordering
3392 deadlock in this case, or run out of journal credits because of
3393 the combined truncate+unlink. Solution is to do object truncate
3394 first in one transaction without parent lock, and then do the
3395 final unlink in a new transaction with the parent lock. This
3396 reduces the lock hold time dramatically.
3399 Frequency : rare, 2.4 kernels only
3401 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3402 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3403 before unmounting their filesystem in order to avoid deadlock.
3404 The kernel_locked() function in 2.4 kernels only checks whether
3405 the kernel lock is held, not whether it is this process that is
3406 holding it as 2.6 kernels do.
3411 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3412 Details : The client ptlrpc code may be trying to reconnect to a down
3413 server before a previous connection attempt has timed out.
3414 Increase the reconnect interval to be longer than the connection
3415 timeout interval to avoid sending duplicate connections to
3419 Frequency : echo_client brw_test command
3421 Description: fix echo_client to work with OST preallocated code
3422 Details : OST preallocation code (5137) didn't take echo_client IO path
3423 into account: echo_client calls filter methods outside of any
3424 OST thread and, hence, there is no per-thread preallocated
3425 pages and buffers to use. Solution: hijack pga pages for IO. As
3426 a byproduct, this avoids unnecessary data copying.
3430 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3431 Description: Client can oops in mdc_commit_close() after open replay
3432 Details : It was possible for the MDS to return an open request with no
3433 transaction number in mds_finish_transno() if the client was
3434 evicted, but without actually returning an error. Clients
3435 would later try to replay that open and may trip an assertion
3436 Simplify the client close codepath, and always return an error
3437 from the MDS in case the open is not successful.
3440 Frequency : rare, 2.6 OSTs only
3442 Description: OST may deadlock under high load on fragmented files
3443 Details : If there was a heavy load and highly-fragmented OST filesystems
3444 it was possible to have all the OST threads deadlock waiting on
3445 allocation of biovecs, because the biovecs were not released
3446 until the entire RPC IO was completed. Instead, release biovecs
3447 as soon as they are complete to ensure forward IO progress.
3449 Severity : enhancement
3451 Description: Support for specifying external journal device at mount
3452 Details : If an OST or MDS device is formatted with an external journal
3453 device, this device major/minor is stored in the ext3 superblock
3454 and may not be valid for failover. Allow detecting and
3455 specifying the external journal at mount time.
3460 Description: Mounting an MDS with pending unlinked files may cause oops
3461 Details : target_finish_recovery() calls mds_postrecov() which returned
3462 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3463 considers this an error and immediately begins cleaning up the
3464 lov, just after starting the mds_lov process
3466 Severity : enhancement
3468 Description: Implement 'lfs df' to report actual free space on per-OST basis
3469 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3470 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3471 usage of objects; '-h' to report in human readable format.
3473 ------------------------------------------------------------------------------
3475 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3482 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3483 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3484 at startup during orphan unlinking. Instead, skip buddy update
3485 before setup, as it will be regenerated after recovery is complete.
3490 Description: If an OST is inactive, its locks might reference stale inodes.
3491 Details : lov_change_cbdata() must iterate over all namespaces, even if
3492 they are inactive to clear inode references from the lock.
3494 Severity : enhancement
3495 Frequency : occasional, if non-standard max_dirty_mb used
3497 Description: Client will block write RPCs if not enough grant
3498 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3499 then the client will block writes while waiting for another RPC
3500 to complete instead of consuming its dirty limit. With change
3501 we get improved performance when max_dirty_mb is small.
3503 Severity : enhancement
3504 Bugzilla : 3389, 6253
3505 Description: Add support for supplementary groups on the MDS.
3506 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3507 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3508 lookups for user supplementary groups into a cache.
3512 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3513 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3514 for liblustre O_EXCL is enforced by MDS.
3518 Description: readdir never returns NULL in liblustre.
3519 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3520 dentry in struct dirent.
3524 Frequency : liblustre only, depends on application IO pattern
3525 Description: liblustre clients evicted if not contacting servers
3526 Details : Don't put liblustre clients into the ping_evictor list, so
3527 they will not be evicted by the pinger ever.
3529 Severity : enhancement
3531 Description: Add ability to evict clients by NID from MDS.
3532 Details : By echoing "nid:$NID" string into
3533 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3534 $NID would be instantly evicted from this MDS and from all active
3535 OSTs connected to it.
3539 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3540 Details : lookup_it_finish() used to query file size from OSTs that was not
3545 Description: service threads change working directory to that of init
3546 Details : Starting lustre service threads may pin the working directory
3547 of the parent thread, making that filesystem busy. Threads
3548 now change to the working directory of init to avoid this.
3552 Frequency : during shutdown only
3553 Description: shutdown with a failed MDS or OST can cause unmount to hang
3554 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3559 Frequency : 2.6 only
3560 Description: chmod/chown may include an extra supplementary group
3561 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3562 supplementary group and if none is specified this is used.
3565 Bugzilla : 5479 (6816)
3567 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3568 Details : If lookup is in progress on a file that is unlinked we might try
3569 to revalidate the inode and fail in revalidate after lookup is
3570 complete and ll_file_open() enqueues the open again but
3571 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3574 Frequency : always, if lconf --abort_recovery used
3576 Description: lconf --abort_recovery fails with 'Operation not supported'
3577 Details : lconf was attempting to abort recovery on the MDT device and not
3580 ------------------------------------------------------------------------------
3582 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3587 Frequency : rare (only unsupported configurations with a node running as an
3589 Bugzilla : 6514, 5137
3590 Description: Mounting a Lustre file system on a node running as an OST could
3592 Details : OSTs now preallocates memory needed to write out data at
3593 startup, instead of when needed, to avoid having to
3594 allocate memory in possibly low memory situations.
3595 Specifically, if the file system is mounted on on OST,
3596 memory pressure could force it to try to write out data,
3597 which it needed to allocate memory to do. Due to the low
3598 memory, it would be unable to do so and the node would
3599 become unresponsive.
3601 Severity : enhancement
3603 Description: Addition of lconf --service command line option
3604 Details : lconf now accepts a '--service <arg>' option, which is
3605 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3607 Severity : enhancement
3609 Description: Failover mode is now the default for OSTs.
3610 Details : By default, OSTs will now run in failover mode. To return to
3611 the old behaviour, add '--failout' to the lmc line for OSTs.
3613 Severity : enhancement
3615 Description: Health checks are now provided for MDS and OSTs
3616 Details : Additional detailed health check information on MSD and OSTs
3617 is now provided through the procfs health_check value.
3620 Frequency : occasional, depends on IO load
3622 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3623 after numerous create/delete cycles
3624 Details : The ext3 inode allocation policy would not allocate new inodes
3625 very well on the OSTs because there are no new directories
3626 being created. Instead we look for groups with free space if
3627 the parent directories are nearly full.
3632 Description: Network or server problems during mount may cause partially
3633 mounted clients instead of returning an error.
3634 Details : The config llog parsing code may overwrite the error return
3635 code during mount error handling, returning success instead
3641 Description: MDS can fail to allocate large reply buffers
3642 Details : After long uptimes the MDS can fail to allocate large reply
3643 buffers (e.g. zconf client mount config records) due to memory
3644 fragmentation or consumption by the buffer cache. Preallocate
3645 some large reply buffers so that these replies can be sent even
3646 under memory pressure.
3650 Frequency : rare (liblustre)
3651 Description: fsx running with liblustre complained that using truncate() to
3652 extend the file doesn't work. This patch corrects that issue.
3653 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3654 ATTR_SIZE and lsm use in llu_setattr_raw.
3658 Frequency : rare, only 2.6 kernels
3659 Description: Unusual file access patterns on the MDS may result in inode
3660 data being lost in very rare circumstances.
3661 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3662 code in the 2.6 kernel caused the inode and/or EA data not to
3663 be read from disk, causing single-file corruption.
3667 Frequency : rare, only 2.6 filesystems using extents
3668 Description: Heavy concurrent write and delete load may cause data corruption.
3669 Details : It was possible under high-load situations to have an extent
3670 metadata block in the block device cache from a just-unlinked
3671 file overwrite a newly-allocated data block. We now unmap any
3672 metadata buffers that alias just-allocated data blocks.
3676 Frequency : filesystems with default stripe_count larger than 77
3677 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3678 Details : lconf specifies an inode size of 4096 bytes when the default
3679 stripe_count is larger than 77. This conflicts with the default
3680 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3681 this case to avoid pinning too much memory for large EAs.
3683 ------------------------------------------------------------------------------
3685 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3690 Frequency : rare (extremely heavy IO load with hundreds of clients)
3692 Description: Client is evicted, gets IO error writing to file
3693 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3694 caused clients to be evicted for AST timeouts. The fixes in
3695 bug 5192 mean we no longer need to have such short AST timeouts
3696 so ldlm_timeout has been increased.
3699 Frequency : occasional during --force or --failover shutdown under load
3700 Bugzilla : 5949, 4834
3701 Description: Server oops/LBUG if stopped with --force or --failover under load
3702 Details : a collection of import/export refcount and cleanup ordering
3703 issues fixed for safer force cleanup
3706 Frequency : only filesystems larger than 120 OSTs
3707 Bugzilla : 5990, 6223
3708 Description: lfs getstripe would oops on a very large filesystem
3709 Details : lov_getconfig used kfree on vmalloc'd memory
3712 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3713 Bugzilla : 6242, 6243
3714 Description: reading from files that had been truncated to a non-zero size
3715 but never opened returned no data
3716 Details : ll_file_read() reads zeros from no-object files to EOF
3721 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3722 Details : There's little harm in aborting MDS/OSS recovery and letting it
3723 try again, so I removed the LASSERT and return an error instead.
3725 Severity : enhancement
3727 Description: New debugging infrastructure for tracking down data corruption
3728 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3729 (b) cover more of the client-side code path, and (c) try to narrow
3730 down where problems occurred
3734 Bugzilla : 3819, 4364, 4397, 6313
3735 Description: Racing close and eviction MDS could cause assertion in mds_close
3736 Details : It was possible to get multiple mfd references during close and
3737 client eviction, leading to one thread referencing a freed mfd.
3739 Severity: : enhancement
3740 Bugzilla : 3262, 6359
3741 Description: Attempts to reconnect to servers are now more aggressive.
3742 Details : This builds on the enhanced upcall-less recovery that was added
3743 in 1.4.2. When trying to reconnect to servers, clients will
3744 now try each server in the failover group every 10 seconds. By
3745 default, clients would previously try one server every 25 seconds.
3750 Description: After recovery, certain operations trigger a failed
3751 assertion on a client.
3752 Details : Failing over an mds, using lconf -d --failover, while a
3753 client was doing a readdir() call would cause the client to
3754 LBUG after recovery completed and the readdir() was resent.
3756 Severity : enhancement
3758 Description: Default groups are now added by lconf
3759 Details : You can now run lconf --group <servicename> without having to
3760 manually add groups with lmc.
3763 Frequency : occasional
3765 Description: Nodes with an elan id of 0 trigger a failed assertion
3768 Frequency : always when accessing e.g. tty/console device nodes
3770 Description: tty and some other devices nodes cannot be used on lustre
3771 Details : file's private_data field is used by device data and lustre
3772 values in there got lost. New field was added to struct file to
3773 store fs-specific private data.
3776 Frequency : when exporting Lustre via NFS
3778 Description: NFSD failed occasionally when looking up a path component
3779 Details : NFSD is looking up ".." which was broken in ext3 directories
3780 that had grown large enough to become hashed.
3783 Frequency : Clusters with multiple interfaces not on the same subnet
3785 Description: Nodes will repeatedly try to reconnect to an interface which it
3786 cannot reach and report an error to the log.
3787 Details : Extra peer list entries will be created by lconf with some peers
3788 unreachable. lconf now validates the peer before adding it.
3791 Frequency : Only if a default stripe is set on the filesystem root.
3793 Description: Setting a default stripe on the filesystem root prevented the
3794 filesystem from being remounted.
3795 Details : The client was sending extra request flags in the root getattr
3796 request and did not allocate a reply buffer for the dir EA.
3799 Frequency : occasional, higher if lots of files are accessed by one client
3800 Bugzilla : 6159, 6097
3801 Description: Client trips assertion regarding lsm mismatch/magic
3802 Details : While revalidating inodes the VFS looks up inodes with ifind()
3803 and in rare cases can find an inode that is being freed.
3804 The ll_test_inode() code will free the lsm during ifind()
3805 when it finds an existing inode and then the VFS later attaches
3806 this free lsm to a new inode.
3810 Bugzilla : 6422, 7030
3811 Description: MDS deadlock between mkdir and client eviction
3812 Details : Creating a new file via mkdir or mknod (starting a transaction
3813 and getting the ns lock) can deadlock with client eviction
3814 (gets ns lock and trying to finish a synchronous transaction).
3817 Frequency : occasional
3818 Description: While starting a server, the fsfilt_ext3 module could not be
3820 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3821 kernels. Previously, lconf would still use the ext3 name
3822 when trying to load modules. Now, it will correctly use
3823 ext3 on 2.4 and ldiskfs on 2.6.
3825 Severity : enhancement
3826 Description: The default stripe count has been changed to 1
3827 Details : The interpretation of the default stripe count (0, to lfs
3828 or lmc) has been changed to mean striping across a single
3829 OST, rather than all available. For general usage we have
3830 found a stripe count of 1 or 2 works best.
3832 Severity : enhancement
3833 Description: Add support for compiling against Cray portals.
3834 Details : Conditional compiling for some areas that are different
3838 Frequency : occasional
3839 Bugzilla : 6409, 6834
3840 Description: Creating files with an explicit stripe count may lead to
3841 a failed assertion on the MDS
3842 Details : If some OSTs are full or unavailable, creating files may
3843 trigger a failed assertion on the MDS. Now, Lustre will
3844 try to use other servers or return an error to the
3848 Frequency : occasional
3850 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3851 caused lock timeout during readahead (since 1.4.2).
3852 Details : Processes doing readahead might match a lock that hasn't been
3853 granted yet if there are overlapping and conflicting lock
3854 requests. The readahead process waits on ungranted lock
3855 (original lock is CBPENDING), while OST waits for that process
3856 to cancel CBPENDING read lock and eventually evicts client.
3858 Severity : enhancement
3860 Description: Initial enabling of flock support for clients
3861 Details : Implements fcntl advisory locking and file status functions.
3862 This feature is provided as an optional mount flag (default
3863 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3864 locking are implemented yet, and those that are are not guaranteed
3865 to be completely correct in production environments.
3866 mount -t lustre -o [flock|noflock] ...
3869 Frequency : occasional
3871 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3872 assertion in the ext3 JBD (journaling) layer.
3873 Details : The b_committed_data struct is protected by the big kernel lock
3874 in 2.4 kernels, serializing journal_commit_transaction() and
3875 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3876 access to this struct. In 2.6 kernels there is finer grained
3877 locking to improve SMP performance of the JBD layer.
3881 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3883 -----------------------------------------------------------------------------
3885 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3887 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3888 versions of Lustre. Please update all servers and clients to
3889 version 1.4.2 or later at the same time. You must also run
3890 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3891 to update the configuration logs.
3893 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3894 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3895 - swab llog records of type '0' so we get proper header size/idx (5861)
3896 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3897 - fix rename of one directory over another leaking an inode (5953)
3898 - avoid SetPageDirty on 2.6 (5981)
3899 - don't re-add just-being-destroyed locks to the waiting list (5653)
3900 - when creating new directories, inherit the parent's custom
3901 striping settings if present parent (3048)
3902 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3903 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3904 - don't hold client locks on temporary worklist from l_lru (5666)
3905 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3906 - automatically evict dead clients (5921)
3907 - Update file size properly in create+truncate+fstat case (6196)
3908 - Do not unhash mountpoint dentries, do not allow removal of
3910 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3911 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3912 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3913 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3914 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3915 - Correctly handle reads of files with no objects (6243)
3916 - lctl recover will also mark a device active if deactivate used (5933)
3918 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3919 - allow --write-conf on an MDS with different nettype than client (5619)
3920 - don't write config llogs to MDS for mounts not from that MDS (5617)
3921 - lconf should create multiple TCP connections from a client (5201)
3922 - init scripts are now turned off by default; run chkconfig --on
3923 lustre and chkconfig --on lustrefs to use them
3924 - upcalls are no longer needed for clients to recover to failover
3926 - add --abort-recovery option to lconf to abort recovery on device
3928 - add support for an arbitrary number of OSTs (3026)
3929 - Quota support protocol changes.
3930 - forward compatibility changes to wire structs (6007)
3931 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3932 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3933 - improved "lustre status" script
3934 - initialize blocksize for non-regular files (6062)
3935 - added --disable-server and --disable-client configure options (5782)
3936 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3937 - Vanilla 2.4.29 support
3938 - increase maximum number of obd devices to 520 (6242)
3939 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3940 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3941 SLES 9 2.6.5 kernels
3942 - limit stripes per file to 160 (the maximum EA size) (6093)
3944 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3947 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3948 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3949 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3950 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3951 - small ext3 extents cleanups and fixes (5733)
3952 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3953 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3954 - avoid needless client->OST connect, fix handle mismatch (5317)
3955 - fix DLM error path that led to out-of-sync client, long delays (5779)
3956 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3957 - fix several locking issues related to i_size (5492,5624,5654,5672)
3958 - don't move pending lock onto export if it is already evicted (5683)
3959 - fix kernel oops when creating .foo in unlinked directory (5548)
3960 - fix deadlock in obdfilter statistics vs. object create (5811)
3961 - use time_{before,after} to avoid timer jiffies wrap (5882)
3962 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3963 - Do not leak request if server was not able to process it (5154)
3964 - If mds_open unable to find parent dir, make that negative lookup(5154)
3965 - don't create new directories with extent-mapping (5909, 5936)
3967 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3968 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3969 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3970 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3971 - support for RHEL4 kernel on i686 (5773)
3972 - provide error messages when incompatible logs are encountered (5898)
3974 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3975 * version 1.4.0.10 (1.4.1 release candidate 1)
3977 - don't keep a lock reference when lock is not granted (4238)
3978 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3979 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3980 - drop import inflight refcount on signal_completed_replay error (5255)
3981 - unlock page after async write error during send (3677)
3982 - handle missing objects in filter_preprw_read properly (5265)
3983 - no transno return for symlink open, don't save no-trasno open (3440)
3984 - don't try to complete elan receive that already failed (4012)
3985 - free RPC server reply state on error (5406)
3986 - clean up thread from ptlrpc_start_thread() on error (5160)
3987 - readahead could read extra page into cache that wasn't ejected (5388)
3988 - prevent races in class_attach/setup/cleanup/detach (5260)
3989 - don't dereference de->d_inode after l_dput of de (5458)
3990 - use "int" for stripe value returned from lock_to_stripe (5544)
3991 - mballoc allocation and error-checking fixes in 2.6 (5504)
3992 - block device patches to fix I/O request sizes in 2.6 (5482)
3993 - look up hostnames for IB nals (5602)
3994 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3995 - don't start multiple acceptors for the same port (5277)
3996 - fix incorrect LASSERT in mds_getattr_name (5635)
3997 - export a proc file for general "ping" checking (5628)
3998 - fix "lfs check" to not block when the MDS is down (5628)
4000 - service request history (4965)
4001 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4002 - create an "evict_client" /proc entry on OSTs, like the MDS has
4003 - fix mount usage message, return errors per mount(8) (5168)
4004 - change grep [] to grep "[]" in tests so they work in more UMLs
4005 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4006 - remove ancient LOV_MAGIC_V0 EA support (5047)
4007 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4008 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4009 - allow readahead allocations to fail when low on memory (5383)
4010 - mmap locking landed again, after considerable improvement (2828)
4011 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4013 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4016 - send OST transaction number in read/write reply to free req (4966)
4017 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4018 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4019 - the watchdog thread now runs as interruptible (5246)
4020 - flock/lockf fixes (but it's still disabled, pending 5135)
4021 - don't use EXT3 constants in llite code (5094)
4022 - memory shortage at startup could cause assertion (5176)
4024 - reorganization of lov code
4025 - single portals codebase
4027 - add extents/mballoc support (5025)
4028 - direct I/O reads in the obdfilter (4048)
4029 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4031 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4034 - send OST transaction number in read/write reply to free req (4966)
4035 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4036 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4037 - the watchdog thread now runs as interruptible (5246)
4038 - handle missing objects in filter_preprw_read properly (5265)
4039 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4040 - drop import inflight refcount on signal_completed_replay error (5255)
4041 - unlock page after async write error during send (3677)
4042 - return original error code on reconstructed replies (3761)
4043 - no transno return for symlink open, don't save no-trasno open (3440)
4045 - add pid to ldlm debugging output (4922)
4046 - bump the watchdog timeouts -- we can't handle 30sec yet
4047 - extra debugging for orphan dentry/inode bug (5259)
4049 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4052 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4053 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4054 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4055 - fix NULL dereference in /proc/sys/portals/routes (4827)
4056 - allow failed mdc_close() operations to be interrupted (4561)
4057 - stop precreate on OST before MDS would time out on it (4778)
4058 - don't send partial-page writes before EOF from client (4410)
4059 - discard client grant for sub-page writes on large-page clients (4520)
4060 - don't free dentries not owned by NFS code, check generation (4806)
4061 - fix lsm leak if mds_create_objects() fails (4801)
4062 - limit debug_daemon file size, always print CERROR messages (4789)
4063 - use transno after validating reply (3892)
4064 - process timed out requests if import state changes (3754)
4065 - update mtime on OST during writes, return in glimpse (4829)
4066 - add mkfsoptions to LDAP (4679)
4067 - use ->max_readahead method instead of zapping global ra (5039)
4068 - don't interrupt __l_wait_event() during strace
4070 - add software watchdogs to catch hung threads quickly (4941)
4071 - make lustrefs init script start after nfs is mounted
4072 - fix CWARN/ERROR duplication (4930)
4073 - return async write errors to application if possible (2248)
4074 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4075 - print NAL number in %x format (4645)
4076 - update barely-supported suse-2.4.21-171 series (4842)
4077 - support for sles 9 %post scripts
4078 - support for building 2.6 kernel-source packages
4079 - support for sles km_* packages
4081 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4084 - ignore -ENOENT errors in osc_destroy (3639)
4085 - notify osc create thread that OSC is being cleaned up (4600)
4086 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4087 - reconstruct ost_handle() like mds_handle() (4657)
4088 - create a new thread to do import eviction to avoid deadlock (3969)
4089 - let lconf resolve symlinked-to devices (4629)
4090 - don't unlink "objects" from directory with default EA (4554)
4091 - hold socknal file ref over connect in case target is down (4394)
4092 - allow more than 32000 subdirectories in a single directory (3244)
4093 - fix blocks count for O_DIRECT writes (3751)
4094 - OST returns ENOSPC from object create when no space left (4539)
4095 - don't send truncate RPC if file size isn't changing (4410)
4096 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4097 - bind to privileged port in socknal and tcpnal (3689)
4099 - rate limit CERROR/CWARN console message to avoid overload (4519)
4100 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4101 - basic mmap support (3918)
4102 - kernel patch series update from b1_4 (4711)
4104 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4107 - avoid crash during MDS cleanup with OST shut down (2775)
4108 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4109 - don't use bad inodes on the MDS (3744)
4110 - dynamic object preallocation to improve recovery speed (4236)
4111 - don't hold spinlock over lock dumping or change debug flags (4401)
4112 - don't zero obd_dev when it is force cleaned (3651)
4113 - print grants to console if they go negative (4431)
4114 - "lctl deactivate" will stop automatic recovery attempts (3406)
4115 - look for existing locks in ldlm_handle_enqueue() (3764)
4116 - don't resolve lock handle twice in recovery avoiding race (4401)
4117 - revalidate should check working dir is a directory (4134)
4119 - don't always mark "slow" obdfilter messages as errors (4418)
4121 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4124 - don't close LustreDB during write_conf until it is done (3860)
4125 - fix typo in lconf for_each_profile (3821)
4126 - allow dumping logs from multiple threads at one time (3820)
4127 - don't allow multiple threads in OSC recovery (3812)
4128 - fix debug_size parameters (3864)
4129 - fix mds_postrecov to initialize import for llog ctxt (3121)
4130 - replace config semaphore with spinlock (3306)
4131 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4132 - don't allow enqueue to complete on a destroyed export (3822)
4133 - down write_lock before checking llog header bitmap (3825)
4134 - recover from lock replay timeout (3764)
4135 - up llog sem before sending rpc (3652)
4136 - reduce ns lock hold times when setting kms (3267)
4137 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4138 - fix NULL deref and obd_dev leak on setup error (3312)
4139 - replace some LBUG about llog ops with error handling (3841)
4140 - don't match INVALID dentries from d_lookup and spin (3784)
4141 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4142 - fix invalid assertion in ptlrpc_set_wait (3880)
4144 - add libwrap support for the TCP acceptor (3996)
4145 - add /proc/sys/portals/routes for non-root route listing (3994)
4146 - allow setting MDS UUID in .xml (2580)
4147 - print the stack of a process that LBUGs (4228)
4149 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4152 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4153 - make sure to unset replay flag from failed open requests (3440)
4154 - if default stripe count is 0, use OST count for inode size (3636)
4155 - update parent mtime/ctime on client for create/unlink (2611)
4156 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4157 - free recovery state on server during a forced cleanup (3571)
4158 - unregister_reply for resent reqs (3063)
4159 - loop back devices mounting and status check on 2.6 (3563)
4160 - fix resource-creation race that can provoke i_size == 0 (3513)
4161 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4162 - more debugging for page-accounting assertion (3746)
4163 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4164 - avoid deadlock after precreation failure (3758)
4165 - fix race and lock order deadlock in orphan handling (3450, 3750)
4166 - add validity checks when grabbing inodes from l_ast_data (3599)
4168 - add /proc/.../recovery_status to obdfilter (3428)
4169 - lightweight CDEBUG infrastructure, debug daemon (3668)
4170 - change default OSC RPC parameters to be better on small clusters
4171 - turn off OST read cache for files smaller than 32MB
4172 - install man pages and include them in rpms (3100)
4173 - add new init script for (un)mounting lustre filesystems (2593)
4174 - run chkconfig in %post for init scripts (3701)
4175 - drop scimac NAL (unmaintained)
4177 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4180 - clean kiobufs before and after use (3485)
4181 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4182 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4183 - change init script to fail more clearly if not run as root (1528)
4184 - allow clients to reconnect during replay (1742)
4185 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4186 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4187 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4188 - fix cond_resched (3554)
4189 - only evict unfinished clients after recovery (3515)
4190 - allow bulk resend, prevent data loss (3570)
4191 - dynamic ptlrpc request buffer allocation (2102)
4192 - don't allow unlinking open directory if it isn't empty (2904)
4193 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4194 - remove extraneous obd dereference causing LASSERT failure (3334)
4195 - don't use get_cycles() when creating temp. files on the mds (3156)
4196 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4197 - handle EEXIST for set-stripe, set proper directory name (3336)
4199 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4200 - fix ksocknal_fmb_callback() error messages (2918)
4202 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4205 - don't copy lvb into (possibly NULL) reply on error (2983)
4206 - don't deref dentry after dput, don't free lvb on error (2922)
4207 - use the kms to determine writeback rpc length (2947)
4208 - increment oti_logcookies when osc is inactive (2948)
4209 - update client's i_blocks count via lvb messages (2543)
4210 - handle intent open/close of special files properly (1557)
4211 - mount MDS with errors=remount-ro, like obdfilter (2009)
4212 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4213 - don't use cancelling-locks' kms values (2947)
4214 - use highest lock extent for kms, not last one (2925)
4215 - don't dereference ERR_PTR() dentry in error handling path (3107)
4216 - fix thread race in portals_debug_dumplog() (3122)
4217 - create lprocfs device entries at setup instead of at attach (1519)
4218 - common AST error handler, don't evict client on completion race (3145)
4219 - zero nameidata in detach_mnt in 2.6 (3118)
4220 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4221 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4222 - initialize RPC timeout timer earlier for 2.6 (3219)
4223 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4224 - print nal/nid for unknown nid (3258)
4225 - additional checks for oscc recovery before doing precreate (3284)
4226 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4227 - don't crash in mdc_close for bad permissions on open (3285)
4228 - zero i_rdev for non-device files (3147)
4229 - clear page->private before handing to FS, better assertion (3119)
4230 - tune the read pipeline (3236)
4231 - fix incorrect decref of invalidated dentry (2350)
4232 - provide read-ahead stats and refine rpc in flight stats (3328)
4233 - don't hold journal transaction open across create RPC (3313)
4234 - update atime on MDS at close time (3265)
4235 - close LDAP connection when recovering to avoid server load (3315)
4236 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4237 - don't leak open file on MDS after open resend (3325)
4238 - serialize filter_precreate and filter_destroy_precreated (3329)
4239 - loop device shouldn't call sync_dev() for nul device (3092)
4240 - clear page cache after eviction (2766)
4241 - resynchronize MDS->OST in background (2824)
4242 - refuse to mount the same filesystem twice on same mountpoint (3394)
4243 - allow llmount to create routes for mounting behind routers (3320)
4244 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4245 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4246 - fix rare problem with rename on htree directories (3417)
4248 - allow default OST striping configuration per directory (1414)
4249 - fix compilation for qswnal for 2.6 kernels (3125)
4250 - increase maximum number of MDS request buffers for large systems
4251 - change liblustreapi to be useful for external progs like lfsck (3098)
4252 - increase local configuration timeout for slow disks (3353)
4253 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4255 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4258 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4259 - don't overwrite extent policy data in reply if lock was blocked (2901)
4260 - drop filter export grants atomically with removal from device (2663)
4261 - del obd_self_export from work_list in class_disconnect_exports (2908)
4262 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4263 - swab reply message in mdc_close, other PPC fixes (2464)
4264 - fix destroying of named logs (2325)
4265 - overwrite old logs when running lconf --write_conf (2264)
4266 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4267 - fix race in target_handle_connect (2898)
4268 - mds_reint_create() should take same inode create lock (2926)
4269 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4270 - don't close files for self_export to avoid uninitialized obd (2936)
4271 - allow MDS with the same name as client node (2939)
4272 - hold dentry reference for closed log files for unlink (2325)
4273 - reserve space for all logs during transactions (2059)
4274 - don't evict page beyond end of stripe extent (2925)
4275 - don't oops on a deleted current working directory (2399)
4276 - handle hard links to targets without a parent properly (2517)
4277 - don't dereference NULL lock when racing during eviction (2867)
4278 - don't grow lock extents when lots of conflicting locks (2919)
4280 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4283 - account for cache space usage on clients to avoid data loss (974)
4284 - lfsck support in lustre kernel code (2349)
4285 - reduce journal credits needed for BRW writes (2370)
4286 - orphan handling to avoid losing space on client/server crashes
4287 - ptlrpcd can be blocked, stopping ALL progress (2477)
4288 - use lock value blocks to assist in proper KMS, faster stat (1021)
4289 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4290 - recovery for initial connections (2355)
4291 - fixes for mds_cleanup_orphans (1934)
4292 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4293 - block all file creations until orphan recovery completes (1901)
4294 - client remove rq_connection from request struct (2423)
4295 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4296 - recovery timer race (2670)
4297 - mdc_close recovey bug (2532)
4298 - ptlrpc cleanup bug (2710)
4299 - mds timeout on local locks (2588)
4300 - namespace lock held during RPCs (2431)
4301 - handle interrupted sync write properly (2503)
4302 - don't try to handle a message that hasn't been replied to (2699)
4303 - client assert failure during cleanup after abort recovery (2701)
4304 - leak mdc device after failed mount (2712)
4305 - ptlrpc_check_set allows timedout requests to complete (2714)
4306 - wait for inflight reqs when ptlrpcd finishes (2710)
4307 - make sure unregistered services are removed from the srv_list
4308 - reset bulk XID's when resending them (caught by 1138 test)
4309 - unregister_bulk after timeout
4310 - fix lconf error (2694)
4311 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4312 - readahead locks pages, leaves pending causing memory pressure (2673)
4313 - increase OST request buffers to 4096 on large machines (2729)
4314 - fix up permission of existing directories in simple_mkdir (2661)
4315 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4316 - don't assign transno to errored transactions (2742)
4317 - don't delete objects on OST if given a bogus objid from MDS (2751)
4318 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4319 - if rq_no_resend, then timeout request after recovery (2432)
4320 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4321 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4322 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4323 - MDS recovery shouldn't delete objects that it has given out (2730)
4324 - if enqueue arrives after completion, don't clobber LVB (2819)
4325 - don't unlock pages twice when trigger_group_io returns error (2814)
4326 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4327 - don't write pages to disk if there was an error (1450)
4328 - don't ping imports that have recovery disabled (2676)
4329 - take buffered bytes into account when balancing socknal conn (2817)
4330 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4331 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4332 - remove little-endian swabbing of llog records (1987)
4333 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4334 - retry reposting request buffers if they fail (1191)
4335 - grow extent at grant time to avoid granting a revoked lock (2809)
4336 - lock revoke doesn't evict page if covered by a second lock (2765)
4337 - disable VM readahead to avoid reading outside lock extents (2805)
4339 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4340 - updated kernel patches for hp-2.4.20 kernel (2681)
4342 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4345 - fix truncated write corruption (2366)
4346 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4348 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4349 - ensure len(uuid) < 37 in lmc (1171)
4350 - fix ia64 OOPS in llog_test (2255)
4351 - zero end of page at obdfilter for partial page writes (2648)
4352 - don't leave stale dentries around after renames (bug 2428)
4353 - fix timeouts when evicting a client with a single lock held (2642)
4354 - set deadline for the initial HELLO message to drain (2634)
4355 - print out dotted-quad IP addresses in the socknal (2302)
4357 - additional debugging for MDS client eviction problem (2443)
4358 - fix mkfsoptions support for osts (2603, 2604)
4360 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4363 - add series for the vanilla 2.6.0 kernel
4364 - add series for the vanilla 2.4.24 kernel
4365 - add series for a cray x86/64 UL kernel drop
4366 - fix xattr patches for the vanilla 2.4.19 series
4368 - generate true UUIDs in lmc (1171)
4369 - have portals stack dumping break in UML (2466)
4370 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4371 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4372 - wake callback waiting threads on client eviction (2460)
4373 - Add --ptldebug and --subsystem to lmc (1719)
4374 - update assertion to allow safe interrupt allocation
4375 - set rq_no_resend for cancel requests (2432)
4376 - recalculate ptlrpcd timeout after resend (2494)
4377 - call vfs_rmdir when removing pending directories (2368)
4378 - fix renaming a file to itself (2429)
4379 - lmc creates a default one-stripe lov (2454)
4380 - expand procfs space to handle large clusters (2326)
4381 - increase UML stack to avoid overflow
4382 - update lconf's list of debug and subsystem masks
4383 - fix lfs find --obd (2510)
4384 - /proc tunable for disabling filter read caching (2591)
4385 - stop rpm packages from altering slapd.conf (2301)
4386 - disable nagle in the socknal under 0conf (2578)
4387 - choose mds inode size based on stripe count (2572)
4388 - fix kernel-source rpm problems (2516)
4390 - add --disable-doc to avoid pdf generation (2421)
4391 - update documentation, tests, type-os, comments
4392 - avoid format warnings on ia64
4393 - remove the TOE NAL
4394 - tiny code cleanups by removing unused fields
4396 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4399 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4400 - remove the most blatant lies from BUILDING file (2371)
4401 - change default debug level to reasonable production setting
4402 - reduce client side cache size to reduce cache flush time
4403 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4404 - make TCP zerocopy and pinger support enabled by default (2476)
4405 - sync writes completed after process exits caused crashes (2319)
4406 - maintain correct mount count on the MDS (2356)
4407 - backout 1557, because 2316 wasn't really fixed
4408 - better file I/O statistics gathering in /proc
4409 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4410 - another kernel patch to fix zero-copy TCP function export
4411 - don't take duplicate lock when processing re-sent getattr (2420)
4412 - lctl uses obd_self_export instead of creating new conn (2353)
4413 - MDS/OST recovery case which requires object creation asserted (2425)
4414 - move lfs from /usr/sbin to /usr/bin in packages
4415 - fix race between mds_client_add and mds_client_free (2417)
4416 - use kmalloc instead of slabs in portals (2430)
4417 - don't create duplicate records when a failover MDS is present (2442)
4418 - remove unnecessary mount age check (2332)
4419 - don't remove directory inodes from locks prematurely (2451)
4420 - don't break if MDS service name is the same as hostname (2103)
4421 - fix races in client write RPC generation when cache full (2482)
4423 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4426 - remove now-unused request->rq_obd (278)
4427 - if an allocation fails, print out how much memory we've used (1933)
4428 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4429 - add the "configurable stack size" patch to most series files (1256)
4430 - ability to write large log records, for 100+ OST configs (2306)
4431 - fix NULL deref when filter_prep fails (2314)
4432 - fix operator precedence error in filter_sync
4433 - dynamic allocation of socknal TX descriptors (2315)
4434 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4435 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4436 - ensure that CWARN messages in Portals always get to the syslog
4437 - __init/__exit are not for prototype decls (ldlm_init/exit)
4438 - x86-64 compile warning fixes
4439 - fix gateway LMC keyword conflict (2318)
4440 - fix MDS lock inversions in getattr/reint paths (1844)
4441 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4442 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4443 - DLM race condition prevented some lock evictions (2328)
4444 - ENOMEM detection and retry on socknal sends (2230)
4445 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4446 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4447 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4448 - better error messages when a client is rejected during recovery (1505)
4449 - avoid cancelling locks which were never granted, after failure (2330)
4450 - fix i_sem/journal inversion in mds_client_add (2333)
4451 - fix truncate/getattr lock cycle deadlock (2334)
4452 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4453 - fix two rare exit paths which could leak an l_lock() ref (2321)
4454 - fencepost error in MDS/OST orphan recovery (2226)
4455 - make log record alignment 8 bytes (1988)
4456 - lstripe now fails when requested offset > ost_count (2237)
4457 - ensure that all kernel series have a complete list.h (1607)
4458 - fix crashes in special-file operations (2316)
4459 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4461 - allow configurable automake binary, for testing new versions
4462 - small update to the lfs documentation
4464 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4466 * fix negative export reference count in fsfilt_sync (2312)
4468 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4469 * release candidate 0.9.1
4471 - orphans are moved into the PENDING directory for possible recovery
4472 - replayed opens now open by fid for orphan/rename safety (1042)
4473 - last close of an orphan inode generates a transno (683)
4474 - chdir() and mount() now pin the directory entry (1020)
4475 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4476 - discard very old requests without processing them (1502)
4477 - remove some common, well-understood CERRORs (1505)
4478 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4479 - clear "grant" flags in OST replies until OST grant code lands (1644)
4480 - fix read performance by not clobbering i_blksize on client (1598)
4481 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4482 - make LRU size a /proc tunable, clears locks when reduced (707)
4483 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4484 - server locks take references on exports now (1558)
4485 - build fixes for 2.4.20-rh trees (1663)
4486 - return an error from lov_create if all OSCs are inactive (1751)
4487 - fix import levels when a reconnect happens without a timeout (1597)
4488 - exit early from mds_open if we get a lookup error (1749)
4489 - partial page read at EOF wouldn't wait for disk before sending (1642)
4490 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4491 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4492 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4493 - write/truncate lock inversion (1639)
4494 - Don't auto-load obdclass, portals modules during cleanup (1495)
4495 - fix timestamps from jumping to "now" (1763)
4496 - extra journal assertions (1648)
4497 - add an extra multiunlink test (1771)
4498 - fix read_record/write_record API (1776)
4499 - fix leak of offset_extent, possible incorrect i_size later (1772)
4500 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4501 - Debugging for the kqswnal_get_idle_tx problems (1820)
4502 - Allow recovery to be attempted multiple times (1536)
4503 - Write out MDS last_rcvd file after it is first created (1600)
4504 - Fix tx_descriptor leak in failed transmit situations (1827)
4505 - ext3 journaling fixes for assertion failure after IO error (1871)
4506 - class_export_put() on freed export after completion AST error (1896)
4507 - Fix revalidate looping in VFS (1322)
4508 - Don't access a freed export during MDS_REINT timeout (1521)
4509 - Add open-unlink recovery support on the MDS (1673,1764)
4510 - Return an error if no MDS data was read from last_rcvd (1946)
4511 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4512 - Don't setup a disk device that doesn't match exported UUID (317)
4513 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4514 - avoid committing NULL handle in force close
4515 - local.sh is now a one-stripe LOV configuration
4516 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4517 - fix ext3 htree duplicate directory entry corruption (1516)
4518 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4519 - update to new LOV EA format (2097)
4520 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4522 2003-06-15 Phil Schwan <phil@clusterfs.com>
4525 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4526 - per-import/export recovery handling (958, 931, 959)
4527 - multiple last-rcvd slots, for serving multiple FSes (949)
4528 - connections are again shared between multiple imp/exports (963, 964)
4529 - "umount -f" would hang if any requests needed to be sent (393, 978)
4530 - avoid pinning large req buffer by copying for queued messages (989)
4531 - add "uuid" to "lctl device" command to help upcalls (991)
4532 - "open" RPCs with transnos would confuse recovery counters (1037)
4533 - do proper endian conversion of all wire messages (288, 340, 891)
4534 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4535 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4536 - fix ptlrpc_connection leak in target_handle_connect (1174)
4537 - fix import refcounting bug in OST and MDS cleanup (1134)
4538 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4539 - fix very unlikely obd_types race condition (501)
4540 - remove osc_open hack for echo_client (1187)
4541 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4542 - a failure in read_inode2 leads to deadlock (1139)
4543 - cancel ack-locks as soon as transaction is committed (1072)
4544 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4545 - make sure to commitrw after any preprw to avoid deadlock (1162)
4546 - failing to execute a file in a lustre FS would lock inode (1203)
4547 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4548 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4549 - obd_uuid2tgt would walk off the end of the list (1255)
4550 - on IA64 the osc would give portals incorrect bulk size (1258)
4551 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4552 - fix lock inversion caused by new llite matching code (1282)
4553 - limit the number of dirty pages on a client to 10MB (1286)
4554 - timed out locks were not being corrected cancelled (1289)
4555 - fix O_DIRECT above 4GB on IA-32 (1292)
4556 * major user-visible changes
4557 - fail out/fail over policy now controlled by the upcall (993)
4559 - add OBD_PING to check server availability and failure (954)
4560 - lustre messages are now sent in sending host order (288, 340, 891)
4561 - add eadatalen to MDS getattr reply (340)
4562 - OST read replies may contain second buffer, with per-page status (593)
4564 2003-03-11 Phil Schwan <phil@clusterfs.com>
4567 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4568 - failed open()s could cause deadlock; fixed (867, 869)
4569 - stop cancelling OST locks when files are closed (481)
4570 - overlapping XID spaces caused network corruption (851, 853)
4571 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4572 - setattr_raw added, to do single-RPC, server-side setattrs
4573 - lmc/lconf syntax change for OST UUIDs
4574 - fix crashy race condition between ptlrpc_free_req and osc_close
4575 - don't use request in mdc_enqueue if we hit a timeout (889)
4576 - don't set the inode i_size for regular files from the MDS (896)
4577 - handle out of order completion AST (842)
4578 - don't LBUG if a lock request times out after receiving AST (913)
4579 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4580 - if a bad lock AST arrives, send an error instead of dropping entirely
4581 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4582 - fix leak in bulk IO when only partially completed (899, 900, 926)
4583 - fix O_DIRECT for ia64 (55)
4584 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4585 - C-z after timeout could hang a process forever; fixed (977)
4587 - client-side I/O cache (678, 924, 929, 941, 970)
4589 - READPAGE and SETATTRs which don't take server-side locks get
4592 2003-02-11 Phil Schwan <phil@clusterfs.com>
4595 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4596 - Fix fsfilt_extN_readpage() to read a full page of directory
4597 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4598 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4599 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4600 - Most intent operations were removed, in favour of a new RPC mode
4601 that does a single RPC to the server and bypasses most of the VFS
4602 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4603 - Aggressively cancel local locks on DLM servers
4604 - mds_reint_unlink sends EA to the client if it's the last nlink.
4605 client uses that EA to unlink OST objects.
4606 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4607 - recursive symlinks were fixed (439)
4608 - fixed NULL deref in DEBUG_REQ
4609 - filter_update_lastobjid no longer calls sync, which annoyed extN
4610 - fixed multi-client small-writes to a single file problem (445)
4611 - fixed mtime updates during file writes (607)
4612 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4613 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4614 - fixed Linux OST size reporting problem (444, 656)
4615 - OST now updates object mtime with writes or setattr (607, 619)
4616 - client verifies file size before zeroing page past EOF (445)
4617 - OST now writes last allocated objid to disk with allocation (108)
4618 - LOV on echo now works (409)
4620 - mds_reint_unlink sends a new buffer, with the EA included. this
4621 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4622 set if a regular file was being unlinked, and it was the last link
4623 - use PtlGet from the target for bulk writes (315)
4624 - OST now updates object mtime with writes or setattr (607, 619)
4625 - LDLM now has a grant-time callback to revalidate locked items, if
4627 - Many MDS operations were reorganized to combat race conditions
4629 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4630 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4632 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4635 - Fully reactivate OST imports after reconnection (512, others)
4636 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4637 - More graceful error handling for truncating on dead OST (515)
4638 - Don't error out unless we're actually accessing dead stripes (474)
4639 - Fix garbage sizes when stripes are missing (410)
4640 - LRU counters were broken, causing constant lock purge (433, 432)
4641 - garbage on read from stripes with failed OSTs (441)
4642 - mark OSCs as active before reconnecting during recovery (438)
4643 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4644 - lfind did not preserve OST order in output (443)
4645 - symlinks cause hung clients, incorrect data (439)
4646 - stop dereferencing request after dropping refcount (457)
4647 - don't LASSERT(spin_is_locked) on non-SMP (455)
4648 - fixes for many rename() bugs
4649 - fstat didn't correctly synchronize attributes (399)
4650 - server must handle lock cancellation during blocking AST prep (487)
4651 - bulk descriptors were free()d too soon (511)
4652 - fix paths in lconf, which would load incorrect modules (451, 507)
4653 - fix confusing lconf 'host not found' error message (386)
4654 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4655 - fix race condition in mdc_blocking_ast() for inode access (526)
4656 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4657 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4658 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4659 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4660 - fix lstripe in several ways (kernel side) (527)
4661 - fix request leak in ldlm_cli_enqueue (262)
4662 - incorrect OSC was marked inactive after OST failure
4663 - call mds_fs_cleanup before unmounting filesystem (524)
4664 - fix races between taking ns_lock and ldlm_lock_change_resource
4665 - fix races updating LOV export open file list
4666 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4667 - fix recovery NULL deref in ldlm_cli_cancel_unused
4668 - fix some DLM races by using new hash table for lock handles (419)
4669 - permit the client to specify desired inodes, at replay
4670 - duplicate requests when we queue them for replay reintegration
4671 - fix last_rcvd offset calculation
4672 - sync after each recovered transaction, so we always make progress
4673 - never, not always, ERESTART requests without transnos
4674 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4676 - skip replay if the MDS says that the client is already connected
4677 - don't check for a recovery-enabled export to match lctl's UUID
4678 - don't INC_USE_COUNT for phantom exports
4679 - don't crash when cleaning up phantom exports (567)
4680 - don't double-finish or set replay data for errored mdc_open requests
4681 - abort requests when they time out, so we don't get old replies
4682 - send/receive replies for AST messages again
4683 - if the client says that it doesn't have the lock, cancel it on the
4685 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4686 mark it as destroyed, it will all work out when decref is called
4687 - fix module use counts (22, 581)
4689 - ASTs now expect a reply (server cancels lock on error reply)
4691 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4694 - fix many simultaneous client startup (392)
4695 - fix dentry->d_it clobbering
4696 - credentials weren't being shipped for readdir/getattr operations
4697 - remove invalid assertions triggered during some concurrent MD
4699 - proper Lustre versions added (336, 389)
4700 - fix memory leak for create error case (398)
4701 - fix LOV locking bug that would get cli/srv out of sync
4702 - fix echo client over LOV (409)
4703 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4704 - fix double-O_EXCL intent crash (424)
4705 - avoid sending multiple lock CANCELs (352)
4707 - MDS can do multi-client recovery (modulo bugs in new code)
4709 - many updates, edits, cleanups
4711 2002-11-18 Phil Schwan <phil@clusterfs.com>
4714 - fix null d_it dereference (346)
4715 - fix full OST/dbench hang (333)
4716 - fix permission problem with file removal (286)
4717 - fix removal of OSCs from LOV when they fail
4718 - fix NULL deref during bulk timeout (214)
4719 - fix problems related to multiple filesystems on one MDS (241)
4720 - fixed serious subtle metadata locking bugs
4721 - free locks on clients when inodes are removed due to memory
4723 - fix inode pointer in lock data (285)
4724 - partial support for multiple MDS on a single host (241)
4725 - data locks weren't cancelled at clear_inode time (290, 311)
4726 - intent locks could lead to unbounded lock growth (205)
4727 - added a maximum lock count, an LRU list, and a flusher
4728 - fix multiple rename (365)
4729 - properly abstracted the echo client
4730 - OSC locked 1 byte too many; fixed
4731 - rewrote brw callback code:
4732 - fixed recovery bugs related to LOVs (306)
4733 - fixed too-many-pages-in-one-write crash (191)
4734 - fixed (again) crash in sync_io_timeout (214)
4735 - probably fixed callback-related race (385)
4737 - Add capability to MDS protocol
4738 - LDLM cancellations and callbacks on different portals
4740 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4743 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4744 - defer open object destruction to close time (601981, 138)
4745 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4746 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4747 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4748 - don't call obd_setattr in ll_file_release for destroyed objects
4749 * protocol change to lustre_msg: move |version| and add |flags|
4750 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4751 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4752 * added replay of create, unlink, link and rename operations during
4753 MDS failover; recovery should be much more robust now
4754 * remove failed OSCs from LOVs (only lov_create uses this so far)
4755 * the lustre-HOWTO was brought (more) up to date (582544)
4757 2002-10-23 Phil Schwan <phil@clusterfs.com>
4760 - in-use dentries weren't being reused properly (617851)
4761 - prevent multiple LDLM setup (599178)
4762 - fix LOV size calculations for truncate (617853)
4763 - fix client handling of MDS intent errors (POSIX)
4764 - fix permission bug in lovstripe.c test (624321)
4765 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4766 - truncate past end of file could corrupt data
4767 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4768 - a race in recovery could return ETIMEDOUT to apps (623947)
4769 - building outside the source directory was fixed
4770 * the lustre-HOWTO was brought (more) up to date (582544)
4771 * major progress was made on recovery functionality
4773 2002-10-10 Phil Schwan <phil@clusterfs.com>
4776 - recovery deadlock fix
4777 - rm -rf causes LBUG fix (617817)
4778 - file open by multiple tasks fix (618962)
4779 - directory permissions bugs (602707 and 620007)
4780 - journal_stop fixed with locking (611313)
4781 - O_APPEND failures resolved (618273, perhaps 614459)
4782 - lconf PATH fix (619770)
4783 - IA64 build fix (621450)
4784 - RPC buffer sizes scale with amount of memory
4786 2002-10-01 Phil Schwan <phil@clusterfs.com>
4789 - locks would be cancelled without throwing away data pages,
4790 resulting in inconsistent data (605627)
4791 - inode attributes were not always being refreshed (605627, 612449)
4792 - lconf now continues to cleanup after lctl reports an error
4793 - MDS now enforces user permissions (602707)
4794 - lprocfs cleanup fixed, but not yet enabled (614157)
4795 - fixed infinite server hang, should a client not respond to an AST
4796 - avoid going into recovery if user calls readlink() with a buffer
4797 that's too small (613941)
4798 - AST RPCs no longer require replies (614867) -- this may be changed
4799 - don't crash server if client sends an IOV that's too big (611336)
4800 - fixed lock conversion deadlock (611892)
4801 - fixed the following of symlinks (614622)
4802 * recovery: the server can remove locks from a client that dies, other
4803 clients can make progress
4804 * more extN patch fixes
4805 * compile-time configurable ptlrpc buffer allocations
4807 - collaborative read cache document
4808 - Lustre Lite Performance CDR document-in-progress
4810 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4813 - fix typo in patch-2.4.18
4815 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4818 - clear ptlrpc request each time in handle_incoming_request()
4819 - unlink of files now destroys the object on the OST
4821 2002-09-19 Peter Braam <braam@clusterfs.com>
4823 * add hard link support
4824 * change obdfile creation method
4825 * kernel patch changed
4827 2002-09-19 Peter Braam <braam@clusterfs.com>
4830 - stack overflow bug in extN fixed
4832 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4834 * documentation updates
4835 - add man pages for config tools
4836 - update tests/README to describe testing with new config tools
4837 - finish metadata API descriptions
4838 * bug fixes and cleanups
4839 - statfs workaround for 16TB limit
4840 - LOV stripe allocation improved, can stripe on subset of OSTs
4841 - LOV file size/IO offset was wrong for files > 4GB in size
4842 - object EA data was being dropped, caused files to be unreadable
4843 - memory overflow with non-LOV OST caused memory corruption
4844 - fixed regression tests to work with new config tools, obdfilter
4845 - fixed bug when directory size became larger than 1 block
4846 - fixed bug (for single client case) when PWD was deleted
4847 - invalidate local directory pages when doing intent-based ops
4848 - avoid LDLM oops when lock callback contained bad data
4850 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4852 * documentation updates
4853 * bug fixes and cleanups
4854 - configuration tools
4857 - 64-bit compile warnings
4858 - 64-bit internal statfs data
4860 * test_brw on persistent OST devices
4862 * lprocfs (disabled)
4864 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4866 * documentation updates
4867 * bug fixes and cleanups
4868 * configuration tools
4870 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4873 * many small fixes to 0_5_4
4874 * io/network handling
4875 * thinkos in MDS operations
4877 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4880 * crucial basic fixes to 0.5.3
4881 * IOR, Iozone work over Elan
4884 2002-08-07 Phil Schwan <phil@clusterfs.com>
4885 * version 0_5_3, our first alpha
4886 * we use the new Portals iovs
4887 * documentation updates
4888 * bug fixes and cleanups
4889 * small changes in the DLM wire protocol
4891 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4892 * version 0_5_1 with some initial stability,
4893 * locking on MD and file I/O.
4894 * documentation updates
4895 * several bug fixes since 0.5.0
4896 * small changes in wire protocol
4898 2002-07-18 Phil Schwan <phil@clusterfs.com>
4900 * delivered as Lustre Light Alpha
4901 * fixed a crash after handling invalid MDS requests
4902 * fixed directory pages for architectures with non-4k pages sizes
4904 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4905 * release version v0_4_4
4906 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4908 * Fixes a number of interruption problems with OST operations.
4909 * Update documentation for portals header changes
4910 * Move all wire protocol structs/defines to lustre_idl.h
4911 * Fixes symlink length bug.
4912 * Add tcpdump to repository.
4914 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4915 * release version v0_4_3
4916 * Fixes statfs for inodes on extN.
4917 * Fixes bug in runtests which would delete /etc/hosts.
4918 * Use 64-bit object IDs wherever possible (not into VFS though)
4919 Remove ost_get_info, which is unused by lustre, and out of date.
4921 2002-07-03 Peter Braam <braam@clusterfs.com>
4922 * release version v0_4_2 Fixes a lookup error (type not passed)
4923 * move forward to head of Portals
4924 * move forward to latest Lustre kernel
4926 2002-06-25 Peter Braam <braam@clusterfs.com>
4927 * release version v0_4_1. Hopefully stable on single node use.