1 tbd Sun Microsystems, Inc.
5 2.6.9-67.0.4.EL (RHEL 4),
6 2.6.16.54-0.2.5 (SLES 10),
7 2.6.18-53.1.13.el5 (RHEL 5),
8 2.6.18.8 vanilla (kernel.org)
9 2.6.22.14 vanilla (kernel.org)
10 * Client support for unpatched kernels:
11 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
12 2.6.16 - 2.6.22 vanilla (kernel.org)
13 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
14 we do not recommend using patchless RHEL4 clients with kernels
15 prior to 2.6.9-55EL (RHEL4U5).
16 * Recommended e2fsprogs version: 1.40.4-cfs1
17 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
18 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
19 removed cwd "./" (refer to Bugzilla 14399).
23 Frequency : rare, on recovery
24 Description: read procfs can produce deadlock in some situation
25 Details : Holding lprocfs lock with send rpc can produce block for destroy
26 obd objects and this also block reconnect with -EALREADY. This isn't
27 fix all lprocfs bugs - but make it rare.
29 Severity : enhancement
31 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
34 Frequency : frequent on X2 node
36 Description: mdc_set_open_replay_data LBUG
37 Details : Set replay data for requests that are eligible for replay.
41 Description: lustre_mgs: operation 101 on unconnected MGS
42 Details : When MGC is disconnected from MGS long enough, MGS will evict the
43 MGC, and late on MGC cannot successfully connect to MGS and a lot
44 of the error messages complaining that MGS is not connected.
46 Severity : enhancement
48 Description: Add server support for vanilla-2.6.22.14.
51 Frequency : rare, depends on device drivers and load
53 Description: MDS or OSS nodes crash due to stack overflow
54 Details : Code changes in 1.6.4 increased the stack usage of some functions.
55 In some cases, in conjunction with device drivers that use a lot
56 of stack, the MDS (or possibly OSS) service threads could overflow
57 the stack. One change which was identified to consume additional
58 stack has been reworked to avoid the extra stack usage.
60 Severity : enhancement
62 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
64 Severity : enhancement
66 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
68 Severity : enhancement
70 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
72 Serverity : enhancement
74 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
77 Frequency : rare on shutdown ost
79 Description: Don't allow skipping OSTs if index has been specified.
80 Details : Don't allow skipping OSTs if index has been specified, make locking
81 in internal create lots better.
85 Description: ASSERTION(!PageDirty(page)) failed
86 Details : Wrong check could lead to an assertion failure under specific
91 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
92 Details : access to bitfield in structure is always rounded to long
93 and this produce problem with not atomic change any bit.
97 Description: Lustre make rpms failed.
98 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
99 builds Lustre rpms from tarball.
101 Severity : enhancement
103 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
106 Frequency : rare on shutdown ost
108 Description: If llog cancel was not send before clean_exports phase, this can
109 produce deadlock in llog code.
110 Details : If llog thread has last reference to obd and call class_import_put
111 this produce deadlock because llog_cleanup_commit_master wait when
112 last llog_commit_thread exited, but this never success because was
113 called from llog_commit_thread.
117 Description: NULL lov_tgts causing MDS oops
118 Details : more safe checks for NULL lov_tgts for avoid oops.
120 Severity : enhancement
122 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
124 Severity : enhancement
126 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
130 Description: make mgs_setparam() handle fsname containing dash
131 Details : fsname containing a dash does not work with lctl conf_param
133 Severity : enhancement
135 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
139 Frequency : rare, in recovery and (or) destroy lovobjid file.
140 Description: rewrite lov objid code.
141 Details : Cleanup for lov objid code, remove scability problems and wrong
142 locking. Fix sending last_id into ost.
144 Severity : enhancement
146 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
148 Severity : enhancement
150 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
151 Details : Use d_move_locked instead of __d_move.
155 Frequency : rare, at shutdown
156 Description: access already free / zero obd_namespace.
157 Details : if client_disconnect_export was called without force flag set,
158 and exist connect request in flight, this can produce access to
159 NULL pointer (or already free pointer) when connect_interpret
160 store ocd flags in obd_namespace.
164 Frequency : only at startup
165 Description: not alloc memory with spinlock held.
166 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
167 if any spinlock held.
172 Description: lfs find does not continue on file error
173 Details : Continue other files processing when a file/dir is absent.
177 Description: Inconsistent usage of lustre_pack_reply()
178 Details : Standardize the usage of lustre_pack_reply() such that it
179 always generate a CERROR on failure.
182 Frequency : very rare
184 Description: Fix replay if there is an un-replied request and open
185 Details : In some cases, older replay request will revert the
186 mcd->mcd_last_xid on MDS which is used to record the client's
189 Severity : enhancement
191 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
193 Severity : enhancement
195 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
197 Severity : enhancement
199 Description: Update to SLES9 kernel-2.6.5-7.287.3.
201 Severity : enhancement
203 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
205 Severity : enhancement
207 Description: Build SLES10 patchless client fails
208 Details : The configure was broken by run ./configure with
209 --with-linux-obj=.... argument for patchless client. When the
210 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
211 can't search header adequately. Use absolute path such as
212 -I($LINUX)/include instead.
214 Severity : enhancement
216 Description: Lustre Page Accounting
217 Details : New macros for page alloc and free which enable accounting
218 of page allocation of Lustre. Use percpu counters to store memory
222 Frequency : only if debugging is disabled
224 Description: LASSERT_{REQ,REP}SWAB macros are buggy
225 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
226 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
227 these macros and replace them with their difinitions instead.
232 Description: interrupt oig_wait produce painc on resend.
233 Details : brw_redo_request can be used for resend requests from ptlrpcd and
234 private set, and this produce situation when rq_ptlrpcd_data not
235 copyed to new allocated request and triggered LBUG on assert
236 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
237 ptlrpcd set if request is changed and can be safety changed to use
240 Severity : enhancement
242 Description: organize the server-side client stats on per-nid basis
243 Details : Change the structure of stats under obdfilter and mds to
251 The "uuid"s file would list the uuids of _active_ exports.
252 And the clear entry is to clear all stats and stale nids.
257 Description: Processes looping in ll_readdir() on Lustre clients finally causing
258 a full node pseudo-hang
259 Details : Concurrent access to the same directory from multiple clients with
260 intensive file creation/removal can cause a client node to spin in
261 ll_readdir(). i_version must be increased every time the lock is
262 cancelled to ensure a revalidate is done.
267 Description: touch file failed when fs is not full
268 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
269 otherwise we can get ENOSP while fs is not full.
272 Frequency : only for Cray XT3
273 Bugzilla : 12829/13455
274 Description: Changing primary group doesn't change the group lustre assigns to
276 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
277 overridden with the primary group provided by the group upcall,
278 whereas the supplied fsgid can be trusted if it is in the list of
279 supplementary groups returned by the group upcall.
283 Description: Root Squash Functionality
284 Details : Implementation of NFS-like root squash capability. Specifically,
285 don't allow someone with root access on a client node to be able
286 to manipulate files owned by root on a server node.
290 Description: Slow trucate/writes to huge files at high offsets.
291 Details : Directly associate cached pages to lock that protect those pages,
292 this allows us to quickly find what pages to write and remove
293 once lock callback is received.
297 Description: Too many locks accumulating on client during NFS usage
298 Details : mds_open improperly used accmode to find out access mode to a
299 file. Also mdc_intent_lock logic to find out if we already have
300 lock similar to just received was flawed since introduction of
301 skiplists - locks are now added to the front of the granted
306 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
308 Details : In obd setup failure handler, obd_stopping will not necessarily to
309 be 1, and obd_set_up should also be checked to make sure whether
310 obd is completely setup.
312 Severity : enhancement
314 Description: Allow masking D_WARNING, D_ERROR messages from console
315 Details : Console messages can now be disabled via lnet.printk.
319 Description: User code with malformed file open parameter crashes client node
320 Details : Before packing join_file req, all the related reference should be
321 checked carefully in case some malformed flags cause fake join_file
327 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
328 grain of quota doing harm to quota's accuracy
329 Details : qunit size will be changed when quota limitation is too low/high;
330 record the pending quota write in order to get more accureate
331 quota; delete the patch for bug12588, which is unnecessary when
332 this patch is landed. This bug also contains fixes for bug 14526,
333 14299, 14601 and 13794.
337 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
338 Details : ldlm_completion_ast() assumes that a lock is granted when the req
339 mode is equal to the granted mode. However, it should also check
340 that LDLM_FL_CP_REQD is not set.
344 Description: Heavy nfs access might result in deadlocks
345 Details : After ELC code landed, it is now improper to enqueue any mds
346 locks under och_sem, because enqueue might want to decide to
347 cancel open locks for same inode we are holding och_sem for.
351 Description: 35% write performance drop with ldiskfs2 when quotas are on
352 Details : Enable ext3 journalled quota by default to improve performance
353 when quotas are turned on.
357 Description: Client eviction while running blogbench
358 Details : A lot of unlink operations with concurrent I/O can lead to a
359 deadlock causing evictions. To address the problem, the number of
360 oustanding OST_DESTROY requests is now throttled to
361 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
362 callbacks are processed in priority.
365 Frequency : RHEL4 only
367 Description: mkfs is very slow on IA64/RHEL4
368 Details : A performance regression has been discovered in the MPT Fusion
369 driver between versions 3.02.73rh and 3.02.99.00rh. As a
370 consequence, we have downgraded the MPT Fusion driver in the RHEL4
371 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
374 Frequency : PPC/PPC64 only
376 Description: conflicts between asm-ppc64/types.h and lustre_types.h
377 Details : fix duplicated definitions between asm-ppc64/types.h and
378 lustre_types.h on PPC.
381 Frequency : PPC/PPC64 only
383 Description: asm-ppc/segment.h does not exist
384 Details : fix compile issue on PPC.
388 Description: data checksumming impacts single node performance
389 Details : add support for several checksum algorithms. Currently, CRC32 and
390 Adler-32 are supported. The checksum type can be changed on the fly
391 through /proc/fs/lustre/osc/*/checksum_type.
395 Description: use adler32 for page checksums
396 Details : when available, use the Adler-32 algorithm instead of CRC32 for
401 Description: better handle error messages in extents code
403 Severity : enhancement
405 Description: SNMP support enhancement
406 Details : Adding total number of sampled request for an MDS node in snmp
409 Severity : enhancement
411 Description: Optimize ldlm waiting list processing for PR extent locks
412 Details : When processing waiting list for read extent lock and meeting read
413 lock that is same or wider to it that is not contended, skip
414 processing rest of the list and immediatelly return current
415 status of conflictness, since we are guaranteed there are no
416 conflicting locks in the rest of the list.
420 Description: Time out and refuse to reconnect
421 Details : When the failover node is the primary node, it is possible
422 to have two identical connections in imp_conn_list. We must
423 compare not conn's pointers but NIDs, otherwise we
424 can defeat connection throttling.
428 Description: Client not clear own cache if answer to reconnect is lost.
429 Details : Client gets evicted from server. Now client also thinks it is
430 disconnected (or gets ENOTCONN on its operation) and decides to
431 reconnect. Server receives reconnect message, but cannot find
432 export. New export is created that is fully valid (new cookie!),
433 but reply is lost and not reported to client. Client reconnects
434 again and gets back a just-created connection, but it is not new
435 so client thinks it was not evicted and does not do recovery.
439 Description: Detect stride IO mode in read-ahead
440 Details : When a client does stride read, read-ahead should detect that and
441 read-ahead pages according to the detected stride pattern.
445 Description: build for x2 fails
446 Details : fix compile issue on Cray systems.
448 --------------------------------------------------------------------------------
450 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
452 * Support for kernels:
453 2.6.5-7.286 (SLES 9),
454 2.6.9-55.0.9.EL (RHEL 4),
455 2.6.16.53-0.8 (SLES 10),
456 2.6.18-8.1.14.el5 (RHEL 5),
457 2.6.18.8 vanilla (kernel.org)
458 * Client support for unpatched kernels:
459 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
460 2.6.16 - 2.6.22 vanilla (kernel.org)
461 * Due to recently discovered recovery problems, we do not recommend
462 using patchless RHEL 4 clients with this or any earlier release.
463 * Recommended e2fsprogs version: 1.40.2-cfs1
464 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
467 Frequency : occasional
469 Description: MDS hang or stay in waiting lock
470 Details : If client receive lock with CBPENDING flag ldlm need send lock
471 cancel as separate rpc, to avoid situation when cancel request
472 can't processed due all i/o threads stay in wait lock.
475 Frequency : occasional
477 Description: Do not fail import if osc_interpret_create gets -EAGAIN
478 Details : If osc_interpret_create got -EAGAIN it immediately exits and
479 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
480 oscc_has_objects and see OSC has no objests and call
481 oscc_internal_create to resend create request.
483 Severity : enhancement
485 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
486 Details : Update which_patch & target file for SLES10 latest kernel.
488 Severity : enhancement
490 Description: add --type and --size parameters to lfs find
491 Details : Enhance lfs find by adding filetype and filesize parameters. Also
492 multiple OBDs can now be specified for the --obd option.
494 Severity : enhancement
496 Description: eliminate client locks in face of contention
497 Details : file contention detection and lockless i/o implementation
500 Severity : enhancement
502 Description: Remove client patches from SLES 10 kernel.
503 Details : This causes SLES 10 clients to behave as patchless clients
504 even on a Lustre-patched (server) kernel.
506 Severity : enhancement
508 Description: use i_size_read and i_size_write in 2.6 port
509 Details : replace inode->i_size access with i_size_read/write()
512 Frequency : when removing large files
514 Description: scheduling issue during removal of large Lustre files
515 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
516 It causes scheduling issues when removing large files (17TB in the
522 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
523 Details : Impossible (EPROTO is returned) to access a directory that has a
524 non-default striping and ACLs.
527 Frequency : only on ppc
529 Description: /proc/fs/lustre/devices broken on ppc
530 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
531 We should make sure the type of 'index' is loff_t and then cast
532 explicitly as needed below. Do not assign an explicitly cast
536 Frequency : only for rhel5
538 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
539 Details : Modify the target file & which_kernel.
542 Frequency : if the uninit_groups feature is enabled on ldiskfs
544 Description: e2fsck reports "invalid unused inodes count"
545 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
546 feature and only a single inode is created in a group then the
547 "bg_unused_inodes" count is incorrectly updated. Creating a
548 second inode in that group would update it correctly.
553 Description: buffer overruns could theoretically occur
554 Details : llapi_semantic_traverse() modifies the "path" argument by
555 appending values to the end of the origin string, and a buffer
556 overrun may occur. Adding buffer overrun check in liblustreapi.
558 Severity : enhancement
560 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
563 Frequency : only if filesystem is inconsistent
565 Description: handle "serious error: objid * already exists" more gracefully
566 Details : If LAST_ID value on disk is smaller than the objects existing in
567 the O/0/d* directories, it indicates disk corruption and causes an
568 LBUG(). If the object is 0-length, then we should use the existing
569 object. This will help to avoid a full fsck in most cases.
571 Severity : enhancement
573 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
574 Details : Modify vm-tunables-rhel4.patch.
576 Severity : enhancement
578 Description: Kernel config for 2.6.18-vanilla.
579 Details : Modify targets/2.6-vanilla.target.in.
580 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
581 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
582 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
583 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
586 Frequency : occasional
588 Description: improve handling recoverable errors
589 Details : If request processed with error which can be recoverable on server
590 request should be resend, otherwise page released from cache and
596 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
597 Details : Modify target file & which_patch.
598 A flaw was found in the IA32 system call emulation provided
599 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
600 value could be stored in the %RAX register, which could trigger an
601 out-of-bounds system call table access. An untrusted local user
602 could exploit this flaw to run code in the kernel
603 (ie a root privilege escalation). (CVE-2007-4573).
607 Description: change order of libsysio includes
608 Details : '#include sysio.h' should always come before '#include xtio.h'
610 Severity : enhancement
612 Description: adapt the lustre_config script to support the upgrade case
613 Details : Add "-u" option for lustre_config script to support upgrading 1.4
614 server targets to 1.6 in parallel.
619 Description: To avoid grant space > avaible space when the disk is almost
620 full. Without this patch you might see the error "grant XXXX >
621 available" or some LBUG about grant, when the disk is almost
623 Details : In filter_check_grant, for non_grant cache write, we should
624 check the left space by if (*left > ungranted + bytes), instead
625 of (*left > ungranted), because only we are sure the left space
626 is enough for another "bytes", then the ungrant space should be
627 increase. In client, we should update cl_avail_grant only there is
628 OBD_MD_FLGRANT in the reply.
633 Description: Update RHEL 4 kernel to fix local root privilege escalation.
634 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
635 described in CVE-2007-4573. This problem could allow untrusted
636 local users to gain root access.
639 Frequency : when using O_DIRECT and quotas
641 Description: Incorrect file ownership on O_DIRECT output files
642 Details : block usage reported by 'lfs quota' does not take into account
643 files that have been written with O_DIRECT.
645 --------------------------------------------------------------------------------
647 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
649 * Support for kernels:
650 2.6.5-7.286 (SLES 9),
651 2.6.9-55.0.2.EL (RHEL 4),
652 2.6.16.46-0.14 (SLES 10),
653 2.6.18-8.1.8.el5 (RHEL 5),
654 2.6.18.8 vanilla (kernel.org)
655 * Client support for unpatched kernels:
656 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
657 2.6.16 - 2.6.21 vanilla (kernel.org)
658 * Due to recently discovered recovery problems, we do not recommend
659 using patchless RHEL 4 clients with this or any earlier release.
660 * Recommended e2fsprogs version: 1.40.2-cfs1
661 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
665 Description: Fix errors in lfs documentation
666 Details : Fixes man pages
668 Severity : enhancement
670 Description: Adaptive timeouts
671 Details : RPC timeouts adapt to changing server load and network
672 conditions to reduce resend attempts and improve recovery time.
674 Severity : enhancement
676 Description: llapi_file_create() does not allow some changes
677 Details : add llapi_file_open() that allows specifying the file creation
678 mode and open flags, and also returns an open file handle.
680 Severity : enhancement
682 Description: df doesn't work properly if diskfs blocksize != 4K
683 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
686 Frequency : always on directories with default striping set
688 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
689 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
691 Severity : enhancement
693 Description: merge and cleanup kernel patches.
694 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
696 Severity : enhancement
698 Description: RedHat Update kernel for RHEL5
699 Details : Kernel config file for RHEL5.
701 Severity : enhancement
703 Description: OSS needs mutliple precreate threads
704 Details : Add ability to start more than one create thread per OSS.
707 Frequency : only with quota on the root user
709 Description: mds_obd_create error creating tmp object
710 Details : When the user sets quota on root, llog will be affected and can't
711 create files and write files.
714 Frequency : Always on ia64 patchless client, and possibly others.
716 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
717 Details : This allows the patchless client to be loaded on architectures
723 Description: disorder of journal start and llog_add cause deadlock.
724 Details : in llog_origin_connect, journal start should happen before
725 llog_add keep the same order as other functions to avoid
728 Severity : enhancement
730 Description: RedHat Update kernel for RHEL5
731 Details : Modify the kernel config file more closer RHEL5.
733 Severity : enhancement
735 Description: Build failure against Centos5 (RHEL5)
736 Details : Define PAGE_SIZE when it isn't present.
739 Frequency : occasionally when using NFS
741 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
742 Details : with NFS, the anon dentry's parent was set to itself in
743 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
744 corresponding dentry other than use rec->ur_name.
746 Severity : enhancement
748 Description: enable data checksumming by default
749 Details : enable checksum by default, allow --disable-checksum
750 configure option and "-o nochecksum" mount option. Checksums
751 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
754 Frequency : Occasionally with failover
756 Description: Client eviction due to failover config
757 Details : after a connection loss, the lustre client should attempt to
758 reconnect to the last active server first before trying the
759 other potential connections.
761 Severity : enhancement
763 Description: client-side metadata stat-ahead during readdir(directory readahead)
764 Details : perform client-side metadata stat-ahead when the client detects
765 readdir and sequential stat of dir entries therein
770 Description: when mds and osts use different quota unit(32bit and 64bit),
771 quota will be released repeatly.
772 Details : void sending multiple quota reqs to mds, which will keep the
773 status between the reqs.
776 Frequency : only with liblustre clients on XT3
778 Description: evictions taking too long
779 Details : allow llrd to evict clients directly on OSTs
783 Description: osts not allocated evenly to files
784 Details : change the condition to increase offset_idx
786 Severity : enhancement
788 Description: self-adjustable client's lru lists
789 Details : use adaptive algorithm for managing client cached locks lru
790 lists according to current server load, other client's work
791 pattern, memory activities, etc. Both, server and client
792 side namespaces provide number of proc tunables for controlling
797 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
798 Details : rewrite inherited code (uuid parsing code from ext2 utils and
799 readdir code from ext3) from scratch preserving functionality.
803 Description: Only those disconnect error should be returned by rq_status.
804 Details : In open/enqueue processs, Some errors, which will cause client
805 disconnected, should be returned by rq_status, while other
806 errors should still be returned by intent, then mdc or llite will
809 Severity : enhancement
811 Description: Tune the kernel for good SCSI performance.
812 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
813 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
817 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
818 Bugzilla : 13547, 13627
819 Description: Data corruption for OSTs that are formatted larger than 2TB
821 Details : When generating the bio request for lustre file writes the
822 sector number would overflow a temporary variable before being
823 used for the IO. The data reads correctly from Lustre (which
824 will overflow in a similar manner) but other file data or
825 filesystem metadata may be corrupted in some cases.
829 Description: "lfs find -obd UUID" prints directories
830 Details : "lfs find -obd UUID" will return all directory names instead
831 of just file names. It is incorrect because the directories
832 do not reside on the OSTs.
835 Bugzilla : 13596, 13614
836 Description: MDS hang after unclean shutdown of lots of clients
837 Details : Never resend AST requests.
841 Frequency : Always, for kernels after 2.6.16
842 Description: Fix warning idr_remove called for id=.. which is not allocated.
843 Details : Last kernels save old s_dev before kill super and not allow
844 to restore from callback - restore it before call kill_anon_super.
848 Description: Fix error on 'ls .' at the top of the Lustre mount.
849 Details : Don't revalidate dentry if it is a root dentry.
851 --------------------------------------------------------------------------------
853 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
855 * Support for kernels:
856 2.6.5-7.286 (SLES 9),
857 2.6.9-55.0.2.EL (RHEL 4),
858 2.6.16.46-0.14 (SLES 10),
859 2.6.18-8.1.8.el5 (RHEL 5),
860 2.6.18.8 vanilla (kernel.org)
861 * Client support for unpatched kernels:
862 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
863 2.6.16 - 2.6.22 vanilla (kernel.org)
864 * Due to recently discovered recovery problems, we do not recommend
865 using patchless RHEL 4 clients with this or any earlier release.
866 * Recommended e2fsprogs version: 1.39.cfs8
867 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
872 Description: block reactivating mgc import until all deactivates complete
873 Details : Fix race when failing back MDT/MGS to itself (testing)
875 Severity : enhancement
877 Description: lfs setstripe enhancement
878 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
881 Frequency : mds/oss recovery
883 Description: llog ctxt is refrenced after it has been freed.
884 Details : llog ctxt refcount was added to avoide the race between ctxt free
885 and llog recovery process. Each llog user must hold ctxt refcount
886 before it access the llog. And the llog ctxt can only be freed
887 when its refcount is zero.
889 Severity : enhancement
891 Description: randomly memory allocation failure util
892 Details : Make lustre randomly failed allocating memory for testing purpose.
895 Frequency : only for SLES10
897 Description: Update kernel patch for SLES10 SP1
898 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
900 Severity : enhancement
902 Description: omit set fsid for export NFS
903 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
904 as FS_REQUIRES_DEV for avoid problems with generate fsid.
907 Frequency : after network failures
909 Description: Add sync option to mount_lustre.c
910 Details : Client loses data written to lustre after a network interruption.
912 Severity : enhancement
914 Description: Error message improvement.
915 Details : Merging of two LCONSOLE_ERROR_MSG into one.
917 Severity : enhancement
919 Description: don't use GFP_* in generic Lustre code.
920 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
922 Severity : enhancement
924 Description: obdclass is limited by single OBD_ALLOC(idarray)
925 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
927 Severity : enhancement
929 Description: updated patchess for new RHEL4 kernel
930 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
932 Severity : enhancement
934 Description: warnings with build patchless client with vanila 2.6.19 and up
935 Details : change old ctl_table style and replace ctl_table/ctl_table_header
936 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
938 Severity : enhancement
940 Description: O_DIRECT bypasses client statistics.
941 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
942 read_ahead_stats were updated. Copied stats section from
943 osc_send_oap_rpc() into async_internal().
946 Frequency : only for Cray XT3
948 Description: peer credits not enough on many OST per OSS systems.
949 Details : Use new lnet way to add credits as we need those for pings and ASTs
952 Frequency : only with liblustre
954 Description: Liblustre is not releasing flock locks on file close.
955 Details : Release flock locks on file close.
958 Frequency : only for RHEL4
960 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
961 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
962 Update target file and kernel config.
966 Description: lustre support for RHEL5
967 Details : Add support for RHEL5.
972 Description: ASSERTION(export != NULL) failed in target_handle_connect
973 Details : Assetion hit is result of rare race between disconnect and connet
974 to same nid. target_handle_connect found old connect cockie and
975 tried to reconnect, but can't find export for this cockie.
980 Description: umount blocks forever on error
981 Details : In result of wrong using obd_no_recov and obd_force flags client
982 can hand if cancel or some other requests is lost.
985 Frequency : Only for SLES
987 Description: sanity_quota fail test_1
988 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
989 file, which makes TSTID[2] inunique.
991 Severity : enhancement
993 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
994 Details : Update target/ChangeLog/which_patch .
996 Severity : enhancement
998 Description: jbd statistics
999 Details : Port older jbd statistics patch for sles10
1004 Description: lvbo_init failed for resource with missing objects.
1005 Details : Fix returning error if we do stat for file with missing/corrupted
1006 objects and i_size set to all sum of size of all avaible objects.
1007 if we to truncate/write to missing object - it is recreated.
1012 Description: Oops in read and write path when failing to allocate lock.
1013 Details : Check if lock allocation failed and return error back.
1016 Frequency : When flocks are used.
1018 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1019 Details : Flock locks might destroy just granted lock if it could be merged
1020 with another existing flock, this is done in completion handler,
1021 so teach ldlm_cli_enquque_fini that this is a valid case for
1027 Description: reply_lock_interpret crash due to race with it and lock cancel.
1028 Details : Do not replay locks that are being cancelled. Do not reference
1029 locks by their address during replay, just by their handle.
1032 Frequency : only with deactivated OSTs
1034 Description: lstripe command fails for valid OST index
1035 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
1036 instead of lov->desc.ld_active_tgt_count.
1038 Severity : enhancement
1040 Description: Build failure against Centos5 (RHEL5)
1041 Details : Use getpagesize() instead of PAGE_SIZE.
1043 --------------------------------------------------------------------------------
1045 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
1047 * Support for kernels:
1048 2.6.5-7.283 (SLES 9),
1049 2.6.9-55.EL (RHEL 4),
1050 2.6.16.46-0.14 (SLES 10),
1051 2.6.18.8 vanilla (kernel.org)
1052 * Client support for unpatched kernels:
1053 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1054 2.6.16 - 2.6.22 vanilla (kernel.org)
1055 * Due to recently discovered recovery problems, we do not recommend
1056 using patchless RHEL 4 clients with this or any earlier release.
1057 * Recommended e2fsprogs version: 1.39.cfs8
1058 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1059 * Starting with this release, the ldiskfs backing filesystem required
1060 by Lustre is now in its own package, lustre-ldiskfs. This package
1061 should be installed. It is versioned separately from Lustre and
1062 may be released separately in future.
1067 Description: server LBUG when shutting down
1068 Details : Block umount forever until the mount refcount is zero rather
1069 than giving up after an arbitrary timeout.
1071 Severity : enhancement
1073 Description: add optional extra BUILD_VERSION info
1074 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1075 to override the lustre version.
1078 Frequency : 2.6.18 servers only
1080 Description: ll_kern_mount() doesn't release the module reference
1081 Details : The ldiskfs module reference count never drops down to 0
1082 because ll_kern_mount() doesn't release the module reference.
1087 Description: server LBUG when using old ost_num_threads parameter
1088 Details : Accept the old ost_num_threads parameter but warn that it
1089 is deprecated, and fix an off-by-one error that caused an LBUG.
1094 Description: Transient SCSI error results in persistent IO issue
1095 Details : iobuf->dr_error is not reinitialized to 0 between two
1099 Frequency : sometimes when underlying device returns I/O errors
1101 Description: OSTs not going read-only during write failures
1102 Details : OSTs are not remounted read-only when the journal commit threads
1103 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1104 instead of the ext3 wrappers.
1107 Frequency : only with 10000 clients or more
1109 Description: poor connect scaling with increasing client count
1110 Details : Don't run filter_grant_sanity_check for more than 100 exports
1111 to improve scaling for large numbers of clients.
1114 Frequency : SLES10 only
1116 Description: sanity-quota.sh quotacheck failed: rc = -22
1117 Details : Quotas cannot be enabled on SLES10.
1120 Frequency : liblustre clients only
1122 Description: getdirentries does not give error when run on compute nodes
1123 Details : getdirentries does not fail when the size specified as an argument
1124 is too small to contain at least one entry
1126 Severity : enhancement
1128 Description: Add LNET router traceability for debug purposes
1129 Details : If a checksum failure occurs with a router as part of the
1130 IO path, the NID of the last router that forwarded the bulk data
1131 is printed so it can be identified.
1136 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1137 Details : Due to a race condition, liblustre clients were occasionally
1138 evicted incorrectly.
1140 Severity : enhancement
1142 Description: lfs setstripe use optional parameters instead of postional
1145 Severity : enhancement
1147 Description: Nanosecond timestamp support for ldiskfs
1148 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1149 resolution timestamps. There is not yet support for this at
1150 the Lustre filesystem level.
1153 Frequency : during server recovery
1155 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1156 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1160 Frequency : nfs export on patchless client
1162 Description: connectathon hang when test nfs export over patchless client
1163 Details : Disconnected dentry cannot be found with lookup, so we do not need
1164 to unhash it or make it invalid
1167 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1168 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1169 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1174 Description: open req refcounting wrong on reconnect
1175 Details : If reconnect happened between getting open reply from server and
1176 call to mdc_set_replay_data in ll_file_open, we will schedule
1177 replay for unreferenced request that we are about to free.
1178 Subsequent close will crash in variety of ways.
1179 Check that request is still eligible for replay in
1180 mdc_set_replay_data().
1185 Description: disable writes to filesystem when reading health_check file
1186 Details : the default for reading the health_check proc file has changed
1187 to NOT do a journal transaction and write to disk, because this
1188 can cause reads of the /proc file to hang and block HA state
1189 checking on a healthy but otherwise heavily loaded system. It
1190 is possible to return to the previous behaviour during configure
1191 with --enable-health-write.
1193 Severity : enhancement
1195 Description: 64-bit inode version
1196 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1197 to the inode. This will be required for version-based recovery.
1202 Description: MDS fails to start if a duplicate client export is detected
1203 Details : in some rare cases it was possible for a client to connect to
1204 an MDS multiple times. Upon recovery the MDS would detect this
1205 and fail during startup. Handle this more gracefully.
1207 Severity : enhancement
1209 Description: Add -o localflock option to simulate old noflock behaviour.
1210 Details : This will achieve local-only flock/fcntl locks coherentness.
1215 Description: log_commit_thread vs filter_destroy race leads to crash
1216 Details : Take import reference before releasing llog record semaphore
1221 Description: Wrong request locking in request set processing
1222 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1223 imp_delayed_list, in this place should be used imp_lock.
1226 Frequency : when reconnecting
1228 Description: Grant leak when OSC reconnect to OST
1229 Details : When osc reconnect ost, OST(filter) should check whether it
1230 should grant more space to client by comparing fed_grant and
1231 cl_avail_grant, and return the granted space to client instead
1232 of "new granted" space, because client will call osc_init_grant
1233 to update the client grant space info.
1236 Frequency : when client reconnects to OST
1238 Description: Grant leak when OSC does a resend and replays bulk write
1239 Details : When osc reconnect to OST, OST(filter) should clear grant info of
1240 bulk write request, because the grant info while be sync between
1241 OSC and OST when reconnect, and we should ignore the grant info
1242 these of resend/replay write req.
1247 Description: Grant space more than avaiable space sometimes.
1248 Details : When then OST is about to be full, if two bulk writing from
1249 different clients came to OST. Accord the avaliable space of the
1250 OST, the first req should be permitted, and the second one
1251 should be denied by ENOSPC. But if the seconde arrived before
1252 the first one is commited. The OST might wrongly permit second
1253 writing, which will cause grant space > avaiable space.
1256 Frequency : when client is evicted
1258 Description: Grant might be wrongly erased when osc is evicted by OST
1259 Details : when the import is evicted by server, it will fork another
1260 thread ptlrpc_invalidate_import_thread to invalidate the
1261 import, where the grant will be set to 0. While the original
1262 thread will update the grant it got when connecting. So if
1263 the former happened latter, the grant will be wrongly errased
1264 because of this race.
1269 Description: Checking Stale with correct fid
1270 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1271 to check whether it is stale, because sometimes, we want the
1272 enqueue happened anyway, and op_data.fid2 will not be initialized.
1274 Severity : enhancement
1276 Description: update patchless client
1277 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1280 Frequency : only with 2.4 kernel
1282 Description: random memory corruption
1283 Details : size of struct ll_inode_info is to big for union inode.u and this
1284 can be cause of random memory corruption.
1289 Description: Memory leak in recovery
1290 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1291 It should also check obd_fail before fsfilt_start, otherwise if
1292 fsfilt_start return -EROFS,(failover mds during mds recovery).
1293 then the req will return with repmsg->transno = 0 and rc = EROFS.
1294 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1295 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1296 be freed no matter whether fsfilt_commit success or not.
1299 Frequency : only with huge numbers of clients
1301 Description: Prevents from taking the superblock lock in llap_from_page for
1303 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1304 for llap_from_page call in ll_removepage() prevents from taking
1305 the superblock lock for a soon died page.
1310 Description: Not check open intent error before release open handle
1311 Details : in some rare cases, the open intent error is not checked before
1312 release open handle, which may cause
1313 ASSERTION(open_req->rq_transno != 0), because it tries to release
1314 the failed open handle.
1319 Description: Set cat log bitmap only after create log success.
1320 Details : in some rare cases, the cat log bitmap is set too early. and it
1321 should be set only after create log success.
1325 Description: Accessing a block bevice can re-enable I/O when Lustre is
1326 tearing down a device.
1327 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1332 Description: service threads may hog cpus when there are a lot of requests
1333 Details : Insert cond_resched to give other threads a chance to use some CPU
1338 Description: the cat log was not initialized in recovery
1339 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1340 unlink log on mds will not be initialized until mds post
1341 recovery. And also in mds post recovery, the unlink log will
1342 initialization will be done asynchronausly, so there will be race
1343 between add unlink log and unlink log initialization.
1347 Description: brw_stats were being printed incorrectly
1348 Details : brw_stats were being printed as log2 but all of them were not
1349 recorded as log2. Also remove some code duplication arising from
1350 filter_tally_{read,write}.
1354 Frequency : rare, only in recovery.
1355 Description: ASSERTION(req->rq_type != LI_POISON) failed
1356 Details : imp_lock should be held while iterating over imp_sending_list for
1357 prevent destroy request after get timeout in ptlrpc_queue_wait.
1361 Description: replay-single.sh test 52 fails
1362 Details : A lock's skiplist need to be cleanup when it being unlinked
1363 from its resource list.
1368 Description: Short directio read returns full requested size rather than
1370 Details : Direct I/O operations should return actual amount of bytes
1371 transferred rather than requested size.
1373 Severity : enhancement
1375 Description: metadata RPC reduction (e.g. for rm performance)
1376 Details : decrease the amount of synchronous RPC between clients and servers
1377 by canceling conflicing lock before the operation on the client
1378 and packing thier handles into the main operation RPC to server.
1380 Severity : enhancement
1382 Description: add #ifdef HAVE_KERNEL_CONFIG_H
1383 Details : kernels from 2.6.19 not need include linux/config.h, but add
1384 include linux/autoconf.h in commpiler command line.
1386 Severity : enhancement
1388 Description: patchless client support for 2.6.22 kernel
1389 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
1390 removed. In this case we need drop using os depended interface to
1391 kmem_cache and use cfs_mem_cache API.
1396 Description: fix mal-formatted messages
1397 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
1401 Frequency : always in liblustre
1402 Description: wrong IS_ERR implementation in liblustre.h
1403 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
1408 Description: Correct condition for output debug message.
1409 Details : inode i_nlink equal zero is not enough for output message about
1410 disk corruption, i_ctime and i_mode should be also checked.
1414 Frequency : always in patchless client
1415 Description: add configure check for truncate_complete_page
1416 Details : improve checks for exported symbols. This allow run check without
1417 sources, but with Module.symvers shipped with kernel distribution.
1418 add check for truncate_complete_page used by patchless client.
1422 Description: sanity.sh test_77h fails with "test_77h file compare failed"
1423 Details : test_77h uses a file which was messed by other test case.
1427 Description: Not Check whether lov_tgts is NULL in some lov functions
1428 Details : Checking whether lov_tgts is NULL in some functions.
1432 Description: replace obdo_alloc() with OBDO_ALLOC macro
1433 Details : nothing special is done in obdo_alloc() function, and for
1434 debugging purpose, it needs to be replaced with macros.
1438 Description: bad return value and errno from fcntl call
1439 Details : In liblustre API, errno should be a negative value if error
1444 Description: ptlrpc_check_set() LBUG
1445 Details : In case of positive reply from server and failed client bulk
1446 callback after bulk transfer shouldn't LBUG, but process this
1447 request as erroneous.
1449 Severity : enhancement
1451 Description: VFS operations stats tool.
1452 Details : Tool which collects stats by tracking value written in pid,
1453 ppid, gid and uses llstat to generate output to plot graph using
1455 Updated lustre/utils/Makefile.am
1456 Added lustre/utils/ltrack_stats.c
1458 Severity : enhancement
1460 Description: 2.6.18 server support (lustre 1.6.1)
1461 Details : Support for 2.6.18 kernels on the server side.
1466 Description: ASSERTION(imp->imp_conn_current) failed
1467 Details : an assertion failure is hit if a client node boots and attempts to
1468 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
1471 Frequency : only for i686
1473 Description: 1.4.11 RC1 build fails for RHEL 4, i686
1474 Details : Fixed config variable for build.
1479 Description: Updated patchess for new RHEL4 kernel
1480 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
1481 Updated patch jbd-stats-2.6.9.patch
1482 Updated patch qsnet-rhel4-2.6.patch
1483 Updated patch quota-deadlock-on-pagelock-core.patch
1484 Updated patch vfs_intent-2.6-rhel4.patch
1485 Updated patch vfs_races-2.6-rhel4.patch
1486 Updated series file 2.6-rhel4-titech.series
1487 Updated series file 2.6-rhel4.series
1488 Updated kernel config files
1493 Description: lquota slave complains LBUG when reconnecting with mds
1495 Details : quota slave depends qctxt->lqc_import to send its quota request.
1496 This pointer will be invalid if mds did failover or broke its
1497 connect to osts, which leads to LBUG.
1500 Frequency : when qunit size is too small(less than 20M)
1502 Description: write is stopped by improper -EDQUOT
1503 Details : If the master is busy and qunit size is small enough(let's say 1M),
1504 the slave can not get quota from master on time, which will lead
1505 slave to trigger a -EQUOTA to client.
1510 Description: Deadlock during metadata tests
1511 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
1512 called with the per-dentry lock held.
1515 Frequency : SLES9 only
1517 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
1518 Details : set CONFIG_NR_CPUS to 128 instead of 8.
1520 Severity : enhancement
1522 Description: remove fs_prep_san_write operation and related patches
1523 Details : remove the ext3-san-jdike patches which are no longer useful.
1528 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
1529 Details : a disk corruption can cause the mballoc code to assert on a
1530 double free or other extent corruptions. Handle these with
1531 ext3_error() instead of assertions.
1534 Frequency : only with mballoc3 code and deep extent trees
1536 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
1537 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
1538 can cause the filesystem to be remounted read-only.
1543 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
1544 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
1545 few external users of this ioctl number is lfsck's call to
1546 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
1547 Add the old ioctl number to the handler so both old and new
1552 Description: parallel lock callbacks
1553 Details : Instead of sending blocking and completion callbacks as separated
1554 requests, adding them to a set and sending in parallel.
1558 Description: Disable most debugging by default
1559 Details : To improve performance, disable most logging (for debug purposes)
1560 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
1561 default, and HA includes fewer messages.
1565 Frequency : only run on patchless client.
1566 Description: use do_facet on sanity.sh for test handling recoverables errors
1567 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1571 Description: hash tables for lists of nids, connections and uuids
1572 Details : Hash tables noticeably help when a lot of clients connect to a
1573 server, to faster identify duplicate connections or reconnects,
1574 also to faster find export to evict in manual eviction case.
1578 Description: Sometimes, when the server evict a client, and the client will
1579 not be evicted as soon as possible.
1580 Details : In enqueue req, the error was returned by intent, instead of
1581 rq_status which make ptlrpc layer not detect this error, and
1582 does not evict the client. So enqueue error should be returned
1586 Frequency : only at startup
1588 Description: Delay client connections to MDT until fist MDT->OST connect
1589 Details : If a client tried to create a new file before the MDT had
1590 connected to any OSTs, the create would return EIO. Now
1591 the client will simply block until the MDT connects to the
1592 first OST and the create can succeed.
1595 Frequency : only for SLES9
1597 Description: Routinely utilize latest Quadrics drivers in CFS releases
1598 Details : Update patch qsnet-suse-2.6.patch.
1601 Frequency : only for sles10
1603 Description: Update patches for SLES 10 SP1 kernel.
1604 Details : Update the patch vfs_intent-2.6-sles10.patch.
1609 Description: Routinely utilize latest Quadrics drivers in CFS releases
1610 Details : Update patch qsnet-rhel4-2.6.patch.
1613 Frequency : at statup only
1615 Description: mds_lov_synchronize race leads to various problems
1616 Details : simultaneous MDT->OST connections at startup can cause the
1617 sync to abort, leaving the OSC in a bad state.
1622 Description: Using wrong pointer in osc_brw_prep_request
1623 Details : Access to array[-1] can produce panic if kernel compiled with
1624 CONFIG_PAGE_ALLOC enabled
1626 Severity : enhancement
1628 Description: Async OSC create to avoid the blocking unnecessarily.
1629 Details : If a OST has no remain object, system will block on the creating
1630 when need to create a new object on this OST. Now, ways use
1631 pre-created objects when available, instead of blocking on an
1632 empty osc while others are not empty. If we must block, we block
1633 for the shortest possible period of time.
1637 Frequency : only in recovery
1638 Description: Mark OST as early accessible if his start SYNC.
1639 Details : osc_precreate return flag early accessible if oscc marked as
1640 OSCC_FLAG_SYNC_IN_PROGRESS.
1645 Description: Sometimes precreate code can triger create object on wrong ost
1646 Details : Wrong protected or not not restored variables aftre precreate loop
1647 can produce creation object on wrong ost.
1650 Frequency : oss recovery
1652 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1654 Details : llog_commit_thread_count should be synced between llog_commit
1655 start and cleanup, so new llog_commit thread should not be started
1656 when llog_commit threads being stopped to avoid accessing some
1659 Severity : enhancement
1661 Description: Add printing inode info into message about error in writepage.
1663 --------------------------------------------------------------------------------
1665 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1670 Frequency : on some architectures
1672 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1673 Details : Uninitialized flags sometimes cause configuration commands to
1677 Frequency : patchless clients only
1679 Description: missing __iget() symbol export
1680 Details : The __iget() symbol export is missing. To avoid the need for
1681 this on patchless clients the deathrow inode reaper is turned
1682 off, and we depend on the VM to clean up old inodes. This
1683 dependency was during via the fix for bug 12181.
1688 Description: sanity.sh fail: test_52b
1689 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1690 incorrect inode's flags to client.
1692 --------------------------------------------------------------------------------
1694 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1696 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1697 INTEROPERATE with older versions automatically. In many cases a
1698 special upgrade step is needed. Please read the
1699 user documentation before upgrading any part of a 1.4.x system.
1700 * WARNING: Lustre configuration and startup changes are required with
1701 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1703 * Support for kernels:
1704 2.4.21-47.0.1.EL (RHEL 3),
1705 2.6.5-7.283 (SLES 9),
1706 2.6.9-42.0.10.EL (RHEL 4),
1707 2.6.12.6 vanilla (kernel.org),
1708 2.6.16.27-0.9 (SLES10)
1709 * Client support for unpatched kernels:
1710 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1711 2.6.16 - 2.6.19 vanilla (kernel.org),
1712 2.6.9-42.0.8EL (RHEL 4)
1713 * Recommended e2fsprogs version: 1.39.cfs6
1714 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1717 Severity : enhancement
1719 Description: Async OSC create to avoid the blocking unnecessarily.
1720 Details : If a OST has no remain object, system will block on the creating
1721 when need to create a new object on this OST. Now, ways use
1722 pre-created objects when available, instead of blocking on an
1723 empty osc while others are not empty. If we must block, we block
1724 for the shortest possible period of time.
1726 Severity : enhancement
1728 Description: MountConf
1729 Details : Lustre configuration is now managed via mkfs and mount
1730 commands instead of lmc and lconf. New obd types (MGS, MGC)
1731 are added for dynamic configuration management. See
1732 https://mail.clusterfs.com/wikis/lustre/MountConf for
1735 Severity : enhancement
1737 Description: dynamic OST addition
1738 Details : OSTs can now be added to a live filesystem
1740 Severity : enhancement
1742 Description: startup order invariance
1743 Details : MDTs and OSTs can be started in any order. Clients only
1744 require the MDT to complete startup.
1746 Severity : enhancement
1748 Description: parallel, asynchronous orphan cleanup
1749 Details : orphan cleanup is now performed in separate threads for each
1750 OST, allowing parallel non-blocking operation.
1752 Severity : enhancement
1754 Description: optimized stripe assignment
1755 Details : stripe assignments are now made based on ost space available,
1756 ost previous usage, and OSS previous usage, in order to try
1757 to optimize storage space and networking resources.
1759 Severity : enhancement
1761 Description: Permanently set tunables
1762 Details : All writable /proc/fs/lustre tunables can now be permanently
1763 set on a per-server basis, at mkfs time or on a live
1766 Severity : enhancement
1768 Description: Lustre message v2
1769 Details : Add lustre message format v2.
1771 Severity : enhancement
1773 Description: client OST exclusion list
1774 Details : Clients can be started with a list of OSTs that should be
1775 declared "inactive" for known non-responsive OSTs.
1779 Description: ENOENT returned for valid filehandle during dbench.
1780 Details : Check if a directory has children when invalidating dentries
1781 associated with an inode during lock cancellation. This fixes
1782 an incorrect ENOENT sometimes seen for valid filehandles during
1783 testing with dbench.
1786 Frequency : SFS test only (otherwise harmless)
1788 Description: SPEC SFS validation failure on NFS v2 over lustre.
1789 Details : Changes the blocksize for regular files to be 2x RPC size,
1790 and not depend on stripe size.
1792 Severity : enhancement
1794 Description: fine-grained SMP locking inside DLM
1795 Details : Improve DLM performance on SMP systems by removing the single
1796 per-namespace lock and replace it with per-resource locks.
1798 Severity : enhancement
1800 Description: don't hold multiple extent locks at one time
1801 Details : To avoid client eviction during large writes, locks are not
1802 held on multiple stripes at one time or for very large writes.
1803 Otherwise, clients can block waiting for a lock on a failed OST
1804 while holding locks on other OSTs and be evicted.
1806 Severity : enhancement
1808 Description: Multiple MD RPCs in flight.
1809 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1810 To avoid overly-overloading MDT, introduce a limit on number of
1811 MDT RPCs in flight for a single client and add /proc controls
1812 to adjust this limit.
1814 Severity : enhancement
1816 Description: client read/write statistics
1817 Details : Add client read/write call usage stats for performance
1818 analysis of user processes.
1819 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1820 file access. extents_stats shows chunk size distribution.
1821 extents_stats_per_process show chunk size distribution per
1824 Severity : enhancement
1826 Description: per-client statistics on server
1827 Details : Add ldlm and operations statistics for each client in
1828 /proc/fs/lustre/mds|obdfilter/*/exports/
1830 Severity : enhancement
1832 Description: improved MDT statistics
1833 Details : Add detailed MDT operations statistics in
1834 /proc/fs/lustre/mds/*/stats
1836 Severity : enhancement
1838 Description: VFS operations stats
1839 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1840 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1845 Description: Fix client-side osc byte counters
1846 Details : The osc read/write byte counters in
1847 /proc/fs/lustre/osc/*/stats are now working
1850 Frequency : always as root on SLES
1852 Description: Failure of copying files with lustre special EAs.
1853 Details : Client side always return success for setxattr call for lustre
1854 special xattr (currently only "trusted.lov").
1859 Description: Refcount LNET uuids
1860 Details : The global LNET uuid list grew linearly with every startup;
1861 refcount repeated list entries instead of always adding to
1864 Severity : enhancement
1866 Description: Dynamic service threads
1867 Details : Within a small range, start extra service threads
1868 automatically when the request queue builds up.
1871 Frequency : mixed-endian client/server environments
1873 Description: mixed-endian crashes
1874 Details : The new msg_v2 system had some failures in mixed-endian
1877 Severity : enhancement
1879 Description: Easy OST removal
1880 Details : OSTs can be permanently deactivated with e.g. 'lctl
1881 conf_param lustre-OST0001.osc.active=0'
1883 Severity : enhancement
1885 Description: MGS proc entries
1886 Details : Added basic proc entries for the MGS showing what filesystems
1889 Severity : enhancement
1891 Description: provide MGS failover
1892 Details : Added config lock reacquisition after MGS server failover.
1894 Severity : enhancement
1896 Description: add Linux 2.4 support
1897 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1901 Description: a large application tries to do I/O to the same resource and dies
1902 in the middle of it.
1903 Details : Check the req->rq_arrival time after the call to
1904 ost_brw_lock_get(), but before we do anything about
1905 processing it & sending the BULK transfer request. This
1906 should help move old stale pending locks off the queue as
1907 quickly as obd_timeout.
1910 Frequency : when an incorrect nid is specified during startup
1912 Description: ptlrpc connect to non-existant node causes kernel crash
1913 Details : LNET can't be re-entered from an event callback, which
1914 happened when we expire a message after the export has been
1915 cleaned up. Instead, hand the zombie cleanup off to another
1918 Severity : enhancement
1920 Description: plain/inodebits lock performance improvement
1921 Details : Grouping plain/inodebits in granted list by their request modes
1922 and bits policy, thus improving the performance of search through
1926 Frequency : only if OST filesystem is corrupted
1928 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1929 Details : for a short time RPCs with bulk IO are in the replay list,
1930 but replay of bulk IOs is unimplemented. If the OST filesystem
1931 is corrupted due to disk cache incoherency and then replay is
1932 started it is possible to trip an assertion. Avoid putting
1933 committed RPCs into the replay list at all to avoid this issue.
1936 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1938 Description: System hang on startup
1939 Details : This bug allowed the liblustre (e.g. catamount) client to
1940 return to the app before handling all startup RPCs. This
1941 could leave the node unresponsive to lustre network traffic
1942 and manifested as a server ptllnd timeout.
1944 Severity : enhancement
1946 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1947 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1948 boolean to control whether to print peer debug info when a
1949 client's RPC times out.
1952 Frequency : only for kernels with patches from Lustre below 1.4.3
1954 Description: Remove old rdonly API
1955 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1958 Frequency : only for devices with external journals
1960 Description: Set external device read-only also
1961 Details : During a commanded failover stop, we set the disk device
1962 read-only while the server shuts down. We now also set any
1963 external journal device read-only at the same time.
1966 Frequency : when upgrading from 1.4 while trying to change parameters
1968 Description: The wrong (new) MDC name was used when setting parameters for
1969 upgraded MDT's. Also allows changing of OSC (and MDC)
1970 parameters if --writeconf is specified at tunefs upgrade time.
1973 Frequency : when setting specific ost indicies
1975 Description: QOS code breaks on skipped indicies
1976 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1977 created with --index need not be sequential.
1979 Severity : enhancement
1981 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1982 Details : The uninit_groups feature works in conjunction with the kernel
1983 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1984 up the pass1 processing of e2fsck. This is a read-only feature
1985 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1986 mount filesystems that have had this feature enabled.
1988 Severity : enhancement
1990 Description: Improve multi-block allocation algorithm to avoid fragmentation
1991 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1992 allocation locality and avoid filesystem fragmentation.
1997 Description: lfs quota fails with deactivated OSTS
1998 Details : With this patch, three improvements are included:
1999 1. detete the softlimit in mds and osts when use "lfs quota".
2000 2. display the inaccurate data in the output of "lfs quota".
2001 3. try to get quota info when "lfs quota" is executed.
2003 ------------------------------------------------------------------------------
2005 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2007 * Support for kernels:
2008 2.4.21-47.0.1.EL (RHEL 3)
2009 2.6.5-7.283 (SLES 9)
2010 2.6.9-42.0.10.EL (RHEL 4)
2011 2.6.12.6 vanilla (kernel.org)
2012 2.6.16.27-0.9 (SLES 10)
2013 * Recommended e2fsprogs version: 1.39.cfs5
2015 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2019 Frequency : occasional, depends on client load and configuration
2020 Bugzilla : 12181, 12203
2021 Description: data loss for recently-modified files
2023 Details : In some cases it is possible that recently written or created
2024 files may not be written to disk in a timely manner (this should
2025 normally be within 30s unless client IO load is very high).
2026 The problem appears as zero-length files or files that are a
2027 multiple of 1MB in size after a client crash or client eviction
2028 that are missing data at the end of the file.
2030 This problem is more likely to be hit on clients where files are
2031 repeatedly created and unlinked in the same directory, clients
2032 have a large amount of RAM, have many CPUs, the filesystem has
2033 many OSTs, the clients are rebooted frequently, and/or the files
2034 are not accessed by other nodes after being written.
2036 The presence of the problem can be detected by looking at
2037 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2038 smaller than the second (nr_unused) then dirty files will not
2039 be flushed automatically to disk. "sync; sleep 10" should be
2040 run several times on the node before unmounting it to update
2041 Lustre (this is also safe to run on nodes without this problem).
2043 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2044 that can cause this same problem, so customers using that kernel
2045 also need to update the kernel in addition to Lustre. In order
2046 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2048 It is normal that files written just before a client crash (less
2049 than 30s) may not yet have been flushed to disk, even for local
2053 Frequency : frequent on thin XT3 nodes
2055 Description: UUID collision on thin XT3 Linux nodes
2056 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2057 randomly, since we relied on an insufficiently-seeded PRNG.
2062 Description: OSS hangs after "All ost request buffers busy"
2063 Details : A deadlock between quota and journal operations caused OSS
2064 hangs after printing "All ost request buffers busy."
2067 Frequency : always on liblustre builds
2069 Description: Cleanup compiler warnings on liblustre
2072 Frequency : always on liblustre builds on XT3
2074 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2078 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2079 > 32000 subdirectories
2080 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2081 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2082 subdirectory count crosses 32000. This will aid e2fsck to
2083 correctly handle more than 32000 subdirectories.
2086 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2088 Description: System hang on startup
2089 Details : This bug allowed the liblustre (e.g. catamount) client to
2090 return to the app before handling all startup RPCs. This
2091 could leave the node unresponsive to lustre network traffic
2092 and manifested as a server ptllnd timeout.
2094 Severity : enhancement
2096 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2097 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2098 boolean to control whether to print peer debug info when a
2099 client's RPC times out.
2104 Description: make O_SYNC working on 2.6 kernels
2105 Details : 2.6 kernels use different method for mark pages for write,
2106 so need add a code to lustre for O_SYNC work.
2111 Description: Failure to close file and release space on NFS
2112 Details : Put inode details into lock acquired in ll_intent_file_open.
2113 Use mdc_intent_lock in ll_intent_open to properly
2114 detect all kind of errors unhandled by mdc_enqueue.
2119 Description: proc file read during shutdown sometimes raced obd removal,
2121 Details : Add lock to prevent obd access after proc file removal.
2124 Frequency : Only for files larger than 4GB on 32-bit clients.
2126 Description: improperly doing page alignment of locks
2127 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2128 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2133 Description: RPCs being resent when they shouldn't be
2134 Details : Some RPCs that should not be resent are being resent. This
2135 can cause inconsistencies in the RPC state machine. Do not
2136 resend such requests.
2139 Frequency : rare, only with NFS export
2141 Description: Crash on NFS re-export node
2142 Details : under very unusual load conditions an assertion is hit in
2143 ll_intent_file_open()
2146 Frequency : only if OST filesystem is corrupted
2148 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2149 Details : for a short time RPCs with bulk IO are in the replay list,
2150 but replay of bulk IOs is unimplemented. If the OST filesystem
2151 is corrupted due to disk cache incoherency and then replay is
2152 started it is possible to trip an assertion. Avoid putting
2153 committed RPCs into the replay list at all to avoid this issue.
2158 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2159 Details : Large single O_DIRECT read and write calls can fail to allocate
2160 a sufficiently large buffer to process the request. In case of
2161 allocation failure the allocation is retried with a smaller
2162 buffer and broken into smaller requests.
2164 Severity : enhancement
2166 Description: Add -o localflock option to simulate old noflock behaviour.
2167 Details : This will achieve local-only flock/fcntl locks coherentness.
2172 Description: versioning check is incomplete
2173 Details : Checking the version difference of client vs. server, report
2174 error if the gap is too big.
2179 Description: add support PG_writeback bit
2180 Details : add support for PG_writeback bit for Lustre, for more carefull
2181 work with page cache in 2.6 kernel. This also fix some deadlocks
2182 and remove hack for work O_SYNC with 2.6 kernel.
2184 Severity : enhancement
2186 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2187 Details : The uninit_groups feature works in conjunction with the kernel
2188 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2189 up the pass1 processing of e2fsck. This is a read-only feature
2190 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2191 mount filesystems that have had this feature enabled.
2193 Severity : enhancement
2195 Description: Improve multi-block allocation algorithm to avoid fragmentation
2196 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2197 allocation locality and avoid filesystem fragmentation.
2199 ------------------------------------------------------------------------------
2201 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2203 * Support for kernels:
2204 2.6.9-42.0.3.EL (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)
2208 2.6.16.21-0.8 (SLES10)
2209 * Recommended e2fsprogs version: 1.39.cfs2-0
2211 * The backwards-compatible /proc/sys/portals symlink has been removed
2212 in this release. Before upgrading, please ensure that you change
2213 any configuration scripts or /etc/sysctl.conf files that access
2214 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2215 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2216 in advance of the upgrade on any system running Lustre 1.4.6 or
2217 newer, since /proc/sys/lnet was added in that version.
2218 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2222 Frequency : only when quota is used
2224 Description: avoid scanning export list for quota master
2225 Details : Change the algorithms to avoid scanning export list in order
2226 to improve the efficiency.
2229 Frequency : MDS failover only, very rarely
2231 Description: "went back in time" messages on mds failover
2232 Details : The greatest transno may be lost when the current operation
2233 finishes with an error (transno==0) and the client's last_rcvd
2234 record is over-written. Save the greatest transno in the
2235 mds_last_transno for this case.
2238 Frequency : always for specific kernels and striping counts
2240 Description: client may get "Matching packet too big" without ACL support
2241 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2242 when trying to access files in certain configurations. The
2243 clients should in fact be denied when mounting because they do
2244 not understand ACLs.
2247 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2249 Description: many clients connecting with IO in progress causes connect timeouts
2250 Details : Avoid synchronous journal commits to avoid delays caused by many
2251 clients connecting/disconnecting when bulk IO is in progress.
2252 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2253 OST_IO_PORTAL to avoid delays behind potentially many pending
2257 Frequency : occasionally with multiple writers to a single file
2259 Description: shared writes to file may result in wrong size reported by stat()
2260 Details : Allow growing of kms when extent lock is cancelled
2263 Frequency : always with random mmap IO to multi-striped file
2265 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2266 Details : Only if the hole is at the end of OST object so that kms is too
2267 small. Fix is to increase kms accordingly in ll_nopage.
2270 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2272 Description: writes to a missing object would leak memory on the OST
2273 Details : If there is an inconsistency between the MDS and OST filesystems,
2274 such that the MDS references an object that doesn't exist, writes
2275 to that object will leak memory due to incorrect cleanup in the
2276 error handling path, eventually running out of memory on the OST.
2281 Description: Creating too long symlink causes lustre errors
2282 Details : Check symlink and name lengths before sending requests to MDS.
2285 Frequency : only if flock is enabled (not on by default)
2287 Description: posix locks not released on fd closure on 2.6.9+
2288 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2289 kernels, this caused such locks not to be released on fd close and
2290 then assertions on fs unmount about still used locks.
2293 Frequency : MDS failover only, very rarely
2295 Description: clients may get ASSERTION(granted_lock != NULL)
2296 Details : When request was taking a long time, and a client was resending
2297 a getattr by name lock request. The were multiple lock requests
2298 with the same client lock handle and
2299 mds_getattr_name->fixup_handle_for_resent_request found one of the
2300 lock handles but later failed with ASSERTION(granted_lock != NULL).
2305 Description: handle->h_buffer_credits > 0, assertion failure
2306 Details : h_buffer_credits is zero after truncate, causing assertion
2307 failure. This patch extends the transaction or creates a new
2311 Frequency : NFS re-export or patchless client
2312 Bugzilla : 11179, 10796
2313 Description: Crash on NFS re-export node (__d_move)
2314 Details : We do not want to hash the dentry if we don't have a lock.
2315 But if this dentry is later used in d_move, we'd hit uninitialised
2316 list head d_hash, so we just do this to init d_hash field but
2317 leave dentry unhashed.
2320 Frequency : NFS re-export or patchless client
2322 Description: NFS exports has problem with symbolic link
2323 Details : lustre client didn't properly install dentry when re-exported
2324 to NFS or running patchless client.
2327 Frequency : NFS re-export or patchless client
2329 Description: Various nfs/patchless fixes.
2330 Details : fixes reuse disconected alias for lookup process - this fixes
2331 warning "find_exported_dentry: npd != pd",
2332 fix permission error with open files at nfs.
2333 fix apply umask when do revalidate.
2336 Frequency : occasional
2338 Description: Crash on NFS re-export node
2339 Details : calling clear_page() on the wrong pointer triggered oops in
2340 generic_mapping_read().
2343 Frequency : rarely, using O_DIRECT IO
2345 Description: unaligned directio crashes client with LASSERT
2346 Details : check for unaligned buffers before trying any requests.
2349 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2351 Description: stale data returned from RAID cache
2352 Details : If only a small amount of IO is done to the RAID device before
2353 reading it again it is possible to get stale data from the RAID
2354 cache instead of reading it from disk.
2357 Frequency : always for sles10 kernel
2359 Description: sles10 support
2360 Details : ll_follow_link: compile fixes and using of nd_set_link
2361 under newer kernels.
2364 Frequency : depends on arch, kernel and compiler version, always on sles10
2367 Description: recursive or deep enough symlinks cause stack overflow
2368 Details : getting rid of large stack-allocated variable in
2372 Frequency : depends on hardware
2374 Description: lustre write performance loss in the SLES10 kernel
2375 Details : the performance loss is caused by using of write barriers in the
2376 ext3 code. The SLES10 kernel turns barrier support on by
2377 default. The fix is to undo that change for ldiskfs.
2379 ------------------------------------------------------------------------------
2381 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2383 * Support for kernels:
2384 2.6.9-42.0.3EL (RHEL 4)
2385 2.6.5-7.276 (SLES 9)
2386 2.4.21-47.0.1.EL (RHEL 3)
2387 2.6.12.6 vanilla (kernel.org)
2391 Frequency : quota enabled and large files being deleted
2393 Description: releasing more than 4GB of quota at once hangs OST
2394 Details : If a user deletes more than 4GB of files on a single OST it
2395 will cause the OST to spin in an infinite loop. Release
2396 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2401 Description: statfs data retrieved from /proc may be stale or zero
2402 Details : When reading per-device statfs data from /proc, in the
2403 {kbytes,files}_{total,free,avail} files, it may appear
2404 as zero or be out of date.
2407 Frequency : systems with MD RAID1 external journal devices
2409 Description: lconf's call to blkid is confused by RAID1 journal devices
2410 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2411 of returning all block devices that match the journal UUID.
2414 Frequency : always, for aggregate stripe size over 4GB
2416 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2417 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2418 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2421 Frequency : always if "lfs find" used on a local file/directory
2423 Description: "lfs find" segfaults if used on a local file/directory
2424 Details : The case where a directory component was not specified wasn't
2425 handled correctly. Handle this properly.
2428 Frequency : always on ppc64
2430 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2431 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2432 perform endianness conversion.
2435 Frequency : rarely (truncate to non-zero file size after write under load)
2436 Bugzilla : 10730, 10687
2437 Description: Files padded with zeros to next 4K multiple
2438 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2439 it is possible that files that are truncated to a non-zero size
2440 immediately after being written are filled with zero bytes beyond
2441 the truncated size. No file data is lost.
2443 Severity : enhancement
2445 Description: Allow recovery/failover for liblustre clients.
2446 Details : liblustre clients were unaware of failover configurations until
2449 Severity : enhancement
2451 Description: user file locks should fail when not mounting with flock option
2452 Details : Set up an error-returning stub in ll_file_operations.lock field
2453 to prevent incorrect behaviour when client is mounted without
2454 flock option. Also, set up properly f_op->flock field for
2458 Frequency : always on ia64
2460 Description: "lfs df" loops on printing out MDS statfs information
2461 Details : The obd_ioctl_data was not initialized and in some systems
2462 this caused a failure during the ioctl that did not return
2463 an error. Initialize the struct and return an error on failure.
2466 Frequency : SLES 9 only
2468 Description: Error of copying files with lustre special EAs as root
2469 Details : Client side always return success for setxattr call for lustre
2470 special xattr (currently only "trusted.lov").
2473 Frequency : rarely on clusters with both ia64+i386 clients
2475 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2476 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2477 clients are doing concurrent writes to the same file it is
2478 possible that the ia64 clients may LASSERT because the OST
2479 extent locks are not PAGE_SIZE aligned. Ensure that grown
2480 locks are always aligned on the request boundary.
2483 Frequency : specific use, occasional
2485 Description: Overwriting in use executable truncates on-disk binary image
2486 Details : If one node attempts to overwrite an executable in use by
2487 another node, we now correctly return ETXTBSY instead of
2488 truncating the file.
2490 Severity : enhancement
2492 Description: Async OSC create to avoid the blocking unnecessarily.
2493 Details : If a OST has no remain object, system will block on the creating
2494 when need to create a new object on this OST. Now, ways use
2495 pre-created objects when available, instead of blocking on an
2496 empty osc while others are not empty. If we must block, we block
2497 for the shortest possible period of time.
2502 Description: chmod on Lustre root is propagated to other clients
2503 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2504 invalid by the follow_mount time.
2507 Frequency : liblustre clients only
2509 Description: Race in 'instant cancel' lock handling could lead to such locks
2510 never to be granted in case of SMP MDS
2511 Details : Do not destroy not yet granted but cbpending locks in
2515 Frequency : replay/resend of open
2517 Description: non null lock assetion failure in mds_intent_policy
2518 Details : Trying to replay/resend lockless open requests resulted in
2519 mds_open() returning 0 with no lock. Now it sets a flag if
2520 a lock is going to be returned.
2522 Severity : enhancement
2524 Description: Checksum enhancements
2525 Details : New checksum enhancements allow for resending RPCs that failed
2528 Severity : enhancement
2530 Description: Tunables on number of dirty pages in cacche
2531 Details : Allow to set limit on number of dirty pages cached.
2536 Description: client crash on unmount - lock still has references
2537 Details : In some error handling cases it was possible to leak a lock
2538 reference on a client while accessing a file. This was not
2539 harmful to the client during operation, but would cause the
2540 client to crash when the filesystem is unmounted.
2543 Frequency : specific case, rare
2545 Description: ETXTBSY on mds though file not in use by client
2546 Details : ETXTBSY is no longer incorrectly returned when attempting to
2547 chmod or chown a directory that the user previously tried to
2548 execute or a currently-executing binary.
2551 Frequency : extremely rare except on liblustre-based clients
2553 Description: Lustre space not freed when files are deleted
2554 Details : Clean up open-unlinked files after client eviction. Previously
2555 the unlink was skipped and the files remained as orphans.
2560 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2561 Details : In some cases it was possible to send a blocking callback to
2562 a client doing a glimpse, even though that client didn't get
2563 a lock granted. When the glimpse lock is cancelled on the OST
2564 the freed lock is left on the waiting list and corrupted the list.
2567 Frequency : all core dumps
2569 Description: Broke core dumps to lustre
2570 Details : Negative dentry may be unhashed if parent does not have UPDATE
2571 lock, but some callers, e.g. do_coredump, expect dentry to be
2572 hashed after successful create, hash it in ll_create_it.
2574 ------------------------------------------------------------------------------
2576 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2578 * Support for kernels:
2579 2.6.9-42.0.2.EL (RHEL 4)
2580 2.6.5-7.276 (SLES 9)
2581 2.4.21-40.EL (RHEL 3)
2582 2.6.12.6 vanilla (kernel.org)
2586 Frequency : always on RHEL 3
2588 Description: Number of open files grows over time
2589 Details : The number of open files grows over time, whether or not
2590 Lustre is started. This was due to a filp leak introduced
2591 by one of our kernel patches.
2593 ------------------------------------------------------------------------------
2595 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2597 * Support for kernels:
2598 2.6.9-42.EL (RHEL 4)
2599 2.6.5-7.267 (SLES 9)
2600 2.4.21-40.EL (RHEL 3)
2601 2.6.12.6 vanilla (kernel.org)
2606 Bugzilla : 5719, 9635, 9792, 9684
2607 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2608 Details : If a server is under heavy load and cannot reply to new
2609 connection requests before the client resends the (re)connect,
2610 the connection handling code can behave badly if two service
2611 threads are concurrently handing separate (re)connections from
2612 the same client. Add better locking to the connection handling
2613 code, and ensure that only a single connection will be processed
2614 for a given client UUID, even if the lock is dropped.
2616 Severity : enhancement
2618 Description: add TCP zero-copy support to kernel
2619 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2620 sends if the hardware supports scatter-gather and checksumming.
2621 This allows socklnd to do client-write and server-read more
2622 efficiently and reduce CPU utilization from skbuf copying.
2625 Frequency : only if NFS exporting from client
2627 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2628 Details : If setting attributes on a file created under NFS that had
2629 never been opened it would be possible to oops the client
2630 if the file had no objects.
2634 Bugzilla : 9326, 10402, 10897
2635 Description: client crash in ptlrpcd_wake() thread when sending async RPC
2636 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2637 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2638 before ptlrpcd_wake() was called and the request was freed.
2641 Frequency : always for liblustre
2643 Description: liblustre client does MDS+OSTs setattr RPC for each write
2644 Details : When doing a write from a liblustre client, the client
2645 incorrectly issued an RPC to the MDS and each OST the file was
2646 striped over in order to update the timestamps. When writing
2647 with small chunks and many clients this could overwhelm the MDS
2648 with RPCs. In all cases it would slow down the write because
2649 these RPCs are unnecessary.
2651 Severity : enhancement
2653 Description: allow number of MDS service threads to be changed at module load
2654 Details : It is now possible to change the number of MDS service threads
2655 running. Adding "options mds mds_num_threads={N}" to the MDS's
2656 /etc/modprobe.conf will set the number of threads for the next
2657 time Lustre is restarted (assuming the "mds" module is also
2658 reloaded at that time). The default number of threads will
2659 stay the same, 32 for most systems.
2664 Description: OST crash if filesystem is unformatted or corrupt
2665 Details : If an OST is started on a device that has never been formatted
2666 or if the filesystem is corrupt and cannot even mount then the
2667 error handling cleanup routines would dereference a NULL pointer.
2672 Description: NULL pointer deref in llap_from_page.
2673 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2674 incorrect) mapping if the page was truncated/reclaimed while it was
2675 searched for. Check for this condition and skip such pages when
2676 doing readahead. Introduce extra check to llap_from_page() to
2677 verify page->mapping->host is non-NULL (so page is not anonymous).
2680 Frequency : Sometimes when using sys_sendfile
2682 Description: "page not covered by a lock" warnings from ll_readpage
2683 Details : sendfile called ll_readpage without right page locks present.
2684 Now we introduced ll_file_sendfile that does necessary locking
2685 around call to generic_file_sendfile() much like we do in
2689 Frequency : with certain MDS communication failures at client mount time
2691 Description: NULL pointer deref after failed client mount
2692 Details : a client connection request may delayed by the network layer
2693 and not be sent until after the PTLRPC layer has timed out the
2694 request. If the client fails the mount immediately it will try
2695 to clean up before the network times out the request. Add a
2696 reference from the request import to the obd device and delay
2697 the cleanup until the network drops the request.
2700 Frequency : occasionally during client (re)connect
2702 Description: assertion failure during client (re)connect
2703 Details : processing a client connection request may be delayed by the
2704 client or server longer than the client connect timeout. This
2705 causes the client to resend the connection request. If the
2706 original connection request is replied in this interval, the
2707 client may trip an assertion failure in ptlrpc_connect_interpret()
2708 which thought it would be the only running connect process.
2711 Frequency : only with obd_echo servers and clients that are rebooted
2713 Description: kernel BUG accessing uninitialized data structure
2714 Details : When running an obd_echo server it did not start the ping_evictor
2715 thread, and when a client was evicted an uninitialized data
2716 structure was accessed. Start the ping_evictor in the RPC
2717 service startup instead of the OBD startup.
2719 Severity : enhancement
2720 Bugzilla : 10193 (patchless)
2721 Description: Remove dependency on various unexported kernel interfaces.
2722 Details : No longer need reparent_to_init, exit_mm, exit_files,
2723 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2726 Frequency : rare (only users of deprecated and unsupported LDAP config)
2728 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2729 Details : LDAP apparently contains 'lustreName' attributes instead of
2730 'name'. A simple remapping of the name is sufficient.
2733 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2735 Description: waiting_locks_callback trips kernel BUG if client is evicted
2736 Details : Running with the dump_on_timeout debug flag turned on makes
2737 it possible that the waiting_locks_callback() can try to dump
2738 the Lustre kernel debug logs from an interrupt handler. Defer
2739 this log dumping to the expired_lock_main() thread.
2741 Severity : enhancement
2743 Description: Support NFS exporting on 2.6 kernels.
2744 Details : Implement non-rawops metadata methods for NFS server to use without
2745 changing NFS server code.
2748 Frequency : very rare (synthetic metadata workload only)
2750 Description: two racing renames might cause an MDS thread to deadlock
2751 Details : Running the "racer" program may cause one MDS thread to rename
2752 a file from being the source of a rename to being the target of
2753 a rename at exactly the same time that another thread is doing
2754 so, and the second thread has already enqueued these locks after
2755 doing a lookup of the target and is trying to relock them in
2756 order. Ensure that we don't try to re-lock the same resource.
2759 Frequency : only very large systems with liblustre clients
2761 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2762 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2763 all OSTs in parallel. This allows the eviction of stale liblustre
2764 clients to proceed much faster than if they were done in series,
2765 and also offers similar improvements for other set_info RPCs.
2770 Description: excessive CPU usage during initial read phase on client
2771 Details : During the initial read phase on a client, it would agressively
2772 retry readahead on the file, consuming too much CPU and impacting
2773 performance (since 1.4.5.8). Improve the readahead algorithm
2774 to avoid this, and also improve some other common cases (read
2775 of small files in particular, where "small" is files smaller than
2776 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2781 Description: MDS crash when receiving packet with unknown intent.
2782 Details : Do not LBUG in unknown intent case, just return -EFAULT
2784 Severity : enhancement
2785 Bugzilla : 9293, 9385
2786 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2787 Details : Do not serialize getattr (non-intent version) and statfs.
2790 Frequency : occasional, when OST network is overloaded/intermittent
2792 Description: client evicted by OST after bulk IO timeout
2793 Details : If a client sends a bulk IO request (read or write) the OST
2794 may evict the client if it is unresposive to its data GET/PUT
2795 request. This is incorrect if the network is overloaded (takes
2796 too long to transfer the RPC data) or dropped the OST GET/PUT
2797 request. There is no need to evict the client at all, since
2798 the pinger and/or lock callbacks will handle this, and the
2799 client can restart the bulk request.
2802 Frequency : Always when mmapping file with no objects
2804 Description: client crashes when mmapping file with no objects
2805 Details : Check that we actually have objects in a file before doing any
2806 operations on objects in ll_vm_open, ll_vm_close and
2812 Description: Request leak when working with deleted CWD
2813 Details : Introduce advanced request refcount tracking for requests
2814 referenced from lustre intent.
2816 Severity : Enhancement
2818 Description: Cache open file handles on client.
2819 Details : MDS now will return special lock along with openhandle, if
2820 requested and client is allowed to hold openhandle, even if unused,
2821 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2822 closing files for every read/write openration.
2824 Severity : Enhancement
2826 Description: Cache open negative dentries on client when possible.
2827 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2828 negative dentries on lock revocation.
2833 Description: Remounting a client read-only wasn't possible with a zconf mount
2834 Details : It wasn't possible to remount a client read-only with llmount.
2836 Severity : enhancement
2837 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2838 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2839 MPICH is included to add Lustre-specific ADIO interfaces.
2840 This is based closely on the UFS ADIO layer and only differs
2841 in file creation, in order to allow the OST striping to be set.
2842 This is user-contributed code and not supported by CFS.
2847 Description: extended inode attributes (immutable, append-only) work improperly
2848 when 2.4 and 2.6 kernels are used on client/server or vice versa
2849 Details : Introduce kernel-independent values for these flags.
2851 Severity : enhancement
2854 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2855 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2856 granularity. Now, Fractional MB granularity is be supported,
2857 this is very useful for low memory system.
2859 Severity : enhancement
2861 Description: Getattr by fid
2862 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2863 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2864 have locks and avoids getattr rpc requests for it on every stat.
2867 Frequency : Always, for filesystems larger than 2TB
2869 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2870 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2871 filesystems larger than 2TB to be mounted with mballoc enabled.
2874 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2876 Description: filesystem corruption for non-standard kernels and very large OSTs
2877 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2878 filesystems larger than 2TB is mounted then the kernel will
2879 silently corrupt the start of the filesystem. CONFIG_LBD is
2880 enabled for all CFS-supported kernels, but the possibility of
2881 this happening with a modified kernel config exists.
2883 Severity : enhancement
2885 Description: add client O_DIRECT support for 2.6 kernels
2886 Details : It is now possible to do O_DIRECT reads and writes to files
2887 in the Lustre client mountpoint on 2.6 kernel clients.
2889 Severity : enhancement
2891 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2892 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2893 parallel, not waiting for response from every OST before sending
2894 a rpc to the next OST.
2899 Description: setattr vs write race when updating file timestamps
2900 Details : Client processes that update a file timestamp into the past
2901 right after writing to the file (e.g. tar) it is possible that
2902 the updated file modification time can be reset to the current
2903 time due to a race between processing the setattr and write RPC.
2905 Severity : enhancement
2907 Description: Bring 'lfs find' closer in line with regular Linux find.
2908 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2909 -print0 options and obtains all the needed info through the lustre
2912 Severity : enhancement
2914 Description: support up to 1024 configured devices on one node
2915 Details : change obd_dev array from statically allocated to dynamically
2916 allocated structs as they are first used to reduce memory usage
2921 Description: Flush dirty partially truncated pages during truncate
2922 Details : Immediatelly flush partially truncated pages in filter_setattr,
2923 this way we completely avoid having any pages in page cache on OST
2924 and can retire ugly workarounds during writes to flush such pages.
2929 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2930 Details : protect inode from truncation within vfs_unlink() context
2931 just take a reference before calling vfs_unlink() and release it
2932 when parent's i_sem is free.
2935 Frequency : always, if extents are used on OSTs
2937 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2938 Details : OSTs using the extents format would not zero the high 16 bits of
2939 the index physical block number. This is not a problem for any
2940 OST filesystems smaller than 16TB, and no kernels support ext3
2941 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2942 new/modified files) and can be fixed for existing filesystems
2943 with e2fsprogs-1.39-cfs1.
2948 Description: import connection selection may be incorrect if timer wraps
2949 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2950 import connections to be ignored if the 32-bit jiffies counter
2951 wraps. Use a 64-bit jiffies counter.
2954 Frequency : during server recovery
2956 Description: crash after server is denying duplicate export
2957 Details : If clients are resending connect requests to the server, the
2958 server refuses to allow a client to connect multiple times.
2959 Fixed a bug in the handling of this case.
2962 Frequency : very large clusters immediately after boot
2964 Description: LNET request buffers exhausted under heavy short-term load
2965 Details : If a large number of client requests are generated on a service
2966 that has previously never seen so many requests it is possible
2967 that the request buffer growth cannot keep up with the spike in
2968 demand. Instead of dropping incoming requests, they are held in
2969 the LND until the RPC service can accept more requests.
2972 Frequency : Sometimes during replay
2974 Description: Assertion failure in ll_local_open after replay.
2975 Details : If replay happened on an open request reply before we were able
2976 to set replay handler, reply will become not swabbed tripping the
2977 assertion in ll_local_open. Now we set the handler right after
2978 recognising of open request
2981 Frequency : very rare
2983 Description: kernel reports "badness in vsnprintf"
2984 Details : Reading from the "recovery_status" /proc file in small chunks
2985 may cause a negative length in lprocfs_obd_rd_recovery_status()
2986 call to vsnprintf() (which is otherwise harmless). Exit early
2987 if there is no more space in the output buffer.
2989 Severity : enhancement
2991 Description: clear OBD RPC statistics by writing to them
2992 Details : It is now possible to clear the OBD RPC statistics by writing
2993 to the "stats" file.
2998 Description: Client mtime is not the same on different clients after utimes
2999 Details : In some cases, the client was using the utimes() syscall on
3000 a file cached on another node. The clients now validate the
3001 ctime from the MDS + OSTs to determine which one is right.
3006 Description: Inability to activate failout mode
3007 Details : lconf script incorrectly assumed that in python string's numeric
3008 value is used in comparisons.
3011 Frequency : always with multiple stripes per file
3013 Description: Inefficient object allocation for mutli-stripe files
3014 Details : When selecting which OSTs to stripe files over, for files with
3015 a stripe count that divides evenly into the number of OSTs,
3016 the MDS is always picking the same starting OST for each file.
3017 Return the OST selection heuristic to the original design.
3022 Description: mount failures may take full timeout to return an error
3023 Details : Under some heavy load conditions it is possible that a
3024 failed mount can wait for the full obd_timeout interval,
3025 possibly several minutes, before reporting an error.
3026 Instead return an error as soon as the status is known.
3028 ------------------------------------------------------------------------------
3030 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
3032 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3033 INTEROPERATE with older versions automatically. Please read the
3034 user documentation before upgrading any part of a live system.
3035 * WARNING: Lustre networking configuration changes are required with
3036 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3039 * Support for kernels:
3040 2.6.9-22.0.2.EL (RHEL 4)
3041 2.6.5-7.244 (SLES 9)
3042 2.6.12.6 vanilla (kernel.org)
3045 Severity : enhancement
3046 Bugzilla : 7981/8208
3047 Description: Introduced Lustre Networking (LNET)
3048 Details : LNET is new networking infrastructure for Lustre, it includes
3049 a reorganized network configuration mode (see the user
3050 documentation for full details) as well as support for routing
3051 between different network fabrics. Lustre Networking Devices
3052 (LNDs) for the supported network fabrics have also been
3053 created for this new infrastructure.
3055 Severity : enhancement
3056 Description: Introduced Access control lists
3057 Details : clients can set ACLs on files and directories in order to have
3058 more fine-grained permissions than the standard Unix UGO+RWX.
3059 The MDS must be started with the "-o acl" mount option.
3061 Severity : enhancement
3062 Description: Introduced filesystem quotas
3063 Details : Administrators may now establish per-user quotas on the
3066 Severity : enhancement
3068 Description: Configuration change for the XT3
3069 The PTLLND is now used to run Lustre over Portals on the XT3
3070 The configure option(s) --with-cray-portals are no longer used.
3071 Rather --with-portals=<path-to-portals-includes> is used to
3072 enable building on the XT3. In addition to enable XT3 specific
3073 features the option --enable-cray-xt3 must be used.
3078 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3079 Details : A race between allocating a new llcd and re-getting the llcd_lock
3080 allowed another thread to grab newly-allocated llcd.
3082 Severity : enhancement
3084 Description: 2.6 OST async journal commit and locking fix to improve performance
3085 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3086 2.6 kernels are now async as they already were in 2.4 kernels so
3087 that they can commit concurrently with the network bulk transfer.
3088 For block-allocated files the filter allocation semaphore is held
3089 to avoid filesystem fragmentation during allocation. BKL lock
3090 removed for 2.6 xattr operations where it is no longer needed.
3095 Description: lconf incorrectly determined whether two IP networks could talk
3096 Details : In some more complicated routing and multiple-network
3097 configurations, lconf will avoid trying to make a network
3098 connection to a disjoint part of the IP space. It was doing the
3099 math incorrectly for one set of cases.
3104 Description: Fix for potential infinite loop processing records in an llog.
3105 Details : If an llog record is corrupted/zeroed, it is possible to loop
3106 forever in llog_process(). Validate the llog record length
3107 and skip the remainder of the block on error.
3110 Frequency : occasional (liblustre only)
3112 Description: liblustre could not open files whose last component is a symlink
3113 Details : sysio_path_walk() would incorrectly pass the open intent to
3114 intermediate path components.
3117 Frequency : rare (liblustre only with non-standard tuning)
3118 Bugzilla : 7201 (7350)
3119 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3120 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3121 after it has been released, only do this for referenced locks
3123 Severity : enhancement
3125 Description: specifying an (invalid) directory default stripe_size of -1
3126 would reset the directory default striping
3127 Details : stripe_size -1 was used internally to signal directory stripe
3128 removal, now use "all default" to signal dir stripe removal
3129 as a directory striping of "all default" is not useful
3132 Frequency : common for large clusters running liblustre clients
3134 Description: doing an ls when liblustre clients are running is slow
3135 Details : sending a glimpse AST to a liblustre client waits for every AST
3136 to time out, as liblustre clients will not respond. Since they
3137 cannot cache data we refresh the OST lock LVB from disk instead.
3139 Severity : enhancement
3141 Description: doing an ls at the same time as file IO can be slow
3142 Details : enqueue and other "small" requests can be blocked behind many
3143 large IO requests. Create a new OST IO portal for non-IO
3144 requests so they can be processed faster.
3147 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3149 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3150 created file with mode 000
3151 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3152 allow an ftruncate() as a non-root user to a file with mode 000.
3153 Lustre now respects this flag to disable mode checks when
3154 truncating a file owned by the user
3157 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3160 Description: Revoking locks from clients that went dead or catatonic might take
3162 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3163 cancellation of such locks instant on servers without waiting for
3164 any reply from clients. Clients drops these locks when cancel
3165 notification from server is received without replying.
3168 Frequency : liblustre-only, when liblustre client dies or becomes busy
3170 Description: Doing ls on Linux clients can take a long time with active
3172 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3173 granting such locks to it in the first place if possible. Locks
3174 are taken by proxy on the OST during the read or write and
3175 dropped immediately afterward. Add connect flags handling, do
3176 not grant locks to liblustre clients for glimpse ASTs.
3178 Severity : enhancement
3180 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3181 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3182 sequential reads and behaves badly with applications doing random
3183 reads. Improve it by reading ahead at least read region, and
3184 avoiding excessive large RPC for small reads.
3186 Severity : enhancement
3188 Description: Creating more than 1000 files for a single job may cause a load
3189 imbalance on the OSTs if there are also a large number of OSTs.
3190 Details : qos_prep_create() uses an OST index reseed value that is an
3191 even multiple of the number of available OSTs so that if the
3192 reseed happens in the middle of the object allocation it will
3193 still utilize the OSTs as uniformly as possible.
3198 Description: OST or MDS may oops in ping_evictor_main()
3199 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3200 but doesn't restart at beginning of obd_exports_timed list
3203 Severity : enhancement
3205 Description: improve by-nid export eviction on the MDS and OST
3206 Details : allow multiple exports with the same NID to be evicted at one
3207 time without re-searching the exports list.
3210 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3212 Description: MDS may oops in groups_free()
3213 Details : in rare race conditions a newly allocated group_info struct is
3214 freed again, and this can be NULL. The 2.4 compatibility code
3215 for groups_free() checked for a NULL pointer, but 2.6 did not.
3218 Frequency : common for liblustre clients doing little filesystem IO
3219 Bugzilla : 9352, 7313
3220 Description: server may evict liblustre clients accessing contended locks
3221 Details : if a client is granted a lock or receives a completion AST
3222 with a blocking AST already set it would not reply to the AST
3223 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3226 Frequency : lfs setstripe, only systems with more than 160 OSTs
3228 Description: unable to set striping with a starting offset beyond OST 160
3229 Details : llapi_create_file() incorrectly limited the starting stripe
3230 index to the maximum single-file stripe count.
3233 Frequency : LDAP users only
3235 Description: lconf did not handle in-kernel recovery with LDAP properly
3236 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3238 Severity : enhancement
3240 Description: bind OST threads to NUMA nodes to improve performance
3241 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3242 node and do their allocations there to localize memory access
3244 Severity : enhancement
3246 Description: llmount can determine client NID directly from Myrinet (GM)
3247 Details : the client NID code from gmnalnid was moved directly into
3248 llmount, removing the need to use this or specifying the
3249 client NID explicitly when mounting GM clients with zeroconf
3252 Frequency : if client is started with down MDS
3254 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3255 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3256 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3257 ptlrpc_import_delayed_req() because we didn't check if the
3258 request was interrupted, and we also didn't break out of the
3259 event loop if there was no timeout
3264 Description: data loss during non-page-aligned writes to a single file from
3265 both multiple nodes and multiple threads on one node at same time
3266 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3267 inconsistency led to false short-reads, that were cached and later
3268 used by ->prepare_write() to fill in partially written page,
3269 leading to data loss.
3272 Frequency : always, if lconf --abort_recovery used
3274 Description: lconf --abort_recovery fails with 'Operation not supported'
3275 Details : lconf was attempting to abort recovery on the MDT device and not
3278 Severity : enhancement
3280 Description: remove cleanup logs
3281 Details : replace lconf-generated cleanup logs with lustre internal
3282 cleanup routines. Eliminates the need for client-cleanup and
3285 Severity : enhancement
3287 Description: add support for EAs (user and system) on lustre filesystems
3288 Details : it is now possible to store extended attributes in the Lustre
3289 client filesystem, and with the user_xattr mount option it
3290 is possible to allow users to store EAs on their files also
3292 Severity : enhancement
3294 Description: Add possibility (config option) to show minimal available OST free
3296 Details : When compiled with --enable-mindf configure option, statfs(2)
3297 (and so, df) will return least minimal free space available from
3298 all OSTs as amount of free space on FS, instead of summary of
3299 free spaces of all OSTs.
3301 Severity : enhancement
3303 Description: do not expand extent locks acquired on OST-side
3304 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3305 by server: they are not cached anyway.
3308 Frequency : when mmap is used/binaries executed from Lustre
3310 Description: Unmmap pages before throwing them away from read cache.
3311 Details : llap_shrink cache now attempts to unmap pages before discarding
3312 them (if unmapping failed - do not discard). SLES9 kernel has
3313 extra checks that trigger if this unmapping is not done first.
3318 Description: lconf didn't resolve symlinks before checking to see whether a
3319 given mountpoint was already in use
3322 Frequency : when migrating failover services
3323 Bugzilla : 6395, 9514
3324 Description: When migrating a subset of services from a node (e.g. failback
3325 from a failover service node) the remaining services would
3326 time out and evict clients.
3327 Details : lconf --force (implied by --failover) sets the global obd_timeout
3328 to 5 seconds in order to quickly disconnect, but this caused
3329 other RPCs to time out too quickly. Do not change the global
3330 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3332 Severity : enhancement
3333 Frequency : if MDS is started with down OST
3334 Bugzilla : 9439,5706
3335 Description: Allow startup/shutdown of an MDS without depending on the
3336 availability of the OSTs.
3337 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3338 Add appropriate locking and lov-osc refcounts for safe
3339 cleaning. Add osc abort_inflight calls in case the
3340 synchronize never started.
3343 Frequency : occasional (Cray XT3 only)
3345 Description: root not authorized to access files in CRAY_PORTALS environment
3346 Details : The client process capabilities were not honoured on the MDS in
3347 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3348 been accessed by an authorized user then root was able to access
3349 the file on the local client also. The root user capabilities
3350 are now allowed on the MDS, as this environment has secure UID.
3353 Frequency : occasional
3355 Description: ldiskfs "too long searching" message happens too often
3356 Details : A debugging message (otherwise harmless) prints too often on
3357 the OST console. This has been reduced to only happen when
3358 there are fragmentation problems on the filesystem.
3363 Description: Division by zero in statfs when all OSCs are inactive
3364 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3365 lov_statfs divides by value returned by lov_get_stripecnt().
3369 Bugzilla : 9489, 3273
3370 Description: First write from each client to each OST was only 4kB in size,
3371 to initialize client writeback cache, which caused sub-optimal
3372 RPCs and poor layout on disk for the first writen file.
3373 Details : Clients now request an initial cache grant at (re)connect time
3374 and so that they can start streaming writes to the cache right
3375 away and always do full-sized RPCs if there is enough data.
3376 If the OST is rebooted the client also re-establishes its grant
3377 so that client cached writes will be honoured under the grant.
3382 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3383 Details : Now I/O RPCs go to different portal number and (presumably) fast
3384 lock requests (and glimses) and other RPCs get their own service
3385 threads pool that should be able to service those RPCs
3388 Severity : enhancement
3390 Description: Ability to exchange lustre version between client and servers and
3391 issue warnings at client side if client is too old. Also for
3392 liblustre clients there is ability to refuse connection of too old
3394 Details : New 'version' field is added to connect data structure that is
3395 filled with version info. That info is later checked by server and
3399 Frequency : rare, liblustre only.
3400 Bugzilla : 9296, 9581
3401 Description: Two simultaneous writes from liblustre at offset within same page
3402 might proceed at the same time overwriting eachother with stale
3404 Details : I/O lock withing llu_file_prwv was released too early, before data
3405 actually was hitting the wire. Extended lock-holding time until
3406 server acknowledges receiving data.
3409 Frequency : extremely rare. Never observed in practice.
3411 Description: avoid generating lustre_handle cookie of 0.
3412 Details : class_handle_hash() generates handle cookies by incrementing
3413 global counter, and can hit 0 occasionaly (this is unlikely, but
3414 not impossible, because initial value of cookie counter is
3415 selected randonly). Value of 0 is used as a sentinel meaning
3416 "unassigned handle" --- avoid it. Also coalesce two critical
3417 sections in this function into one.
3419 Severity : enhancement
3421 Description: allow liblustre clients to delegate truncate locking to OST
3422 Details : To avoid overhead of locking, liblustre client instructs OST to
3423 take extent lock in ost_punch() on client's behalf. New connection
3424 flag is added to handle backward compatibility.
3426 Severity : enhancement
3427 Bugzilla : 4928, 7341, 9758
3428 Description: allow number of OST service threads to be specified
3429 Details : a module parameter allows the number of OST service threads
3430 to be specified via "options ost ost_num_threads={N}" in the
3431 OSS's /etc/modules.conf or /etc/modprobe.conf.
3435 Bugzilla : 6146, 9635, 9895
3436 Description: servers crash with bad pointer in target_handle_connect()
3437 Details : In rare cases when a client is reconnecting it was possible that
3438 the connection request was the last reference for that export.
3439 We would temporarily drop the export reference and get a new
3440 one, but this may have been the last reference and the export
3441 was just destroyed. Get new reference before dropping old one.
3443 Severity : enhancement
3444 Frequency : if client is started with failover MDS
3446 Description: Allow multiple MDS hostnames in the mount command
3447 Details : Try to read the configuration from all specified MDS
3448 hostnames during a client mount in case the "primary"
3451 Severity : enhancement
3453 Description: Stop sending data to evicted clients as soon as possible.
3454 Details : Check if the client we are about to send or are sending data to
3455 was evicted already. (Check is done every second of waiting,
3456 for which l_wait_event interface was extended to allow checking
3457 of exit condition at specified intervals).
3460 Frequency : rare, normally only when NFS exporting is done from client
3462 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3464 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3467 Frequency : always, if srand() is called before liblustre initialization
3469 Description: Liblustre uses system PRNG disturbing its usage by user application
3470 Details : Introduce internal to lustre fast and high-quality PRNG for
3471 lustre usage and make liblustre and some other places in generic
3472 lustre code to use it.
3474 Severity : enhancement
3475 Bugzilla : 9477, 9557, 9870
3476 Description: Verify that the MDS configuration logs are updated when xml is
3477 Details : Check if the .xml configuration logs are newer than the config
3478 logs stored on the MDS and report an error if this is the case.
3479 Request --write-conf, or allow starting with --old_conf.
3481 Severity : enhancement
3483 Description: Handle symlinks in the path when checking if Lustre is mounted.
3484 Details : Resolve intermediate symlinks when checking if a client has
3485 mounted a filesystem to avoid duplicate client mounts.
3490 Description: lconf can hit an error exception but still return success.
3491 Details : The lconf command catches the Command error exception at the top
3492 level script context and will exit with the associated exit
3493 status, but doesn't ensure that this exit status is non-zero.
3498 Description: failure of ptlrpc thread startup can cause oops
3499 Details : Starting a ptlrpc service thread can fail if there are a large
3500 number of threads or the server memory is very fragmented.
3501 Handle this without oopsing.
3504 Frequency : always, only if liblustre and non-default acceptor port was used
3506 Description: liblustre cannot connect to servers with non-default acceptor port
3507 Details : tcpnal_set_default_params() was not called and was therefore
3508 ignoring the environment varaible TCPNAL_PORT, as well as other
3509 TCPNAL_ environment variables
3514 Description: two objects could be created on the same OST for a single file
3515 Details : If an OST is down, in some cases it was possible to create two
3516 objects on a single OST for a single file. No problems other
3517 than potential performance impact and spurious error messages.
3521 Bugzilla : 5681, 9562
3522 Description: Client may oops in ll_unhash_aliases
3523 Details : Client dcache may become inconsistent in race condition.
3524 In some cases "getcwd" can fail if the current directory is
3530 Description: Inode refcounting problems in NFS export code
3531 Details : link_raw functions used to call d_instantiate without obtaining
3532 extra inode reference first.
3536 Bugzilla : 9942, 9903
3537 Description: Referencing freed requests leading to crash, memleaks with NFS.
3538 Details : We used to require that call to ll_revalidate_it was always
3539 followed by ll_lookup_it. Also with revalidate_special() it is
3540 possible to call ll_revalidate_it() twice for the same dentry
3541 even if first occurence returned success. This fix changes semantic
3542 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3543 reference on a request referred from the intent.
3544 ll_intent_release() then releases such a request.
3547 Frequency : rare, normally benchmark loads only
3549 Description: unlinked inodes were kept in memory on the client
3550 Details : If a client is repeatedly creating and unlinking files it
3551 can accumulate a lot of stale inodes in the inode slab cache.
3552 If there is no other client load running this can cause the
3553 client node to run out of memory. Instead flush old inodes
3554 from client cache that have the same inode number as a new inode.
3557 Frequency : SLES9 2.6.5 kernel and long filenames only
3558 Bugzilla : 9969, 10379
3559 Description: utime reports stale NFS file handle
3560 Details : SLES9 uses out-of-dentry names in some cases, which confused
3561 the lustre dentry revalidation. Change it to always use the
3565 Frequency : rare, unless heavy write-truncate concurrency is continuous
3566 Bugzilla : 4180, 6984, 7171, 9963, 9331
3567 Description: OST becomes very slow and/or deadlocked during object unlink
3568 Details : filter_destroy() was holding onto the parent directory lock
3569 while truncating+unlinking objects. For very large objects this
3570 may block other threads for a long time and slow overall OST
3571 responsiveness. It may also be possible to get a lock ordering
3572 deadlock in this case, or run out of journal credits because of
3573 the combined truncate+unlink. Solution is to do object truncate
3574 first in one transaction without parent lock, and then do the
3575 final unlink in a new transaction with the parent lock. This
3576 reduces the lock hold time dramatically.
3579 Frequency : rare, 2.4 kernels only
3581 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3582 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3583 before unmounting their filesystem in order to avoid deadlock.
3584 The kernel_locked() function in 2.4 kernels only checks whether
3585 the kernel lock is held, not whether it is this process that is
3586 holding it as 2.6 kernels do.
3591 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3592 Details : The client ptlrpc code may be trying to reconnect to a down
3593 server before a previous connection attempt has timed out.
3594 Increase the reconnect interval to be longer than the connection
3595 timeout interval to avoid sending duplicate connections to
3599 Frequency : echo_client brw_test command
3601 Description: fix echo_client to work with OST preallocated code
3602 Details : OST preallocation code (5137) didn't take echo_client IO path
3603 into account: echo_client calls filter methods outside of any
3604 OST thread and, hence, there is no per-thread preallocated
3605 pages and buffers to use. Solution: hijack pga pages for IO. As
3606 a byproduct, this avoids unnecessary data copying.
3610 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3611 Description: Client can oops in mdc_commit_close() after open replay
3612 Details : It was possible for the MDS to return an open request with no
3613 transaction number in mds_finish_transno() if the client was
3614 evicted, but without actually returning an error. Clients
3615 would later try to replay that open and may trip an assertion
3616 Simplify the client close codepath, and always return an error
3617 from the MDS in case the open is not successful.
3620 Frequency : rare, 2.6 OSTs only
3622 Description: OST may deadlock under high load on fragmented files
3623 Details : If there was a heavy load and highly-fragmented OST filesystems
3624 it was possible to have all the OST threads deadlock waiting on
3625 allocation of biovecs, because the biovecs were not released
3626 until the entire RPC IO was completed. Instead, release biovecs
3627 as soon as they are complete to ensure forward IO progress.
3629 Severity : enhancement
3631 Description: Support for specifying external journal device at mount
3632 Details : If an OST or MDS device is formatted with an external journal
3633 device, this device major/minor is stored in the ext3 superblock
3634 and may not be valid for failover. Allow detecting and
3635 specifying the external journal at mount time.
3640 Description: Mounting an MDS with pending unlinked files may cause oops
3641 Details : target_finish_recovery() calls mds_postrecov() which returned
3642 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3643 considers this an error and immediately begins cleaning up the
3644 lov, just after starting the mds_lov process
3646 Severity : enhancement
3648 Description: Implement 'lfs df' to report actual free space on per-OST basis
3649 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3650 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3651 usage of objects; '-h' to report in human readable format.
3653 ------------------------------------------------------------------------------
3655 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3662 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3663 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3664 at startup during orphan unlinking. Instead, skip buddy update
3665 before setup, as it will be regenerated after recovery is complete.
3670 Description: If an OST is inactive, its locks might reference stale inodes.
3671 Details : lov_change_cbdata() must iterate over all namespaces, even if
3672 they are inactive to clear inode references from the lock.
3674 Severity : enhancement
3675 Frequency : occasional, if non-standard max_dirty_mb used
3677 Description: Client will block write RPCs if not enough grant
3678 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3679 then the client will block writes while waiting for another RPC
3680 to complete instead of consuming its dirty limit. With change
3681 we get improved performance when max_dirty_mb is small.
3683 Severity : enhancement
3684 Bugzilla : 3389, 6253
3685 Description: Add support for supplementary groups on the MDS.
3686 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3687 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3688 lookups for user supplementary groups into a cache.
3692 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3693 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3694 for liblustre O_EXCL is enforced by MDS.
3698 Description: readdir never returns NULL in liblustre.
3699 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3700 dentry in struct dirent.
3704 Frequency : liblustre only, depends on application IO pattern
3705 Description: liblustre clients evicted if not contacting servers
3706 Details : Don't put liblustre clients into the ping_evictor list, so
3707 they will not be evicted by the pinger ever.
3709 Severity : enhancement
3711 Description: Add ability to evict clients by NID from MDS.
3712 Details : By echoing "nid:$NID" string into
3713 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3714 $NID would be instantly evicted from this MDS and from all active
3715 OSTs connected to it.
3719 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3720 Details : lookup_it_finish() used to query file size from OSTs that was not
3725 Description: service threads change working directory to that of init
3726 Details : Starting lustre service threads may pin the working directory
3727 of the parent thread, making that filesystem busy. Threads
3728 now change to the working directory of init to avoid this.
3732 Frequency : during shutdown only
3733 Description: shutdown with a failed MDS or OST can cause unmount to hang
3734 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3739 Frequency : 2.6 only
3740 Description: chmod/chown may include an extra supplementary group
3741 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3742 supplementary group and if none is specified this is used.
3745 Bugzilla : 5479 (6816)
3747 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3748 Details : If lookup is in progress on a file that is unlinked we might try
3749 to revalidate the inode and fail in revalidate after lookup is
3750 complete and ll_file_open() enqueues the open again but
3751 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3754 Frequency : always, if lconf --abort_recovery used
3756 Description: lconf --abort_recovery fails with 'Operation not supported'
3757 Details : lconf was attempting to abort recovery on the MDT device and not
3760 ------------------------------------------------------------------------------
3762 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3767 Frequency : rare (only unsupported configurations with a node running as an
3769 Bugzilla : 6514, 5137
3770 Description: Mounting a Lustre file system on a node running as an OST could
3772 Details : OSTs now preallocates memory needed to write out data at
3773 startup, instead of when needed, to avoid having to
3774 allocate memory in possibly low memory situations.
3775 Specifically, if the file system is mounted on on OST,
3776 memory pressure could force it to try to write out data,
3777 which it needed to allocate memory to do. Due to the low
3778 memory, it would be unable to do so and the node would
3779 become unresponsive.
3781 Severity : enhancement
3783 Description: Addition of lconf --service command line option
3784 Details : lconf now accepts a '--service <arg>' option, which is
3785 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3787 Severity : enhancement
3789 Description: Failover mode is now the default for OSTs.
3790 Details : By default, OSTs will now run in failover mode. To return to
3791 the old behaviour, add '--failout' to the lmc line for OSTs.
3793 Severity : enhancement
3795 Description: Health checks are now provided for MDS and OSTs
3796 Details : Additional detailed health check information on MSD and OSTs
3797 is now provided through the procfs health_check value.
3800 Frequency : occasional, depends on IO load
3802 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3803 after numerous create/delete cycles
3804 Details : The ext3 inode allocation policy would not allocate new inodes
3805 very well on the OSTs because there are no new directories
3806 being created. Instead we look for groups with free space if
3807 the parent directories are nearly full.
3812 Description: Network or server problems during mount may cause partially
3813 mounted clients instead of returning an error.
3814 Details : The config llog parsing code may overwrite the error return
3815 code during mount error handling, returning success instead
3821 Description: MDS can fail to allocate large reply buffers
3822 Details : After long uptimes the MDS can fail to allocate large reply
3823 buffers (e.g. zconf client mount config records) due to memory
3824 fragmentation or consumption by the buffer cache. Preallocate
3825 some large reply buffers so that these replies can be sent even
3826 under memory pressure.
3830 Frequency : rare (liblustre)
3831 Description: fsx running with liblustre complained that using truncate() to
3832 extend the file doesn't work. This patch corrects that issue.
3833 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3834 ATTR_SIZE and lsm use in llu_setattr_raw.
3838 Frequency : rare, only 2.6 kernels
3839 Description: Unusual file access patterns on the MDS may result in inode
3840 data being lost in very rare circumstances.
3841 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3842 code in the 2.6 kernel caused the inode and/or EA data not to
3843 be read from disk, causing single-file corruption.
3847 Frequency : rare, only 2.6 filesystems using extents
3848 Description: Heavy concurrent write and delete load may cause data corruption.
3849 Details : It was possible under high-load situations to have an extent
3850 metadata block in the block device cache from a just-unlinked
3851 file overwrite a newly-allocated data block. We now unmap any
3852 metadata buffers that alias just-allocated data blocks.
3856 Frequency : filesystems with default stripe_count larger than 77
3857 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3858 Details : lconf specifies an inode size of 4096 bytes when the default
3859 stripe_count is larger than 77. This conflicts with the default
3860 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3861 this case to avoid pinning too much memory for large EAs.
3863 ------------------------------------------------------------------------------
3865 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3870 Frequency : rare (extremely heavy IO load with hundreds of clients)
3872 Description: Client is evicted, gets IO error writing to file
3873 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3874 caused clients to be evicted for AST timeouts. The fixes in
3875 bug 5192 mean we no longer need to have such short AST timeouts
3876 so ldlm_timeout has been increased.
3879 Frequency : occasional during --force or --failover shutdown under load
3880 Bugzilla : 5949, 4834
3881 Description: Server oops/LBUG if stopped with --force or --failover under load
3882 Details : a collection of import/export refcount and cleanup ordering
3883 issues fixed for safer force cleanup
3886 Frequency : only filesystems larger than 120 OSTs
3887 Bugzilla : 5990, 6223
3888 Description: lfs getstripe would oops on a very large filesystem
3889 Details : lov_getconfig used kfree on vmalloc'd memory
3892 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3893 Bugzilla : 6242, 6243
3894 Description: reading from files that had been truncated to a non-zero size
3895 but never opened returned no data
3896 Details : ll_file_read() reads zeros from no-object files to EOF
3901 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3902 Details : There's little harm in aborting MDS/OSS recovery and letting it
3903 try again, so I removed the LASSERT and return an error instead.
3905 Severity : enhancement
3907 Description: New debugging infrastructure for tracking down data corruption
3908 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3909 (b) cover more of the client-side code path, and (c) try to narrow
3910 down where problems occurred
3914 Bugzilla : 3819, 4364, 4397, 6313
3915 Description: Racing close and eviction MDS could cause assertion in mds_close
3916 Details : It was possible to get multiple mfd references during close and
3917 client eviction, leading to one thread referencing a freed mfd.
3919 Severity: : enhancement
3920 Bugzilla : 3262, 6359
3921 Description: Attempts to reconnect to servers are now more aggressive.
3922 Details : This builds on the enhanced upcall-less recovery that was added
3923 in 1.4.2. When trying to reconnect to servers, clients will
3924 now try each server in the failover group every 10 seconds. By
3925 default, clients would previously try one server every 25 seconds.
3930 Description: After recovery, certain operations trigger a failed
3931 assertion on a client.
3932 Details : Failing over an mds, using lconf -d --failover, while a
3933 client was doing a readdir() call would cause the client to
3934 LBUG after recovery completed and the readdir() was resent.
3936 Severity : enhancement
3938 Description: Default groups are now added by lconf
3939 Details : You can now run lconf --group <servicename> without having to
3940 manually add groups with lmc.
3943 Frequency : occasional
3945 Description: Nodes with an elan id of 0 trigger a failed assertion
3948 Frequency : always when accessing e.g. tty/console device nodes
3950 Description: tty and some other devices nodes cannot be used on lustre
3951 Details : file's private_data field is used by device data and lustre
3952 values in there got lost. New field was added to struct file to
3953 store fs-specific private data.
3956 Frequency : when exporting Lustre via NFS
3958 Description: NFSD failed occasionally when looking up a path component
3959 Details : NFSD is looking up ".." which was broken in ext3 directories
3960 that had grown large enough to become hashed.
3963 Frequency : Clusters with multiple interfaces not on the same subnet
3965 Description: Nodes will repeatedly try to reconnect to an interface which it
3966 cannot reach and report an error to the log.
3967 Details : Extra peer list entries will be created by lconf with some peers
3968 unreachable. lconf now validates the peer before adding it.
3971 Frequency : Only if a default stripe is set on the filesystem root.
3973 Description: Setting a default stripe on the filesystem root prevented the
3974 filesystem from being remounted.
3975 Details : The client was sending extra request flags in the root getattr
3976 request and did not allocate a reply buffer for the dir EA.
3979 Frequency : occasional, higher if lots of files are accessed by one client
3980 Bugzilla : 6159, 6097
3981 Description: Client trips assertion regarding lsm mismatch/magic
3982 Details : While revalidating inodes the VFS looks up inodes with ifind()
3983 and in rare cases can find an inode that is being freed.
3984 The ll_test_inode() code will free the lsm during ifind()
3985 when it finds an existing inode and then the VFS later attaches
3986 this free lsm to a new inode.
3990 Bugzilla : 6422, 7030
3991 Description: MDS deadlock between mkdir and client eviction
3992 Details : Creating a new file via mkdir or mknod (starting a transaction
3993 and getting the ns lock) can deadlock with client eviction
3994 (gets ns lock and trying to finish a synchronous transaction).
3997 Frequency : occasional
3998 Description: While starting a server, the fsfilt_ext3 module could not be
4000 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4001 kernels. Previously, lconf would still use the ext3 name
4002 when trying to load modules. Now, it will correctly use
4003 ext3 on 2.4 and ldiskfs on 2.6.
4005 Severity : enhancement
4006 Description: The default stripe count has been changed to 1
4007 Details : The interpretation of the default stripe count (0, to lfs
4008 or lmc) has been changed to mean striping across a single
4009 OST, rather than all available. For general usage we have
4010 found a stripe count of 1 or 2 works best.
4012 Severity : enhancement
4013 Description: Add support for compiling against Cray portals.
4014 Details : Conditional compiling for some areas that are different
4018 Frequency : occasional
4019 Bugzilla : 6409, 6834
4020 Description: Creating files with an explicit stripe count may lead to
4021 a failed assertion on the MDS
4022 Details : If some OSTs are full or unavailable, creating files may
4023 trigger a failed assertion on the MDS. Now, Lustre will
4024 try to use other servers or return an error to the
4028 Frequency : occasional
4030 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4031 caused lock timeout during readahead (since 1.4.2).
4032 Details : Processes doing readahead might match a lock that hasn't been
4033 granted yet if there are overlapping and conflicting lock
4034 requests. The readahead process waits on ungranted lock
4035 (original lock is CBPENDING), while OST waits for that process
4036 to cancel CBPENDING read lock and eventually evicts client.
4038 Severity : enhancement
4040 Description: Initial enabling of flock support for clients
4041 Details : Implements fcntl advisory locking and file status functions.
4042 This feature is provided as an optional mount flag (default
4043 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4044 locking are implemented yet, and those that are are not guaranteed
4045 to be completely correct in production environments.
4046 mount -t lustre -o [flock|noflock] ...
4049 Frequency : occasional
4051 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4052 assertion in the ext3 JBD (journaling) layer.
4053 Details : The b_committed_data struct is protected by the big kernel lock
4054 in 2.4 kernels, serializing journal_commit_transaction() and
4055 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4056 access to this struct. In 2.6 kernels there is finer grained
4057 locking to improve SMP performance of the JBD layer.
4061 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4063 -----------------------------------------------------------------------------
4065 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4067 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4068 versions of Lustre. Please update all servers and clients to
4069 version 1.4.2 or later at the same time. You must also run
4070 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4071 to update the configuration logs.
4073 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4074 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4075 - swab llog records of type '0' so we get proper header size/idx (5861)
4076 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4077 - fix rename of one directory over another leaking an inode (5953)
4078 - avoid SetPageDirty on 2.6 (5981)
4079 - don't re-add just-being-destroyed locks to the waiting list (5653)
4080 - when creating new directories, inherit the parent's custom
4081 striping settings if present parent (3048)
4082 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4083 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4084 - don't hold client locks on temporary worklist from l_lru (5666)
4085 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4086 - automatically evict dead clients (5921)
4087 - Update file size properly in create+truncate+fstat case (6196)
4088 - Do not unhash mountpoint dentries, do not allow removal of
4090 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4091 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4092 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4093 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4094 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4095 - Correctly handle reads of files with no objects (6243)
4096 - lctl recover will also mark a device active if deactivate used (5933)
4098 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4099 - allow --write-conf on an MDS with different nettype than client (5619)
4100 - don't write config llogs to MDS for mounts not from that MDS (5617)
4101 - lconf should create multiple TCP connections from a client (5201)
4102 - init scripts are now turned off by default; run chkconfig --on
4103 lustre and chkconfig --on lustrefs to use them
4104 - upcalls are no longer needed for clients to recover to failover
4106 - add --abort-recovery option to lconf to abort recovery on device
4108 - add support for an arbitrary number of OSTs (3026)
4109 - Quota support protocol changes.
4110 - forward compatibility changes to wire structs (6007)
4111 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4112 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4113 - improved "lustre status" script
4114 - initialize blocksize for non-regular files (6062)
4115 - added --disable-server and --disable-client configure options (5782)
4116 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4117 - Vanilla 2.4.29 support
4118 - increase maximum number of obd devices to 520 (6242)
4119 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4120 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4121 SLES 9 2.6.5 kernels
4122 - limit stripes per file to 160 (the maximum EA size) (6093)
4124 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4127 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4128 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4129 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4130 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4131 - small ext3 extents cleanups and fixes (5733)
4132 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4133 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4134 - avoid needless client->OST connect, fix handle mismatch (5317)
4135 - fix DLM error path that led to out-of-sync client, long delays (5779)
4136 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4137 - fix several locking issues related to i_size (5492,5624,5654,5672)
4138 - don't move pending lock onto export if it is already evicted (5683)
4139 - fix kernel oops when creating .foo in unlinked directory (5548)
4140 - fix deadlock in obdfilter statistics vs. object create (5811)
4141 - use time_{before,after} to avoid timer jiffies wrap (5882)
4142 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4143 - Do not leak request if server was not able to process it (5154)
4144 - If mds_open unable to find parent dir, make that negative lookup(5154)
4145 - don't create new directories with extent-mapping (5909, 5936)
4147 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4148 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4149 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4150 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4151 - support for RHEL4 kernel on i686 (5773)
4152 - provide error messages when incompatible logs are encountered (5898)
4154 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4155 * version 1.4.0.10 (1.4.1 release candidate 1)
4157 - don't keep a lock reference when lock is not granted (4238)
4158 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4159 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4160 - drop import inflight refcount on signal_completed_replay error (5255)
4161 - unlock page after async write error during send (3677)
4162 - handle missing objects in filter_preprw_read properly (5265)
4163 - no transno return for symlink open, don't save no-trasno open (3440)
4164 - don't try to complete elan receive that already failed (4012)
4165 - free RPC server reply state on error (5406)
4166 - clean up thread from ptlrpc_start_thread() on error (5160)
4167 - readahead could read extra page into cache that wasn't ejected (5388)
4168 - prevent races in class_attach/setup/cleanup/detach (5260)
4169 - don't dereference de->d_inode after l_dput of de (5458)
4170 - use "int" for stripe value returned from lock_to_stripe (5544)
4171 - mballoc allocation and error-checking fixes in 2.6 (5504)
4172 - block device patches to fix I/O request sizes in 2.6 (5482)
4173 - look up hostnames for IB nals (5602)
4174 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4175 - don't start multiple acceptors for the same port (5277)
4176 - fix incorrect LASSERT in mds_getattr_name (5635)
4177 - export a proc file for general "ping" checking (5628)
4178 - fix "lfs check" to not block when the MDS is down (5628)
4180 - service request history (4965)
4181 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4182 - create an "evict_client" /proc entry on OSTs, like the MDS has
4183 - fix mount usage message, return errors per mount(8) (5168)
4184 - change grep [] to grep "[]" in tests so they work in more UMLs
4185 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4186 - remove ancient LOV_MAGIC_V0 EA support (5047)
4187 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4188 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4189 - allow readahead allocations to fail when low on memory (5383)
4190 - mmap locking landed again, after considerable improvement (2828)
4191 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4193 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4196 - send OST transaction number in read/write reply to free req (4966)
4197 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4198 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4199 - the watchdog thread now runs as interruptible (5246)
4200 - flock/lockf fixes (but it's still disabled, pending 5135)
4201 - don't use EXT3 constants in llite code (5094)
4202 - memory shortage at startup could cause assertion (5176)
4204 - reorganization of lov code
4205 - single portals codebase
4207 - add extents/mballoc support (5025)
4208 - direct I/O reads in the obdfilter (4048)
4209 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4211 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4214 - send OST transaction number in read/write reply to free req (4966)
4215 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4216 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4217 - the watchdog thread now runs as interruptible (5246)
4218 - handle missing objects in filter_preprw_read properly (5265)
4219 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4220 - drop import inflight refcount on signal_completed_replay error (5255)
4221 - unlock page after async write error during send (3677)
4222 - return original error code on reconstructed replies (3761)
4223 - no transno return for symlink open, don't save no-trasno open (3440)
4225 - add pid to ldlm debugging output (4922)
4226 - bump the watchdog timeouts -- we can't handle 30sec yet
4227 - extra debugging for orphan dentry/inode bug (5259)
4229 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4232 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4233 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4234 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4235 - fix NULL dereference in /proc/sys/portals/routes (4827)
4236 - allow failed mdc_close() operations to be interrupted (4561)
4237 - stop precreate on OST before MDS would time out on it (4778)
4238 - don't send partial-page writes before EOF from client (4410)
4239 - discard client grant for sub-page writes on large-page clients (4520)
4240 - don't free dentries not owned by NFS code, check generation (4806)
4241 - fix lsm leak if mds_create_objects() fails (4801)
4242 - limit debug_daemon file size, always print CERROR messages (4789)
4243 - use transno after validating reply (3892)
4244 - process timed out requests if import state changes (3754)
4245 - update mtime on OST during writes, return in glimpse (4829)
4246 - add mkfsoptions to LDAP (4679)
4247 - use ->max_readahead method instead of zapping global ra (5039)
4248 - don't interrupt __l_wait_event() during strace
4250 - add software watchdogs to catch hung threads quickly (4941)
4251 - make lustrefs init script start after nfs is mounted
4252 - fix CWARN/ERROR duplication (4930)
4253 - return async write errors to application if possible (2248)
4254 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4255 - print NAL number in %x format (4645)
4256 - update barely-supported suse-2.4.21-171 series (4842)
4257 - support for sles 9 %post scripts
4258 - support for building 2.6 kernel-source packages
4259 - support for sles km_* packages
4261 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4264 - ignore -ENOENT errors in osc_destroy (3639)
4265 - notify osc create thread that OSC is being cleaned up (4600)
4266 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4267 - reconstruct ost_handle() like mds_handle() (4657)
4268 - create a new thread to do import eviction to avoid deadlock (3969)
4269 - let lconf resolve symlinked-to devices (4629)
4270 - don't unlink "objects" from directory with default EA (4554)
4271 - hold socknal file ref over connect in case target is down (4394)
4272 - allow more than 32000 subdirectories in a single directory (3244)
4273 - fix blocks count for O_DIRECT writes (3751)
4274 - OST returns ENOSPC from object create when no space left (4539)
4275 - don't send truncate RPC if file size isn't changing (4410)
4276 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4277 - bind to privileged port in socknal and tcpnal (3689)
4279 - rate limit CERROR/CWARN console message to avoid overload (4519)
4280 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4281 - basic mmap support (3918)
4282 - kernel patch series update from b1_4 (4711)
4284 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4287 - avoid crash during MDS cleanup with OST shut down (2775)
4288 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4289 - don't use bad inodes on the MDS (3744)
4290 - dynamic object preallocation to improve recovery speed (4236)
4291 - don't hold spinlock over lock dumping or change debug flags (4401)
4292 - don't zero obd_dev when it is force cleaned (3651)
4293 - print grants to console if they go negative (4431)
4294 - "lctl deactivate" will stop automatic recovery attempts (3406)
4295 - look for existing locks in ldlm_handle_enqueue() (3764)
4296 - don't resolve lock handle twice in recovery avoiding race (4401)
4297 - revalidate should check working dir is a directory (4134)
4299 - don't always mark "slow" obdfilter messages as errors (4418)
4301 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4304 - don't close LustreDB during write_conf until it is done (3860)
4305 - fix typo in lconf for_each_profile (3821)
4306 - allow dumping logs from multiple threads at one time (3820)
4307 - don't allow multiple threads in OSC recovery (3812)
4308 - fix debug_size parameters (3864)
4309 - fix mds_postrecov to initialize import for llog ctxt (3121)
4310 - replace config semaphore with spinlock (3306)
4311 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4312 - don't allow enqueue to complete on a destroyed export (3822)
4313 - down write_lock before checking llog header bitmap (3825)
4314 - recover from lock replay timeout (3764)
4315 - up llog sem before sending rpc (3652)
4316 - reduce ns lock hold times when setting kms (3267)
4317 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4318 - fix NULL deref and obd_dev leak on setup error (3312)
4319 - replace some LBUG about llog ops with error handling (3841)
4320 - don't match INVALID dentries from d_lookup and spin (3784)
4321 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4322 - fix invalid assertion in ptlrpc_set_wait (3880)
4324 - add libwrap support for the TCP acceptor (3996)
4325 - add /proc/sys/portals/routes for non-root route listing (3994)
4326 - allow setting MDS UUID in .xml (2580)
4327 - print the stack of a process that LBUGs (4228)
4329 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4332 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4333 - make sure to unset replay flag from failed open requests (3440)
4334 - if default stripe count is 0, use OST count for inode size (3636)
4335 - update parent mtime/ctime on client for create/unlink (2611)
4336 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4337 - free recovery state on server during a forced cleanup (3571)
4338 - unregister_reply for resent reqs (3063)
4339 - loop back devices mounting and status check on 2.6 (3563)
4340 - fix resource-creation race that can provoke i_size == 0 (3513)
4341 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4342 - more debugging for page-accounting assertion (3746)
4343 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4344 - avoid deadlock after precreation failure (3758)
4345 - fix race and lock order deadlock in orphan handling (3450, 3750)
4346 - add validity checks when grabbing inodes from l_ast_data (3599)
4348 - add /proc/.../recovery_status to obdfilter (3428)
4349 - lightweight CDEBUG infrastructure, debug daemon (3668)
4350 - change default OSC RPC parameters to be better on small clusters
4351 - turn off OST read cache for files smaller than 32MB
4352 - install man pages and include them in rpms (3100)
4353 - add new init script for (un)mounting lustre filesystems (2593)
4354 - run chkconfig in %post for init scripts (3701)
4355 - drop scimac NAL (unmaintained)
4357 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4360 - clean kiobufs before and after use (3485)
4361 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4362 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4363 - change init script to fail more clearly if not run as root (1528)
4364 - allow clients to reconnect during replay (1742)
4365 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4366 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4367 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4368 - fix cond_resched (3554)
4369 - only evict unfinished clients after recovery (3515)
4370 - allow bulk resend, prevent data loss (3570)
4371 - dynamic ptlrpc request buffer allocation (2102)
4372 - don't allow unlinking open directory if it isn't empty (2904)
4373 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4374 - remove extraneous obd dereference causing LASSERT failure (3334)
4375 - don't use get_cycles() when creating temp. files on the mds (3156)
4376 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4377 - handle EEXIST for set-stripe, set proper directory name (3336)
4379 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4380 - fix ksocknal_fmb_callback() error messages (2918)
4382 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4385 - don't copy lvb into (possibly NULL) reply on error (2983)
4386 - don't deref dentry after dput, don't free lvb on error (2922)
4387 - use the kms to determine writeback rpc length (2947)
4388 - increment oti_logcookies when osc is inactive (2948)
4389 - update client's i_blocks count via lvb messages (2543)
4390 - handle intent open/close of special files properly (1557)
4391 - mount MDS with errors=remount-ro, like obdfilter (2009)
4392 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4393 - don't use cancelling-locks' kms values (2947)
4394 - use highest lock extent for kms, not last one (2925)
4395 - don't dereference ERR_PTR() dentry in error handling path (3107)
4396 - fix thread race in portals_debug_dumplog() (3122)
4397 - create lprocfs device entries at setup instead of at attach (1519)
4398 - common AST error handler, don't evict client on completion race (3145)
4399 - zero nameidata in detach_mnt in 2.6 (3118)
4400 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4401 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4402 - initialize RPC timeout timer earlier for 2.6 (3219)
4403 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4404 - print nal/nid for unknown nid (3258)
4405 - additional checks for oscc recovery before doing precreate (3284)
4406 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4407 - don't crash in mdc_close for bad permissions on open (3285)
4408 - zero i_rdev for non-device files (3147)
4409 - clear page->private before handing to FS, better assertion (3119)
4410 - tune the read pipeline (3236)
4411 - fix incorrect decref of invalidated dentry (2350)
4412 - provide read-ahead stats and refine rpc in flight stats (3328)
4413 - don't hold journal transaction open across create RPC (3313)
4414 - update atime on MDS at close time (3265)
4415 - close LDAP connection when recovering to avoid server load (3315)
4416 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4417 - don't leak open file on MDS after open resend (3325)
4418 - serialize filter_precreate and filter_destroy_precreated (3329)
4419 - loop device shouldn't call sync_dev() for nul device (3092)
4420 - clear page cache after eviction (2766)
4421 - resynchronize MDS->OST in background (2824)
4422 - refuse to mount the same filesystem twice on same mountpoint (3394)
4423 - allow llmount to create routes for mounting behind routers (3320)
4424 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4425 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4426 - fix rare problem with rename on htree directories (3417)
4428 - allow default OST striping configuration per directory (1414)
4429 - fix compilation for qswnal for 2.6 kernels (3125)
4430 - increase maximum number of MDS request buffers for large systems
4431 - change liblustreapi to be useful for external progs like lfsck (3098)
4432 - increase local configuration timeout for slow disks (3353)
4433 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4435 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4438 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4439 - don't overwrite extent policy data in reply if lock was blocked (2901)
4440 - drop filter export grants atomically with removal from device (2663)
4441 - del obd_self_export from work_list in class_disconnect_exports (2908)
4442 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4443 - swab reply message in mdc_close, other PPC fixes (2464)
4444 - fix destroying of named logs (2325)
4445 - overwrite old logs when running lconf --write_conf (2264)
4446 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4447 - fix race in target_handle_connect (2898)
4448 - mds_reint_create() should take same inode create lock (2926)
4449 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4450 - don't close files for self_export to avoid uninitialized obd (2936)
4451 - allow MDS with the same name as client node (2939)
4452 - hold dentry reference for closed log files for unlink (2325)
4453 - reserve space for all logs during transactions (2059)
4454 - don't evict page beyond end of stripe extent (2925)
4455 - don't oops on a deleted current working directory (2399)
4456 - handle hard links to targets without a parent properly (2517)
4457 - don't dereference NULL lock when racing during eviction (2867)
4458 - don't grow lock extents when lots of conflicting locks (2919)
4460 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4463 - account for cache space usage on clients to avoid data loss (974)
4464 - lfsck support in lustre kernel code (2349)
4465 - reduce journal credits needed for BRW writes (2370)
4466 - orphan handling to avoid losing space on client/server crashes
4467 - ptlrpcd can be blocked, stopping ALL progress (2477)
4468 - use lock value blocks to assist in proper KMS, faster stat (1021)
4469 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4470 - recovery for initial connections (2355)
4471 - fixes for mds_cleanup_orphans (1934)
4472 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4473 - block all file creations until orphan recovery completes (1901)
4474 - client remove rq_connection from request struct (2423)
4475 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4476 - recovery timer race (2670)
4477 - mdc_close recovey bug (2532)
4478 - ptlrpc cleanup bug (2710)
4479 - mds timeout on local locks (2588)
4480 - namespace lock held during RPCs (2431)
4481 - handle interrupted sync write properly (2503)
4482 - don't try to handle a message that hasn't been replied to (2699)
4483 - client assert failure during cleanup after abort recovery (2701)
4484 - leak mdc device after failed mount (2712)
4485 - ptlrpc_check_set allows timedout requests to complete (2714)
4486 - wait for inflight reqs when ptlrpcd finishes (2710)
4487 - make sure unregistered services are removed from the srv_list
4488 - reset bulk XID's when resending them (caught by 1138 test)
4489 - unregister_bulk after timeout
4490 - fix lconf error (2694)
4491 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4492 - readahead locks pages, leaves pending causing memory pressure (2673)
4493 - increase OST request buffers to 4096 on large machines (2729)
4494 - fix up permission of existing directories in simple_mkdir (2661)
4495 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4496 - don't assign transno to errored transactions (2742)
4497 - don't delete objects on OST if given a bogus objid from MDS (2751)
4498 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4499 - if rq_no_resend, then timeout request after recovery (2432)
4500 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4501 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4502 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4503 - MDS recovery shouldn't delete objects that it has given out (2730)
4504 - if enqueue arrives after completion, don't clobber LVB (2819)
4505 - don't unlock pages twice when trigger_group_io returns error (2814)
4506 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4507 - don't write pages to disk if there was an error (1450)
4508 - don't ping imports that have recovery disabled (2676)
4509 - take buffered bytes into account when balancing socknal conn (2817)
4510 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4511 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4512 - remove little-endian swabbing of llog records (1987)
4513 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4514 - retry reposting request buffers if they fail (1191)
4515 - grow extent at grant time to avoid granting a revoked lock (2809)
4516 - lock revoke doesn't evict page if covered by a second lock (2765)
4517 - disable VM readahead to avoid reading outside lock extents (2805)
4519 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4520 - updated kernel patches for hp-2.4.20 kernel (2681)
4522 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4525 - fix truncated write corruption (2366)
4526 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4528 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4529 - ensure len(uuid) < 37 in lmc (1171)
4530 - fix ia64 OOPS in llog_test (2255)
4531 - zero end of page at obdfilter for partial page writes (2648)
4532 - don't leave stale dentries around after renames (bug 2428)
4533 - fix timeouts when evicting a client with a single lock held (2642)
4534 - set deadline for the initial HELLO message to drain (2634)
4535 - print out dotted-quad IP addresses in the socknal (2302)
4537 - additional debugging for MDS client eviction problem (2443)
4538 - fix mkfsoptions support for osts (2603, 2604)
4540 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4543 - add series for the vanilla 2.6.0 kernel
4544 - add series for the vanilla 2.4.24 kernel
4545 - add series for a cray x86/64 UL kernel drop
4546 - fix xattr patches for the vanilla 2.4.19 series
4548 - generate true UUIDs in lmc (1171)
4549 - have portals stack dumping break in UML (2466)
4550 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4551 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4552 - wake callback waiting threads on client eviction (2460)
4553 - Add --ptldebug and --subsystem to lmc (1719)
4554 - update assertion to allow safe interrupt allocation
4555 - set rq_no_resend for cancel requests (2432)
4556 - recalculate ptlrpcd timeout after resend (2494)
4557 - call vfs_rmdir when removing pending directories (2368)
4558 - fix renaming a file to itself (2429)
4559 - lmc creates a default one-stripe lov (2454)
4560 - expand procfs space to handle large clusters (2326)
4561 - increase UML stack to avoid overflow
4562 - update lconf's list of debug and subsystem masks
4563 - fix lfs find --obd (2510)
4564 - /proc tunable for disabling filter read caching (2591)
4565 - stop rpm packages from altering slapd.conf (2301)
4566 - disable nagle in the socknal under 0conf (2578)
4567 - choose mds inode size based on stripe count (2572)
4568 - fix kernel-source rpm problems (2516)
4570 - add --disable-doc to avoid pdf generation (2421)
4571 - update documentation, tests, type-os, comments
4572 - avoid format warnings on ia64
4573 - remove the TOE NAL
4574 - tiny code cleanups by removing unused fields
4576 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4579 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4580 - remove the most blatant lies from BUILDING file (2371)
4581 - change default debug level to reasonable production setting
4582 - reduce client side cache size to reduce cache flush time
4583 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4584 - make TCP zerocopy and pinger support enabled by default (2476)
4585 - sync writes completed after process exits caused crashes (2319)
4586 - maintain correct mount count on the MDS (2356)
4587 - backout 1557, because 2316 wasn't really fixed
4588 - better file I/O statistics gathering in /proc
4589 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4590 - another kernel patch to fix zero-copy TCP function export
4591 - don't take duplicate lock when processing re-sent getattr (2420)
4592 - lctl uses obd_self_export instead of creating new conn (2353)
4593 - MDS/OST recovery case which requires object creation asserted (2425)
4594 - move lfs from /usr/sbin to /usr/bin in packages
4595 - fix race between mds_client_add and mds_client_free (2417)
4596 - use kmalloc instead of slabs in portals (2430)
4597 - don't create duplicate records when a failover MDS is present (2442)
4598 - remove unnecessary mount age check (2332)
4599 - don't remove directory inodes from locks prematurely (2451)
4600 - don't break if MDS service name is the same as hostname (2103)
4601 - fix races in client write RPC generation when cache full (2482)
4603 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4606 - remove now-unused request->rq_obd (278)
4607 - if an allocation fails, print out how much memory we've used (1933)
4608 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4609 - add the "configurable stack size" patch to most series files (1256)
4610 - ability to write large log records, for 100+ OST configs (2306)
4611 - fix NULL deref when filter_prep fails (2314)
4612 - fix operator precedence error in filter_sync
4613 - dynamic allocation of socknal TX descriptors (2315)
4614 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4615 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4616 - ensure that CWARN messages in Portals always get to the syslog
4617 - __init/__exit are not for prototype decls (ldlm_init/exit)
4618 - x86-64 compile warning fixes
4619 - fix gateway LMC keyword conflict (2318)
4620 - fix MDS lock inversions in getattr/reint paths (1844)
4621 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4622 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4623 - DLM race condition prevented some lock evictions (2328)
4624 - ENOMEM detection and retry on socknal sends (2230)
4625 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4626 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4627 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4628 - better error messages when a client is rejected during recovery (1505)
4629 - avoid cancelling locks which were never granted, after failure (2330)
4630 - fix i_sem/journal inversion in mds_client_add (2333)
4631 - fix truncate/getattr lock cycle deadlock (2334)
4632 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4633 - fix two rare exit paths which could leak an l_lock() ref (2321)
4634 - fencepost error in MDS/OST orphan recovery (2226)
4635 - make log record alignment 8 bytes (1988)
4636 - lstripe now fails when requested offset > ost_count (2237)
4637 - ensure that all kernel series have a complete list.h (1607)
4638 - fix crashes in special-file operations (2316)
4639 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4641 - allow configurable automake binary, for testing new versions
4642 - small update to the lfs documentation
4644 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4646 * fix negative export reference count in fsfilt_sync (2312)
4648 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4649 * release candidate 0.9.1
4651 - orphans are moved into the PENDING directory for possible recovery
4652 - replayed opens now open by fid for orphan/rename safety (1042)
4653 - last close of an orphan inode generates a transno (683)
4654 - chdir() and mount() now pin the directory entry (1020)
4655 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4656 - discard very old requests without processing them (1502)
4657 - remove some common, well-understood CERRORs (1505)
4658 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4659 - clear "grant" flags in OST replies until OST grant code lands (1644)
4660 - fix read performance by not clobbering i_blksize on client (1598)
4661 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4662 - make LRU size a /proc tunable, clears locks when reduced (707)
4663 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4664 - server locks take references on exports now (1558)
4665 - build fixes for 2.4.20-rh trees (1663)
4666 - return an error from lov_create if all OSCs are inactive (1751)
4667 - fix import levels when a reconnect happens without a timeout (1597)
4668 - exit early from mds_open if we get a lookup error (1749)
4669 - partial page read at EOF wouldn't wait for disk before sending (1642)
4670 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4671 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4672 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4673 - write/truncate lock inversion (1639)
4674 - Don't auto-load obdclass, portals modules during cleanup (1495)
4675 - fix timestamps from jumping to "now" (1763)
4676 - extra journal assertions (1648)
4677 - add an extra multiunlink test (1771)
4678 - fix read_record/write_record API (1776)
4679 - fix leak of offset_extent, possible incorrect i_size later (1772)
4680 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4681 - Debugging for the kqswnal_get_idle_tx problems (1820)
4682 - Allow recovery to be attempted multiple times (1536)
4683 - Write out MDS last_rcvd file after it is first created (1600)
4684 - Fix tx_descriptor leak in failed transmit situations (1827)
4685 - ext3 journaling fixes for assertion failure after IO error (1871)
4686 - class_export_put() on freed export after completion AST error (1896)
4687 - Fix revalidate looping in VFS (1322)
4688 - Don't access a freed export during MDS_REINT timeout (1521)
4689 - Add open-unlink recovery support on the MDS (1673,1764)
4690 - Return an error if no MDS data was read from last_rcvd (1946)
4691 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4692 - Don't setup a disk device that doesn't match exported UUID (317)
4693 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4694 - avoid committing NULL handle in force close
4695 - local.sh is now a one-stripe LOV configuration
4696 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4697 - fix ext3 htree duplicate directory entry corruption (1516)
4698 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4699 - update to new LOV EA format (2097)
4700 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4702 2003-06-15 Phil Schwan <phil@clusterfs.com>
4705 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4706 - per-import/export recovery handling (958, 931, 959)
4707 - multiple last-rcvd slots, for serving multiple FSes (949)
4708 - connections are again shared between multiple imp/exports (963, 964)
4709 - "umount -f" would hang if any requests needed to be sent (393, 978)
4710 - avoid pinning large req buffer by copying for queued messages (989)
4711 - add "uuid" to "lctl device" command to help upcalls (991)
4712 - "open" RPCs with transnos would confuse recovery counters (1037)
4713 - do proper endian conversion of all wire messages (288, 340, 891)
4714 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4715 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4716 - fix ptlrpc_connection leak in target_handle_connect (1174)
4717 - fix import refcounting bug in OST and MDS cleanup (1134)
4718 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4719 - fix very unlikely obd_types race condition (501)
4720 - remove osc_open hack for echo_client (1187)
4721 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4722 - a failure in read_inode2 leads to deadlock (1139)
4723 - cancel ack-locks as soon as transaction is committed (1072)
4724 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4725 - make sure to commitrw after any preprw to avoid deadlock (1162)
4726 - failing to execute a file in a lustre FS would lock inode (1203)
4727 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4728 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4729 - obd_uuid2tgt would walk off the end of the list (1255)
4730 - on IA64 the osc would give portals incorrect bulk size (1258)
4731 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4732 - fix lock inversion caused by new llite matching code (1282)
4733 - limit the number of dirty pages on a client to 10MB (1286)
4734 - timed out locks were not being corrected cancelled (1289)
4735 - fix O_DIRECT above 4GB on IA-32 (1292)
4736 * major user-visible changes
4737 - fail out/fail over policy now controlled by the upcall (993)
4739 - add OBD_PING to check server availability and failure (954)
4740 - lustre messages are now sent in sending host order (288, 340, 891)
4741 - add eadatalen to MDS getattr reply (340)
4742 - OST read replies may contain second buffer, with per-page status (593)
4744 2003-03-11 Phil Schwan <phil@clusterfs.com>
4747 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4748 - failed open()s could cause deadlock; fixed (867, 869)
4749 - stop cancelling OST locks when files are closed (481)
4750 - overlapping XID spaces caused network corruption (851, 853)
4751 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4752 - setattr_raw added, to do single-RPC, server-side setattrs
4753 - lmc/lconf syntax change for OST UUIDs
4754 - fix crashy race condition between ptlrpc_free_req and osc_close
4755 - don't use request in mdc_enqueue if we hit a timeout (889)
4756 - don't set the inode i_size for regular files from the MDS (896)
4757 - handle out of order completion AST (842)
4758 - don't LBUG if a lock request times out after receiving AST (913)
4759 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4760 - if a bad lock AST arrives, send an error instead of dropping entirely
4761 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4762 - fix leak in bulk IO when only partially completed (899, 900, 926)
4763 - fix O_DIRECT for ia64 (55)
4764 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4765 - C-z after timeout could hang a process forever; fixed (977)
4767 - client-side I/O cache (678, 924, 929, 941, 970)
4769 - READPAGE and SETATTRs which don't take server-side locks get
4772 2003-02-11 Phil Schwan <phil@clusterfs.com>
4775 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4776 - Fix fsfilt_extN_readpage() to read a full page of directory
4777 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4778 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4779 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4780 - Most intent operations were removed, in favour of a new RPC mode
4781 that does a single RPC to the server and bypasses most of the VFS
4782 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4783 - Aggressively cancel local locks on DLM servers
4784 - mds_reint_unlink sends EA to the client if it's the last nlink.
4785 client uses that EA to unlink OST objects.
4786 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4787 - recursive symlinks were fixed (439)
4788 - fixed NULL deref in DEBUG_REQ
4789 - filter_update_lastobjid no longer calls sync, which annoyed extN
4790 - fixed multi-client small-writes to a single file problem (445)
4791 - fixed mtime updates during file writes (607)
4792 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4793 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4794 - fixed Linux OST size reporting problem (444, 656)
4795 - OST now updates object mtime with writes or setattr (607, 619)
4796 - client verifies file size before zeroing page past EOF (445)
4797 - OST now writes last allocated objid to disk with allocation (108)
4798 - LOV on echo now works (409)
4800 - mds_reint_unlink sends a new buffer, with the EA included. this
4801 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4802 set if a regular file was being unlinked, and it was the last link
4803 - use PtlGet from the target for bulk writes (315)
4804 - OST now updates object mtime with writes or setattr (607, 619)
4805 - LDLM now has a grant-time callback to revalidate locked items, if
4807 - Many MDS operations were reorganized to combat race conditions
4809 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4810 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4812 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4815 - Fully reactivate OST imports after reconnection (512, others)
4816 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4817 - More graceful error handling for truncating on dead OST (515)
4818 - Don't error out unless we're actually accessing dead stripes (474)
4819 - Fix garbage sizes when stripes are missing (410)
4820 - LRU counters were broken, causing constant lock purge (433, 432)
4821 - garbage on read from stripes with failed OSTs (441)
4822 - mark OSCs as active before reconnecting during recovery (438)
4823 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4824 - lfind did not preserve OST order in output (443)
4825 - symlinks cause hung clients, incorrect data (439)
4826 - stop dereferencing request after dropping refcount (457)
4827 - don't LASSERT(spin_is_locked) on non-SMP (455)
4828 - fixes for many rename() bugs
4829 - fstat didn't correctly synchronize attributes (399)
4830 - server must handle lock cancellation during blocking AST prep (487)
4831 - bulk descriptors were free()d too soon (511)
4832 - fix paths in lconf, which would load incorrect modules (451, 507)
4833 - fix confusing lconf 'host not found' error message (386)
4834 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4835 - fix race condition in mdc_blocking_ast() for inode access (526)
4836 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4837 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4838 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4839 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4840 - fix lstripe in several ways (kernel side) (527)
4841 - fix request leak in ldlm_cli_enqueue (262)
4842 - incorrect OSC was marked inactive after OST failure
4843 - call mds_fs_cleanup before unmounting filesystem (524)
4844 - fix races between taking ns_lock and ldlm_lock_change_resource
4845 - fix races updating LOV export open file list
4846 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4847 - fix recovery NULL deref in ldlm_cli_cancel_unused
4848 - fix some DLM races by using new hash table for lock handles (419)
4849 - permit the client to specify desired inodes, at replay
4850 - duplicate requests when we queue them for replay reintegration
4851 - fix last_rcvd offset calculation
4852 - sync after each recovered transaction, so we always make progress
4853 - never, not always, ERESTART requests without transnos
4854 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4856 - skip replay if the MDS says that the client is already connected
4857 - don't check for a recovery-enabled export to match lctl's UUID
4858 - don't INC_USE_COUNT for phantom exports
4859 - don't crash when cleaning up phantom exports (567)
4860 - don't double-finish or set replay data for errored mdc_open requests
4861 - abort requests when they time out, so we don't get old replies
4862 - send/receive replies for AST messages again
4863 - if the client says that it doesn't have the lock, cancel it on the
4865 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4866 mark it as destroyed, it will all work out when decref is called
4867 - fix module use counts (22, 581)
4869 - ASTs now expect a reply (server cancels lock on error reply)
4871 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4874 - fix many simultaneous client startup (392)
4875 - fix dentry->d_it clobbering
4876 - credentials weren't being shipped for readdir/getattr operations
4877 - remove invalid assertions triggered during some concurrent MD
4879 - proper Lustre versions added (336, 389)
4880 - fix memory leak for create error case (398)
4881 - fix LOV locking bug that would get cli/srv out of sync
4882 - fix echo client over LOV (409)
4883 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4884 - fix double-O_EXCL intent crash (424)
4885 - avoid sending multiple lock CANCELs (352)
4887 - MDS can do multi-client recovery (modulo bugs in new code)
4889 - many updates, edits, cleanups
4891 2002-11-18 Phil Schwan <phil@clusterfs.com>
4894 - fix null d_it dereference (346)
4895 - fix full OST/dbench hang (333)
4896 - fix permission problem with file removal (286)
4897 - fix removal of OSCs from LOV when they fail
4898 - fix NULL deref during bulk timeout (214)
4899 - fix problems related to multiple filesystems on one MDS (241)
4900 - fixed serious subtle metadata locking bugs
4901 - free locks on clients when inodes are removed due to memory
4903 - fix inode pointer in lock data (285)
4904 - partial support for multiple MDS on a single host (241)
4905 - data locks weren't cancelled at clear_inode time (290, 311)
4906 - intent locks could lead to unbounded lock growth (205)
4907 - added a maximum lock count, an LRU list, and a flusher
4908 - fix multiple rename (365)
4909 - properly abstracted the echo client
4910 - OSC locked 1 byte too many; fixed
4911 - rewrote brw callback code:
4912 - fixed recovery bugs related to LOVs (306)
4913 - fixed too-many-pages-in-one-write crash (191)
4914 - fixed (again) crash in sync_io_timeout (214)
4915 - probably fixed callback-related race (385)
4917 - Add capability to MDS protocol
4918 - LDLM cancellations and callbacks on different portals
4920 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4923 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4924 - defer open object destruction to close time (601981, 138)
4925 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4926 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4927 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4928 - don't call obd_setattr in ll_file_release for destroyed objects
4929 * protocol change to lustre_msg: move |version| and add |flags|
4930 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4931 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4932 * added replay of create, unlink, link and rename operations during
4933 MDS failover; recovery should be much more robust now
4934 * remove failed OSCs from LOVs (only lov_create uses this so far)
4935 * the lustre-HOWTO was brought (more) up to date (582544)
4937 2002-10-23 Phil Schwan <phil@clusterfs.com>
4940 - in-use dentries weren't being reused properly (617851)
4941 - prevent multiple LDLM setup (599178)
4942 - fix LOV size calculations for truncate (617853)
4943 - fix client handling of MDS intent errors (POSIX)
4944 - fix permission bug in lovstripe.c test (624321)
4945 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4946 - truncate past end of file could corrupt data
4947 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4948 - a race in recovery could return ETIMEDOUT to apps (623947)
4949 - building outside the source directory was fixed
4950 * the lustre-HOWTO was brought (more) up to date (582544)
4951 * major progress was made on recovery functionality
4953 2002-10-10 Phil Schwan <phil@clusterfs.com>
4956 - recovery deadlock fix
4957 - rm -rf causes LBUG fix (617817)
4958 - file open by multiple tasks fix (618962)
4959 - directory permissions bugs (602707 and 620007)
4960 - journal_stop fixed with locking (611313)
4961 - O_APPEND failures resolved (618273, perhaps 614459)
4962 - lconf PATH fix (619770)
4963 - IA64 build fix (621450)
4964 - RPC buffer sizes scale with amount of memory
4966 2002-10-01 Phil Schwan <phil@clusterfs.com>
4969 - locks would be cancelled without throwing away data pages,
4970 resulting in inconsistent data (605627)
4971 - inode attributes were not always being refreshed (605627, 612449)
4972 - lconf now continues to cleanup after lctl reports an error
4973 - MDS now enforces user permissions (602707)
4974 - lprocfs cleanup fixed, but not yet enabled (614157)
4975 - fixed infinite server hang, should a client not respond to an AST
4976 - avoid going into recovery if user calls readlink() with a buffer
4977 that's too small (613941)
4978 - AST RPCs no longer require replies (614867) -- this may be changed
4979 - don't crash server if client sends an IOV that's too big (611336)
4980 - fixed lock conversion deadlock (611892)
4981 - fixed the following of symlinks (614622)
4982 * recovery: the server can remove locks from a client that dies, other
4983 clients can make progress
4984 * more extN patch fixes
4985 * compile-time configurable ptlrpc buffer allocations
4987 - collaborative read cache document
4988 - Lustre Lite Performance CDR document-in-progress
4990 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4993 - fix typo in patch-2.4.18
4995 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4998 - clear ptlrpc request each time in handle_incoming_request()
4999 - unlink of files now destroys the object on the OST
5001 2002-09-19 Peter Braam <braam@clusterfs.com>
5003 * add hard link support
5004 * change obdfile creation method
5005 * kernel patch changed
5007 2002-09-19 Peter Braam <braam@clusterfs.com>
5010 - stack overflow bug in extN fixed
5012 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5014 * documentation updates
5015 - add man pages for config tools
5016 - update tests/README to describe testing with new config tools
5017 - finish metadata API descriptions
5018 * bug fixes and cleanups
5019 - statfs workaround for 16TB limit
5020 - LOV stripe allocation improved, can stripe on subset of OSTs
5021 - LOV file size/IO offset was wrong for files > 4GB in size
5022 - object EA data was being dropped, caused files to be unreadable
5023 - memory overflow with non-LOV OST caused memory corruption
5024 - fixed regression tests to work with new config tools, obdfilter
5025 - fixed bug when directory size became larger than 1 block
5026 - fixed bug (for single client case) when PWD was deleted
5027 - invalidate local directory pages when doing intent-based ops
5028 - avoid LDLM oops when lock callback contained bad data
5030 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5032 * documentation updates
5033 * bug fixes and cleanups
5034 - configuration tools
5037 - 64-bit compile warnings
5038 - 64-bit internal statfs data
5040 * test_brw on persistent OST devices
5042 * lprocfs (disabled)
5044 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5046 * documentation updates
5047 * bug fixes and cleanups
5048 * configuration tools
5050 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5053 * many small fixes to 0_5_4
5054 * io/network handling
5055 * thinkos in MDS operations
5057 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5060 * crucial basic fixes to 0.5.3
5061 * IOR, Iozone work over Elan
5064 2002-08-07 Phil Schwan <phil@clusterfs.com>
5065 * version 0_5_3, our first alpha
5066 * we use the new Portals iovs
5067 * documentation updates
5068 * bug fixes and cleanups
5069 * small changes in the DLM wire protocol
5071 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5072 * version 0_5_1 with some initial stability,
5073 * locking on MD and file I/O.
5074 * documentation updates
5075 * several bug fixes since 0.5.0
5076 * small changes in wire protocol
5078 2002-07-18 Phil Schwan <phil@clusterfs.com>
5080 * delivered as Lustre Light Alpha
5081 * fixed a crash after handling invalid MDS requests
5082 * fixed directory pages for architectures with non-4k pages sizes
5084 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5085 * release version v0_4_4
5086 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5088 * Fixes a number of interruption problems with OST operations.
5089 * Update documentation for portals header changes
5090 * Move all wire protocol structs/defines to lustre_idl.h
5091 * Fixes symlink length bug.
5092 * Add tcpdump to repository.
5094 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5095 * release version v0_4_3
5096 * Fixes statfs for inodes on extN.
5097 * Fixes bug in runtests which would delete /etc/hosts.
5098 * Use 64-bit object IDs wherever possible (not into VFS though)
5099 Remove ost_get_info, which is unused by lustre, and out of date.
5101 2002-07-03 Peter Braam <braam@clusterfs.com>
5102 * release version v0_4_2 Fixes a lookup error (type not passed)
5103 * move forward to head of Portals
5104 * move forward to latest Lustre kernel
5106 2002-06-25 Peter Braam <braam@clusterfs.com>
5107 * release version v0_4_1. Hopefully stable on single node use.