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 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.22 vanilla (kernel.org)
12 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
13 we do not recommend using patchless RHEL4 clients with kernels
14 prior to 2.6.9-55EL (RHEL4U5).
15 * Recommended e2fsprogs version: 1.40.4-cfs1
16 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
17 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
18 removed cwd "./" (refer to Bugzilla 14399).
22 Description: lustre_mgs: operation 101 on unconnected MGS
23 Details : When MGC is disconnected from MGS long enough, MGS will evict the
24 MGC, and late on MGC cannot successfully connect to MGS and a lot
25 of the error messages complaining that MGS is not connected.
27 Severity : enhancement
29 Description: Add server support for vanilla-2.6.22.14.
32 Frequency : rare, depends on device drivers and load
34 Description: MDS or OSS nodes crash due to stack overflow
35 Details : Code changes in 1.6.4 increased the stack usage of some functions.
36 In some cases, in conjunction with device drivers that use a lot
37 of stack, the MDS (or possibly OSS) service threads could overflow
38 the stack. One change which was identified to consume additional
39 stack has been reworked to avoid the extra stack usage.
41 Severity : enhancement
43 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
45 Severity : enhancement
47 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
49 Severity : enhancement
51 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
53 Serverity : enhancement
55 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
58 Frequency : rare on shutdown ost
60 Description: Don't allow skipping OSTs if index has been specified.
61 Details : Don't allow skipping OSTs if index has been specified, make locking
62 in internal create lots better.
66 Description: ASSERTION(!PageDirty(page)) failed
67 Details : Wrong check could lead to an assertion failure under specific
72 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
73 Details : access to bitfield in structure is always rounded to long
74 and this produce problem with not atomic change any bit.
78 Description: Lustre make rpms failed.
79 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
80 builds Lustre rpms from tarball.
82 Severity : enhancement
84 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
87 Frequency : rare on shutdown ost
89 Description: If llog cancel was not send before clean_exports phase, this can
90 produce deadlock in llog code.
91 Details : If llog thread has last reference to obd and call class_import_put
92 this produce deadlock because llog_cleanup_commit_master wait when
93 last llog_commit_thread exited, but this never success because was
94 called from llog_commit_thread.
98 Description: NULL lov_tgts causing MDS oops
99 Details : more safe checks for NULL lov_tgts for avoid oops.
101 Severity : enhancement
103 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
105 Severity : enhancement
107 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
111 Description: make mgs_setparam() handle fsname containing dash
112 Details : fsname containing a dash does not work with lctl conf_param
114 Severity : enhancement
116 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
120 Frequency : rare, in recovery and (or) destroy lovobjid file.
121 Description: rewrite lov objid code.
122 Details : Cleanup for lov objid code, remove scability problems and wrong
123 locking. Fix sending last_id into ost.
125 Severity : enhancement
127 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
129 Severity : enhancement
131 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
132 Details : Use d_move_locked instead of __d_move.
136 Frequency : rare, at shutdown
137 Description: access already free / zero obd_namespace.
138 Details : if client_disconnect_export was called without force flag set,
139 and exist connect request in flight, this can produce access to
140 NULL pointer (or already free pointer) when connect_interpret
141 store ocd flags in obd_namespace.
145 Frequency : only at startup
146 Description: not alloc memory with spinlock held.
147 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
148 if any spinlock held.
153 Description: lfs find does not continue on file error
154 Details : Continue other files processing when a file/dir is absent.
158 Description: Inconsistent usage of lustre_pack_reply()
159 Details : Standardize the usage of lustre_pack_reply() such that it
160 always generate a CERROR on failure.
163 Frequency : very rare
165 Description: Fix replay if there is an un-replied request and open
166 Details : In some cases, older replay request will revert the
167 mcd->mcd_last_xid on MDS which is used to record the client's
170 Severity : enhancement
172 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
174 Severity : enhancement
176 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
178 Severity : enhancement
180 Description: Update to SLES9 kernel-2.6.5-7.287.3.
182 Severity : enhancement
184 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
186 Severity : enhancement
188 Description: Build SLES10 patchless client fails
189 Details : The configure was broken by run ./configure with
190 --with-linux-obj=.... argument for patchless client. When the
191 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
192 can't search header adequately. Use absolute path such as
193 -I($LINUX)/include instead.
195 Severity : enhancement
197 Description: Lustre Page Accounting
198 Details : New macros for page alloc and free which enable accounting
199 of page allocation of Lustre. Use percpu counters to store memory
203 Frequency : only if debugging is disabled
205 Description: LASSERT_{REQ,REP}SWAB macros are buggy
206 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
207 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
208 these macros and replace them with their difinitions instead.
213 Description: interrupt oig_wait produce painc on resend.
214 Details : brw_redo_request can be used for resend requests from ptlrpcd and
215 private set, and this produce situation when rq_ptlrpcd_data not
216 copyed to new allocated request and triggered LBUG on assert
217 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
218 ptlrpcd set if request is changed and can be safety changed to use
221 Severity : enhancement
223 Description: organize the server-side client stats on per-nid basis
224 Details : Change the structure of stats under obdfilter and mds to
232 The "uuid"s file would list the uuids of _active_ exports.
233 And the clear entry is to clear all stats and stale nids.
238 Description: Processes looping in ll_readdir() on Lustre clients finally causing
239 a full node pseudo-hang
240 Details : Concurrent access to the same directory from multiple clients with
241 intensive file creation/removal can cause a client node to spin in
242 ll_readdir(). i_version must be increased every time the lock is
243 cancelled to ensure a revalidate is done.
248 Description: touch file failed when fs is not full
249 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
250 otherwise we can get ENOSP while fs is not full.
253 Frequency : only for Cray XT3
254 Bugzilla : 12829/13455
255 Description: Changing primary group doesn't change the group lustre assigns to
257 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
258 overridden with the primary group provided by the group upcall,
259 whereas the supplied fsgid can be trusted if it is in the list of
260 supplementary groups returned by the group upcall.
264 Description: Root Squash Functionality
265 Details : Implementation of NFS-like root squash capability. Specifically,
266 don't allow someone with root access on a client node to be able
267 to manipulate files owned by root on a server node.
271 Description: Slow trucate/writes to huge files at high offsets.
272 Details : Directly associate cached pages to lock that protect those pages,
273 this allows us to quickly find what pages to write and remove
274 once lock callback is received.
278 Description: Too many locks accumulating on client during NFS usage
279 Details : mds_open improperly used accmode to find out access mode to a
280 file. Also mdc_intent_lock logic to find out if we already have
281 lock similar to just received was flawed since introduction of
282 skiplists - locks are now added to the front of the granted
287 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
289 Details : In obd setup failure handler, obd_stopping will not necessarily to
290 be 1, and obd_set_up should also be checked to make sure whether
291 obd is completely setup.
293 Severity : enhancement
295 Description: Allow masking D_WARNING, D_ERROR messages from console
296 Details : Console messages can now be disabled via lnet.printk.
300 Description: User code with malformed file open parameter crashes client node
301 Details : Before packing join_file req, all the related reference should be
302 checked carefully in case some malformed flags cause fake join_file
308 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
309 grain of quota doing harm to quota's accuracy
310 Details : qunit size will be changed when quota limitation is too low/high;
311 record the pending quota write in order to get more accureate
312 quota; delete the patch for bug12588, which is unnecessary when
313 this patch is landed. This bug also contains fixes for bug 14526,
314 14299, 14601 and 13794.
318 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
319 Details : ldlm_completion_ast() assumes that a lock is granted when the req
320 mode is equal to the granted mode. However, it should also check
321 that LDLM_FL_CP_REQD is not set.
325 Description: Heavy nfs access might result in deadlocks
326 Details : After ELC code landed, it is now improper to enqueue any mds
327 locks under och_sem, because enqueue might want to decide to
328 cancel open locks for same inode we are holding och_sem for.
332 Description: 35% write performance drop with ldiskfs2 when quotas are on
333 Details : Enable ext3 journalled quota by default to improve performance
334 when quotas are turned on.
338 Description: Client eviction while running blogbench
339 Details : A lot of unlink operations with concurrent I/O can lead to a
340 deadlock causing evictions. To address the problem, the number of
341 oustanding OST_DESTROY requests is now throttled to
342 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
343 callbacks are processed in priority.
346 Frequency : RHEL4 only
348 Description: mkfs is very slow on IA64/RHEL4
349 Details : A performance regression has been discovered in the MPT Fusion
350 driver between versions 3.02.73rh and 3.02.99.00rh. As a
351 consequence, we have downgraded the MPT Fusion driver in the RHEL4
352 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
355 Frequency : PPC/PPC64 only
357 Description: conflicts between asm-ppc64/types.h and lustre_types.h
358 Details : fix duplicated definitions between asm-ppc64/types.h and
359 lustre_types.h on PPC.
362 Frequency : PPC/PPC64 only
364 Description: asm-ppc/segment.h does not exist
365 Details : fix compile issue on PPC.
369 Description: data checksumming impacts single node performance
370 Details : add support for several checksum algorithms. Currently, CRC32 and
371 Adler-32 are supported. The checksum type can be changed on the fly
372 through /proc/fs/lustre/osc/*/checksum_type.
376 Description: use adler32 for page checksums
377 Details : when available, use the Adler-32 algorithm instead of CRC32 for
382 Description: better handle error messages in extents code
384 Severity : enhancement
386 Description: SNMP support enhancement
387 Details : Adding total number of sampled request for an MDS node in snmp
390 Severity : enhancement
392 Description: Optimize ldlm waiting list processing for PR extent locks
393 Details : When processing waiting list for read extent lock and meeting read
394 lock that is same or wider to it that is not contended, skip
395 processing rest of the list and immediatelly return current
396 status of conflictness, since we are guaranteed there are no
397 conflicting locks in the rest of the list.
401 Description: Time out and refuse to reconnect
402 Details : When the failover node is the primary node, it is possible
403 to have two identical connections in imp_conn_list. We must
404 compare not conn's pointers but NIDs, otherwise we
405 can defeat connection throttling.
409 Description: Client not clear own cache if answer to reconnect is lost.
410 Details : Client gets evicted from server. Now client also thinks it is
411 disconnected (or gets ENOTCONN on its operation) and decides to
412 reconnect. Server receives reconnect message, but cannot find
413 export. New export is created that is fully valid (new cookie!),
414 but reply is lost and not reported to client. Client reconnects
415 again and gets back a just-created connection, but it is not new
416 so client thinks it was not evicted and does not do recovery.
420 Description: Detect stride IO mode in read-ahead
421 Details : When a client does stride read, read-ahead should detect that and
422 read-ahead pages according to the detected stride pattern.
424 --------------------------------------------------------------------------------
426 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
428 * Support for kernels:
429 2.6.5-7.286 (SLES 9),
430 2.6.9-55.0.9.EL (RHEL 4),
431 2.6.16.53-0.8 (SLES 10),
432 2.6.18-8.1.14.el5 (RHEL 5),
433 2.6.18.8 vanilla (kernel.org)
434 * Client support for unpatched kernels:
435 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
436 2.6.16 - 2.6.22 vanilla (kernel.org)
437 * Due to recently discovered recovery problems, we do not recommend
438 using patchless RHEL 4 clients with this or any earlier release.
439 * Recommended e2fsprogs version: 1.40.2-cfs1
440 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
443 Frequency : occasional
445 Description: MDS hang or stay in waiting lock
446 Details : If client receive lock with CBPENDING flag ldlm need send lock
447 cancel as separate rpc, to avoid situation when cancel request
448 can't processed due all i/o threads stay in wait lock.
451 Frequency : occasional
453 Description: Do not fail import if osc_interpret_create gets -EAGAIN
454 Details : If osc_interpret_create got -EAGAIN it immediately exits and
455 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
456 oscc_has_objects and see OSC has no objests and call
457 oscc_internal_create to resend create request.
459 Severity : enhancement
461 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
462 Details : Update which_patch & target file for SLES10 latest kernel.
464 Severity : enhancement
466 Description: add --type and --size parameters to lfs find
467 Details : Enhance lfs find by adding filetype and filesize parameters. Also
468 multiple OBDs can now be specified for the --obd option.
470 Severity : enhancement
472 Description: eliminate client locks in face of contention
473 Details : file contention detection and lockless i/o implementation
476 Severity : enhancement
478 Description: Remove client patches from SLES 10 kernel.
479 Details : This causes SLES 10 clients to behave as patchless clients
480 even on a Lustre-patched (server) kernel.
482 Severity : enhancement
484 Description: use i_size_read and i_size_write in 2.6 port
485 Details : replace inode->i_size access with i_size_read/write()
488 Frequency : when removing large files
490 Description: scheduling issue during removal of large Lustre files
491 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
492 It causes scheduling issues when removing large files (17TB in the
498 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
499 Details : Impossible (EPROTO is returned) to access a directory that has a
500 non-default striping and ACLs.
503 Frequency : only on ppc
505 Description: /proc/fs/lustre/devices broken on ppc
506 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
507 We should make sure the type of 'index' is loff_t and then cast
508 explicitly as needed below. Do not assign an explicitly cast
512 Frequency : only for rhel5
514 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
515 Details : Modify the target file & which_kernel.
518 Frequency : if the uninit_groups feature is enabled on ldiskfs
520 Description: e2fsck reports "invalid unused inodes count"
521 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
522 feature and only a single inode is created in a group then the
523 "bg_unused_inodes" count is incorrectly updated. Creating a
524 second inode in that group would update it correctly.
529 Description: buffer overruns could theoretically occur
530 Details : llapi_semantic_traverse() modifies the "path" argument by
531 appending values to the end of the origin string, and a buffer
532 overrun may occur. Adding buffer overrun check in liblustreapi.
534 Severity : enhancement
536 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
539 Frequency : only if filesystem is inconsistent
541 Description: handle "serious error: objid * already exists" more gracefully
542 Details : If LAST_ID value on disk is smaller than the objects existing in
543 the O/0/d* directories, it indicates disk corruption and causes an
544 LBUG(). If the object is 0-length, then we should use the existing
545 object. This will help to avoid a full fsck in most cases.
547 Severity : enhancement
549 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
550 Details : Modify vm-tunables-rhel4.patch.
552 Severity : enhancement
554 Description: Kernel config for 2.6.18-vanilla.
555 Details : Modify targets/2.6-vanilla.target.in.
556 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
557 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
558 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
559 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
562 Frequency : occasional
564 Description: improve handling recoverable errors
565 Details : If request processed with error which can be recoverable on server
566 request should be resend, otherwise page released from cache and
572 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
573 Details : Modify target file & which_patch.
574 A flaw was found in the IA32 system call emulation provided
575 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
576 value could be stored in the %RAX register, which could trigger an
577 out-of-bounds system call table access. An untrusted local user
578 could exploit this flaw to run code in the kernel
579 (ie a root privilege escalation). (CVE-2007-4573).
583 Description: change order of libsysio includes
584 Details : '#include sysio.h' should always come before '#include xtio.h'
586 Severity : enhancement
588 Description: adapt the lustre_config script to support the upgrade case
589 Details : Add "-u" option for lustre_config script to support upgrading 1.4
590 server targets to 1.6 in parallel.
595 Description: To avoid grant space > avaible space when the disk is almost
596 full. Without this patch you might see the error "grant XXXX >
597 available" or some LBUG about grant, when the disk is almost
599 Details : In filter_check_grant, for non_grant cache write, we should
600 check the left space by if (*left > ungranted + bytes), instead
601 of (*left > ungranted), because only we are sure the left space
602 is enough for another "bytes", then the ungrant space should be
603 increase. In client, we should update cl_avail_grant only there is
604 OBD_MD_FLGRANT in the reply.
609 Description: Update RHEL 4 kernel to fix local root privilege escalation.
610 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
611 described in CVE-2007-4573. This problem could allow untrusted
612 local users to gain root access.
615 Frequency : when using O_DIRECT and quotas
617 Description: Incorrect file ownership on O_DIRECT output files
618 Details : block usage reported by 'lfs quota' does not take into account
619 files that have been written with O_DIRECT.
621 --------------------------------------------------------------------------------
623 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
625 * Support for kernels:
626 2.6.5-7.286 (SLES 9),
627 2.6.9-55.0.2.EL (RHEL 4),
628 2.6.16.46-0.14 (SLES 10),
629 2.6.18-8.1.8.el5 (RHEL 5),
630 2.6.18.8 vanilla (kernel.org)
631 * Client support for unpatched kernels:
632 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
633 2.6.16 - 2.6.21 vanilla (kernel.org)
634 * Due to recently discovered recovery problems, we do not recommend
635 using patchless RHEL 4 clients with this or any earlier release.
636 * Recommended e2fsprogs version: 1.40.2-cfs1
637 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
641 Description: Fix errors in lfs documentation
642 Details : Fixes man pages
644 Severity : enhancement
646 Description: Adaptive timeouts
647 Details : RPC timeouts adapt to changing server load and network
648 conditions to reduce resend attempts and improve recovery time.
650 Severity : enhancement
652 Description: llapi_file_create() does not allow some changes
653 Details : add llapi_file_open() that allows specifying the file creation
654 mode and open flags, and also returns an open file handle.
656 Severity : enhancement
658 Description: df doesn't work properly if diskfs blocksize != 4K
659 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
662 Frequency : always on directories with default striping set
664 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
665 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
667 Severity : enhancement
669 Description: merge and cleanup kernel patches.
670 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
672 Severity : enhancement
674 Description: RedHat Update kernel for RHEL5
675 Details : Kernel config file for RHEL5.
677 Severity : enhancement
679 Description: OSS needs mutliple precreate threads
680 Details : Add ability to start more than one create thread per OSS.
683 Frequency : only with quota on the root user
685 Description: mds_obd_create error creating tmp object
686 Details : When the user sets quota on root, llog will be affected and can't
687 create files and write files.
690 Frequency : Always on ia64 patchless client, and possibly others.
692 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
693 Details : This allows the patchless client to be loaded on architectures
699 Description: disorder of journal start and llog_add cause deadlock.
700 Details : in llog_origin_connect, journal start should happen before
701 llog_add keep the same order as other functions to avoid
704 Severity : enhancement
706 Description: RedHat Update kernel for RHEL5
707 Details : Modify the kernel config file more closer RHEL5.
709 Severity : enhancement
711 Description: Build failure against Centos5 (RHEL5)
712 Details : Define PAGE_SIZE when it isn't present.
715 Frequency : occasionally when using NFS
717 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
718 Details : with NFS, the anon dentry's parent was set to itself in
719 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
720 corresponding dentry other than use rec->ur_name.
722 Severity : enhancement
724 Description: enable data checksumming by default
725 Details : enable checksum by default, allow --disable-checksum
726 configure option and "-o nochecksum" mount option. Checksums
727 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
730 Frequency : Occasionally with failover
732 Description: Client eviction due to failover config
733 Details : after a connection loss, the lustre client should attempt to
734 reconnect to the last active server first before trying the
735 other potential connections.
737 Severity : enhancement
739 Description: client-side metadata stat-ahead during readdir(directory readahead)
740 Details : perform client-side metadata stat-ahead when the client detects
741 readdir and sequential stat of dir entries therein
746 Description: when mds and osts use different quota unit(32bit and 64bit),
747 quota will be released repeatly.
748 Details : void sending multiple quota reqs to mds, which will keep the
749 status between the reqs.
752 Frequency : only with liblustre clients on XT3
754 Description: evictions taking too long
755 Details : allow llrd to evict clients directly on OSTs
759 Description: osts not allocated evenly to files
760 Details : change the condition to increase offset_idx
762 Severity : enhancement
764 Description: self-adjustable client's lru lists
765 Details : use adaptive algorithm for managing client cached locks lru
766 lists according to current server load, other client's work
767 pattern, memory activities, etc. Both, server and client
768 side namespaces provide number of proc tunables for controlling
773 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
774 Details : rewrite inherited code (uuid parsing code from ext2 utils and
775 readdir code from ext3) from scratch preserving functionality.
779 Description: Only those disconnect error should be returned by rq_status.
780 Details : In open/enqueue processs, Some errors, which will cause client
781 disconnected, should be returned by rq_status, while other
782 errors should still be returned by intent, then mdc or llite will
785 Severity : enhancement
787 Description: Tune the kernel for good SCSI performance.
788 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
789 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
793 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
794 Bugzilla : 13547, 13627
795 Description: Data corruption for OSTs that are formatted larger than 2TB
797 Details : When generating the bio request for lustre file writes the
798 sector number would overflow a temporary variable before being
799 used for the IO. The data reads correctly from Lustre (which
800 will overflow in a similar manner) but other file data or
801 filesystem metadata may be corrupted in some cases.
805 Description: "lfs find -obd UUID" prints directories
806 Details : "lfs find -obd UUID" will return all directory names instead
807 of just file names. It is incorrect because the directories
808 do not reside on the OSTs.
811 Bugzilla : 13596, 13614
812 Description: MDS hang after unclean shutdown of lots of clients
813 Details : Never resend AST requests.
817 Frequency : Always, for kernels after 2.6.16
818 Description: Fix warning idr_remove called for id=.. which is not allocated.
819 Details : Last kernels save old s_dev before kill super and not allow
820 to restore from callback - restore it before call kill_anon_super.
824 Description: Fix error on 'ls .' at the top of the Lustre mount.
825 Details : Don't revalidate dentry if it is a root dentry.
827 --------------------------------------------------------------------------------
829 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
831 * Support for kernels:
832 2.6.5-7.286 (SLES 9),
833 2.6.9-55.0.2.EL (RHEL 4),
834 2.6.16.46-0.14 (SLES 10),
835 2.6.18-8.1.8.el5 (RHEL 5),
836 2.6.18.8 vanilla (kernel.org)
837 * Client support for unpatched kernels:
838 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
839 2.6.16 - 2.6.22 vanilla (kernel.org)
840 * Due to recently discovered recovery problems, we do not recommend
841 using patchless RHEL 4 clients with this or any earlier release.
842 * Recommended e2fsprogs version: 1.39.cfs8
843 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
848 Description: block reactivating mgc import until all deactivates complete
849 Details : Fix race when failing back MDT/MGS to itself (testing)
851 Severity : enhancement
853 Description: lfs setstripe enhancement
854 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
857 Frequency : mds/oss recovery
859 Description: llog ctxt is refrenced after it has been freed.
860 Details : llog ctxt refcount was added to avoide the race between ctxt free
861 and llog recovery process. Each llog user must hold ctxt refcount
862 before it access the llog. And the llog ctxt can only be freed
863 when its refcount is zero.
865 Severity : enhancement
867 Description: randomly memory allocation failure util
868 Details : Make lustre randomly failed allocating memory for testing purpose.
871 Frequency : only for SLES10
873 Description: Update kernel patch for SLES10 SP1
874 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
876 Severity : enhancement
878 Description: omit set fsid for export NFS
879 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
880 as FS_REQUIRES_DEV for avoid problems with generate fsid.
883 Frequency : after network failures
885 Description: Add sync option to mount_lustre.c
886 Details : Client loses data written to lustre after a network interruption.
888 Severity : enhancement
890 Description: Error message improvement.
891 Details : Merging of two LCONSOLE_ERROR_MSG into one.
893 Severity : enhancement
895 Description: don't use GFP_* in generic Lustre code.
896 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
898 Severity : enhancement
900 Description: obdclass is limited by single OBD_ALLOC(idarray)
901 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
903 Severity : enhancement
905 Description: updated patchess for new RHEL4 kernel
906 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
908 Severity : enhancement
910 Description: warnings with build patchless client with vanila 2.6.19 and up
911 Details : change old ctl_table style and replace ctl_table/ctl_table_header
912 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
914 Severity : enhancement
916 Description: O_DIRECT bypasses client statistics.
917 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
918 read_ahead_stats were updated. Copied stats section from
919 osc_send_oap_rpc() into async_internal().
922 Frequency : only for Cray XT3
924 Description: peer credits not enough on many OST per OSS systems.
925 Details : Use new lnet way to add credits as we need those for pings and ASTs
928 Frequency : only with liblustre
930 Description: Liblustre is not releasing flock locks on file close.
931 Details : Release flock locks on file close.
934 Frequency : only for RHEL4
936 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
937 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
938 Update target file and kernel config.
942 Description: lustre support for RHEL5
943 Details : Add support for RHEL5.
948 Description: ASSERTION(export != NULL) failed in target_handle_connect
949 Details : Assetion hit is result of rare race between disconnect and connet
950 to same nid. target_handle_connect found old connect cockie and
951 tried to reconnect, but can't find export for this cockie.
956 Description: umount blocks forever on error
957 Details : In result of wrong using obd_no_recov and obd_force flags client
958 can hand if cancel or some other requests is lost.
961 Frequency : Only for SLES
963 Description: sanity_quota fail test_1
964 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
965 file, which makes TSTID[2] inunique.
967 Severity : enhancement
969 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
970 Details : Update target/ChangeLog/which_patch .
972 Severity : enhancement
974 Description: jbd statistics
975 Details : Port older jbd statistics patch for sles10
980 Description: lvbo_init failed for resource with missing objects.
981 Details : Fix returning error if we do stat for file with missing/corrupted
982 objects and i_size set to all sum of size of all avaible objects.
983 if we to truncate/write to missing object - it is recreated.
988 Description: Oops in read and write path when failing to allocate lock.
989 Details : Check if lock allocation failed and return error back.
992 Frequency : When flocks are used.
994 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
995 Details : Flock locks might destroy just granted lock if it could be merged
996 with another existing flock, this is done in completion handler,
997 so teach ldlm_cli_enquque_fini that this is a valid case for
1003 Description: reply_lock_interpret crash due to race with it and lock cancel.
1004 Details : Do not replay locks that are being cancelled. Do not reference
1005 locks by their address during replay, just by their handle.
1008 Frequency : only with deactivated OSTs
1010 Description: lstripe command fails for valid OST index
1011 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
1012 instead of lov->desc.ld_active_tgt_count.
1014 Severity : enhancement
1016 Description: Build failure against Centos5 (RHEL5)
1017 Details : Use getpagesize() instead of PAGE_SIZE.
1019 --------------------------------------------------------------------------------
1021 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
1023 * Support for kernels:
1024 2.6.5-7.283 (SLES 9),
1025 2.6.9-55.EL (RHEL 4),
1026 2.6.16.46-0.14 (SLES 10),
1027 2.6.18.8 vanilla (kernel.org)
1028 * Client support for unpatched kernels:
1029 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1030 2.6.16 - 2.6.22 vanilla (kernel.org)
1031 * Due to recently discovered recovery problems, we do not recommend
1032 using patchless RHEL 4 clients with this or any earlier release.
1033 * Recommended e2fsprogs version: 1.39.cfs8
1034 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1035 * Starting with this release, the ldiskfs backing filesystem required
1036 by Lustre is now in its own package, lustre-ldiskfs. This package
1037 should be installed. It is versioned separately from Lustre and
1038 may be released separately in future.
1043 Description: server LBUG when shutting down
1044 Details : Block umount forever until the mount refcount is zero rather
1045 than giving up after an arbitrary timeout.
1047 Severity : enhancement
1049 Description: add optional extra BUILD_VERSION info
1050 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1051 to override the lustre version.
1054 Frequency : 2.6.18 servers only
1056 Description: ll_kern_mount() doesn't release the module reference
1057 Details : The ldiskfs module reference count never drops down to 0
1058 because ll_kern_mount() doesn't release the module reference.
1063 Description: server LBUG when using old ost_num_threads parameter
1064 Details : Accept the old ost_num_threads parameter but warn that it
1065 is deprecated, and fix an off-by-one error that caused an LBUG.
1070 Description: Transient SCSI error results in persistent IO issue
1071 Details : iobuf->dr_error is not reinitialized to 0 between two
1075 Frequency : sometimes when underlying device returns I/O errors
1077 Description: OSTs not going read-only during write failures
1078 Details : OSTs are not remounted read-only when the journal commit threads
1079 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1080 instead of the ext3 wrappers.
1083 Frequency : only with 10000 clients or more
1085 Description: poor connect scaling with increasing client count
1086 Details : Don't run filter_grant_sanity_check for more than 100 exports
1087 to improve scaling for large numbers of clients.
1090 Frequency : SLES10 only
1092 Description: sanity-quota.sh quotacheck failed: rc = -22
1093 Details : Quotas cannot be enabled on SLES10.
1096 Frequency : liblustre clients only
1098 Description: getdirentries does not give error when run on compute nodes
1099 Details : getdirentries does not fail when the size specified as an argument
1100 is too small to contain at least one entry
1102 Severity : enhancement
1104 Description: Add LNET router traceability for debug purposes
1105 Details : If a checksum failure occurs with a router as part of the
1106 IO path, the NID of the last router that forwarded the bulk data
1107 is printed so it can be identified.
1112 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1113 Details : Due to a race condition, liblustre clients were occasionally
1114 evicted incorrectly.
1116 Severity : enhancement
1118 Description: lfs setstripe use optional parameters instead of postional
1121 Severity : enhancement
1123 Description: Nanosecond timestamp support for ldiskfs
1124 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1125 resolution timestamps. There is not yet support for this at
1126 the Lustre filesystem level.
1129 Frequency : during server recovery
1131 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1132 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1136 Frequency : nfs export on patchless client
1138 Description: connectathon hang when test nfs export over patchless client
1139 Details : Disconnected dentry cannot be found with lookup, so we do not need
1140 to unhash it or make it invalid
1143 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1144 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1145 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1150 Description: open req refcounting wrong on reconnect
1151 Details : If reconnect happened between getting open reply from server and
1152 call to mdc_set_replay_data in ll_file_open, we will schedule
1153 replay for unreferenced request that we are about to free.
1154 Subsequent close will crash in variety of ways.
1155 Check that request is still eligible for replay in
1156 mdc_set_replay_data().
1161 Description: disable writes to filesystem when reading health_check file
1162 Details : the default for reading the health_check proc file has changed
1163 to NOT do a journal transaction and write to disk, because this
1164 can cause reads of the /proc file to hang and block HA state
1165 checking on a healthy but otherwise heavily loaded system. It
1166 is possible to return to the previous behaviour during configure
1167 with --enable-health-write.
1169 Severity : enhancement
1171 Description: 64-bit inode version
1172 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1173 to the inode. This will be required for version-based recovery.
1178 Description: MDS fails to start if a duplicate client export is detected
1179 Details : in some rare cases it was possible for a client to connect to
1180 an MDS multiple times. Upon recovery the MDS would detect this
1181 and fail during startup. Handle this more gracefully.
1183 Severity : enhancement
1185 Description: Add -o localflock option to simulate old noflock behaviour.
1186 Details : This will achieve local-only flock/fcntl locks coherentness.
1191 Description: log_commit_thread vs filter_destroy race leads to crash
1192 Details : Take import reference before releasing llog record semaphore
1197 Description: Wrong request locking in request set processing
1198 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1199 imp_delayed_list, in this place should be used imp_lock.
1202 Frequency : when reconnecting
1204 Description: Grant leak when OSC reconnect to OST
1205 Details : When osc reconnect ost, OST(filter) should check whether it
1206 should grant more space to client by comparing fed_grant and
1207 cl_avail_grant, and return the granted space to client instead
1208 of "new granted" space, because client will call osc_init_grant
1209 to update the client grant space info.
1212 Frequency : when client reconnects to OST
1214 Description: Grant leak when OSC does a resend and replays bulk write
1215 Details : When osc reconnect to OST, OST(filter) should clear grant info of
1216 bulk write request, because the grant info while be sync between
1217 OSC and OST when reconnect, and we should ignore the grant info
1218 these of resend/replay write req.
1223 Description: Grant space more than avaiable space sometimes.
1224 Details : When then OST is about to be full, if two bulk writing from
1225 different clients came to OST. Accord the avaliable space of the
1226 OST, the first req should be permitted, and the second one
1227 should be denied by ENOSPC. But if the seconde arrived before
1228 the first one is commited. The OST might wrongly permit second
1229 writing, which will cause grant space > avaiable space.
1232 Frequency : when client is evicted
1234 Description: Grant might be wrongly erased when osc is evicted by OST
1235 Details : when the import is evicted by server, it will fork another
1236 thread ptlrpc_invalidate_import_thread to invalidate the
1237 import, where the grant will be set to 0. While the original
1238 thread will update the grant it got when connecting. So if
1239 the former happened latter, the grant will be wrongly errased
1240 because of this race.
1245 Description: Checking Stale with correct fid
1246 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1247 to check whether it is stale, because sometimes, we want the
1248 enqueue happened anyway, and op_data.fid2 will not be initialized.
1250 Severity : enhancement
1252 Description: update patchless client
1253 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1256 Frequency : only with 2.4 kernel
1258 Description: random memory corruption
1259 Details : size of struct ll_inode_info is to big for union inode.u and this
1260 can be cause of random memory corruption.
1265 Description: Memory leak in recovery
1266 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1267 It should also check obd_fail before fsfilt_start, otherwise if
1268 fsfilt_start return -EROFS,(failover mds during mds recovery).
1269 then the req will return with repmsg->transno = 0 and rc = EROFS.
1270 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1271 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1272 be freed no matter whether fsfilt_commit success or not.
1275 Frequency : only with huge numbers of clients
1277 Description: Prevents from taking the superblock lock in llap_from_page for
1279 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1280 for llap_from_page call in ll_removepage() prevents from taking
1281 the superblock lock for a soon died page.
1286 Description: Not check open intent error before release open handle
1287 Details : in some rare cases, the open intent error is not checked before
1288 release open handle, which may cause
1289 ASSERTION(open_req->rq_transno != 0), because it tries to release
1290 the failed open handle.
1295 Description: Set cat log bitmap only after create log success.
1296 Details : in some rare cases, the cat log bitmap is set too early. and it
1297 should be set only after create log success.
1301 Description: Accessing a block bevice can re-enable I/O when Lustre is
1302 tearing down a device.
1303 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1308 Description: service threads may hog cpus when there are a lot of requests
1309 Details : Insert cond_resched to give other threads a chance to use some CPU
1314 Description: the cat log was not initialized in recovery
1315 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1316 unlink log on mds will not be initialized until mds post
1317 recovery. And also in mds post recovery, the unlink log will
1318 initialization will be done asynchronausly, so there will be race
1319 between add unlink log and unlink log initialization.
1323 Description: brw_stats were being printed incorrectly
1324 Details : brw_stats were being printed as log2 but all of them were not
1325 recorded as log2. Also remove some code duplication arising from
1326 filter_tally_{read,write}.
1330 Frequency : rare, only in recovery.
1331 Description: ASSERTION(req->rq_type != LI_POISON) failed
1332 Details : imp_lock should be held while iterating over imp_sending_list for
1333 prevent destroy request after get timeout in ptlrpc_queue_wait.
1337 Description: replay-single.sh test 52 fails
1338 Details : A lock's skiplist need to be cleanup when it being unlinked
1339 from its resource list.
1344 Description: Short directio read returns full requested size rather than
1346 Details : Direct I/O operations should return actual amount of bytes
1347 transferred rather than requested size.
1349 Severity : enhancement
1351 Description: metadata RPC reduction (e.g. for rm performance)
1352 Details : decrease the amount of synchronous RPC between clients and servers
1353 by canceling conflicing lock before the operation on the client
1354 and packing thier handles into the main operation RPC to server.
1356 Severity : enhancement
1358 Description: add #ifdef HAVE_KERNEL_CONFIG_H
1359 Details : kernels from 2.6.19 not need include linux/config.h, but add
1360 include linux/autoconf.h in commpiler command line.
1362 Severity : enhancement
1364 Description: patchless client support for 2.6.22 kernel
1365 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
1366 removed. In this case we need drop using os depended interface to
1367 kmem_cache and use cfs_mem_cache API.
1372 Description: fix mal-formatted messages
1373 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
1377 Frequency : always in liblustre
1378 Description: wrong IS_ERR implementation in liblustre.h
1379 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
1384 Description: Correct condition for output debug message.
1385 Details : inode i_nlink equal zero is not enough for output message about
1386 disk corruption, i_ctime and i_mode should be also checked.
1390 Frequency : always in patchless client
1391 Description: add configure check for truncate_complete_page
1392 Details : improve checks for exported symbols. This allow run check without
1393 sources, but with Module.symvers shipped with kernel distribution.
1394 add check for truncate_complete_page used by patchless client.
1398 Description: sanity.sh test_77h fails with "test_77h file compare failed"
1399 Details : test_77h uses a file which was messed by other test case.
1403 Description: Not Check whether lov_tgts is NULL in some lov functions
1404 Details : Checking whether lov_tgts is NULL in some functions.
1408 Description: replace obdo_alloc() with OBDO_ALLOC macro
1409 Details : nothing special is done in obdo_alloc() function, and for
1410 debugging purpose, it needs to be replaced with macros.
1414 Description: bad return value and errno from fcntl call
1415 Details : In liblustre API, errno should be a negative value if error
1420 Description: ptlrpc_check_set() LBUG
1421 Details : In case of positive reply from server and failed client bulk
1422 callback after bulk transfer shouldn't LBUG, but process this
1423 request as erroneous.
1425 Severity : enhancement
1427 Description: VFS operations stats tool.
1428 Details : Tool which collects stats by tracking value written in pid,
1429 ppid, gid and uses llstat to generate output to plot graph using
1431 Updated lustre/utils/Makefile.am
1432 Added lustre/utils/ltrack_stats.c
1434 Severity : enhancement
1436 Description: 2.6.18 server support (lustre 1.6.1)
1437 Details : Support for 2.6.18 kernels on the server side.
1442 Description: ASSERTION(imp->imp_conn_current) failed
1443 Details : an assertion failure is hit if a client node boots and attempts to
1444 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
1447 Frequency : only for i686
1449 Description: 1.4.11 RC1 build fails for RHEL 4, i686
1450 Details : Fixed config variable for build.
1455 Description: Updated patchess for new RHEL4 kernel
1456 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
1457 Updated patch jbd-stats-2.6.9.patch
1458 Updated patch qsnet-rhel4-2.6.patch
1459 Updated patch quota-deadlock-on-pagelock-core.patch
1460 Updated patch vfs_intent-2.6-rhel4.patch
1461 Updated patch vfs_races-2.6-rhel4.patch
1462 Updated series file 2.6-rhel4-titech.series
1463 Updated series file 2.6-rhel4.series
1464 Updated kernel config files
1469 Description: lquota slave complains LBUG when reconnecting with mds
1471 Details : quota slave depends qctxt->lqc_import to send its quota request.
1472 This pointer will be invalid if mds did failover or broke its
1473 connect to osts, which leads to LBUG.
1476 Frequency : when qunit size is too small(less than 20M)
1478 Description: write is stopped by improper -EDQUOT
1479 Details : If the master is busy and qunit size is small enough(let's say 1M),
1480 the slave can not get quota from master on time, which will lead
1481 slave to trigger a -EQUOTA to client.
1486 Description: Deadlock during metadata tests
1487 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
1488 called with the per-dentry lock held.
1491 Frequency : SLES9 only
1493 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
1494 Details : set CONFIG_NR_CPUS to 128 instead of 8.
1496 Severity : enhancement
1498 Description: remove fs_prep_san_write operation and related patches
1499 Details : remove the ext3-san-jdike patches which are no longer useful.
1504 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
1505 Details : a disk corruption can cause the mballoc code to assert on a
1506 double free or other extent corruptions. Handle these with
1507 ext3_error() instead of assertions.
1510 Frequency : only with mballoc3 code and deep extent trees
1512 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
1513 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
1514 can cause the filesystem to be remounted read-only.
1519 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
1520 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
1521 few external users of this ioctl number is lfsck's call to
1522 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
1523 Add the old ioctl number to the handler so both old and new
1528 Description: parallel lock callbacks
1529 Details : Instead of sending blocking and completion callbacks as separated
1530 requests, adding them to a set and sending in parallel.
1534 Description: Disable most debugging by default
1535 Details : To improve performance, disable most logging (for debug purposes)
1536 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
1537 default, and HA includes fewer messages.
1541 Frequency : only run on patchless client.
1542 Description: use do_facet on sanity.sh for test handling recoverables errors
1543 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1547 Description: hash tables for lists of nids, connections and uuids
1548 Details : Hash tables noticeably help when a lot of clients connect to a
1549 server, to faster identify duplicate connections or reconnects,
1550 also to faster find export to evict in manual eviction case.
1554 Description: Sometimes, when the server evict a client, and the client will
1555 not be evicted as soon as possible.
1556 Details : In enqueue req, the error was returned by intent, instead of
1557 rq_status which make ptlrpc layer not detect this error, and
1558 does not evict the client. So enqueue error should be returned
1562 Frequency : only at startup
1564 Description: Delay client connections to MDT until fist MDT->OST connect
1565 Details : If a client tried to create a new file before the MDT had
1566 connected to any OSTs, the create would return EIO. Now
1567 the client will simply block until the MDT connects to the
1568 first OST and the create can succeed.
1571 Frequency : only for SLES9
1573 Description: Routinely utilize latest Quadrics drivers in CFS releases
1574 Details : Update patch qsnet-suse-2.6.patch.
1577 Frequency : only for sles10
1579 Description: Update patches for SLES 10 SP1 kernel.
1580 Details : Update the patch vfs_intent-2.6-sles10.patch.
1585 Description: Routinely utilize latest Quadrics drivers in CFS releases
1586 Details : Update patch qsnet-rhel4-2.6.patch.
1589 Frequency : at statup only
1591 Description: mds_lov_synchronize race leads to various problems
1592 Details : simultaneous MDT->OST connections at startup can cause the
1593 sync to abort, leaving the OSC in a bad state.
1598 Description: Using wrong pointer in osc_brw_prep_request
1599 Details : Access to array[-1] can produce panic if kernel compiled with
1600 CONFIG_PAGE_ALLOC enabled
1602 Severity : enhancement
1604 Description: Async OSC create to avoid the blocking unnecessarily.
1605 Details : If a OST has no remain object, system will block on the creating
1606 when need to create a new object on this OST. Now, ways use
1607 pre-created objects when available, instead of blocking on an
1608 empty osc while others are not empty. If we must block, we block
1609 for the shortest possible period of time.
1613 Frequency : only in recovery
1614 Description: Mark OST as early accessible if his start SYNC.
1615 Details : osc_precreate return flag early accessible if oscc marked as
1616 OSCC_FLAG_SYNC_IN_PROGRESS.
1621 Description: Sometimes precreate code can triger create object on wrong ost
1622 Details : Wrong protected or not not restored variables aftre precreate loop
1623 can produce creation object on wrong ost.
1626 Frequency : oss recovery
1628 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1630 Details : llog_commit_thread_count should be synced between llog_commit
1631 start and cleanup, so new llog_commit thread should not be started
1632 when llog_commit threads being stopped to avoid accessing some
1635 Severity : enhancement
1637 Description: Add printing inode info into message about error in writepage.
1639 --------------------------------------------------------------------------------
1641 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1646 Frequency : on some architectures
1648 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1649 Details : Uninitialized flags sometimes cause configuration commands to
1653 Frequency : patchless clients only
1655 Description: missing __iget() symbol export
1656 Details : The __iget() symbol export is missing. To avoid the need for
1657 this on patchless clients the deathrow inode reaper is turned
1658 off, and we depend on the VM to clean up old inodes. This
1659 dependency was during via the fix for bug 12181.
1664 Description: sanity.sh fail: test_52b
1665 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1666 incorrect inode's flags to client.
1668 --------------------------------------------------------------------------------
1670 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1672 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1673 INTEROPERATE with older versions automatically. In many cases a
1674 special upgrade step is needed. Please read the
1675 user documentation before upgrading any part of a 1.4.x system.
1676 * WARNING: Lustre configuration and startup changes are required with
1677 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1679 * Support for kernels:
1680 2.4.21-47.0.1.EL (RHEL 3),
1681 2.6.5-7.283 (SLES 9),
1682 2.6.9-42.0.10.EL (RHEL 4),
1683 2.6.12.6 vanilla (kernel.org),
1684 2.6.16.27-0.9 (SLES10)
1685 * Client support for unpatched kernels:
1686 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1687 2.6.16 - 2.6.19 vanilla (kernel.org),
1688 2.6.9-42.0.8EL (RHEL 4)
1689 * Recommended e2fsprogs version: 1.39.cfs6
1690 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1693 Severity : enhancement
1695 Description: Async OSC create to avoid the blocking unnecessarily.
1696 Details : If a OST has no remain object, system will block on the creating
1697 when need to create a new object on this OST. Now, ways use
1698 pre-created objects when available, instead of blocking on an
1699 empty osc while others are not empty. If we must block, we block
1700 for the shortest possible period of time.
1702 Severity : enhancement
1704 Description: MountConf
1705 Details : Lustre configuration is now managed via mkfs and mount
1706 commands instead of lmc and lconf. New obd types (MGS, MGC)
1707 are added for dynamic configuration management. See
1708 https://mail.clusterfs.com/wikis/lustre/MountConf for
1711 Severity : enhancement
1713 Description: dynamic OST addition
1714 Details : OSTs can now be added to a live filesystem
1716 Severity : enhancement
1718 Description: startup order invariance
1719 Details : MDTs and OSTs can be started in any order. Clients only
1720 require the MDT to complete startup.
1722 Severity : enhancement
1724 Description: parallel, asynchronous orphan cleanup
1725 Details : orphan cleanup is now performed in separate threads for each
1726 OST, allowing parallel non-blocking operation.
1728 Severity : enhancement
1730 Description: optimized stripe assignment
1731 Details : stripe assignments are now made based on ost space available,
1732 ost previous usage, and OSS previous usage, in order to try
1733 to optimize storage space and networking resources.
1735 Severity : enhancement
1737 Description: Permanently set tunables
1738 Details : All writable /proc/fs/lustre tunables can now be permanently
1739 set on a per-server basis, at mkfs time or on a live
1742 Severity : enhancement
1744 Description: Lustre message v2
1745 Details : Add lustre message format v2.
1747 Severity : enhancement
1749 Description: client OST exclusion list
1750 Details : Clients can be started with a list of OSTs that should be
1751 declared "inactive" for known non-responsive OSTs.
1755 Description: ENOENT returned for valid filehandle during dbench.
1756 Details : Check if a directory has children when invalidating dentries
1757 associated with an inode during lock cancellation. This fixes
1758 an incorrect ENOENT sometimes seen for valid filehandles during
1759 testing with dbench.
1762 Frequency : SFS test only (otherwise harmless)
1764 Description: SPEC SFS validation failure on NFS v2 over lustre.
1765 Details : Changes the blocksize for regular files to be 2x RPC size,
1766 and not depend on stripe size.
1768 Severity : enhancement
1770 Description: fine-grained SMP locking inside DLM
1771 Details : Improve DLM performance on SMP systems by removing the single
1772 per-namespace lock and replace it with per-resource locks.
1774 Severity : enhancement
1776 Description: don't hold multiple extent locks at one time
1777 Details : To avoid client eviction during large writes, locks are not
1778 held on multiple stripes at one time or for very large writes.
1779 Otherwise, clients can block waiting for a lock on a failed OST
1780 while holding locks on other OSTs and be evicted.
1782 Severity : enhancement
1784 Description: Multiple MD RPCs in flight.
1785 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1786 To avoid overly-overloading MDT, introduce a limit on number of
1787 MDT RPCs in flight for a single client and add /proc controls
1788 to adjust this limit.
1790 Severity : enhancement
1792 Description: client read/write statistics
1793 Details : Add client read/write call usage stats for performance
1794 analysis of user processes.
1795 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1796 file access. extents_stats shows chunk size distribution.
1797 extents_stats_per_process show chunk size distribution per
1800 Severity : enhancement
1802 Description: per-client statistics on server
1803 Details : Add ldlm and operations statistics for each client in
1804 /proc/fs/lustre/mds|obdfilter/*/exports/
1806 Severity : enhancement
1808 Description: improved MDT statistics
1809 Details : Add detailed MDT operations statistics in
1810 /proc/fs/lustre/mds/*/stats
1812 Severity : enhancement
1814 Description: VFS operations stats
1815 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1816 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1821 Description: Fix client-side osc byte counters
1822 Details : The osc read/write byte counters in
1823 /proc/fs/lustre/osc/*/stats are now working
1826 Frequency : always as root on SLES
1828 Description: Failure of copying files with lustre special EAs.
1829 Details : Client side always return success for setxattr call for lustre
1830 special xattr (currently only "trusted.lov").
1835 Description: Refcount LNET uuids
1836 Details : The global LNET uuid list grew linearly with every startup;
1837 refcount repeated list entries instead of always adding to
1840 Severity : enhancement
1842 Description: Dynamic service threads
1843 Details : Within a small range, start extra service threads
1844 automatically when the request queue builds up.
1847 Frequency : mixed-endian client/server environments
1849 Description: mixed-endian crashes
1850 Details : The new msg_v2 system had some failures in mixed-endian
1853 Severity : enhancement
1855 Description: Easy OST removal
1856 Details : OSTs can be permanently deactivated with e.g. 'lctl
1857 conf_param lustre-OST0001.osc.active=0'
1859 Severity : enhancement
1861 Description: MGS proc entries
1862 Details : Added basic proc entries for the MGS showing what filesystems
1865 Severity : enhancement
1867 Description: provide MGS failover
1868 Details : Added config lock reacquisition after MGS server failover.
1870 Severity : enhancement
1872 Description: add Linux 2.4 support
1873 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1877 Description: a large application tries to do I/O to the same resource and dies
1878 in the middle of it.
1879 Details : Check the req->rq_arrival time after the call to
1880 ost_brw_lock_get(), but before we do anything about
1881 processing it & sending the BULK transfer request. This
1882 should help move old stale pending locks off the queue as
1883 quickly as obd_timeout.
1886 Frequency : when an incorrect nid is specified during startup
1888 Description: ptlrpc connect to non-existant node causes kernel crash
1889 Details : LNET can't be re-entered from an event callback, which
1890 happened when we expire a message after the export has been
1891 cleaned up. Instead, hand the zombie cleanup off to another
1894 Severity : enhancement
1896 Description: plain/inodebits lock performance improvement
1897 Details : Grouping plain/inodebits in granted list by their request modes
1898 and bits policy, thus improving the performance of search through
1902 Frequency : only if OST filesystem is corrupted
1904 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1905 Details : for a short time RPCs with bulk IO are in the replay list,
1906 but replay of bulk IOs is unimplemented. If the OST filesystem
1907 is corrupted due to disk cache incoherency and then replay is
1908 started it is possible to trip an assertion. Avoid putting
1909 committed RPCs into the replay list at all to avoid this issue.
1912 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1914 Description: System hang on startup
1915 Details : This bug allowed the liblustre (e.g. catamount) client to
1916 return to the app before handling all startup RPCs. This
1917 could leave the node unresponsive to lustre network traffic
1918 and manifested as a server ptllnd timeout.
1920 Severity : enhancement
1922 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1923 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1924 boolean to control whether to print peer debug info when a
1925 client's RPC times out.
1928 Frequency : only for kernels with patches from Lustre below 1.4.3
1930 Description: Remove old rdonly API
1931 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1934 Frequency : only for devices with external journals
1936 Description: Set external device read-only also
1937 Details : During a commanded failover stop, we set the disk device
1938 read-only while the server shuts down. We now also set any
1939 external journal device read-only at the same time.
1942 Frequency : when upgrading from 1.4 while trying to change parameters
1944 Description: The wrong (new) MDC name was used when setting parameters for
1945 upgraded MDT's. Also allows changing of OSC (and MDC)
1946 parameters if --writeconf is specified at tunefs upgrade time.
1949 Frequency : when setting specific ost indicies
1951 Description: QOS code breaks on skipped indicies
1952 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1953 created with --index need not be sequential.
1955 Severity : enhancement
1957 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1958 Details : The uninit_groups feature works in conjunction with the kernel
1959 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1960 up the pass1 processing of e2fsck. This is a read-only feature
1961 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1962 mount filesystems that have had this feature enabled.
1964 Severity : enhancement
1966 Description: Improve multi-block allocation algorithm to avoid fragmentation
1967 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1968 allocation locality and avoid filesystem fragmentation.
1973 Description: lfs quota fails with deactivated OSTS
1974 Details : With this patch, three improvements are included:
1975 1. detete the softlimit in mds and osts when use "lfs quota".
1976 2. display the inaccurate data in the output of "lfs quota".
1977 3. try to get quota info when "lfs quota" is executed.
1979 ------------------------------------------------------------------------------
1981 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1983 * Support for kernels:
1984 2.4.21-47.0.1.EL (RHEL 3)
1985 2.6.5-7.283 (SLES 9)
1986 2.6.9-42.0.10.EL (RHEL 4)
1987 2.6.12.6 vanilla (kernel.org)
1988 2.6.16.27-0.9 (SLES 10)
1989 * Recommended e2fsprogs version: 1.39.cfs5
1991 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1995 Frequency : occasional, depends on client load and configuration
1996 Bugzilla : 12181, 12203
1997 Description: data loss for recently-modified files
1999 Details : In some cases it is possible that recently written or created
2000 files may not be written to disk in a timely manner (this should
2001 normally be within 30s unless client IO load is very high).
2002 The problem appears as zero-length files or files that are a
2003 multiple of 1MB in size after a client crash or client eviction
2004 that are missing data at the end of the file.
2006 This problem is more likely to be hit on clients where files are
2007 repeatedly created and unlinked in the same directory, clients
2008 have a large amount of RAM, have many CPUs, the filesystem has
2009 many OSTs, the clients are rebooted frequently, and/or the files
2010 are not accessed by other nodes after being written.
2012 The presence of the problem can be detected by looking at
2013 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2014 smaller than the second (nr_unused) then dirty files will not
2015 be flushed automatically to disk. "sync; sleep 10" should be
2016 run several times on the node before unmounting it to update
2017 Lustre (this is also safe to run on nodes without this problem).
2019 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2020 that can cause this same problem, so customers using that kernel
2021 also need to update the kernel in addition to Lustre. In order
2022 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2024 It is normal that files written just before a client crash (less
2025 than 30s) may not yet have been flushed to disk, even for local
2029 Frequency : frequent on thin XT3 nodes
2031 Description: UUID collision on thin XT3 Linux nodes
2032 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2033 randomly, since we relied on an insufficiently-seeded PRNG.
2038 Description: OSS hangs after "All ost request buffers busy"
2039 Details : A deadlock between quota and journal operations caused OSS
2040 hangs after printing "All ost request buffers busy."
2043 Frequency : always on liblustre builds
2045 Description: Cleanup compiler warnings on liblustre
2048 Frequency : always on liblustre builds on XT3
2050 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2054 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2055 > 32000 subdirectories
2056 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2057 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2058 subdirectory count crosses 32000. This will aid e2fsck to
2059 correctly handle more than 32000 subdirectories.
2062 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2064 Description: System hang on startup
2065 Details : This bug allowed the liblustre (e.g. catamount) client to
2066 return to the app before handling all startup RPCs. This
2067 could leave the node unresponsive to lustre network traffic
2068 and manifested as a server ptllnd timeout.
2070 Severity : enhancement
2072 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2073 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2074 boolean to control whether to print peer debug info when a
2075 client's RPC times out.
2080 Description: make O_SYNC working on 2.6 kernels
2081 Details : 2.6 kernels use different method for mark pages for write,
2082 so need add a code to lustre for O_SYNC work.
2087 Description: Failure to close file and release space on NFS
2088 Details : Put inode details into lock acquired in ll_intent_file_open.
2089 Use mdc_intent_lock in ll_intent_open to properly
2090 detect all kind of errors unhandled by mdc_enqueue.
2095 Description: proc file read during shutdown sometimes raced obd removal,
2097 Details : Add lock to prevent obd access after proc file removal.
2100 Frequency : Only for files larger than 4GB on 32-bit clients.
2102 Description: improperly doing page alignment of locks
2103 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2104 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2109 Description: RPCs being resent when they shouldn't be
2110 Details : Some RPCs that should not be resent are being resent. This
2111 can cause inconsistencies in the RPC state machine. Do not
2112 resend such requests.
2115 Frequency : rare, only with NFS export
2117 Description: Crash on NFS re-export node
2118 Details : under very unusual load conditions an assertion is hit in
2119 ll_intent_file_open()
2122 Frequency : only if OST filesystem is corrupted
2124 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2125 Details : for a short time RPCs with bulk IO are in the replay list,
2126 but replay of bulk IOs is unimplemented. If the OST filesystem
2127 is corrupted due to disk cache incoherency and then replay is
2128 started it is possible to trip an assertion. Avoid putting
2129 committed RPCs into the replay list at all to avoid this issue.
2134 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2135 Details : Large single O_DIRECT read and write calls can fail to allocate
2136 a sufficiently large buffer to process the request. In case of
2137 allocation failure the allocation is retried with a smaller
2138 buffer and broken into smaller requests.
2140 Severity : enhancement
2142 Description: Add -o localflock option to simulate old noflock behaviour.
2143 Details : This will achieve local-only flock/fcntl locks coherentness.
2148 Description: versioning check is incomplete
2149 Details : Checking the version difference of client vs. server, report
2150 error if the gap is too big.
2155 Description: add support PG_writeback bit
2156 Details : add support for PG_writeback bit for Lustre, for more carefull
2157 work with page cache in 2.6 kernel. This also fix some deadlocks
2158 and remove hack for work O_SYNC with 2.6 kernel.
2160 Severity : enhancement
2162 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2163 Details : The uninit_groups feature works in conjunction with the kernel
2164 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2165 up the pass1 processing of e2fsck. This is a read-only feature
2166 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2167 mount filesystems that have had this feature enabled.
2169 Severity : enhancement
2171 Description: Improve multi-block allocation algorithm to avoid fragmentation
2172 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2173 allocation locality and avoid filesystem fragmentation.
2175 ------------------------------------------------------------------------------
2177 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2179 * Support for kernels:
2180 2.6.9-42.0.3.EL (RHEL 4)
2181 2.6.5-7.276 (SLES 9)
2182 2.4.21-47.0.1.EL (RHEL 3)
2183 2.6.12.6 vanilla (kernel.org)
2184 2.6.16.21-0.8 (SLES10)
2185 * Recommended e2fsprogs version: 1.39.cfs2-0
2187 * The backwards-compatible /proc/sys/portals symlink has been removed
2188 in this release. Before upgrading, please ensure that you change
2189 any configuration scripts or /etc/sysctl.conf files that access
2190 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2191 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2192 in advance of the upgrade on any system running Lustre 1.4.6 or
2193 newer, since /proc/sys/lnet was added in that version.
2194 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2198 Frequency : only when quota is used
2200 Description: avoid scanning export list for quota master
2201 Details : Change the algorithms to avoid scanning export list in order
2202 to improve the efficiency.
2205 Frequency : MDS failover only, very rarely
2207 Description: "went back in time" messages on mds failover
2208 Details : The greatest transno may be lost when the current operation
2209 finishes with an error (transno==0) and the client's last_rcvd
2210 record is over-written. Save the greatest transno in the
2211 mds_last_transno for this case.
2214 Frequency : always for specific kernels and striping counts
2216 Description: client may get "Matching packet too big" without ACL support
2217 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2218 when trying to access files in certain configurations. The
2219 clients should in fact be denied when mounting because they do
2220 not understand ACLs.
2223 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2225 Description: many clients connecting with IO in progress causes connect timeouts
2226 Details : Avoid synchronous journal commits to avoid delays caused by many
2227 clients connecting/disconnecting when bulk IO is in progress.
2228 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2229 OST_IO_PORTAL to avoid delays behind potentially many pending
2233 Frequency : occasionally with multiple writers to a single file
2235 Description: shared writes to file may result in wrong size reported by stat()
2236 Details : Allow growing of kms when extent lock is cancelled
2239 Frequency : always with random mmap IO to multi-striped file
2241 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2242 Details : Only if the hole is at the end of OST object so that kms is too
2243 small. Fix is to increase kms accordingly in ll_nopage.
2246 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2248 Description: writes to a missing object would leak memory on the OST
2249 Details : If there is an inconsistency between the MDS and OST filesystems,
2250 such that the MDS references an object that doesn't exist, writes
2251 to that object will leak memory due to incorrect cleanup in the
2252 error handling path, eventually running out of memory on the OST.
2257 Description: Creating too long symlink causes lustre errors
2258 Details : Check symlink and name lengths before sending requests to MDS.
2261 Frequency : only if flock is enabled (not on by default)
2263 Description: posix locks not released on fd closure on 2.6.9+
2264 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2265 kernels, this caused such locks not to be released on fd close and
2266 then assertions on fs unmount about still used locks.
2269 Frequency : MDS failover only, very rarely
2271 Description: clients may get ASSERTION(granted_lock != NULL)
2272 Details : When request was taking a long time, and a client was resending
2273 a getattr by name lock request. The were multiple lock requests
2274 with the same client lock handle and
2275 mds_getattr_name->fixup_handle_for_resent_request found one of the
2276 lock handles but later failed with ASSERTION(granted_lock != NULL).
2281 Description: handle->h_buffer_credits > 0, assertion failure
2282 Details : h_buffer_credits is zero after truncate, causing assertion
2283 failure. This patch extends the transaction or creates a new
2287 Frequency : NFS re-export or patchless client
2288 Bugzilla : 11179, 10796
2289 Description: Crash on NFS re-export node (__d_move)
2290 Details : We do not want to hash the dentry if we don't have a lock.
2291 But if this dentry is later used in d_move, we'd hit uninitialised
2292 list head d_hash, so we just do this to init d_hash field but
2293 leave dentry unhashed.
2296 Frequency : NFS re-export or patchless client
2298 Description: NFS exports has problem with symbolic link
2299 Details : lustre client didn't properly install dentry when re-exported
2300 to NFS or running patchless client.
2303 Frequency : NFS re-export or patchless client
2305 Description: Various nfs/patchless fixes.
2306 Details : fixes reuse disconected alias for lookup process - this fixes
2307 warning "find_exported_dentry: npd != pd",
2308 fix permission error with open files at nfs.
2309 fix apply umask when do revalidate.
2312 Frequency : occasional
2314 Description: Crash on NFS re-export node
2315 Details : calling clear_page() on the wrong pointer triggered oops in
2316 generic_mapping_read().
2319 Frequency : rarely, using O_DIRECT IO
2321 Description: unaligned directio crashes client with LASSERT
2322 Details : check for unaligned buffers before trying any requests.
2325 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2327 Description: stale data returned from RAID cache
2328 Details : If only a small amount of IO is done to the RAID device before
2329 reading it again it is possible to get stale data from the RAID
2330 cache instead of reading it from disk.
2333 Frequency : always for sles10 kernel
2335 Description: sles10 support
2336 Details : ll_follow_link: compile fixes and using of nd_set_link
2337 under newer kernels.
2340 Frequency : depends on arch, kernel and compiler version, always on sles10
2343 Description: recursive or deep enough symlinks cause stack overflow
2344 Details : getting rid of large stack-allocated variable in
2348 Frequency : depends on hardware
2350 Description: lustre write performance loss in the SLES10 kernel
2351 Details : the performance loss is caused by using of write barriers in the
2352 ext3 code. The SLES10 kernel turns barrier support on by
2353 default. The fix is to undo that change for ldiskfs.
2355 ------------------------------------------------------------------------------
2357 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2359 * Support for kernels:
2360 2.6.9-42.0.3EL (RHEL 4)
2361 2.6.5-7.276 (SLES 9)
2362 2.4.21-47.0.1.EL (RHEL 3)
2363 2.6.12.6 vanilla (kernel.org)
2367 Frequency : quota enabled and large files being deleted
2369 Description: releasing more than 4GB of quota at once hangs OST
2370 Details : If a user deletes more than 4GB of files on a single OST it
2371 will cause the OST to spin in an infinite loop. Release
2372 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2377 Description: statfs data retrieved from /proc may be stale or zero
2378 Details : When reading per-device statfs data from /proc, in the
2379 {kbytes,files}_{total,free,avail} files, it may appear
2380 as zero or be out of date.
2383 Frequency : systems with MD RAID1 external journal devices
2385 Description: lconf's call to blkid is confused by RAID1 journal devices
2386 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2387 of returning all block devices that match the journal UUID.
2390 Frequency : always, for aggregate stripe size over 4GB
2392 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2393 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2394 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2397 Frequency : always if "lfs find" used on a local file/directory
2399 Description: "lfs find" segfaults if used on a local file/directory
2400 Details : The case where a directory component was not specified wasn't
2401 handled correctly. Handle this properly.
2404 Frequency : always on ppc64
2406 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2407 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2408 perform endianness conversion.
2411 Frequency : rarely (truncate to non-zero file size after write under load)
2412 Bugzilla : 10730, 10687
2413 Description: Files padded with zeros to next 4K multiple
2414 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2415 it is possible that files that are truncated to a non-zero size
2416 immediately after being written are filled with zero bytes beyond
2417 the truncated size. No file data is lost.
2419 Severity : enhancement
2421 Description: Allow recovery/failover for liblustre clients.
2422 Details : liblustre clients were unaware of failover configurations until
2425 Severity : enhancement
2427 Description: user file locks should fail when not mounting with flock option
2428 Details : Set up an error-returning stub in ll_file_operations.lock field
2429 to prevent incorrect behaviour when client is mounted without
2430 flock option. Also, set up properly f_op->flock field for
2434 Frequency : always on ia64
2436 Description: "lfs df" loops on printing out MDS statfs information
2437 Details : The obd_ioctl_data was not initialized and in some systems
2438 this caused a failure during the ioctl that did not return
2439 an error. Initialize the struct and return an error on failure.
2442 Frequency : SLES 9 only
2444 Description: Error of copying files with lustre special EAs as root
2445 Details : Client side always return success for setxattr call for lustre
2446 special xattr (currently only "trusted.lov").
2449 Frequency : rarely on clusters with both ia64+i386 clients
2451 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2452 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2453 clients are doing concurrent writes to the same file it is
2454 possible that the ia64 clients may LASSERT because the OST
2455 extent locks are not PAGE_SIZE aligned. Ensure that grown
2456 locks are always aligned on the request boundary.
2459 Frequency : specific use, occasional
2461 Description: Overwriting in use executable truncates on-disk binary image
2462 Details : If one node attempts to overwrite an executable in use by
2463 another node, we now correctly return ETXTBSY instead of
2464 truncating the file.
2466 Severity : enhancement
2468 Description: Async OSC create to avoid the blocking unnecessarily.
2469 Details : If a OST has no remain object, system will block on the creating
2470 when need to create a new object on this OST. Now, ways use
2471 pre-created objects when available, instead of blocking on an
2472 empty osc while others are not empty. If we must block, we block
2473 for the shortest possible period of time.
2478 Description: chmod on Lustre root is propagated to other clients
2479 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2480 invalid by the follow_mount time.
2483 Frequency : liblustre clients only
2485 Description: Race in 'instant cancel' lock handling could lead to such locks
2486 never to be granted in case of SMP MDS
2487 Details : Do not destroy not yet granted but cbpending locks in
2491 Frequency : replay/resend of open
2493 Description: non null lock assetion failure in mds_intent_policy
2494 Details : Trying to replay/resend lockless open requests resulted in
2495 mds_open() returning 0 with no lock. Now it sets a flag if
2496 a lock is going to be returned.
2498 Severity : enhancement
2500 Description: Checksum enhancements
2501 Details : New checksum enhancements allow for resending RPCs that failed
2504 Severity : enhancement
2506 Description: Tunables on number of dirty pages in cacche
2507 Details : Allow to set limit on number of dirty pages cached.
2512 Description: client crash on unmount - lock still has references
2513 Details : In some error handling cases it was possible to leak a lock
2514 reference on a client while accessing a file. This was not
2515 harmful to the client during operation, but would cause the
2516 client to crash when the filesystem is unmounted.
2519 Frequency : specific case, rare
2521 Description: ETXTBSY on mds though file not in use by client
2522 Details : ETXTBSY is no longer incorrectly returned when attempting to
2523 chmod or chown a directory that the user previously tried to
2524 execute or a currently-executing binary.
2527 Frequency : extremely rare except on liblustre-based clients
2529 Description: Lustre space not freed when files are deleted
2530 Details : Clean up open-unlinked files after client eviction. Previously
2531 the unlink was skipped and the files remained as orphans.
2536 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2537 Details : In some cases it was possible to send a blocking callback to
2538 a client doing a glimpse, even though that client didn't get
2539 a lock granted. When the glimpse lock is cancelled on the OST
2540 the freed lock is left on the waiting list and corrupted the list.
2543 Frequency : all core dumps
2545 Description: Broke core dumps to lustre
2546 Details : Negative dentry may be unhashed if parent does not have UPDATE
2547 lock, but some callers, e.g. do_coredump, expect dentry to be
2548 hashed after successful create, hash it in ll_create_it.
2550 ------------------------------------------------------------------------------
2552 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2554 * Support for kernels:
2555 2.6.9-42.0.2.EL (RHEL 4)
2556 2.6.5-7.276 (SLES 9)
2557 2.4.21-40.EL (RHEL 3)
2558 2.6.12.6 vanilla (kernel.org)
2562 Frequency : always on RHEL 3
2564 Description: Number of open files grows over time
2565 Details : The number of open files grows over time, whether or not
2566 Lustre is started. This was due to a filp leak introduced
2567 by one of our kernel patches.
2569 ------------------------------------------------------------------------------
2571 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2573 * Support for kernels:
2574 2.6.9-42.EL (RHEL 4)
2575 2.6.5-7.267 (SLES 9)
2576 2.4.21-40.EL (RHEL 3)
2577 2.6.12.6 vanilla (kernel.org)
2582 Bugzilla : 5719, 9635, 9792, 9684
2583 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2584 Details : If a server is under heavy load and cannot reply to new
2585 connection requests before the client resends the (re)connect,
2586 the connection handling code can behave badly if two service
2587 threads are concurrently handing separate (re)connections from
2588 the same client. Add better locking to the connection handling
2589 code, and ensure that only a single connection will be processed
2590 for a given client UUID, even if the lock is dropped.
2592 Severity : enhancement
2594 Description: add TCP zero-copy support to kernel
2595 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2596 sends if the hardware supports scatter-gather and checksumming.
2597 This allows socklnd to do client-write and server-read more
2598 efficiently and reduce CPU utilization from skbuf copying.
2601 Frequency : only if NFS exporting from client
2603 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2604 Details : If setting attributes on a file created under NFS that had
2605 never been opened it would be possible to oops the client
2606 if the file had no objects.
2610 Bugzilla : 9326, 10402, 10897
2611 Description: client crash in ptlrpcd_wake() thread when sending async RPC
2612 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2613 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2614 before ptlrpcd_wake() was called and the request was freed.
2617 Frequency : always for liblustre
2619 Description: liblustre client does MDS+OSTs setattr RPC for each write
2620 Details : When doing a write from a liblustre client, the client
2621 incorrectly issued an RPC to the MDS and each OST the file was
2622 striped over in order to update the timestamps. When writing
2623 with small chunks and many clients this could overwhelm the MDS
2624 with RPCs. In all cases it would slow down the write because
2625 these RPCs are unnecessary.
2627 Severity : enhancement
2629 Description: allow number of MDS service threads to be changed at module load
2630 Details : It is now possible to change the number of MDS service threads
2631 running. Adding "options mds mds_num_threads={N}" to the MDS's
2632 /etc/modprobe.conf will set the number of threads for the next
2633 time Lustre is restarted (assuming the "mds" module is also
2634 reloaded at that time). The default number of threads will
2635 stay the same, 32 for most systems.
2640 Description: OST crash if filesystem is unformatted or corrupt
2641 Details : If an OST is started on a device that has never been formatted
2642 or if the filesystem is corrupt and cannot even mount then the
2643 error handling cleanup routines would dereference a NULL pointer.
2648 Description: NULL pointer deref in llap_from_page.
2649 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2650 incorrect) mapping if the page was truncated/reclaimed while it was
2651 searched for. Check for this condition and skip such pages when
2652 doing readahead. Introduce extra check to llap_from_page() to
2653 verify page->mapping->host is non-NULL (so page is not anonymous).
2656 Frequency : Sometimes when using sys_sendfile
2658 Description: "page not covered by a lock" warnings from ll_readpage
2659 Details : sendfile called ll_readpage without right page locks present.
2660 Now we introduced ll_file_sendfile that does necessary locking
2661 around call to generic_file_sendfile() much like we do in
2665 Frequency : with certain MDS communication failures at client mount time
2667 Description: NULL pointer deref after failed client mount
2668 Details : a client connection request may delayed by the network layer
2669 and not be sent until after the PTLRPC layer has timed out the
2670 request. If the client fails the mount immediately it will try
2671 to clean up before the network times out the request. Add a
2672 reference from the request import to the obd device and delay
2673 the cleanup until the network drops the request.
2676 Frequency : occasionally during client (re)connect
2678 Description: assertion failure during client (re)connect
2679 Details : processing a client connection request may be delayed by the
2680 client or server longer than the client connect timeout. This
2681 causes the client to resend the connection request. If the
2682 original connection request is replied in this interval, the
2683 client may trip an assertion failure in ptlrpc_connect_interpret()
2684 which thought it would be the only running connect process.
2687 Frequency : only with obd_echo servers and clients that are rebooted
2689 Description: kernel BUG accessing uninitialized data structure
2690 Details : When running an obd_echo server it did not start the ping_evictor
2691 thread, and when a client was evicted an uninitialized data
2692 structure was accessed. Start the ping_evictor in the RPC
2693 service startup instead of the OBD startup.
2695 Severity : enhancement
2696 Bugzilla : 10193 (patchless)
2697 Description: Remove dependency on various unexported kernel interfaces.
2698 Details : No longer need reparent_to_init, exit_mm, exit_files,
2699 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2702 Frequency : rare (only users of deprecated and unsupported LDAP config)
2704 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2705 Details : LDAP apparently contains 'lustreName' attributes instead of
2706 'name'. A simple remapping of the name is sufficient.
2709 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2711 Description: waiting_locks_callback trips kernel BUG if client is evicted
2712 Details : Running with the dump_on_timeout debug flag turned on makes
2713 it possible that the waiting_locks_callback() can try to dump
2714 the Lustre kernel debug logs from an interrupt handler. Defer
2715 this log dumping to the expired_lock_main() thread.
2717 Severity : enhancement
2719 Description: Support NFS exporting on 2.6 kernels.
2720 Details : Implement non-rawops metadata methods for NFS server to use without
2721 changing NFS server code.
2724 Frequency : very rare (synthetic metadata workload only)
2726 Description: two racing renames might cause an MDS thread to deadlock
2727 Details : Running the "racer" program may cause one MDS thread to rename
2728 a file from being the source of a rename to being the target of
2729 a rename at exactly the same time that another thread is doing
2730 so, and the second thread has already enqueued these locks after
2731 doing a lookup of the target and is trying to relock them in
2732 order. Ensure that we don't try to re-lock the same resource.
2735 Frequency : only very large systems with liblustre clients
2737 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2738 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2739 all OSTs in parallel. This allows the eviction of stale liblustre
2740 clients to proceed much faster than if they were done in series,
2741 and also offers similar improvements for other set_info RPCs.
2746 Description: excessive CPU usage during initial read phase on client
2747 Details : During the initial read phase on a client, it would agressively
2748 retry readahead on the file, consuming too much CPU and impacting
2749 performance (since 1.4.5.8). Improve the readahead algorithm
2750 to avoid this, and also improve some other common cases (read
2751 of small files in particular, where "small" is files smaller than
2752 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2757 Description: MDS crash when receiving packet with unknown intent.
2758 Details : Do not LBUG in unknown intent case, just return -EFAULT
2760 Severity : enhancement
2761 Bugzilla : 9293, 9385
2762 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2763 Details : Do not serialize getattr (non-intent version) and statfs.
2766 Frequency : occasional, when OST network is overloaded/intermittent
2768 Description: client evicted by OST after bulk IO timeout
2769 Details : If a client sends a bulk IO request (read or write) the OST
2770 may evict the client if it is unresposive to its data GET/PUT
2771 request. This is incorrect if the network is overloaded (takes
2772 too long to transfer the RPC data) or dropped the OST GET/PUT
2773 request. There is no need to evict the client at all, since
2774 the pinger and/or lock callbacks will handle this, and the
2775 client can restart the bulk request.
2778 Frequency : Always when mmapping file with no objects
2780 Description: client crashes when mmapping file with no objects
2781 Details : Check that we actually have objects in a file before doing any
2782 operations on objects in ll_vm_open, ll_vm_close and
2788 Description: Request leak when working with deleted CWD
2789 Details : Introduce advanced request refcount tracking for requests
2790 referenced from lustre intent.
2792 Severity : Enhancement
2794 Description: Cache open file handles on client.
2795 Details : MDS now will return special lock along with openhandle, if
2796 requested and client is allowed to hold openhandle, even if unused,
2797 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2798 closing files for every read/write openration.
2800 Severity : Enhancement
2802 Description: Cache open negative dentries on client when possible.
2803 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2804 negative dentries on lock revocation.
2809 Description: Remounting a client read-only wasn't possible with a zconf mount
2810 Details : It wasn't possible to remount a client read-only with llmount.
2812 Severity : enhancement
2813 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2814 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2815 MPICH is included to add Lustre-specific ADIO interfaces.
2816 This is based closely on the UFS ADIO layer and only differs
2817 in file creation, in order to allow the OST striping to be set.
2818 This is user-contributed code and not supported by CFS.
2823 Description: extended inode attributes (immutable, append-only) work improperly
2824 when 2.4 and 2.6 kernels are used on client/server or vice versa
2825 Details : Introduce kernel-independent values for these flags.
2827 Severity : enhancement
2830 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2831 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2832 granularity. Now, Fractional MB granularity is be supported,
2833 this is very useful for low memory system.
2835 Severity : enhancement
2837 Description: Getattr by fid
2838 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2839 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2840 have locks and avoids getattr rpc requests for it on every stat.
2843 Frequency : Always, for filesystems larger than 2TB
2845 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2846 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2847 filesystems larger than 2TB to be mounted with mballoc enabled.
2850 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2852 Description: filesystem corruption for non-standard kernels and very large OSTs
2853 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2854 filesystems larger than 2TB is mounted then the kernel will
2855 silently corrupt the start of the filesystem. CONFIG_LBD is
2856 enabled for all CFS-supported kernels, but the possibility of
2857 this happening with a modified kernel config exists.
2859 Severity : enhancement
2861 Description: add client O_DIRECT support for 2.6 kernels
2862 Details : It is now possible to do O_DIRECT reads and writes to files
2863 in the Lustre client mountpoint on 2.6 kernel clients.
2865 Severity : enhancement
2867 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2868 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2869 parallel, not waiting for response from every OST before sending
2870 a rpc to the next OST.
2875 Description: setattr vs write race when updating file timestamps
2876 Details : Client processes that update a file timestamp into the past
2877 right after writing to the file (e.g. tar) it is possible that
2878 the updated file modification time can be reset to the current
2879 time due to a race between processing the setattr and write RPC.
2881 Severity : enhancement
2883 Description: Bring 'lfs find' closer in line with regular Linux find.
2884 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2885 -print0 options and obtains all the needed info through the lustre
2888 Severity : enhancement
2890 Description: support up to 1024 configured devices on one node
2891 Details : change obd_dev array from statically allocated to dynamically
2892 allocated structs as they are first used to reduce memory usage
2897 Description: Flush dirty partially truncated pages during truncate
2898 Details : Immediatelly flush partially truncated pages in filter_setattr,
2899 this way we completely avoid having any pages in page cache on OST
2900 and can retire ugly workarounds during writes to flush such pages.
2905 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2906 Details : protect inode from truncation within vfs_unlink() context
2907 just take a reference before calling vfs_unlink() and release it
2908 when parent's i_sem is free.
2911 Frequency : always, if extents are used on OSTs
2913 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2914 Details : OSTs using the extents format would not zero the high 16 bits of
2915 the index physical block number. This is not a problem for any
2916 OST filesystems smaller than 16TB, and no kernels support ext3
2917 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2918 new/modified files) and can be fixed for existing filesystems
2919 with e2fsprogs-1.39-cfs1.
2924 Description: import connection selection may be incorrect if timer wraps
2925 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2926 import connections to be ignored if the 32-bit jiffies counter
2927 wraps. Use a 64-bit jiffies counter.
2930 Frequency : during server recovery
2932 Description: crash after server is denying duplicate export
2933 Details : If clients are resending connect requests to the server, the
2934 server refuses to allow a client to connect multiple times.
2935 Fixed a bug in the handling of this case.
2938 Frequency : very large clusters immediately after boot
2940 Description: LNET request buffers exhausted under heavy short-term load
2941 Details : If a large number of client requests are generated on a service
2942 that has previously never seen so many requests it is possible
2943 that the request buffer growth cannot keep up with the spike in
2944 demand. Instead of dropping incoming requests, they are held in
2945 the LND until the RPC service can accept more requests.
2948 Frequency : Sometimes during replay
2950 Description: Assertion failure in ll_local_open after replay.
2951 Details : If replay happened on an open request reply before we were able
2952 to set replay handler, reply will become not swabbed tripping the
2953 assertion in ll_local_open. Now we set the handler right after
2954 recognising of open request
2957 Frequency : very rare
2959 Description: kernel reports "badness in vsnprintf"
2960 Details : Reading from the "recovery_status" /proc file in small chunks
2961 may cause a negative length in lprocfs_obd_rd_recovery_status()
2962 call to vsnprintf() (which is otherwise harmless). Exit early
2963 if there is no more space in the output buffer.
2965 Severity : enhancement
2967 Description: clear OBD RPC statistics by writing to them
2968 Details : It is now possible to clear the OBD RPC statistics by writing
2969 to the "stats" file.
2974 Description: Client mtime is not the same on different clients after utimes
2975 Details : In some cases, the client was using the utimes() syscall on
2976 a file cached on another node. The clients now validate the
2977 ctime from the MDS + OSTs to determine which one is right.
2982 Description: Inability to activate failout mode
2983 Details : lconf script incorrectly assumed that in python string's numeric
2984 value is used in comparisons.
2987 Frequency : always with multiple stripes per file
2989 Description: Inefficient object allocation for mutli-stripe files
2990 Details : When selecting which OSTs to stripe files over, for files with
2991 a stripe count that divides evenly into the number of OSTs,
2992 the MDS is always picking the same starting OST for each file.
2993 Return the OST selection heuristic to the original design.
2998 Description: mount failures may take full timeout to return an error
2999 Details : Under some heavy load conditions it is possible that a
3000 failed mount can wait for the full obd_timeout interval,
3001 possibly several minutes, before reporting an error.
3002 Instead return an error as soon as the status is known.
3004 ------------------------------------------------------------------------------
3006 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
3008 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3009 INTEROPERATE with older versions automatically. Please read the
3010 user documentation before upgrading any part of a live system.
3011 * WARNING: Lustre networking configuration changes are required with
3012 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3015 * Support for kernels:
3016 2.6.9-22.0.2.EL (RHEL 4)
3017 2.6.5-7.244 (SLES 9)
3018 2.6.12.6 vanilla (kernel.org)
3021 Severity : enhancement
3022 Bugzilla : 7981/8208
3023 Description: Introduced Lustre Networking (LNET)
3024 Details : LNET is new networking infrastructure for Lustre, it includes
3025 a reorganized network configuration mode (see the user
3026 documentation for full details) as well as support for routing
3027 between different network fabrics. Lustre Networking Devices
3028 (LNDs) for the supported network fabrics have also been
3029 created for this new infrastructure.
3031 Severity : enhancement
3032 Description: Introduced Access control lists
3033 Details : clients can set ACLs on files and directories in order to have
3034 more fine-grained permissions than the standard Unix UGO+RWX.
3035 The MDS must be started with the "-o acl" mount option.
3037 Severity : enhancement
3038 Description: Introduced filesystem quotas
3039 Details : Administrators may now establish per-user quotas on the
3042 Severity : enhancement
3044 Description: Configuration change for the XT3
3045 The PTLLND is now used to run Lustre over Portals on the XT3
3046 The configure option(s) --with-cray-portals are no longer used.
3047 Rather --with-portals=<path-to-portals-includes> is used to
3048 enable building on the XT3. In addition to enable XT3 specific
3049 features the option --enable-cray-xt3 must be used.
3054 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3055 Details : A race between allocating a new llcd and re-getting the llcd_lock
3056 allowed another thread to grab newly-allocated llcd.
3058 Severity : enhancement
3060 Description: 2.6 OST async journal commit and locking fix to improve performance
3061 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3062 2.6 kernels are now async as they already were in 2.4 kernels so
3063 that they can commit concurrently with the network bulk transfer.
3064 For block-allocated files the filter allocation semaphore is held
3065 to avoid filesystem fragmentation during allocation. BKL lock
3066 removed for 2.6 xattr operations where it is no longer needed.
3071 Description: lconf incorrectly determined whether two IP networks could talk
3072 Details : In some more complicated routing and multiple-network
3073 configurations, lconf will avoid trying to make a network
3074 connection to a disjoint part of the IP space. It was doing the
3075 math incorrectly for one set of cases.
3080 Description: Fix for potential infinite loop processing records in an llog.
3081 Details : If an llog record is corrupted/zeroed, it is possible to loop
3082 forever in llog_process(). Validate the llog record length
3083 and skip the remainder of the block on error.
3086 Frequency : occasional (liblustre only)
3088 Description: liblustre could not open files whose last component is a symlink
3089 Details : sysio_path_walk() would incorrectly pass the open intent to
3090 intermediate path components.
3093 Frequency : rare (liblustre only with non-standard tuning)
3094 Bugzilla : 7201 (7350)
3095 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3096 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3097 after it has been released, only do this for referenced locks
3099 Severity : enhancement
3101 Description: specifying an (invalid) directory default stripe_size of -1
3102 would reset the directory default striping
3103 Details : stripe_size -1 was used internally to signal directory stripe
3104 removal, now use "all default" to signal dir stripe removal
3105 as a directory striping of "all default" is not useful
3108 Frequency : common for large clusters running liblustre clients
3110 Description: doing an ls when liblustre clients are running is slow
3111 Details : sending a glimpse AST to a liblustre client waits for every AST
3112 to time out, as liblustre clients will not respond. Since they
3113 cannot cache data we refresh the OST lock LVB from disk instead.
3115 Severity : enhancement
3117 Description: doing an ls at the same time as file IO can be slow
3118 Details : enqueue and other "small" requests can be blocked behind many
3119 large IO requests. Create a new OST IO portal for non-IO
3120 requests so they can be processed faster.
3123 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3125 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3126 created file with mode 000
3127 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3128 allow an ftruncate() as a non-root user to a file with mode 000.
3129 Lustre now respects this flag to disable mode checks when
3130 truncating a file owned by the user
3133 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3136 Description: Revoking locks from clients that went dead or catatonic might take
3138 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3139 cancellation of such locks instant on servers without waiting for
3140 any reply from clients. Clients drops these locks when cancel
3141 notification from server is received without replying.
3144 Frequency : liblustre-only, when liblustre client dies or becomes busy
3146 Description: Doing ls on Linux clients can take a long time with active
3148 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3149 granting such locks to it in the first place if possible. Locks
3150 are taken by proxy on the OST during the read or write and
3151 dropped immediately afterward. Add connect flags handling, do
3152 not grant locks to liblustre clients for glimpse ASTs.
3154 Severity : enhancement
3156 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3157 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3158 sequential reads and behaves badly with applications doing random
3159 reads. Improve it by reading ahead at least read region, and
3160 avoiding excessive large RPC for small reads.
3162 Severity : enhancement
3164 Description: Creating more than 1000 files for a single job may cause a load
3165 imbalance on the OSTs if there are also a large number of OSTs.
3166 Details : qos_prep_create() uses an OST index reseed value that is an
3167 even multiple of the number of available OSTs so that if the
3168 reseed happens in the middle of the object allocation it will
3169 still utilize the OSTs as uniformly as possible.
3174 Description: OST or MDS may oops in ping_evictor_main()
3175 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3176 but doesn't restart at beginning of obd_exports_timed list
3179 Severity : enhancement
3181 Description: improve by-nid export eviction on the MDS and OST
3182 Details : allow multiple exports with the same NID to be evicted at one
3183 time without re-searching the exports list.
3186 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3188 Description: MDS may oops in groups_free()
3189 Details : in rare race conditions a newly allocated group_info struct is
3190 freed again, and this can be NULL. The 2.4 compatibility code
3191 for groups_free() checked for a NULL pointer, but 2.6 did not.
3194 Frequency : common for liblustre clients doing little filesystem IO
3195 Bugzilla : 9352, 7313
3196 Description: server may evict liblustre clients accessing contended locks
3197 Details : if a client is granted a lock or receives a completion AST
3198 with a blocking AST already set it would not reply to the AST
3199 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3202 Frequency : lfs setstripe, only systems with more than 160 OSTs
3204 Description: unable to set striping with a starting offset beyond OST 160
3205 Details : llapi_create_file() incorrectly limited the starting stripe
3206 index to the maximum single-file stripe count.
3209 Frequency : LDAP users only
3211 Description: lconf did not handle in-kernel recovery with LDAP properly
3212 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3214 Severity : enhancement
3216 Description: bind OST threads to NUMA nodes to improve performance
3217 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3218 node and do their allocations there to localize memory access
3220 Severity : enhancement
3222 Description: llmount can determine client NID directly from Myrinet (GM)
3223 Details : the client NID code from gmnalnid was moved directly into
3224 llmount, removing the need to use this or specifying the
3225 client NID explicitly when mounting GM clients with zeroconf
3228 Frequency : if client is started with down MDS
3230 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3231 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3232 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3233 ptlrpc_import_delayed_req() because we didn't check if the
3234 request was interrupted, and we also didn't break out of the
3235 event loop if there was no timeout
3240 Description: data loss during non-page-aligned writes to a single file from
3241 both multiple nodes and multiple threads on one node at same time
3242 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3243 inconsistency led to false short-reads, that were cached and later
3244 used by ->prepare_write() to fill in partially written page,
3245 leading to data loss.
3248 Frequency : always, if lconf --abort_recovery used
3250 Description: lconf --abort_recovery fails with 'Operation not supported'
3251 Details : lconf was attempting to abort recovery on the MDT device and not
3254 Severity : enhancement
3256 Description: remove cleanup logs
3257 Details : replace lconf-generated cleanup logs with lustre internal
3258 cleanup routines. Eliminates the need for client-cleanup and
3261 Severity : enhancement
3263 Description: add support for EAs (user and system) on lustre filesystems
3264 Details : it is now possible to store extended attributes in the Lustre
3265 client filesystem, and with the user_xattr mount option it
3266 is possible to allow users to store EAs on their files also
3268 Severity : enhancement
3270 Description: Add possibility (config option) to show minimal available OST free
3272 Details : When compiled with --enable-mindf configure option, statfs(2)
3273 (and so, df) will return least minimal free space available from
3274 all OSTs as amount of free space on FS, instead of summary of
3275 free spaces of all OSTs.
3277 Severity : enhancement
3279 Description: do not expand extent locks acquired on OST-side
3280 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3281 by server: they are not cached anyway.
3284 Frequency : when mmap is used/binaries executed from Lustre
3286 Description: Unmmap pages before throwing them away from read cache.
3287 Details : llap_shrink cache now attempts to unmap pages before discarding
3288 them (if unmapping failed - do not discard). SLES9 kernel has
3289 extra checks that trigger if this unmapping is not done first.
3294 Description: lconf didn't resolve symlinks before checking to see whether a
3295 given mountpoint was already in use
3298 Frequency : when migrating failover services
3299 Bugzilla : 6395, 9514
3300 Description: When migrating a subset of services from a node (e.g. failback
3301 from a failover service node) the remaining services would
3302 time out and evict clients.
3303 Details : lconf --force (implied by --failover) sets the global obd_timeout
3304 to 5 seconds in order to quickly disconnect, but this caused
3305 other RPCs to time out too quickly. Do not change the global
3306 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3308 Severity : enhancement
3309 Frequency : if MDS is started with down OST
3310 Bugzilla : 9439,5706
3311 Description: Allow startup/shutdown of an MDS without depending on the
3312 availability of the OSTs.
3313 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3314 Add appropriate locking and lov-osc refcounts for safe
3315 cleaning. Add osc abort_inflight calls in case the
3316 synchronize never started.
3319 Frequency : occasional (Cray XT3 only)
3321 Description: root not authorized to access files in CRAY_PORTALS environment
3322 Details : The client process capabilities were not honoured on the MDS in
3323 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3324 been accessed by an authorized user then root was able to access
3325 the file on the local client also. The root user capabilities
3326 are now allowed on the MDS, as this environment has secure UID.
3329 Frequency : occasional
3331 Description: ldiskfs "too long searching" message happens too often
3332 Details : A debugging message (otherwise harmless) prints too often on
3333 the OST console. This has been reduced to only happen when
3334 there are fragmentation problems on the filesystem.
3339 Description: Division by zero in statfs when all OSCs are inactive
3340 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3341 lov_statfs divides by value returned by lov_get_stripecnt().
3345 Bugzilla : 9489, 3273
3346 Description: First write from each client to each OST was only 4kB in size,
3347 to initialize client writeback cache, which caused sub-optimal
3348 RPCs and poor layout on disk for the first writen file.
3349 Details : Clients now request an initial cache grant at (re)connect time
3350 and so that they can start streaming writes to the cache right
3351 away and always do full-sized RPCs if there is enough data.
3352 If the OST is rebooted the client also re-establishes its grant
3353 so that client cached writes will be honoured under the grant.
3358 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3359 Details : Now I/O RPCs go to different portal number and (presumably) fast
3360 lock requests (and glimses) and other RPCs get their own service
3361 threads pool that should be able to service those RPCs
3364 Severity : enhancement
3366 Description: Ability to exchange lustre version between client and servers and
3367 issue warnings at client side if client is too old. Also for
3368 liblustre clients there is ability to refuse connection of too old
3370 Details : New 'version' field is added to connect data structure that is
3371 filled with version info. That info is later checked by server and
3375 Frequency : rare, liblustre only.
3376 Bugzilla : 9296, 9581
3377 Description: Two simultaneous writes from liblustre at offset within same page
3378 might proceed at the same time overwriting eachother with stale
3380 Details : I/O lock withing llu_file_prwv was released too early, before data
3381 actually was hitting the wire. Extended lock-holding time until
3382 server acknowledges receiving data.
3385 Frequency : extremely rare. Never observed in practice.
3387 Description: avoid generating lustre_handle cookie of 0.
3388 Details : class_handle_hash() generates handle cookies by incrementing
3389 global counter, and can hit 0 occasionaly (this is unlikely, but
3390 not impossible, because initial value of cookie counter is
3391 selected randonly). Value of 0 is used as a sentinel meaning
3392 "unassigned handle" --- avoid it. Also coalesce two critical
3393 sections in this function into one.
3395 Severity : enhancement
3397 Description: allow liblustre clients to delegate truncate locking to OST
3398 Details : To avoid overhead of locking, liblustre client instructs OST to
3399 take extent lock in ost_punch() on client's behalf. New connection
3400 flag is added to handle backward compatibility.
3402 Severity : enhancement
3403 Bugzilla : 4928, 7341, 9758
3404 Description: allow number of OST service threads to be specified
3405 Details : a module parameter allows the number of OST service threads
3406 to be specified via "options ost ost_num_threads={N}" in the
3407 OSS's /etc/modules.conf or /etc/modprobe.conf.
3411 Bugzilla : 6146, 9635, 9895
3412 Description: servers crash with bad pointer in target_handle_connect()
3413 Details : In rare cases when a client is reconnecting it was possible that
3414 the connection request was the last reference for that export.
3415 We would temporarily drop the export reference and get a new
3416 one, but this may have been the last reference and the export
3417 was just destroyed. Get new reference before dropping old one.
3419 Severity : enhancement
3420 Frequency : if client is started with failover MDS
3422 Description: Allow multiple MDS hostnames in the mount command
3423 Details : Try to read the configuration from all specified MDS
3424 hostnames during a client mount in case the "primary"
3427 Severity : enhancement
3429 Description: Stop sending data to evicted clients as soon as possible.
3430 Details : Check if the client we are about to send or are sending data to
3431 was evicted already. (Check is done every second of waiting,
3432 for which l_wait_event interface was extended to allow checking
3433 of exit condition at specified intervals).
3436 Frequency : rare, normally only when NFS exporting is done from client
3438 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3440 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3443 Frequency : always, if srand() is called before liblustre initialization
3445 Description: Liblustre uses system PRNG disturbing its usage by user application
3446 Details : Introduce internal to lustre fast and high-quality PRNG for
3447 lustre usage and make liblustre and some other places in generic
3448 lustre code to use it.
3450 Severity : enhancement
3451 Bugzilla : 9477, 9557, 9870
3452 Description: Verify that the MDS configuration logs are updated when xml is
3453 Details : Check if the .xml configuration logs are newer than the config
3454 logs stored on the MDS and report an error if this is the case.
3455 Request --write-conf, or allow starting with --old_conf.
3457 Severity : enhancement
3459 Description: Handle symlinks in the path when checking if Lustre is mounted.
3460 Details : Resolve intermediate symlinks when checking if a client has
3461 mounted a filesystem to avoid duplicate client mounts.
3466 Description: lconf can hit an error exception but still return success.
3467 Details : The lconf command catches the Command error exception at the top
3468 level script context and will exit with the associated exit
3469 status, but doesn't ensure that this exit status is non-zero.
3474 Description: failure of ptlrpc thread startup can cause oops
3475 Details : Starting a ptlrpc service thread can fail if there are a large
3476 number of threads or the server memory is very fragmented.
3477 Handle this without oopsing.
3480 Frequency : always, only if liblustre and non-default acceptor port was used
3482 Description: liblustre cannot connect to servers with non-default acceptor port
3483 Details : tcpnal_set_default_params() was not called and was therefore
3484 ignoring the environment varaible TCPNAL_PORT, as well as other
3485 TCPNAL_ environment variables
3490 Description: two objects could be created on the same OST for a single file
3491 Details : If an OST is down, in some cases it was possible to create two
3492 objects on a single OST for a single file. No problems other
3493 than potential performance impact and spurious error messages.
3497 Bugzilla : 5681, 9562
3498 Description: Client may oops in ll_unhash_aliases
3499 Details : Client dcache may become inconsistent in race condition.
3500 In some cases "getcwd" can fail if the current directory is
3506 Description: Inode refcounting problems in NFS export code
3507 Details : link_raw functions used to call d_instantiate without obtaining
3508 extra inode reference first.
3512 Bugzilla : 9942, 9903
3513 Description: Referencing freed requests leading to crash, memleaks with NFS.
3514 Details : We used to require that call to ll_revalidate_it was always
3515 followed by ll_lookup_it. Also with revalidate_special() it is
3516 possible to call ll_revalidate_it() twice for the same dentry
3517 even if first occurence returned success. This fix changes semantic
3518 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3519 reference on a request referred from the intent.
3520 ll_intent_release() then releases such a request.
3523 Frequency : rare, normally benchmark loads only
3525 Description: unlinked inodes were kept in memory on the client
3526 Details : If a client is repeatedly creating and unlinking files it
3527 can accumulate a lot of stale inodes in the inode slab cache.
3528 If there is no other client load running this can cause the
3529 client node to run out of memory. Instead flush old inodes
3530 from client cache that have the same inode number as a new inode.
3533 Frequency : SLES9 2.6.5 kernel and long filenames only
3534 Bugzilla : 9969, 10379
3535 Description: utime reports stale NFS file handle
3536 Details : SLES9 uses out-of-dentry names in some cases, which confused
3537 the lustre dentry revalidation. Change it to always use the
3541 Frequency : rare, unless heavy write-truncate concurrency is continuous
3542 Bugzilla : 4180, 6984, 7171, 9963, 9331
3543 Description: OST becomes very slow and/or deadlocked during object unlink
3544 Details : filter_destroy() was holding onto the parent directory lock
3545 while truncating+unlinking objects. For very large objects this
3546 may block other threads for a long time and slow overall OST
3547 responsiveness. It may also be possible to get a lock ordering
3548 deadlock in this case, or run out of journal credits because of
3549 the combined truncate+unlink. Solution is to do object truncate
3550 first in one transaction without parent lock, and then do the
3551 final unlink in a new transaction with the parent lock. This
3552 reduces the lock hold time dramatically.
3555 Frequency : rare, 2.4 kernels only
3557 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3558 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3559 before unmounting their filesystem in order to avoid deadlock.
3560 The kernel_locked() function in 2.4 kernels only checks whether
3561 the kernel lock is held, not whether it is this process that is
3562 holding it as 2.6 kernels do.
3567 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3568 Details : The client ptlrpc code may be trying to reconnect to a down
3569 server before a previous connection attempt has timed out.
3570 Increase the reconnect interval to be longer than the connection
3571 timeout interval to avoid sending duplicate connections to
3575 Frequency : echo_client brw_test command
3577 Description: fix echo_client to work with OST preallocated code
3578 Details : OST preallocation code (5137) didn't take echo_client IO path
3579 into account: echo_client calls filter methods outside of any
3580 OST thread and, hence, there is no per-thread preallocated
3581 pages and buffers to use. Solution: hijack pga pages for IO. As
3582 a byproduct, this avoids unnecessary data copying.
3586 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3587 Description: Client can oops in mdc_commit_close() after open replay
3588 Details : It was possible for the MDS to return an open request with no
3589 transaction number in mds_finish_transno() if the client was
3590 evicted, but without actually returning an error. Clients
3591 would later try to replay that open and may trip an assertion
3592 Simplify the client close codepath, and always return an error
3593 from the MDS in case the open is not successful.
3596 Frequency : rare, 2.6 OSTs only
3598 Description: OST may deadlock under high load on fragmented files
3599 Details : If there was a heavy load and highly-fragmented OST filesystems
3600 it was possible to have all the OST threads deadlock waiting on
3601 allocation of biovecs, because the biovecs were not released
3602 until the entire RPC IO was completed. Instead, release biovecs
3603 as soon as they are complete to ensure forward IO progress.
3605 Severity : enhancement
3607 Description: Support for specifying external journal device at mount
3608 Details : If an OST or MDS device is formatted with an external journal
3609 device, this device major/minor is stored in the ext3 superblock
3610 and may not be valid for failover. Allow detecting and
3611 specifying the external journal at mount time.
3616 Description: Mounting an MDS with pending unlinked files may cause oops
3617 Details : target_finish_recovery() calls mds_postrecov() which returned
3618 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3619 considers this an error and immediately begins cleaning up the
3620 lov, just after starting the mds_lov process
3622 Severity : enhancement
3624 Description: Implement 'lfs df' to report actual free space on per-OST basis
3625 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3626 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3627 usage of objects; '-h' to report in human readable format.
3629 ------------------------------------------------------------------------------
3631 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3638 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3639 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3640 at startup during orphan unlinking. Instead, skip buddy update
3641 before setup, as it will be regenerated after recovery is complete.
3646 Description: If an OST is inactive, its locks might reference stale inodes.
3647 Details : lov_change_cbdata() must iterate over all namespaces, even if
3648 they are inactive to clear inode references from the lock.
3650 Severity : enhancement
3651 Frequency : occasional, if non-standard max_dirty_mb used
3653 Description: Client will block write RPCs if not enough grant
3654 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3655 then the client will block writes while waiting for another RPC
3656 to complete instead of consuming its dirty limit. With change
3657 we get improved performance when max_dirty_mb is small.
3659 Severity : enhancement
3660 Bugzilla : 3389, 6253
3661 Description: Add support for supplementary groups on the MDS.
3662 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3663 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3664 lookups for user supplementary groups into a cache.
3668 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3669 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3670 for liblustre O_EXCL is enforced by MDS.
3674 Description: readdir never returns NULL in liblustre.
3675 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3676 dentry in struct dirent.
3680 Frequency : liblustre only, depends on application IO pattern
3681 Description: liblustre clients evicted if not contacting servers
3682 Details : Don't put liblustre clients into the ping_evictor list, so
3683 they will not be evicted by the pinger ever.
3685 Severity : enhancement
3687 Description: Add ability to evict clients by NID from MDS.
3688 Details : By echoing "nid:$NID" string into
3689 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3690 $NID would be instantly evicted from this MDS and from all active
3691 OSTs connected to it.
3695 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3696 Details : lookup_it_finish() used to query file size from OSTs that was not
3701 Description: service threads change working directory to that of init
3702 Details : Starting lustre service threads may pin the working directory
3703 of the parent thread, making that filesystem busy. Threads
3704 now change to the working directory of init to avoid this.
3708 Frequency : during shutdown only
3709 Description: shutdown with a failed MDS or OST can cause unmount to hang
3710 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3715 Frequency : 2.6 only
3716 Description: chmod/chown may include an extra supplementary group
3717 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3718 supplementary group and if none is specified this is used.
3721 Bugzilla : 5479 (6816)
3723 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3724 Details : If lookup is in progress on a file that is unlinked we might try
3725 to revalidate the inode and fail in revalidate after lookup is
3726 complete and ll_file_open() enqueues the open again but
3727 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3730 Frequency : always, if lconf --abort_recovery used
3732 Description: lconf --abort_recovery fails with 'Operation not supported'
3733 Details : lconf was attempting to abort recovery on the MDT device and not
3736 ------------------------------------------------------------------------------
3738 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3743 Frequency : rare (only unsupported configurations with a node running as an
3745 Bugzilla : 6514, 5137
3746 Description: Mounting a Lustre file system on a node running as an OST could
3748 Details : OSTs now preallocates memory needed to write out data at
3749 startup, instead of when needed, to avoid having to
3750 allocate memory in possibly low memory situations.
3751 Specifically, if the file system is mounted on on OST,
3752 memory pressure could force it to try to write out data,
3753 which it needed to allocate memory to do. Due to the low
3754 memory, it would be unable to do so and the node would
3755 become unresponsive.
3757 Severity : enhancement
3759 Description: Addition of lconf --service command line option
3760 Details : lconf now accepts a '--service <arg>' option, which is
3761 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3763 Severity : enhancement
3765 Description: Failover mode is now the default for OSTs.
3766 Details : By default, OSTs will now run in failover mode. To return to
3767 the old behaviour, add '--failout' to the lmc line for OSTs.
3769 Severity : enhancement
3771 Description: Health checks are now provided for MDS and OSTs
3772 Details : Additional detailed health check information on MSD and OSTs
3773 is now provided through the procfs health_check value.
3776 Frequency : occasional, depends on IO load
3778 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3779 after numerous create/delete cycles
3780 Details : The ext3 inode allocation policy would not allocate new inodes
3781 very well on the OSTs because there are no new directories
3782 being created. Instead we look for groups with free space if
3783 the parent directories are nearly full.
3788 Description: Network or server problems during mount may cause partially
3789 mounted clients instead of returning an error.
3790 Details : The config llog parsing code may overwrite the error return
3791 code during mount error handling, returning success instead
3797 Description: MDS can fail to allocate large reply buffers
3798 Details : After long uptimes the MDS can fail to allocate large reply
3799 buffers (e.g. zconf client mount config records) due to memory
3800 fragmentation or consumption by the buffer cache. Preallocate
3801 some large reply buffers so that these replies can be sent even
3802 under memory pressure.
3806 Frequency : rare (liblustre)
3807 Description: fsx running with liblustre complained that using truncate() to
3808 extend the file doesn't work. This patch corrects that issue.
3809 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3810 ATTR_SIZE and lsm use in llu_setattr_raw.
3814 Frequency : rare, only 2.6 kernels
3815 Description: Unusual file access patterns on the MDS may result in inode
3816 data being lost in very rare circumstances.
3817 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3818 code in the 2.6 kernel caused the inode and/or EA data not to
3819 be read from disk, causing single-file corruption.
3823 Frequency : rare, only 2.6 filesystems using extents
3824 Description: Heavy concurrent write and delete load may cause data corruption.
3825 Details : It was possible under high-load situations to have an extent
3826 metadata block in the block device cache from a just-unlinked
3827 file overwrite a newly-allocated data block. We now unmap any
3828 metadata buffers that alias just-allocated data blocks.
3832 Frequency : filesystems with default stripe_count larger than 77
3833 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3834 Details : lconf specifies an inode size of 4096 bytes when the default
3835 stripe_count is larger than 77. This conflicts with the default
3836 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3837 this case to avoid pinning too much memory for large EAs.
3839 ------------------------------------------------------------------------------
3841 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3846 Frequency : rare (extremely heavy IO load with hundreds of clients)
3848 Description: Client is evicted, gets IO error writing to file
3849 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3850 caused clients to be evicted for AST timeouts. The fixes in
3851 bug 5192 mean we no longer need to have such short AST timeouts
3852 so ldlm_timeout has been increased.
3855 Frequency : occasional during --force or --failover shutdown under load
3856 Bugzilla : 5949, 4834
3857 Description: Server oops/LBUG if stopped with --force or --failover under load
3858 Details : a collection of import/export refcount and cleanup ordering
3859 issues fixed for safer force cleanup
3862 Frequency : only filesystems larger than 120 OSTs
3863 Bugzilla : 5990, 6223
3864 Description: lfs getstripe would oops on a very large filesystem
3865 Details : lov_getconfig used kfree on vmalloc'd memory
3868 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3869 Bugzilla : 6242, 6243
3870 Description: reading from files that had been truncated to a non-zero size
3871 but never opened returned no data
3872 Details : ll_file_read() reads zeros from no-object files to EOF
3877 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3878 Details : There's little harm in aborting MDS/OSS recovery and letting it
3879 try again, so I removed the LASSERT and return an error instead.
3881 Severity : enhancement
3883 Description: New debugging infrastructure for tracking down data corruption
3884 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3885 (b) cover more of the client-side code path, and (c) try to narrow
3886 down where problems occurred
3890 Bugzilla : 3819, 4364, 4397, 6313
3891 Description: Racing close and eviction MDS could cause assertion in mds_close
3892 Details : It was possible to get multiple mfd references during close and
3893 client eviction, leading to one thread referencing a freed mfd.
3895 Severity: : enhancement
3896 Bugzilla : 3262, 6359
3897 Description: Attempts to reconnect to servers are now more aggressive.
3898 Details : This builds on the enhanced upcall-less recovery that was added
3899 in 1.4.2. When trying to reconnect to servers, clients will
3900 now try each server in the failover group every 10 seconds. By
3901 default, clients would previously try one server every 25 seconds.
3906 Description: After recovery, certain operations trigger a failed
3907 assertion on a client.
3908 Details : Failing over an mds, using lconf -d --failover, while a
3909 client was doing a readdir() call would cause the client to
3910 LBUG after recovery completed and the readdir() was resent.
3912 Severity : enhancement
3914 Description: Default groups are now added by lconf
3915 Details : You can now run lconf --group <servicename> without having to
3916 manually add groups with lmc.
3919 Frequency : occasional
3921 Description: Nodes with an elan id of 0 trigger a failed assertion
3924 Frequency : always when accessing e.g. tty/console device nodes
3926 Description: tty and some other devices nodes cannot be used on lustre
3927 Details : file's private_data field is used by device data and lustre
3928 values in there got lost. New field was added to struct file to
3929 store fs-specific private data.
3932 Frequency : when exporting Lustre via NFS
3934 Description: NFSD failed occasionally when looking up a path component
3935 Details : NFSD is looking up ".." which was broken in ext3 directories
3936 that had grown large enough to become hashed.
3939 Frequency : Clusters with multiple interfaces not on the same subnet
3941 Description: Nodes will repeatedly try to reconnect to an interface which it
3942 cannot reach and report an error to the log.
3943 Details : Extra peer list entries will be created by lconf with some peers
3944 unreachable. lconf now validates the peer before adding it.
3947 Frequency : Only if a default stripe is set on the filesystem root.
3949 Description: Setting a default stripe on the filesystem root prevented the
3950 filesystem from being remounted.
3951 Details : The client was sending extra request flags in the root getattr
3952 request and did not allocate a reply buffer for the dir EA.
3955 Frequency : occasional, higher if lots of files are accessed by one client
3956 Bugzilla : 6159, 6097
3957 Description: Client trips assertion regarding lsm mismatch/magic
3958 Details : While revalidating inodes the VFS looks up inodes with ifind()
3959 and in rare cases can find an inode that is being freed.
3960 The ll_test_inode() code will free the lsm during ifind()
3961 when it finds an existing inode and then the VFS later attaches
3962 this free lsm to a new inode.
3966 Bugzilla : 6422, 7030
3967 Description: MDS deadlock between mkdir and client eviction
3968 Details : Creating a new file via mkdir or mknod (starting a transaction
3969 and getting the ns lock) can deadlock with client eviction
3970 (gets ns lock and trying to finish a synchronous transaction).
3973 Frequency : occasional
3974 Description: While starting a server, the fsfilt_ext3 module could not be
3976 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3977 kernels. Previously, lconf would still use the ext3 name
3978 when trying to load modules. Now, it will correctly use
3979 ext3 on 2.4 and ldiskfs on 2.6.
3981 Severity : enhancement
3982 Description: The default stripe count has been changed to 1
3983 Details : The interpretation of the default stripe count (0, to lfs
3984 or lmc) has been changed to mean striping across a single
3985 OST, rather than all available. For general usage we have
3986 found a stripe count of 1 or 2 works best.
3988 Severity : enhancement
3989 Description: Add support for compiling against Cray portals.
3990 Details : Conditional compiling for some areas that are different
3994 Frequency : occasional
3995 Bugzilla : 6409, 6834
3996 Description: Creating files with an explicit stripe count may lead to
3997 a failed assertion on the MDS
3998 Details : If some OSTs are full or unavailable, creating files may
3999 trigger a failed assertion on the MDS. Now, Lustre will
4000 try to use other servers or return an error to the
4004 Frequency : occasional
4006 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4007 caused lock timeout during readahead (since 1.4.2).
4008 Details : Processes doing readahead might match a lock that hasn't been
4009 granted yet if there are overlapping and conflicting lock
4010 requests. The readahead process waits on ungranted lock
4011 (original lock is CBPENDING), while OST waits for that process
4012 to cancel CBPENDING read lock and eventually evicts client.
4014 Severity : enhancement
4016 Description: Initial enabling of flock support for clients
4017 Details : Implements fcntl advisory locking and file status functions.
4018 This feature is provided as an optional mount flag (default
4019 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4020 locking are implemented yet, and those that are are not guaranteed
4021 to be completely correct in production environments.
4022 mount -t lustre -o [flock|noflock] ...
4025 Frequency : occasional
4027 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4028 assertion in the ext3 JBD (journaling) layer.
4029 Details : The b_committed_data struct is protected by the big kernel lock
4030 in 2.4 kernels, serializing journal_commit_transaction() and
4031 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4032 access to this struct. In 2.6 kernels there is finer grained
4033 locking to improve SMP performance of the JBD layer.
4037 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4039 -----------------------------------------------------------------------------
4041 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4043 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4044 versions of Lustre. Please update all servers and clients to
4045 version 1.4.2 or later at the same time. You must also run
4046 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4047 to update the configuration logs.
4049 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4050 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4051 - swab llog records of type '0' so we get proper header size/idx (5861)
4052 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4053 - fix rename of one directory over another leaking an inode (5953)
4054 - avoid SetPageDirty on 2.6 (5981)
4055 - don't re-add just-being-destroyed locks to the waiting list (5653)
4056 - when creating new directories, inherit the parent's custom
4057 striping settings if present parent (3048)
4058 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4059 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4060 - don't hold client locks on temporary worklist from l_lru (5666)
4061 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4062 - automatically evict dead clients (5921)
4063 - Update file size properly in create+truncate+fstat case (6196)
4064 - Do not unhash mountpoint dentries, do not allow removal of
4066 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4067 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4068 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4069 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4070 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4071 - Correctly handle reads of files with no objects (6243)
4072 - lctl recover will also mark a device active if deactivate used (5933)
4074 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4075 - allow --write-conf on an MDS with different nettype than client (5619)
4076 - don't write config llogs to MDS for mounts not from that MDS (5617)
4077 - lconf should create multiple TCP connections from a client (5201)
4078 - init scripts are now turned off by default; run chkconfig --on
4079 lustre and chkconfig --on lustrefs to use them
4080 - upcalls are no longer needed for clients to recover to failover
4082 - add --abort-recovery option to lconf to abort recovery on device
4084 - add support for an arbitrary number of OSTs (3026)
4085 - Quota support protocol changes.
4086 - forward compatibility changes to wire structs (6007)
4087 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4088 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4089 - improved "lustre status" script
4090 - initialize blocksize for non-regular files (6062)
4091 - added --disable-server and --disable-client configure options (5782)
4092 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4093 - Vanilla 2.4.29 support
4094 - increase maximum number of obd devices to 520 (6242)
4095 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4096 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4097 SLES 9 2.6.5 kernels
4098 - limit stripes per file to 160 (the maximum EA size) (6093)
4100 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4103 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4104 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4105 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4106 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4107 - small ext3 extents cleanups and fixes (5733)
4108 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4109 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4110 - avoid needless client->OST connect, fix handle mismatch (5317)
4111 - fix DLM error path that led to out-of-sync client, long delays (5779)
4112 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4113 - fix several locking issues related to i_size (5492,5624,5654,5672)
4114 - don't move pending lock onto export if it is already evicted (5683)
4115 - fix kernel oops when creating .foo in unlinked directory (5548)
4116 - fix deadlock in obdfilter statistics vs. object create (5811)
4117 - use time_{before,after} to avoid timer jiffies wrap (5882)
4118 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4119 - Do not leak request if server was not able to process it (5154)
4120 - If mds_open unable to find parent dir, make that negative lookup(5154)
4121 - don't create new directories with extent-mapping (5909, 5936)
4123 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4124 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4125 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4126 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4127 - support for RHEL4 kernel on i686 (5773)
4128 - provide error messages when incompatible logs are encountered (5898)
4130 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4131 * version 1.4.0.10 (1.4.1 release candidate 1)
4133 - don't keep a lock reference when lock is not granted (4238)
4134 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4135 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4136 - drop import inflight refcount on signal_completed_replay error (5255)
4137 - unlock page after async write error during send (3677)
4138 - handle missing objects in filter_preprw_read properly (5265)
4139 - no transno return for symlink open, don't save no-trasno open (3440)
4140 - don't try to complete elan receive that already failed (4012)
4141 - free RPC server reply state on error (5406)
4142 - clean up thread from ptlrpc_start_thread() on error (5160)
4143 - readahead could read extra page into cache that wasn't ejected (5388)
4144 - prevent races in class_attach/setup/cleanup/detach (5260)
4145 - don't dereference de->d_inode after l_dput of de (5458)
4146 - use "int" for stripe value returned from lock_to_stripe (5544)
4147 - mballoc allocation and error-checking fixes in 2.6 (5504)
4148 - block device patches to fix I/O request sizes in 2.6 (5482)
4149 - look up hostnames for IB nals (5602)
4150 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4151 - don't start multiple acceptors for the same port (5277)
4152 - fix incorrect LASSERT in mds_getattr_name (5635)
4153 - export a proc file for general "ping" checking (5628)
4154 - fix "lfs check" to not block when the MDS is down (5628)
4156 - service request history (4965)
4157 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4158 - create an "evict_client" /proc entry on OSTs, like the MDS has
4159 - fix mount usage message, return errors per mount(8) (5168)
4160 - change grep [] to grep "[]" in tests so they work in more UMLs
4161 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4162 - remove ancient LOV_MAGIC_V0 EA support (5047)
4163 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4164 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4165 - allow readahead allocations to fail when low on memory (5383)
4166 - mmap locking landed again, after considerable improvement (2828)
4167 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4169 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4172 - send OST transaction number in read/write reply to free req (4966)
4173 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4174 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4175 - the watchdog thread now runs as interruptible (5246)
4176 - flock/lockf fixes (but it's still disabled, pending 5135)
4177 - don't use EXT3 constants in llite code (5094)
4178 - memory shortage at startup could cause assertion (5176)
4180 - reorganization of lov code
4181 - single portals codebase
4183 - add extents/mballoc support (5025)
4184 - direct I/O reads in the obdfilter (4048)
4185 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4187 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4190 - send OST transaction number in read/write reply to free req (4966)
4191 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4192 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4193 - the watchdog thread now runs as interruptible (5246)
4194 - handle missing objects in filter_preprw_read properly (5265)
4195 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4196 - drop import inflight refcount on signal_completed_replay error (5255)
4197 - unlock page after async write error during send (3677)
4198 - return original error code on reconstructed replies (3761)
4199 - no transno return for symlink open, don't save no-trasno open (3440)
4201 - add pid to ldlm debugging output (4922)
4202 - bump the watchdog timeouts -- we can't handle 30sec yet
4203 - extra debugging for orphan dentry/inode bug (5259)
4205 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4208 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4209 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4210 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4211 - fix NULL dereference in /proc/sys/portals/routes (4827)
4212 - allow failed mdc_close() operations to be interrupted (4561)
4213 - stop precreate on OST before MDS would time out on it (4778)
4214 - don't send partial-page writes before EOF from client (4410)
4215 - discard client grant for sub-page writes on large-page clients (4520)
4216 - don't free dentries not owned by NFS code, check generation (4806)
4217 - fix lsm leak if mds_create_objects() fails (4801)
4218 - limit debug_daemon file size, always print CERROR messages (4789)
4219 - use transno after validating reply (3892)
4220 - process timed out requests if import state changes (3754)
4221 - update mtime on OST during writes, return in glimpse (4829)
4222 - add mkfsoptions to LDAP (4679)
4223 - use ->max_readahead method instead of zapping global ra (5039)
4224 - don't interrupt __l_wait_event() during strace
4226 - add software watchdogs to catch hung threads quickly (4941)
4227 - make lustrefs init script start after nfs is mounted
4228 - fix CWARN/ERROR duplication (4930)
4229 - return async write errors to application if possible (2248)
4230 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4231 - print NAL number in %x format (4645)
4232 - update barely-supported suse-2.4.21-171 series (4842)
4233 - support for sles 9 %post scripts
4234 - support for building 2.6 kernel-source packages
4235 - support for sles km_* packages
4237 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4240 - ignore -ENOENT errors in osc_destroy (3639)
4241 - notify osc create thread that OSC is being cleaned up (4600)
4242 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4243 - reconstruct ost_handle() like mds_handle() (4657)
4244 - create a new thread to do import eviction to avoid deadlock (3969)
4245 - let lconf resolve symlinked-to devices (4629)
4246 - don't unlink "objects" from directory with default EA (4554)
4247 - hold socknal file ref over connect in case target is down (4394)
4248 - allow more than 32000 subdirectories in a single directory (3244)
4249 - fix blocks count for O_DIRECT writes (3751)
4250 - OST returns ENOSPC from object create when no space left (4539)
4251 - don't send truncate RPC if file size isn't changing (4410)
4252 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4253 - bind to privileged port in socknal and tcpnal (3689)
4255 - rate limit CERROR/CWARN console message to avoid overload (4519)
4256 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4257 - basic mmap support (3918)
4258 - kernel patch series update from b1_4 (4711)
4260 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4263 - avoid crash during MDS cleanup with OST shut down (2775)
4264 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4265 - don't use bad inodes on the MDS (3744)
4266 - dynamic object preallocation to improve recovery speed (4236)
4267 - don't hold spinlock over lock dumping or change debug flags (4401)
4268 - don't zero obd_dev when it is force cleaned (3651)
4269 - print grants to console if they go negative (4431)
4270 - "lctl deactivate" will stop automatic recovery attempts (3406)
4271 - look for existing locks in ldlm_handle_enqueue() (3764)
4272 - don't resolve lock handle twice in recovery avoiding race (4401)
4273 - revalidate should check working dir is a directory (4134)
4275 - don't always mark "slow" obdfilter messages as errors (4418)
4277 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4280 - don't close LustreDB during write_conf until it is done (3860)
4281 - fix typo in lconf for_each_profile (3821)
4282 - allow dumping logs from multiple threads at one time (3820)
4283 - don't allow multiple threads in OSC recovery (3812)
4284 - fix debug_size parameters (3864)
4285 - fix mds_postrecov to initialize import for llog ctxt (3121)
4286 - replace config semaphore with spinlock (3306)
4287 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4288 - don't allow enqueue to complete on a destroyed export (3822)
4289 - down write_lock before checking llog header bitmap (3825)
4290 - recover from lock replay timeout (3764)
4291 - up llog sem before sending rpc (3652)
4292 - reduce ns lock hold times when setting kms (3267)
4293 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4294 - fix NULL deref and obd_dev leak on setup error (3312)
4295 - replace some LBUG about llog ops with error handling (3841)
4296 - don't match INVALID dentries from d_lookup and spin (3784)
4297 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4298 - fix invalid assertion in ptlrpc_set_wait (3880)
4300 - add libwrap support for the TCP acceptor (3996)
4301 - add /proc/sys/portals/routes for non-root route listing (3994)
4302 - allow setting MDS UUID in .xml (2580)
4303 - print the stack of a process that LBUGs (4228)
4305 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4308 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4309 - make sure to unset replay flag from failed open requests (3440)
4310 - if default stripe count is 0, use OST count for inode size (3636)
4311 - update parent mtime/ctime on client for create/unlink (2611)
4312 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4313 - free recovery state on server during a forced cleanup (3571)
4314 - unregister_reply for resent reqs (3063)
4315 - loop back devices mounting and status check on 2.6 (3563)
4316 - fix resource-creation race that can provoke i_size == 0 (3513)
4317 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4318 - more debugging for page-accounting assertion (3746)
4319 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4320 - avoid deadlock after precreation failure (3758)
4321 - fix race and lock order deadlock in orphan handling (3450, 3750)
4322 - add validity checks when grabbing inodes from l_ast_data (3599)
4324 - add /proc/.../recovery_status to obdfilter (3428)
4325 - lightweight CDEBUG infrastructure, debug daemon (3668)
4326 - change default OSC RPC parameters to be better on small clusters
4327 - turn off OST read cache for files smaller than 32MB
4328 - install man pages and include them in rpms (3100)
4329 - add new init script for (un)mounting lustre filesystems (2593)
4330 - run chkconfig in %post for init scripts (3701)
4331 - drop scimac NAL (unmaintained)
4333 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4336 - clean kiobufs before and after use (3485)
4337 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4338 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4339 - change init script to fail more clearly if not run as root (1528)
4340 - allow clients to reconnect during replay (1742)
4341 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4342 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4343 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4344 - fix cond_resched (3554)
4345 - only evict unfinished clients after recovery (3515)
4346 - allow bulk resend, prevent data loss (3570)
4347 - dynamic ptlrpc request buffer allocation (2102)
4348 - don't allow unlinking open directory if it isn't empty (2904)
4349 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4350 - remove extraneous obd dereference causing LASSERT failure (3334)
4351 - don't use get_cycles() when creating temp. files on the mds (3156)
4352 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4353 - handle EEXIST for set-stripe, set proper directory name (3336)
4355 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4356 - fix ksocknal_fmb_callback() error messages (2918)
4358 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4361 - don't copy lvb into (possibly NULL) reply on error (2983)
4362 - don't deref dentry after dput, don't free lvb on error (2922)
4363 - use the kms to determine writeback rpc length (2947)
4364 - increment oti_logcookies when osc is inactive (2948)
4365 - update client's i_blocks count via lvb messages (2543)
4366 - handle intent open/close of special files properly (1557)
4367 - mount MDS with errors=remount-ro, like obdfilter (2009)
4368 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4369 - don't use cancelling-locks' kms values (2947)
4370 - use highest lock extent for kms, not last one (2925)
4371 - don't dereference ERR_PTR() dentry in error handling path (3107)
4372 - fix thread race in portals_debug_dumplog() (3122)
4373 - create lprocfs device entries at setup instead of at attach (1519)
4374 - common AST error handler, don't evict client on completion race (3145)
4375 - zero nameidata in detach_mnt in 2.6 (3118)
4376 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4377 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4378 - initialize RPC timeout timer earlier for 2.6 (3219)
4379 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4380 - print nal/nid for unknown nid (3258)
4381 - additional checks for oscc recovery before doing precreate (3284)
4382 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4383 - don't crash in mdc_close for bad permissions on open (3285)
4384 - zero i_rdev for non-device files (3147)
4385 - clear page->private before handing to FS, better assertion (3119)
4386 - tune the read pipeline (3236)
4387 - fix incorrect decref of invalidated dentry (2350)
4388 - provide read-ahead stats and refine rpc in flight stats (3328)
4389 - don't hold journal transaction open across create RPC (3313)
4390 - update atime on MDS at close time (3265)
4391 - close LDAP connection when recovering to avoid server load (3315)
4392 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4393 - don't leak open file on MDS after open resend (3325)
4394 - serialize filter_precreate and filter_destroy_precreated (3329)
4395 - loop device shouldn't call sync_dev() for nul device (3092)
4396 - clear page cache after eviction (2766)
4397 - resynchronize MDS->OST in background (2824)
4398 - refuse to mount the same filesystem twice on same mountpoint (3394)
4399 - allow llmount to create routes for mounting behind routers (3320)
4400 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4401 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4402 - fix rare problem with rename on htree directories (3417)
4404 - allow default OST striping configuration per directory (1414)
4405 - fix compilation for qswnal for 2.6 kernels (3125)
4406 - increase maximum number of MDS request buffers for large systems
4407 - change liblustreapi to be useful for external progs like lfsck (3098)
4408 - increase local configuration timeout for slow disks (3353)
4409 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4411 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4414 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4415 - don't overwrite extent policy data in reply if lock was blocked (2901)
4416 - drop filter export grants atomically with removal from device (2663)
4417 - del obd_self_export from work_list in class_disconnect_exports (2908)
4418 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4419 - swab reply message in mdc_close, other PPC fixes (2464)
4420 - fix destroying of named logs (2325)
4421 - overwrite old logs when running lconf --write_conf (2264)
4422 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4423 - fix race in target_handle_connect (2898)
4424 - mds_reint_create() should take same inode create lock (2926)
4425 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4426 - don't close files for self_export to avoid uninitialized obd (2936)
4427 - allow MDS with the same name as client node (2939)
4428 - hold dentry reference for closed log files for unlink (2325)
4429 - reserve space for all logs during transactions (2059)
4430 - don't evict page beyond end of stripe extent (2925)
4431 - don't oops on a deleted current working directory (2399)
4432 - handle hard links to targets without a parent properly (2517)
4433 - don't dereference NULL lock when racing during eviction (2867)
4434 - don't grow lock extents when lots of conflicting locks (2919)
4436 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4439 - account for cache space usage on clients to avoid data loss (974)
4440 - lfsck support in lustre kernel code (2349)
4441 - reduce journal credits needed for BRW writes (2370)
4442 - orphan handling to avoid losing space on client/server crashes
4443 - ptlrpcd can be blocked, stopping ALL progress (2477)
4444 - use lock value blocks to assist in proper KMS, faster stat (1021)
4445 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4446 - recovery for initial connections (2355)
4447 - fixes for mds_cleanup_orphans (1934)
4448 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4449 - block all file creations until orphan recovery completes (1901)
4450 - client remove rq_connection from request struct (2423)
4451 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4452 - recovery timer race (2670)
4453 - mdc_close recovey bug (2532)
4454 - ptlrpc cleanup bug (2710)
4455 - mds timeout on local locks (2588)
4456 - namespace lock held during RPCs (2431)
4457 - handle interrupted sync write properly (2503)
4458 - don't try to handle a message that hasn't been replied to (2699)
4459 - client assert failure during cleanup after abort recovery (2701)
4460 - leak mdc device after failed mount (2712)
4461 - ptlrpc_check_set allows timedout requests to complete (2714)
4462 - wait for inflight reqs when ptlrpcd finishes (2710)
4463 - make sure unregistered services are removed from the srv_list
4464 - reset bulk XID's when resending them (caught by 1138 test)
4465 - unregister_bulk after timeout
4466 - fix lconf error (2694)
4467 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4468 - readahead locks pages, leaves pending causing memory pressure (2673)
4469 - increase OST request buffers to 4096 on large machines (2729)
4470 - fix up permission of existing directories in simple_mkdir (2661)
4471 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4472 - don't assign transno to errored transactions (2742)
4473 - don't delete objects on OST if given a bogus objid from MDS (2751)
4474 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4475 - if rq_no_resend, then timeout request after recovery (2432)
4476 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4477 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4478 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4479 - MDS recovery shouldn't delete objects that it has given out (2730)
4480 - if enqueue arrives after completion, don't clobber LVB (2819)
4481 - don't unlock pages twice when trigger_group_io returns error (2814)
4482 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4483 - don't write pages to disk if there was an error (1450)
4484 - don't ping imports that have recovery disabled (2676)
4485 - take buffered bytes into account when balancing socknal conn (2817)
4486 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4487 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4488 - remove little-endian swabbing of llog records (1987)
4489 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4490 - retry reposting request buffers if they fail (1191)
4491 - grow extent at grant time to avoid granting a revoked lock (2809)
4492 - lock revoke doesn't evict page if covered by a second lock (2765)
4493 - disable VM readahead to avoid reading outside lock extents (2805)
4495 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4496 - updated kernel patches for hp-2.4.20 kernel (2681)
4498 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4501 - fix truncated write corruption (2366)
4502 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4504 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4505 - ensure len(uuid) < 37 in lmc (1171)
4506 - fix ia64 OOPS in llog_test (2255)
4507 - zero end of page at obdfilter for partial page writes (2648)
4508 - don't leave stale dentries around after renames (bug 2428)
4509 - fix timeouts when evicting a client with a single lock held (2642)
4510 - set deadline for the initial HELLO message to drain (2634)
4511 - print out dotted-quad IP addresses in the socknal (2302)
4513 - additional debugging for MDS client eviction problem (2443)
4514 - fix mkfsoptions support for osts (2603, 2604)
4516 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4519 - add series for the vanilla 2.6.0 kernel
4520 - add series for the vanilla 2.4.24 kernel
4521 - add series for a cray x86/64 UL kernel drop
4522 - fix xattr patches for the vanilla 2.4.19 series
4524 - generate true UUIDs in lmc (1171)
4525 - have portals stack dumping break in UML (2466)
4526 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4527 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4528 - wake callback waiting threads on client eviction (2460)
4529 - Add --ptldebug and --subsystem to lmc (1719)
4530 - update assertion to allow safe interrupt allocation
4531 - set rq_no_resend for cancel requests (2432)
4532 - recalculate ptlrpcd timeout after resend (2494)
4533 - call vfs_rmdir when removing pending directories (2368)
4534 - fix renaming a file to itself (2429)
4535 - lmc creates a default one-stripe lov (2454)
4536 - expand procfs space to handle large clusters (2326)
4537 - increase UML stack to avoid overflow
4538 - update lconf's list of debug and subsystem masks
4539 - fix lfs find --obd (2510)
4540 - /proc tunable for disabling filter read caching (2591)
4541 - stop rpm packages from altering slapd.conf (2301)
4542 - disable nagle in the socknal under 0conf (2578)
4543 - choose mds inode size based on stripe count (2572)
4544 - fix kernel-source rpm problems (2516)
4546 - add --disable-doc to avoid pdf generation (2421)
4547 - update documentation, tests, type-os, comments
4548 - avoid format warnings on ia64
4549 - remove the TOE NAL
4550 - tiny code cleanups by removing unused fields
4552 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4555 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4556 - remove the most blatant lies from BUILDING file (2371)
4557 - change default debug level to reasonable production setting
4558 - reduce client side cache size to reduce cache flush time
4559 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4560 - make TCP zerocopy and pinger support enabled by default (2476)
4561 - sync writes completed after process exits caused crashes (2319)
4562 - maintain correct mount count on the MDS (2356)
4563 - backout 1557, because 2316 wasn't really fixed
4564 - better file I/O statistics gathering in /proc
4565 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4566 - another kernel patch to fix zero-copy TCP function export
4567 - don't take duplicate lock when processing re-sent getattr (2420)
4568 - lctl uses obd_self_export instead of creating new conn (2353)
4569 - MDS/OST recovery case which requires object creation asserted (2425)
4570 - move lfs from /usr/sbin to /usr/bin in packages
4571 - fix race between mds_client_add and mds_client_free (2417)
4572 - use kmalloc instead of slabs in portals (2430)
4573 - don't create duplicate records when a failover MDS is present (2442)
4574 - remove unnecessary mount age check (2332)
4575 - don't remove directory inodes from locks prematurely (2451)
4576 - don't break if MDS service name is the same as hostname (2103)
4577 - fix races in client write RPC generation when cache full (2482)
4579 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4582 - remove now-unused request->rq_obd (278)
4583 - if an allocation fails, print out how much memory we've used (1933)
4584 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4585 - add the "configurable stack size" patch to most series files (1256)
4586 - ability to write large log records, for 100+ OST configs (2306)
4587 - fix NULL deref when filter_prep fails (2314)
4588 - fix operator precedence error in filter_sync
4589 - dynamic allocation of socknal TX descriptors (2315)
4590 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4591 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4592 - ensure that CWARN messages in Portals always get to the syslog
4593 - __init/__exit are not for prototype decls (ldlm_init/exit)
4594 - x86-64 compile warning fixes
4595 - fix gateway LMC keyword conflict (2318)
4596 - fix MDS lock inversions in getattr/reint paths (1844)
4597 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4598 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4599 - DLM race condition prevented some lock evictions (2328)
4600 - ENOMEM detection and retry on socknal sends (2230)
4601 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4602 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4603 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4604 - better error messages when a client is rejected during recovery (1505)
4605 - avoid cancelling locks which were never granted, after failure (2330)
4606 - fix i_sem/journal inversion in mds_client_add (2333)
4607 - fix truncate/getattr lock cycle deadlock (2334)
4608 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4609 - fix two rare exit paths which could leak an l_lock() ref (2321)
4610 - fencepost error in MDS/OST orphan recovery (2226)
4611 - make log record alignment 8 bytes (1988)
4612 - lstripe now fails when requested offset > ost_count (2237)
4613 - ensure that all kernel series have a complete list.h (1607)
4614 - fix crashes in special-file operations (2316)
4615 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4617 - allow configurable automake binary, for testing new versions
4618 - small update to the lfs documentation
4620 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4622 * fix negative export reference count in fsfilt_sync (2312)
4624 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4625 * release candidate 0.9.1
4627 - orphans are moved into the PENDING directory for possible recovery
4628 - replayed opens now open by fid for orphan/rename safety (1042)
4629 - last close of an orphan inode generates a transno (683)
4630 - chdir() and mount() now pin the directory entry (1020)
4631 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4632 - discard very old requests without processing them (1502)
4633 - remove some common, well-understood CERRORs (1505)
4634 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4635 - clear "grant" flags in OST replies until OST grant code lands (1644)
4636 - fix read performance by not clobbering i_blksize on client (1598)
4637 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4638 - make LRU size a /proc tunable, clears locks when reduced (707)
4639 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4640 - server locks take references on exports now (1558)
4641 - build fixes for 2.4.20-rh trees (1663)
4642 - return an error from lov_create if all OSCs are inactive (1751)
4643 - fix import levels when a reconnect happens without a timeout (1597)
4644 - exit early from mds_open if we get a lookup error (1749)
4645 - partial page read at EOF wouldn't wait for disk before sending (1642)
4646 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4647 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4648 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4649 - write/truncate lock inversion (1639)
4650 - Don't auto-load obdclass, portals modules during cleanup (1495)
4651 - fix timestamps from jumping to "now" (1763)
4652 - extra journal assertions (1648)
4653 - add an extra multiunlink test (1771)
4654 - fix read_record/write_record API (1776)
4655 - fix leak of offset_extent, possible incorrect i_size later (1772)
4656 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4657 - Debugging for the kqswnal_get_idle_tx problems (1820)
4658 - Allow recovery to be attempted multiple times (1536)
4659 - Write out MDS last_rcvd file after it is first created (1600)
4660 - Fix tx_descriptor leak in failed transmit situations (1827)
4661 - ext3 journaling fixes for assertion failure after IO error (1871)
4662 - class_export_put() on freed export after completion AST error (1896)
4663 - Fix revalidate looping in VFS (1322)
4664 - Don't access a freed export during MDS_REINT timeout (1521)
4665 - Add open-unlink recovery support on the MDS (1673,1764)
4666 - Return an error if no MDS data was read from last_rcvd (1946)
4667 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4668 - Don't setup a disk device that doesn't match exported UUID (317)
4669 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4670 - avoid committing NULL handle in force close
4671 - local.sh is now a one-stripe LOV configuration
4672 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4673 - fix ext3 htree duplicate directory entry corruption (1516)
4674 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4675 - update to new LOV EA format (2097)
4676 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4678 2003-06-15 Phil Schwan <phil@clusterfs.com>
4681 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4682 - per-import/export recovery handling (958, 931, 959)
4683 - multiple last-rcvd slots, for serving multiple FSes (949)
4684 - connections are again shared between multiple imp/exports (963, 964)
4685 - "umount -f" would hang if any requests needed to be sent (393, 978)
4686 - avoid pinning large req buffer by copying for queued messages (989)
4687 - add "uuid" to "lctl device" command to help upcalls (991)
4688 - "open" RPCs with transnos would confuse recovery counters (1037)
4689 - do proper endian conversion of all wire messages (288, 340, 891)
4690 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4691 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4692 - fix ptlrpc_connection leak in target_handle_connect (1174)
4693 - fix import refcounting bug in OST and MDS cleanup (1134)
4694 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4695 - fix very unlikely obd_types race condition (501)
4696 - remove osc_open hack for echo_client (1187)
4697 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4698 - a failure in read_inode2 leads to deadlock (1139)
4699 - cancel ack-locks as soon as transaction is committed (1072)
4700 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4701 - make sure to commitrw after any preprw to avoid deadlock (1162)
4702 - failing to execute a file in a lustre FS would lock inode (1203)
4703 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4704 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4705 - obd_uuid2tgt would walk off the end of the list (1255)
4706 - on IA64 the osc would give portals incorrect bulk size (1258)
4707 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4708 - fix lock inversion caused by new llite matching code (1282)
4709 - limit the number of dirty pages on a client to 10MB (1286)
4710 - timed out locks were not being corrected cancelled (1289)
4711 - fix O_DIRECT above 4GB on IA-32 (1292)
4712 * major user-visible changes
4713 - fail out/fail over policy now controlled by the upcall (993)
4715 - add OBD_PING to check server availability and failure (954)
4716 - lustre messages are now sent in sending host order (288, 340, 891)
4717 - add eadatalen to MDS getattr reply (340)
4718 - OST read replies may contain second buffer, with per-page status (593)
4720 2003-03-11 Phil Schwan <phil@clusterfs.com>
4723 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4724 - failed open()s could cause deadlock; fixed (867, 869)
4725 - stop cancelling OST locks when files are closed (481)
4726 - overlapping XID spaces caused network corruption (851, 853)
4727 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4728 - setattr_raw added, to do single-RPC, server-side setattrs
4729 - lmc/lconf syntax change for OST UUIDs
4730 - fix crashy race condition between ptlrpc_free_req and osc_close
4731 - don't use request in mdc_enqueue if we hit a timeout (889)
4732 - don't set the inode i_size for regular files from the MDS (896)
4733 - handle out of order completion AST (842)
4734 - don't LBUG if a lock request times out after receiving AST (913)
4735 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4736 - if a bad lock AST arrives, send an error instead of dropping entirely
4737 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4738 - fix leak in bulk IO when only partially completed (899, 900, 926)
4739 - fix O_DIRECT for ia64 (55)
4740 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4741 - C-z after timeout could hang a process forever; fixed (977)
4743 - client-side I/O cache (678, 924, 929, 941, 970)
4745 - READPAGE and SETATTRs which don't take server-side locks get
4748 2003-02-11 Phil Schwan <phil@clusterfs.com>
4751 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4752 - Fix fsfilt_extN_readpage() to read a full page of directory
4753 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4754 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4755 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4756 - Most intent operations were removed, in favour of a new RPC mode
4757 that does a single RPC to the server and bypasses most of the VFS
4758 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4759 - Aggressively cancel local locks on DLM servers
4760 - mds_reint_unlink sends EA to the client if it's the last nlink.
4761 client uses that EA to unlink OST objects.
4762 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4763 - recursive symlinks were fixed (439)
4764 - fixed NULL deref in DEBUG_REQ
4765 - filter_update_lastobjid no longer calls sync, which annoyed extN
4766 - fixed multi-client small-writes to a single file problem (445)
4767 - fixed mtime updates during file writes (607)
4768 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4769 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4770 - fixed Linux OST size reporting problem (444, 656)
4771 - OST now updates object mtime with writes or setattr (607, 619)
4772 - client verifies file size before zeroing page past EOF (445)
4773 - OST now writes last allocated objid to disk with allocation (108)
4774 - LOV on echo now works (409)
4776 - mds_reint_unlink sends a new buffer, with the EA included. this
4777 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4778 set if a regular file was being unlinked, and it was the last link
4779 - use PtlGet from the target for bulk writes (315)
4780 - OST now updates object mtime with writes or setattr (607, 619)
4781 - LDLM now has a grant-time callback to revalidate locked items, if
4783 - Many MDS operations were reorganized to combat race conditions
4785 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4786 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4788 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4791 - Fully reactivate OST imports after reconnection (512, others)
4792 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4793 - More graceful error handling for truncating on dead OST (515)
4794 - Don't error out unless we're actually accessing dead stripes (474)
4795 - Fix garbage sizes when stripes are missing (410)
4796 - LRU counters were broken, causing constant lock purge (433, 432)
4797 - garbage on read from stripes with failed OSTs (441)
4798 - mark OSCs as active before reconnecting during recovery (438)
4799 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4800 - lfind did not preserve OST order in output (443)
4801 - symlinks cause hung clients, incorrect data (439)
4802 - stop dereferencing request after dropping refcount (457)
4803 - don't LASSERT(spin_is_locked) on non-SMP (455)
4804 - fixes for many rename() bugs
4805 - fstat didn't correctly synchronize attributes (399)
4806 - server must handle lock cancellation during blocking AST prep (487)
4807 - bulk descriptors were free()d too soon (511)
4808 - fix paths in lconf, which would load incorrect modules (451, 507)
4809 - fix confusing lconf 'host not found' error message (386)
4810 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4811 - fix race condition in mdc_blocking_ast() for inode access (526)
4812 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4813 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4814 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4815 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4816 - fix lstripe in several ways (kernel side) (527)
4817 - fix request leak in ldlm_cli_enqueue (262)
4818 - incorrect OSC was marked inactive after OST failure
4819 - call mds_fs_cleanup before unmounting filesystem (524)
4820 - fix races between taking ns_lock and ldlm_lock_change_resource
4821 - fix races updating LOV export open file list
4822 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4823 - fix recovery NULL deref in ldlm_cli_cancel_unused
4824 - fix some DLM races by using new hash table for lock handles (419)
4825 - permit the client to specify desired inodes, at replay
4826 - duplicate requests when we queue them for replay reintegration
4827 - fix last_rcvd offset calculation
4828 - sync after each recovered transaction, so we always make progress
4829 - never, not always, ERESTART requests without transnos
4830 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4832 - skip replay if the MDS says that the client is already connected
4833 - don't check for a recovery-enabled export to match lctl's UUID
4834 - don't INC_USE_COUNT for phantom exports
4835 - don't crash when cleaning up phantom exports (567)
4836 - don't double-finish or set replay data for errored mdc_open requests
4837 - abort requests when they time out, so we don't get old replies
4838 - send/receive replies for AST messages again
4839 - if the client says that it doesn't have the lock, cancel it on the
4841 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4842 mark it as destroyed, it will all work out when decref is called
4843 - fix module use counts (22, 581)
4845 - ASTs now expect a reply (server cancels lock on error reply)
4847 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4850 - fix many simultaneous client startup (392)
4851 - fix dentry->d_it clobbering
4852 - credentials weren't being shipped for readdir/getattr operations
4853 - remove invalid assertions triggered during some concurrent MD
4855 - proper Lustre versions added (336, 389)
4856 - fix memory leak for create error case (398)
4857 - fix LOV locking bug that would get cli/srv out of sync
4858 - fix echo client over LOV (409)
4859 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4860 - fix double-O_EXCL intent crash (424)
4861 - avoid sending multiple lock CANCELs (352)
4863 - MDS can do multi-client recovery (modulo bugs in new code)
4865 - many updates, edits, cleanups
4867 2002-11-18 Phil Schwan <phil@clusterfs.com>
4870 - fix null d_it dereference (346)
4871 - fix full OST/dbench hang (333)
4872 - fix permission problem with file removal (286)
4873 - fix removal of OSCs from LOV when they fail
4874 - fix NULL deref during bulk timeout (214)
4875 - fix problems related to multiple filesystems on one MDS (241)
4876 - fixed serious subtle metadata locking bugs
4877 - free locks on clients when inodes are removed due to memory
4879 - fix inode pointer in lock data (285)
4880 - partial support for multiple MDS on a single host (241)
4881 - data locks weren't cancelled at clear_inode time (290, 311)
4882 - intent locks could lead to unbounded lock growth (205)
4883 - added a maximum lock count, an LRU list, and a flusher
4884 - fix multiple rename (365)
4885 - properly abstracted the echo client
4886 - OSC locked 1 byte too many; fixed
4887 - rewrote brw callback code:
4888 - fixed recovery bugs related to LOVs (306)
4889 - fixed too-many-pages-in-one-write crash (191)
4890 - fixed (again) crash in sync_io_timeout (214)
4891 - probably fixed callback-related race (385)
4893 - Add capability to MDS protocol
4894 - LDLM cancellations and callbacks on different portals
4896 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4899 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4900 - defer open object destruction to close time (601981, 138)
4901 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4902 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4903 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4904 - don't call obd_setattr in ll_file_release for destroyed objects
4905 * protocol change to lustre_msg: move |version| and add |flags|
4906 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4907 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4908 * added replay of create, unlink, link and rename operations during
4909 MDS failover; recovery should be much more robust now
4910 * remove failed OSCs from LOVs (only lov_create uses this so far)
4911 * the lustre-HOWTO was brought (more) up to date (582544)
4913 2002-10-23 Phil Schwan <phil@clusterfs.com>
4916 - in-use dentries weren't being reused properly (617851)
4917 - prevent multiple LDLM setup (599178)
4918 - fix LOV size calculations for truncate (617853)
4919 - fix client handling of MDS intent errors (POSIX)
4920 - fix permission bug in lovstripe.c test (624321)
4921 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4922 - truncate past end of file could corrupt data
4923 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4924 - a race in recovery could return ETIMEDOUT to apps (623947)
4925 - building outside the source directory was fixed
4926 * the lustre-HOWTO was brought (more) up to date (582544)
4927 * major progress was made on recovery functionality
4929 2002-10-10 Phil Schwan <phil@clusterfs.com>
4932 - recovery deadlock fix
4933 - rm -rf causes LBUG fix (617817)
4934 - file open by multiple tasks fix (618962)
4935 - directory permissions bugs (602707 and 620007)
4936 - journal_stop fixed with locking (611313)
4937 - O_APPEND failures resolved (618273, perhaps 614459)
4938 - lconf PATH fix (619770)
4939 - IA64 build fix (621450)
4940 - RPC buffer sizes scale with amount of memory
4942 2002-10-01 Phil Schwan <phil@clusterfs.com>
4945 - locks would be cancelled without throwing away data pages,
4946 resulting in inconsistent data (605627)
4947 - inode attributes were not always being refreshed (605627, 612449)
4948 - lconf now continues to cleanup after lctl reports an error
4949 - MDS now enforces user permissions (602707)
4950 - lprocfs cleanup fixed, but not yet enabled (614157)
4951 - fixed infinite server hang, should a client not respond to an AST
4952 - avoid going into recovery if user calls readlink() with a buffer
4953 that's too small (613941)
4954 - AST RPCs no longer require replies (614867) -- this may be changed
4955 - don't crash server if client sends an IOV that's too big (611336)
4956 - fixed lock conversion deadlock (611892)
4957 - fixed the following of symlinks (614622)
4958 * recovery: the server can remove locks from a client that dies, other
4959 clients can make progress
4960 * more extN patch fixes
4961 * compile-time configurable ptlrpc buffer allocations
4963 - collaborative read cache document
4964 - Lustre Lite Performance CDR document-in-progress
4966 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4969 - fix typo in patch-2.4.18
4971 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4974 - clear ptlrpc request each time in handle_incoming_request()
4975 - unlink of files now destroys the object on the OST
4977 2002-09-19 Peter Braam <braam@clusterfs.com>
4979 * add hard link support
4980 * change obdfile creation method
4981 * kernel patch changed
4983 2002-09-19 Peter Braam <braam@clusterfs.com>
4986 - stack overflow bug in extN fixed
4988 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4990 * documentation updates
4991 - add man pages for config tools
4992 - update tests/README to describe testing with new config tools
4993 - finish metadata API descriptions
4994 * bug fixes and cleanups
4995 - statfs workaround for 16TB limit
4996 - LOV stripe allocation improved, can stripe on subset of OSTs
4997 - LOV file size/IO offset was wrong for files > 4GB in size
4998 - object EA data was being dropped, caused files to be unreadable
4999 - memory overflow with non-LOV OST caused memory corruption
5000 - fixed regression tests to work with new config tools, obdfilter
5001 - fixed bug when directory size became larger than 1 block
5002 - fixed bug (for single client case) when PWD was deleted
5003 - invalidate local directory pages when doing intent-based ops
5004 - avoid LDLM oops when lock callback contained bad data
5006 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5008 * documentation updates
5009 * bug fixes and cleanups
5010 - configuration tools
5013 - 64-bit compile warnings
5014 - 64-bit internal statfs data
5016 * test_brw on persistent OST devices
5018 * lprocfs (disabled)
5020 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5022 * documentation updates
5023 * bug fixes and cleanups
5024 * configuration tools
5026 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5029 * many small fixes to 0_5_4
5030 * io/network handling
5031 * thinkos in MDS operations
5033 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5036 * crucial basic fixes to 0.5.3
5037 * IOR, Iozone work over Elan
5040 2002-08-07 Phil Schwan <phil@clusterfs.com>
5041 * version 0_5_3, our first alpha
5042 * we use the new Portals iovs
5043 * documentation updates
5044 * bug fixes and cleanups
5045 * small changes in the DLM wire protocol
5047 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5048 * version 0_5_1 with some initial stability,
5049 * locking on MD and file I/O.
5050 * documentation updates
5051 * several bug fixes since 0.5.0
5052 * small changes in wire protocol
5054 2002-07-18 Phil Schwan <phil@clusterfs.com>
5056 * delivered as Lustre Light Alpha
5057 * fixed a crash after handling invalid MDS requests
5058 * fixed directory pages for architectures with non-4k pages sizes
5060 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5061 * release version v0_4_4
5062 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5064 * Fixes a number of interruption problems with OST operations.
5065 * Update documentation for portals header changes
5066 * Move all wire protocol structs/defines to lustre_idl.h
5067 * Fixes symlink length bug.
5068 * Add tcpdump to repository.
5070 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5071 * release version v0_4_3
5072 * Fixes statfs for inodes on extN.
5073 * Fixes bug in runtests which would delete /etc/hosts.
5074 * Use 64-bit object IDs wherever possible (not into VFS though)
5075 Remove ost_get_info, which is unused by lustre, and out of date.
5077 2002-07-03 Peter Braam <braam@clusterfs.com>
5078 * release version v0_4_2 Fixes a lookup error (type not passed)
5079 * move forward to head of Portals
5080 * move forward to latest Lustre kernel
5082 2002-06-25 Peter Braam <braam@clusterfs.com>
5083 * release version v0_4_1. Hopefully stable on single node use.