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).
20 Severity : enhancement
22 Description: Add server support for vanilla-2.6.22.14.
25 Frequency : rare, depends on device drivers and load
27 Description: MDS or OSS nodes crash due to stack overflow
28 Details : Code changes in 1.6.4 increased the stack usage of some functions.
29 In some cases, in conjunction with device drivers that use a lot
30 of stack, the MDS (or possibly OSS) service threads could overflow
31 the stack. One change which was identified to consume additional
32 stack has been reworked to avoid the extra stack usage.
34 Severity : enhancement
36 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
38 Severity : enhancement
40 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
42 Severity : enhancement
44 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
46 Serverity : enhancement
48 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
51 Frequency : rare on shutdown ost
53 Description: Don't allow skipping OSTs if index has been specified.
54 Details : Don't allow skipping OSTs if index has been specified, make locking
55 in internal create lots better.
59 Description: ASSERTION(!PageDirty(page)) failed
60 Details : Wrong check could lead to an assertion failure under specific
65 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
66 Details : access to bitfield in structure is always rounded to long
67 and this produce problem with not atomic change any bit.
71 Description: Lustre make rpms failed.
72 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
73 builds Lustre rpms from tarball.
75 Severity : enhancement
77 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
80 Frequency : rare on shutdown ost
82 Description: If llog cancel was not send before clean_exports phase, this can
83 produce deadlock in llog code.
84 Details : If llog thread has last reference to obd and call class_import_put
85 this produce deadlock because llog_cleanup_commit_master wait when
86 last llog_commit_thread exited, but this never success because was
87 called from llog_commit_thread.
91 Description: NULL lov_tgts causing MDS oops
92 Details : more safe checks for NULL lov_tgts for avoid oops.
94 Severity : enhancement
96 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
98 Severity : enhancement
100 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
104 Description: make mgs_setparam() handle fsname containing dash
105 Details : fsname containing a dash does not work with lctl conf_param
107 Severity : enhancement
109 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
113 Frequency : rare, in recovery and (or) destroy lovobjid file.
114 Description: rewrite lov objid code.
115 Details : Cleanup for lov objid code, remove scability problems and wrong
116 locking. Fix sending last_id into ost.
118 Severity : enhancement
120 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
122 Severity : enhancement
124 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
125 Details : Use d_move_locked instead of __d_move.
129 Frequency : rare, at shutdown
130 Description: access already free / zero obd_namespace.
131 Details : if client_disconnect_export was called without force flag set,
132 and exist connect request in flight, this can produce access to
133 NULL pointer (or already free pointer) when connect_interpret
134 store ocd flags in obd_namespace.
138 Frequency : only at startup
139 Description: not alloc memory with spinlock held.
140 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
141 if any spinlock held.
146 Description: lfs find does not continue on file error
147 Details : Continue other files processing when a file/dir is absent.
151 Description: Inconsistent usage of lustre_pack_reply()
152 Details : Standardize the usage of lustre_pack_reply() such that it
153 always generate a CERROR on failure.
156 Frequency : very rare
158 Description: Fix replay if there is an un-replied request and open
159 Details : In some cases, older replay request will revert the
160 mcd->mcd_last_xid on MDS which is used to record the client's
163 Severity : enhancement
165 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
167 Severity : enhancement
169 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
171 Severity : enhancement
173 Description: Update to SLES9 kernel-2.6.5-7.287.3.
175 Severity : enhancement
177 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
179 Severity : enhancement
181 Description: Build SLES10 patchless client fails
182 Details : The configure was broken by run ./configure with
183 --with-linux-obj=.... argument for patchless client. When the
184 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
185 can't search header adequately. Use absolute path such as
186 -I($LINUX)/include instead.
188 Severity : enhancement
190 Description: Lustre Page Accounting
191 Details : New macros for page alloc and free which enable accounting
192 of page allocation of Lustre. Use percpu counters to store memory
196 Frequency : only if debugging is disabled
198 Description: LASSERT_{REQ,REP}SWAB macros are buggy
199 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
200 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
201 these macros and replace them with their difinitions instead.
206 Description: interrupt oig_wait produce painc on resend.
207 Details : brw_redo_request can be used for resend requests from ptlrpcd and
208 private set, and this produce situation when rq_ptlrpcd_data not
209 copyed to new allocated request and triggered LBUG on assert
210 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
211 ptlrpcd set if request is changed and can be safety changed to use
214 Severity : enhancement
216 Description: organize the server-side client stats on per-nid basis
217 Details : Change the structure of stats under obdfilter and mds to
225 The "uuid"s file would list the uuids of _active_ exports.
226 And the clear entry is to clear all stats and stale nids.
231 Description: Processes looping in ll_readdir() on Lustre clients finally causing
232 a full node pseudo-hang
233 Details : Concurrent access to the same directory from multiple clients with
234 intensive file creation/removal can cause a client node to spin in
235 ll_readdir(). i_version must be increased every time the lock is
236 cancelled to ensure a revalidate is done.
241 Description: touch file failed when fs is not full
242 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
243 otherwise we can get ENOSP while fs is not full.
246 Frequency : only for Cray XT3
247 Bugzilla : 12829/13455
248 Description: Changing primary group doesn't change the group lustre assigns to
250 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
251 overridden with the primary group provided by the group upcall,
252 whereas the supplied fsgid can be trusted if it is in the list of
253 supplementary groups returned by the group upcall.
257 Description: Root Squash Functionality
258 Details : Implementation of NFS-like root squash capability. Specifically,
259 don't allow someone with root access on a client node to be able
260 to manipulate files owned by root on a server node.
264 Description: Slow trucate/writes to huge files at high offsets.
265 Details : Directly associate cached pages to lock that protect those pages,
266 this allows us to quickly find what pages to write and remove
267 once lock callback is received.
271 Description: Too many locks accumulating on client during NFS usage
272 Details : mds_open improperly used accmode to find out access mode to a
273 file. Also mdc_intent_lock logic to find out if we already have
274 lock similar to just received was flawed since introduction of
275 skiplists - locks are now added to the front of the granted
280 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
282 Details : In obd setup failure handler, obd_stopping will not necessarily to
283 be 1, and obd_set_up should also be checked to make sure whether
284 obd is completely setup.
286 Severity : enhancement
288 Description: Allow masking D_WARNING, D_ERROR messages from console
289 Details : Console messages can now be disabled via lnet.printk.
293 Description: User code with malformed file open parameter crashes client node
294 Details : Before packing join_file req, all the related reference should be
295 checked carefully in case some malformed flags cause fake join_file
301 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
302 grain of quota doing harm to quota's accuracy
303 Details : qunit size will be changed when quota limitation is too low/high;
304 record the pending quota write in order to get more accureate
305 quota; delete the patch for bug12588, which is unnecessary when
306 this patch is landed. This bug also contains fixes for bug 14526,
307 14299, 14601 and 13794.
311 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
312 Details : ldlm_completion_ast() assumes that a lock is granted when the req
313 mode is equal to the granted mode. However, it should also check
314 that LDLM_FL_CP_REQD is not set.
318 Description: Heavy nfs access might result in deadlocks
319 Details : After ELC code landed, it is now improper to enqueue any mds
320 locks under och_sem, because enqueue might want to decide to
321 cancel open locks for same inode we are holding och_sem for.
325 Description: 35% write performance drop with ldiskfs2 when quotas are on
326 Details : Enable ext3 journalled quota by default to improve performance
327 when quotas are turned on.
331 Description: Client eviction while running blogbench
332 Details : A lot of unlink operations with concurrent I/O can lead to a
333 deadlock causing evictions. To address the problem, the number of
334 oustanding OST_DESTROY requests is now throttled to
335 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
336 callbacks are processed in priority.
339 Frequency : RHEL4 only
341 Description: mkfs is very slow on IA64/RHEL4
342 Details : A performance regression has been discovered in the MPT Fusion
343 driver between versions 3.02.73rh and 3.02.99.00rh. As a
344 consequence, we have downgraded the MPT Fusion driver in the RHEL4
345 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
348 Frequency : PPC/PPC64 only
350 Description: conflicts between asm-ppc64/types.h and lustre_types.h
351 Details : fix duplicated definitions between asm-ppc64/types.h and
352 lustre_types.h on PPC.
355 Frequency : PPC/PPC64 only
357 Description: asm-ppc/segment.h does not exist
358 Details : fix compile issue on PPC.
362 Description: data checksumming impacts single node performance
363 Details : add support for several checksum algorithms. Currently, CRC32 and
364 Adler-32 are supported. The checksum type can be changed on the fly
365 through /proc/fs/lustre/osc/*/checksum_type.
369 Description: use adler32 for page checksums
370 Details : when available, use the Adler-32 algorithm instead of CRC32 for
375 Description: better handle error messages in extents code
377 Severity : enhancement
379 Description: SNMP support enhancement
380 Details : Adding total number of sampled request for an MDS node in snmp
383 Severity : enhancement
385 Description: Optimize ldlm waiting list processing for PR extent locks
386 Details : When processing waiting list for read extent lock and meeting read
387 lock that is same or wider to it that is not contended, skip
388 processing rest of the list and immediatelly return current
389 status of conflictness, since we are guaranteed there are no
390 conflicting locks in the rest of the list.
394 Description: Time out and refuse to reconnect
395 Details : When the failover node is the primary node, it is possible
396 to have two identical connections in imp_conn_list. We must
397 compare not conn's pointers but NIDs, otherwise we
398 can defeat connection throttling.
402 Description: Client not clear own cache if answer to reconnect is lost.
403 Details : Client gets evicted from server. Now client also thinks it is
404 disconnected (or gets ENOTCONN on its operation) and decides to
405 reconnect. Server receives reconnect message, but cannot find
406 export. New export is created that is fully valid (new cookie!),
407 but reply is lost and not reported to client. Client reconnects
408 again and gets back a just-created connection, but it is not new
409 so client thinks it was not evicted and does not do recovery.
413 Description: Detect stride IO mode in read-ahead
414 Details : When a client does stride read, read-ahead should detect that and
415 read-ahead pages according to the detected stride pattern.
417 --------------------------------------------------------------------------------
419 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
421 * Support for kernels:
422 2.6.5-7.286 (SLES 9),
423 2.6.9-55.0.9.EL (RHEL 4),
424 2.6.16.53-0.8 (SLES 10),
425 2.6.18-8.1.14.el5 (RHEL 5),
426 2.6.18.8 vanilla (kernel.org)
427 * Client support for unpatched kernels:
428 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
429 2.6.16 - 2.6.22 vanilla (kernel.org)
430 * Due to recently discovered recovery problems, we do not recommend
431 using patchless RHEL 4 clients with this or any earlier release.
432 * Recommended e2fsprogs version: 1.40.2-cfs1
433 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
436 Frequency : occasional
438 Description: MDS hang or stay in waiting lock
439 Details : If client receive lock with CBPENDING flag ldlm need send lock
440 cancel as separate rpc, to avoid situation when cancel request
441 can't processed due all i/o threads stay in wait lock.
444 Frequency : occasional
446 Description: Do not fail import if osc_interpret_create gets -EAGAIN
447 Details : If osc_interpret_create got -EAGAIN it immediately exits and
448 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
449 oscc_has_objects and see OSC has no objests and call
450 oscc_internal_create to resend create request.
452 Severity : enhancement
454 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
455 Details : Update which_patch & target file for SLES10 latest kernel.
457 Severity : enhancement
459 Description: add --type and --size parameters to lfs find
460 Details : Enhance lfs find by adding filetype and filesize parameters. Also
461 multiple OBDs can now be specified for the --obd option.
463 Severity : enhancement
465 Description: eliminate client locks in face of contention
466 Details : file contention detection and lockless i/o implementation
469 Severity : enhancement
471 Description: Remove client patches from SLES 10 kernel.
472 Details : This causes SLES 10 clients to behave as patchless clients
473 even on a Lustre-patched (server) kernel.
475 Severity : enhancement
477 Description: use i_size_read and i_size_write in 2.6 port
478 Details : replace inode->i_size access with i_size_read/write()
481 Frequency : when removing large files
483 Description: scheduling issue during removal of large Lustre files
484 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
485 It causes scheduling issues when removing large files (17TB in the
491 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
492 Details : Impossible (EPROTO is returned) to access a directory that has a
493 non-default striping and ACLs.
496 Frequency : only on ppc
498 Description: /proc/fs/lustre/devices broken on ppc
499 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
500 We should make sure the type of 'index' is loff_t and then cast
501 explicitly as needed below. Do not assign an explicitly cast
505 Frequency : only for rhel5
507 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
508 Details : Modify the target file & which_kernel.
511 Frequency : if the uninit_groups feature is enabled on ldiskfs
513 Description: e2fsck reports "invalid unused inodes count"
514 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
515 feature and only a single inode is created in a group then the
516 "bg_unused_inodes" count is incorrectly updated. Creating a
517 second inode in that group would update it correctly.
522 Description: buffer overruns could theoretically occur
523 Details : llapi_semantic_traverse() modifies the "path" argument by
524 appending values to the end of the origin string, and a buffer
525 overrun may occur. Adding buffer overrun check in liblustreapi.
527 Severity : enhancement
529 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
532 Frequency : only if filesystem is inconsistent
534 Description: handle "serious error: objid * already exists" more gracefully
535 Details : If LAST_ID value on disk is smaller than the objects existing in
536 the O/0/d* directories, it indicates disk corruption and causes an
537 LBUG(). If the object is 0-length, then we should use the existing
538 object. This will help to avoid a full fsck in most cases.
540 Severity : enhancement
542 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
543 Details : Modify vm-tunables-rhel4.patch.
545 Severity : enhancement
547 Description: Kernel config for 2.6.18-vanilla.
548 Details : Modify targets/2.6-vanilla.target.in.
549 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
550 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
551 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
552 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
555 Frequency : occasional
557 Description: improve handling recoverable errors
558 Details : If request processed with error which can be recoverable on server
559 request should be resend, otherwise page released from cache and
565 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
566 Details : Modify target file & which_patch.
567 A flaw was found in the IA32 system call emulation provided
568 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
569 value could be stored in the %RAX register, which could trigger an
570 out-of-bounds system call table access. An untrusted local user
571 could exploit this flaw to run code in the kernel
572 (ie a root privilege escalation). (CVE-2007-4573).
576 Description: change order of libsysio includes
577 Details : '#include sysio.h' should always come before '#include xtio.h'
579 Severity : enhancement
581 Description: adapt the lustre_config script to support the upgrade case
582 Details : Add "-u" option for lustre_config script to support upgrading 1.4
583 server targets to 1.6 in parallel.
588 Description: To avoid grant space > avaible space when the disk is almost
589 full. Without this patch you might see the error "grant XXXX >
590 available" or some LBUG about grant, when the disk is almost
592 Details : In filter_check_grant, for non_grant cache write, we should
593 check the left space by if (*left > ungranted + bytes), instead
594 of (*left > ungranted), because only we are sure the left space
595 is enough for another "bytes", then the ungrant space should be
596 increase. In client, we should update cl_avail_grant only there is
597 OBD_MD_FLGRANT in the reply.
602 Description: Update RHEL 4 kernel to fix local root privilege escalation.
603 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
604 described in CVE-2007-4573. This problem could allow untrusted
605 local users to gain root access.
608 Frequency : when using O_DIRECT and quotas
610 Description: Incorrect file ownership on O_DIRECT output files
611 Details : block usage reported by 'lfs quota' does not take into account
612 files that have been written with O_DIRECT.
614 --------------------------------------------------------------------------------
616 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
618 * Support for kernels:
619 2.6.5-7.286 (SLES 9),
620 2.6.9-55.0.2.EL (RHEL 4),
621 2.6.16.46-0.14 (SLES 10),
622 2.6.18-8.1.8.el5 (RHEL 5),
623 2.6.18.8 vanilla (kernel.org)
624 * Client support for unpatched kernels:
625 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
626 2.6.16 - 2.6.21 vanilla (kernel.org)
627 * Due to recently discovered recovery problems, we do not recommend
628 using patchless RHEL 4 clients with this or any earlier release.
629 * Recommended e2fsprogs version: 1.40.2-cfs1
630 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
634 Description: Fix errors in lfs documentation
635 Details : Fixes man pages
637 Severity : enhancement
639 Description: Adaptive timeouts
640 Details : RPC timeouts adapt to changing server load and network
641 conditions to reduce resend attempts and improve recovery time.
643 Severity : enhancement
645 Description: llapi_file_create() does not allow some changes
646 Details : add llapi_file_open() that allows specifying the file creation
647 mode and open flags, and also returns an open file handle.
649 Severity : enhancement
651 Description: df doesn't work properly if diskfs blocksize != 4K
652 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
655 Frequency : always on directories with default striping set
657 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
658 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
660 Severity : enhancement
662 Description: merge and cleanup kernel patches.
663 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
665 Severity : enhancement
667 Description: RedHat Update kernel for RHEL5
668 Details : Kernel config file for RHEL5.
670 Severity : enhancement
672 Description: OSS needs mutliple precreate threads
673 Details : Add ability to start more than one create thread per OSS.
676 Frequency : only with quota on the root user
678 Description: mds_obd_create error creating tmp object
679 Details : When the user sets quota on root, llog will be affected and can't
680 create files and write files.
683 Frequency : Always on ia64 patchless client, and possibly others.
685 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
686 Details : This allows the patchless client to be loaded on architectures
692 Description: disorder of journal start and llog_add cause deadlock.
693 Details : in llog_origin_connect, journal start should happen before
694 llog_add keep the same order as other functions to avoid
697 Severity : enhancement
699 Description: RedHat Update kernel for RHEL5
700 Details : Modify the kernel config file more closer RHEL5.
702 Severity : enhancement
704 Description: Build failure against Centos5 (RHEL5)
705 Details : Define PAGE_SIZE when it isn't present.
708 Frequency : occasionally when using NFS
710 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
711 Details : with NFS, the anon dentry's parent was set to itself in
712 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
713 corresponding dentry other than use rec->ur_name.
715 Severity : enhancement
717 Description: enable data checksumming by default
718 Details : enable checksum by default, allow --disable-checksum
719 configure option and "-o nochecksum" mount option. Checksums
720 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
723 Frequency : Occasionally with failover
725 Description: Client eviction due to failover config
726 Details : after a connection loss, the lustre client should attempt to
727 reconnect to the last active server first before trying the
728 other potential connections.
730 Severity : enhancement
732 Description: client-side metadata stat-ahead during readdir(directory readahead)
733 Details : perform client-side metadata stat-ahead when the client detects
734 readdir and sequential stat of dir entries therein
739 Description: when mds and osts use different quota unit(32bit and 64bit),
740 quota will be released repeatly.
741 Details : void sending multiple quota reqs to mds, which will keep the
742 status between the reqs.
745 Frequency : only with liblustre clients on XT3
747 Description: evictions taking too long
748 Details : allow llrd to evict clients directly on OSTs
752 Description: osts not allocated evenly to files
753 Details : change the condition to increase offset_idx
755 Severity : enhancement
757 Description: self-adjustable client's lru lists
758 Details : use adaptive algorithm for managing client cached locks lru
759 lists according to current server load, other client's work
760 pattern, memory activities, etc. Both, server and client
761 side namespaces provide number of proc tunables for controlling
766 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
767 Details : rewrite inherited code (uuid parsing code from ext2 utils and
768 readdir code from ext3) from scratch preserving functionality.
772 Description: Only those disconnect error should be returned by rq_status.
773 Details : In open/enqueue processs, Some errors, which will cause client
774 disconnected, should be returned by rq_status, while other
775 errors should still be returned by intent, then mdc or llite will
778 Severity : enhancement
780 Description: Tune the kernel for good SCSI performance.
781 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
782 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
786 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
787 Bugzilla : 13547, 13627
788 Description: Data corruption for OSTs that are formatted larger than 2TB
790 Details : When generating the bio request for lustre file writes the
791 sector number would overflow a temporary variable before being
792 used for the IO. The data reads correctly from Lustre (which
793 will overflow in a similar manner) but other file data or
794 filesystem metadata may be corrupted in some cases.
798 Description: "lfs find -obd UUID" prints directories
799 Details : "lfs find -obd UUID" will return all directory names instead
800 of just file names. It is incorrect because the directories
801 do not reside on the OSTs.
804 Bugzilla : 13596, 13614
805 Description: MDS hang after unclean shutdown of lots of clients
806 Details : Never resend AST requests.
810 Frequency : Always, for kernels after 2.6.16
811 Description: Fix warning idr_remove called for id=.. which is not allocated.
812 Details : Last kernels save old s_dev before kill super and not allow
813 to restore from callback - restore it before call kill_anon_super.
817 Description: Fix error on 'ls .' at the top of the Lustre mount.
818 Details : Don't revalidate dentry if it is a root dentry.
820 --------------------------------------------------------------------------------
822 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
824 * Support for kernels:
825 2.6.5-7.286 (SLES 9),
826 2.6.9-55.0.2.EL (RHEL 4),
827 2.6.16.46-0.14 (SLES 10),
828 2.6.18-8.1.8.el5 (RHEL 5),
829 2.6.18.8 vanilla (kernel.org)
830 * Client support for unpatched kernels:
831 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
832 2.6.16 - 2.6.22 vanilla (kernel.org)
833 * Due to recently discovered recovery problems, we do not recommend
834 using patchless RHEL 4 clients with this or any earlier release.
835 * Recommended e2fsprogs version: 1.39.cfs8
836 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
841 Description: block reactivating mgc import until all deactivates complete
842 Details : Fix race when failing back MDT/MGS to itself (testing)
844 Severity : enhancement
846 Description: lfs setstripe enhancement
847 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
850 Frequency : mds/oss recovery
852 Description: llog ctxt is refrenced after it has been freed.
853 Details : llog ctxt refcount was added to avoide the race between ctxt free
854 and llog recovery process. Each llog user must hold ctxt refcount
855 before it access the llog. And the llog ctxt can only be freed
856 when its refcount is zero.
858 Severity : enhancement
860 Description: randomly memory allocation failure util
861 Details : Make lustre randomly failed allocating memory for testing purpose.
864 Frequency : only for SLES10
866 Description: Update kernel patch for SLES10 SP1
867 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
869 Severity : enhancement
871 Description: omit set fsid for export NFS
872 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
873 as FS_REQUIRES_DEV for avoid problems with generate fsid.
876 Frequency : after network failures
878 Description: Add sync option to mount_lustre.c
879 Details : Client loses data written to lustre after a network interruption.
881 Severity : enhancement
883 Description: Error message improvement.
884 Details : Merging of two LCONSOLE_ERROR_MSG into one.
886 Severity : enhancement
888 Description: don't use GFP_* in generic Lustre code.
889 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
891 Severity : enhancement
893 Description: obdclass is limited by single OBD_ALLOC(idarray)
894 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
896 Severity : enhancement
898 Description: updated patchess for new RHEL4 kernel
899 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
901 Severity : enhancement
903 Description: warnings with build patchless client with vanila 2.6.19 and up
904 Details : change old ctl_table style and replace ctl_table/ctl_table_header
905 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
907 Severity : enhancement
909 Description: O_DIRECT bypasses client statistics.
910 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
911 read_ahead_stats were updated. Copied stats section from
912 osc_send_oap_rpc() into async_internal().
915 Frequency : only for Cray XT3
917 Description: peer credits not enough on many OST per OSS systems.
918 Details : Use new lnet way to add credits as we need those for pings and ASTs
921 Frequency : only with liblustre
923 Description: Liblustre is not releasing flock locks on file close.
924 Details : Release flock locks on file close.
927 Frequency : only for RHEL4
929 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
930 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
931 Update target file and kernel config.
935 Description: lustre support for RHEL5
936 Details : Add support for RHEL5.
941 Description: ASSERTION(export != NULL) failed in target_handle_connect
942 Details : Assetion hit is result of rare race between disconnect and connet
943 to same nid. target_handle_connect found old connect cockie and
944 tried to reconnect, but can't find export for this cockie.
949 Description: umount blocks forever on error
950 Details : In result of wrong using obd_no_recov and obd_force flags client
951 can hand if cancel or some other requests is lost.
954 Frequency : Only for SLES
956 Description: sanity_quota fail test_1
957 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
958 file, which makes TSTID[2] inunique.
960 Severity : enhancement
962 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
963 Details : Update target/ChangeLog/which_patch .
965 Severity : enhancement
967 Description: jbd statistics
968 Details : Port older jbd statistics patch for sles10
973 Description: lvbo_init failed for resource with missing objects.
974 Details : Fix returning error if we do stat for file with missing/corrupted
975 objects and i_size set to all sum of size of all avaible objects.
976 if we to truncate/write to missing object - it is recreated.
981 Description: Oops in read and write path when failing to allocate lock.
982 Details : Check if lock allocation failed and return error back.
985 Frequency : When flocks are used.
987 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
988 Details : Flock locks might destroy just granted lock if it could be merged
989 with another existing flock, this is done in completion handler,
990 so teach ldlm_cli_enquque_fini that this is a valid case for
996 Description: reply_lock_interpret crash due to race with it and lock cancel.
997 Details : Do not replay locks that are being cancelled. Do not reference
998 locks by their address during replay, just by their handle.
1001 Frequency : only with deactivated OSTs
1003 Description: lstripe command fails for valid OST index
1004 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
1005 instead of lov->desc.ld_active_tgt_count.
1007 Severity : enhancement
1009 Description: Build failure against Centos5 (RHEL5)
1010 Details : Use getpagesize() instead of PAGE_SIZE.
1012 --------------------------------------------------------------------------------
1014 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
1016 * Support for kernels:
1017 2.6.5-7.283 (SLES 9),
1018 2.6.9-55.EL (RHEL 4),
1019 2.6.16.46-0.14 (SLES 10),
1020 2.6.18.8 vanilla (kernel.org)
1021 * Client support for unpatched kernels:
1022 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1023 2.6.16 - 2.6.22 vanilla (kernel.org)
1024 * Due to recently discovered recovery problems, we do not recommend
1025 using patchless RHEL 4 clients with this or any earlier release.
1026 * Recommended e2fsprogs version: 1.39.cfs8
1027 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1028 * Starting with this release, the ldiskfs backing filesystem required
1029 by Lustre is now in its own package, lustre-ldiskfs. This package
1030 should be installed. It is versioned separately from Lustre and
1031 may be released separately in future.
1036 Description: server LBUG when shutting down
1037 Details : Block umount forever until the mount refcount is zero rather
1038 than giving up after an arbitrary timeout.
1040 Severity : enhancement
1042 Description: add optional extra BUILD_VERSION info
1043 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1044 to override the lustre version.
1047 Frequency : 2.6.18 servers only
1049 Description: ll_kern_mount() doesn't release the module reference
1050 Details : The ldiskfs module reference count never drops down to 0
1051 because ll_kern_mount() doesn't release the module reference.
1056 Description: server LBUG when using old ost_num_threads parameter
1057 Details : Accept the old ost_num_threads parameter but warn that it
1058 is deprecated, and fix an off-by-one error that caused an LBUG.
1063 Description: Transient SCSI error results in persistent IO issue
1064 Details : iobuf->dr_error is not reinitialized to 0 between two
1068 Frequency : sometimes when underlying device returns I/O errors
1070 Description: OSTs not going read-only during write failures
1071 Details : OSTs are not remounted read-only when the journal commit threads
1072 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1073 instead of the ext3 wrappers.
1076 Frequency : only with 10000 clients or more
1078 Description: poor connect scaling with increasing client count
1079 Details : Don't run filter_grant_sanity_check for more than 100 exports
1080 to improve scaling for large numbers of clients.
1083 Frequency : SLES10 only
1085 Description: sanity-quota.sh quotacheck failed: rc = -22
1086 Details : Quotas cannot be enabled on SLES10.
1089 Frequency : liblustre clients only
1091 Description: getdirentries does not give error when run on compute nodes
1092 Details : getdirentries does not fail when the size specified as an argument
1093 is too small to contain at least one entry
1095 Severity : enhancement
1097 Description: Add LNET router traceability for debug purposes
1098 Details : If a checksum failure occurs with a router as part of the
1099 IO path, the NID of the last router that forwarded the bulk data
1100 is printed so it can be identified.
1105 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1106 Details : Due to a race condition, liblustre clients were occasionally
1107 evicted incorrectly.
1109 Severity : enhancement
1111 Description: lfs setstripe use optional parameters instead of postional
1114 Severity : enhancement
1116 Description: Nanosecond timestamp support for ldiskfs
1117 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1118 resolution timestamps. There is not yet support for this at
1119 the Lustre filesystem level.
1122 Frequency : during server recovery
1124 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1125 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1129 Frequency : nfs export on patchless client
1131 Description: connectathon hang when test nfs export over patchless client
1132 Details : Disconnected dentry cannot be found with lookup, so we do not need
1133 to unhash it or make it invalid
1136 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1137 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1138 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1143 Description: open req refcounting wrong on reconnect
1144 Details : If reconnect happened between getting open reply from server and
1145 call to mdc_set_replay_data in ll_file_open, we will schedule
1146 replay for unreferenced request that we are about to free.
1147 Subsequent close will crash in variety of ways.
1148 Check that request is still eligible for replay in
1149 mdc_set_replay_data().
1154 Description: disable writes to filesystem when reading health_check file
1155 Details : the default for reading the health_check proc file has changed
1156 to NOT do a journal transaction and write to disk, because this
1157 can cause reads of the /proc file to hang and block HA state
1158 checking on a healthy but otherwise heavily loaded system. It
1159 is possible to return to the previous behaviour during configure
1160 with --enable-health-write.
1162 Severity : enhancement
1164 Description: 64-bit inode version
1165 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1166 to the inode. This will be required for version-based recovery.
1171 Description: MDS fails to start if a duplicate client export is detected
1172 Details : in some rare cases it was possible for a client to connect to
1173 an MDS multiple times. Upon recovery the MDS would detect this
1174 and fail during startup. Handle this more gracefully.
1176 Severity : enhancement
1178 Description: Add -o localflock option to simulate old noflock behaviour.
1179 Details : This will achieve local-only flock/fcntl locks coherentness.
1184 Description: log_commit_thread vs filter_destroy race leads to crash
1185 Details : Take import reference before releasing llog record semaphore
1190 Description: Wrong request locking in request set processing
1191 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1192 imp_delayed_list, in this place should be used imp_lock.
1195 Frequency : when reconnecting
1197 Description: Grant leak when OSC reconnect to OST
1198 Details : When osc reconnect ost, OST(filter) should check whether it
1199 should grant more space to client by comparing fed_grant and
1200 cl_avail_grant, and return the granted space to client instead
1201 of "new granted" space, because client will call osc_init_grant
1202 to update the client grant space info.
1205 Frequency : when client reconnects to OST
1207 Description: Grant leak when OSC does a resend and replays bulk write
1208 Details : When osc reconnect to OST, OST(filter) should clear grant info of
1209 bulk write request, because the grant info while be sync between
1210 OSC and OST when reconnect, and we should ignore the grant info
1211 these of resend/replay write req.
1216 Description: Grant space more than avaiable space sometimes.
1217 Details : When then OST is about to be full, if two bulk writing from
1218 different clients came to OST. Accord the avaliable space of the
1219 OST, the first req should be permitted, and the second one
1220 should be denied by ENOSPC. But if the seconde arrived before
1221 the first one is commited. The OST might wrongly permit second
1222 writing, which will cause grant space > avaiable space.
1225 Frequency : when client is evicted
1227 Description: Grant might be wrongly erased when osc is evicted by OST
1228 Details : when the import is evicted by server, it will fork another
1229 thread ptlrpc_invalidate_import_thread to invalidate the
1230 import, where the grant will be set to 0. While the original
1231 thread will update the grant it got when connecting. So if
1232 the former happened latter, the grant will be wrongly errased
1233 because of this race.
1238 Description: Checking Stale with correct fid
1239 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1240 to check whether it is stale, because sometimes, we want the
1241 enqueue happened anyway, and op_data.fid2 will not be initialized.
1243 Severity : enhancement
1245 Description: update patchless client
1246 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1249 Frequency : only with 2.4 kernel
1251 Description: random memory corruption
1252 Details : size of struct ll_inode_info is to big for union inode.u and this
1253 can be cause of random memory corruption.
1258 Description: Memory leak in recovery
1259 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1260 It should also check obd_fail before fsfilt_start, otherwise if
1261 fsfilt_start return -EROFS,(failover mds during mds recovery).
1262 then the req will return with repmsg->transno = 0 and rc = EROFS.
1263 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1264 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1265 be freed no matter whether fsfilt_commit success or not.
1268 Frequency : only with huge numbers of clients
1270 Description: Prevents from taking the superblock lock in llap_from_page for
1272 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1273 for llap_from_page call in ll_removepage() prevents from taking
1274 the superblock lock for a soon died page.
1279 Description: Not check open intent error before release open handle
1280 Details : in some rare cases, the open intent error is not checked before
1281 release open handle, which may cause
1282 ASSERTION(open_req->rq_transno != 0), because it tries to release
1283 the failed open handle.
1288 Description: Set cat log bitmap only after create log success.
1289 Details : in some rare cases, the cat log bitmap is set too early. and it
1290 should be set only after create log success.
1294 Description: Accessing a block bevice can re-enable I/O when Lustre is
1295 tearing down a device.
1296 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1301 Description: service threads may hog cpus when there are a lot of requests
1302 Details : Insert cond_resched to give other threads a chance to use some CPU
1307 Description: the cat log was not initialized in recovery
1308 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1309 unlink log on mds will not be initialized until mds post
1310 recovery. And also in mds post recovery, the unlink log will
1311 initialization will be done asynchronausly, so there will be race
1312 between add unlink log and unlink log initialization.
1316 Description: brw_stats were being printed incorrectly
1317 Details : brw_stats were being printed as log2 but all of them were not
1318 recorded as log2. Also remove some code duplication arising from
1319 filter_tally_{read,write}.
1323 Frequency : rare, only in recovery.
1324 Description: ASSERTION(req->rq_type != LI_POISON) failed
1325 Details : imp_lock should be held while iterating over imp_sending_list for
1326 prevent destroy request after get timeout in ptlrpc_queue_wait.
1330 Description: replay-single.sh test 52 fails
1331 Details : A lock's skiplist need to be cleanup when it being unlinked
1332 from its resource list.
1337 Description: Short directio read returns full requested size rather than
1339 Details : Direct I/O operations should return actual amount of bytes
1340 transferred rather than requested size.
1342 Severity : enhancement
1344 Description: metadata RPC reduction (e.g. for rm performance)
1345 Details : decrease the amount of synchronous RPC between clients and servers
1346 by canceling conflicing lock before the operation on the client
1347 and packing thier handles into the main operation RPC to server.
1349 Severity : enhancement
1351 Description: add #ifdef HAVE_KERNEL_CONFIG_H
1352 Details : kernels from 2.6.19 not need include linux/config.h, but add
1353 include linux/autoconf.h in commpiler command line.
1355 Severity : enhancement
1357 Description: patchless client support for 2.6.22 kernel
1358 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
1359 removed. In this case we need drop using os depended interface to
1360 kmem_cache and use cfs_mem_cache API.
1365 Description: fix mal-formatted messages
1366 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
1370 Frequency : always in liblustre
1371 Description: wrong IS_ERR implementation in liblustre.h
1372 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
1377 Description: Correct condition for output debug message.
1378 Details : inode i_nlink equal zero is not enough for output message about
1379 disk corruption, i_ctime and i_mode should be also checked.
1383 Frequency : always in patchless client
1384 Description: add configure check for truncate_complete_page
1385 Details : improve checks for exported symbols. This allow run check without
1386 sources, but with Module.symvers shipped with kernel distribution.
1387 add check for truncate_complete_page used by patchless client.
1391 Description: sanity.sh test_77h fails with "test_77h file compare failed"
1392 Details : test_77h uses a file which was messed by other test case.
1396 Description: Not Check whether lov_tgts is NULL in some lov functions
1397 Details : Checking whether lov_tgts is NULL in some functions.
1401 Description: replace obdo_alloc() with OBDO_ALLOC macro
1402 Details : nothing special is done in obdo_alloc() function, and for
1403 debugging purpose, it needs to be replaced with macros.
1407 Description: bad return value and errno from fcntl call
1408 Details : In liblustre API, errno should be a negative value if error
1413 Description: ptlrpc_check_set() LBUG
1414 Details : In case of positive reply from server and failed client bulk
1415 callback after bulk transfer shouldn't LBUG, but process this
1416 request as erroneous.
1418 Severity : enhancement
1420 Description: VFS operations stats tool.
1421 Details : Tool which collects stats by tracking value written in pid,
1422 ppid, gid and uses llstat to generate output to plot graph using
1424 Updated lustre/utils/Makefile.am
1425 Added lustre/utils/ltrack_stats.c
1427 Severity : enhancement
1429 Description: 2.6.18 server support (lustre 1.6.1)
1430 Details : Support for 2.6.18 kernels on the server side.
1435 Description: ASSERTION(imp->imp_conn_current) failed
1436 Details : an assertion failure is hit if a client node boots and attempts to
1437 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
1440 Frequency : only for i686
1442 Description: 1.4.11 RC1 build fails for RHEL 4, i686
1443 Details : Fixed config variable for build.
1448 Description: Updated patchess for new RHEL4 kernel
1449 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
1450 Updated patch jbd-stats-2.6.9.patch
1451 Updated patch qsnet-rhel4-2.6.patch
1452 Updated patch quota-deadlock-on-pagelock-core.patch
1453 Updated patch vfs_intent-2.6-rhel4.patch
1454 Updated patch vfs_races-2.6-rhel4.patch
1455 Updated series file 2.6-rhel4-titech.series
1456 Updated series file 2.6-rhel4.series
1457 Updated kernel config files
1462 Description: lquota slave complains LBUG when reconnecting with mds
1464 Details : quota slave depends qctxt->lqc_import to send its quota request.
1465 This pointer will be invalid if mds did failover or broke its
1466 connect to osts, which leads to LBUG.
1469 Frequency : when qunit size is too small(less than 20M)
1471 Description: write is stopped by improper -EDQUOT
1472 Details : If the master is busy and qunit size is small enough(let's say 1M),
1473 the slave can not get quota from master on time, which will lead
1474 slave to trigger a -EQUOTA to client.
1479 Description: Deadlock during metadata tests
1480 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
1481 called with the per-dentry lock held.
1484 Frequency : SLES9 only
1486 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
1487 Details : set CONFIG_NR_CPUS to 128 instead of 8.
1489 Severity : enhancement
1491 Description: remove fs_prep_san_write operation and related patches
1492 Details : remove the ext3-san-jdike patches which are no longer useful.
1497 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
1498 Details : a disk corruption can cause the mballoc code to assert on a
1499 double free or other extent corruptions. Handle these with
1500 ext3_error() instead of assertions.
1503 Frequency : only with mballoc3 code and deep extent trees
1505 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
1506 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
1507 can cause the filesystem to be remounted read-only.
1512 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
1513 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
1514 few external users of this ioctl number is lfsck's call to
1515 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
1516 Add the old ioctl number to the handler so both old and new
1521 Description: parallel lock callbacks
1522 Details : Instead of sending blocking and completion callbacks as separated
1523 requests, adding them to a set and sending in parallel.
1527 Description: Disable most debugging by default
1528 Details : To improve performance, disable most logging (for debug purposes)
1529 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
1530 default, and HA includes fewer messages.
1534 Frequency : only run on patchless client.
1535 Description: use do_facet on sanity.sh for test handling recoverables errors
1536 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1540 Description: hash tables for lists of nids, connections and uuids
1541 Details : Hash tables noticeably help when a lot of clients connect to a
1542 server, to faster identify duplicate connections or reconnects,
1543 also to faster find export to evict in manual eviction case.
1547 Description: Sometimes, when the server evict a client, and the client will
1548 not be evicted as soon as possible.
1549 Details : In enqueue req, the error was returned by intent, instead of
1550 rq_status which make ptlrpc layer not detect this error, and
1551 does not evict the client. So enqueue error should be returned
1555 Frequency : only at startup
1557 Description: Delay client connections to MDT until fist MDT->OST connect
1558 Details : If a client tried to create a new file before the MDT had
1559 connected to any OSTs, the create would return EIO. Now
1560 the client will simply block until the MDT connects to the
1561 first OST and the create can succeed.
1564 Frequency : only for SLES9
1566 Description: Routinely utilize latest Quadrics drivers in CFS releases
1567 Details : Update patch qsnet-suse-2.6.patch.
1570 Frequency : only for sles10
1572 Description: Update patches for SLES 10 SP1 kernel.
1573 Details : Update the patch vfs_intent-2.6-sles10.patch.
1578 Description: Routinely utilize latest Quadrics drivers in CFS releases
1579 Details : Update patch qsnet-rhel4-2.6.patch.
1582 Frequency : at statup only
1584 Description: mds_lov_synchronize race leads to various problems
1585 Details : simultaneous MDT->OST connections at startup can cause the
1586 sync to abort, leaving the OSC in a bad state.
1591 Description: Using wrong pointer in osc_brw_prep_request
1592 Details : Access to array[-1] can produce panic if kernel compiled with
1593 CONFIG_PAGE_ALLOC enabled
1595 Severity : enhancement
1597 Description: Async OSC create to avoid the blocking unnecessarily.
1598 Details : If a OST has no remain object, system will block on the creating
1599 when need to create a new object on this OST. Now, ways use
1600 pre-created objects when available, instead of blocking on an
1601 empty osc while others are not empty. If we must block, we block
1602 for the shortest possible period of time.
1606 Frequency : only in recovery
1607 Description: Mark OST as early accessible if his start SYNC.
1608 Details : osc_precreate return flag early accessible if oscc marked as
1609 OSCC_FLAG_SYNC_IN_PROGRESS.
1614 Description: Sometimes precreate code can triger create object on wrong ost
1615 Details : Wrong protected or not not restored variables aftre precreate loop
1616 can produce creation object on wrong ost.
1619 Frequency : oss recovery
1621 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1623 Details : llog_commit_thread_count should be synced between llog_commit
1624 start and cleanup, so new llog_commit thread should not be started
1625 when llog_commit threads being stopped to avoid accessing some
1628 Severity : enhancement
1630 Description: Add printing inode info into message about error in writepage.
1632 --------------------------------------------------------------------------------
1634 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1639 Frequency : on some architectures
1641 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1642 Details : Uninitialized flags sometimes cause configuration commands to
1646 Frequency : patchless clients only
1648 Description: missing __iget() symbol export
1649 Details : The __iget() symbol export is missing. To avoid the need for
1650 this on patchless clients the deathrow inode reaper is turned
1651 off, and we depend on the VM to clean up old inodes. This
1652 dependency was during via the fix for bug 12181.
1657 Description: sanity.sh fail: test_52b
1658 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1659 incorrect inode's flags to client.
1661 --------------------------------------------------------------------------------
1663 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1665 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1666 INTEROPERATE with older versions automatically. In many cases a
1667 special upgrade step is needed. Please read the
1668 user documentation before upgrading any part of a 1.4.x system.
1669 * WARNING: Lustre configuration and startup changes are required with
1670 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1672 * Support for kernels:
1673 2.4.21-47.0.1.EL (RHEL 3),
1674 2.6.5-7.283 (SLES 9),
1675 2.6.9-42.0.10.EL (RHEL 4),
1676 2.6.12.6 vanilla (kernel.org),
1677 2.6.16.27-0.9 (SLES10)
1678 * Client support for unpatched kernels:
1679 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1680 2.6.16 - 2.6.19 vanilla (kernel.org),
1681 2.6.9-42.0.8EL (RHEL 4)
1682 * Recommended e2fsprogs version: 1.39.cfs6
1683 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1686 Severity : enhancement
1688 Description: Async OSC create to avoid the blocking unnecessarily.
1689 Details : If a OST has no remain object, system will block on the creating
1690 when need to create a new object on this OST. Now, ways use
1691 pre-created objects when available, instead of blocking on an
1692 empty osc while others are not empty. If we must block, we block
1693 for the shortest possible period of time.
1695 Severity : enhancement
1697 Description: MountConf
1698 Details : Lustre configuration is now managed via mkfs and mount
1699 commands instead of lmc and lconf. New obd types (MGS, MGC)
1700 are added for dynamic configuration management. See
1701 https://mail.clusterfs.com/wikis/lustre/MountConf for
1704 Severity : enhancement
1706 Description: dynamic OST addition
1707 Details : OSTs can now be added to a live filesystem
1709 Severity : enhancement
1711 Description: startup order invariance
1712 Details : MDTs and OSTs can be started in any order. Clients only
1713 require the MDT to complete startup.
1715 Severity : enhancement
1717 Description: parallel, asynchronous orphan cleanup
1718 Details : orphan cleanup is now performed in separate threads for each
1719 OST, allowing parallel non-blocking operation.
1721 Severity : enhancement
1723 Description: optimized stripe assignment
1724 Details : stripe assignments are now made based on ost space available,
1725 ost previous usage, and OSS previous usage, in order to try
1726 to optimize storage space and networking resources.
1728 Severity : enhancement
1730 Description: Permanently set tunables
1731 Details : All writable /proc/fs/lustre tunables can now be permanently
1732 set on a per-server basis, at mkfs time or on a live
1735 Severity : enhancement
1737 Description: Lustre message v2
1738 Details : Add lustre message format v2.
1740 Severity : enhancement
1742 Description: client OST exclusion list
1743 Details : Clients can be started with a list of OSTs that should be
1744 declared "inactive" for known non-responsive OSTs.
1748 Description: ENOENT returned for valid filehandle during dbench.
1749 Details : Check if a directory has children when invalidating dentries
1750 associated with an inode during lock cancellation. This fixes
1751 an incorrect ENOENT sometimes seen for valid filehandles during
1752 testing with dbench.
1755 Frequency : SFS test only (otherwise harmless)
1757 Description: SPEC SFS validation failure on NFS v2 over lustre.
1758 Details : Changes the blocksize for regular files to be 2x RPC size,
1759 and not depend on stripe size.
1761 Severity : enhancement
1763 Description: fine-grained SMP locking inside DLM
1764 Details : Improve DLM performance on SMP systems by removing the single
1765 per-namespace lock and replace it with per-resource locks.
1767 Severity : enhancement
1769 Description: don't hold multiple extent locks at one time
1770 Details : To avoid client eviction during large writes, locks are not
1771 held on multiple stripes at one time or for very large writes.
1772 Otherwise, clients can block waiting for a lock on a failed OST
1773 while holding locks on other OSTs and be evicted.
1775 Severity : enhancement
1777 Description: Multiple MD RPCs in flight.
1778 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1779 To avoid overly-overloading MDT, introduce a limit on number of
1780 MDT RPCs in flight for a single client and add /proc controls
1781 to adjust this limit.
1783 Severity : enhancement
1785 Description: client read/write statistics
1786 Details : Add client read/write call usage stats for performance
1787 analysis of user processes.
1788 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1789 file access. extents_stats shows chunk size distribution.
1790 extents_stats_per_process show chunk size distribution per
1793 Severity : enhancement
1795 Description: per-client statistics on server
1796 Details : Add ldlm and operations statistics for each client in
1797 /proc/fs/lustre/mds|obdfilter/*/exports/
1799 Severity : enhancement
1801 Description: improved MDT statistics
1802 Details : Add detailed MDT operations statistics in
1803 /proc/fs/lustre/mds/*/stats
1805 Severity : enhancement
1807 Description: VFS operations stats
1808 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1809 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1814 Description: Fix client-side osc byte counters
1815 Details : The osc read/write byte counters in
1816 /proc/fs/lustre/osc/*/stats are now working
1819 Frequency : always as root on SLES
1821 Description: Failure of copying files with lustre special EAs.
1822 Details : Client side always return success for setxattr call for lustre
1823 special xattr (currently only "trusted.lov").
1828 Description: Refcount LNET uuids
1829 Details : The global LNET uuid list grew linearly with every startup;
1830 refcount repeated list entries instead of always adding to
1833 Severity : enhancement
1835 Description: Dynamic service threads
1836 Details : Within a small range, start extra service threads
1837 automatically when the request queue builds up.
1840 Frequency : mixed-endian client/server environments
1842 Description: mixed-endian crashes
1843 Details : The new msg_v2 system had some failures in mixed-endian
1846 Severity : enhancement
1848 Description: Easy OST removal
1849 Details : OSTs can be permanently deactivated with e.g. 'lctl
1850 conf_param lustre-OST0001.osc.active=0'
1852 Severity : enhancement
1854 Description: MGS proc entries
1855 Details : Added basic proc entries for the MGS showing what filesystems
1858 Severity : enhancement
1860 Description: provide MGS failover
1861 Details : Added config lock reacquisition after MGS server failover.
1863 Severity : enhancement
1865 Description: add Linux 2.4 support
1866 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1870 Description: a large application tries to do I/O to the same resource and dies
1871 in the middle of it.
1872 Details : Check the req->rq_arrival time after the call to
1873 ost_brw_lock_get(), but before we do anything about
1874 processing it & sending the BULK transfer request. This
1875 should help move old stale pending locks off the queue as
1876 quickly as obd_timeout.
1879 Frequency : when an incorrect nid is specified during startup
1881 Description: ptlrpc connect to non-existant node causes kernel crash
1882 Details : LNET can't be re-entered from an event callback, which
1883 happened when we expire a message after the export has been
1884 cleaned up. Instead, hand the zombie cleanup off to another
1887 Severity : enhancement
1889 Description: plain/inodebits lock performance improvement
1890 Details : Grouping plain/inodebits in granted list by their request modes
1891 and bits policy, thus improving the performance of search through
1895 Frequency : only if OST filesystem is corrupted
1897 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1898 Details : for a short time RPCs with bulk IO are in the replay list,
1899 but replay of bulk IOs is unimplemented. If the OST filesystem
1900 is corrupted due to disk cache incoherency and then replay is
1901 started it is possible to trip an assertion. Avoid putting
1902 committed RPCs into the replay list at all to avoid this issue.
1905 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1907 Description: System hang on startup
1908 Details : This bug allowed the liblustre (e.g. catamount) client to
1909 return to the app before handling all startup RPCs. This
1910 could leave the node unresponsive to lustre network traffic
1911 and manifested as a server ptllnd timeout.
1913 Severity : enhancement
1915 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1916 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1917 boolean to control whether to print peer debug info when a
1918 client's RPC times out.
1921 Frequency : only for kernels with patches from Lustre below 1.4.3
1923 Description: Remove old rdonly API
1924 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1927 Frequency : only for devices with external journals
1929 Description: Set external device read-only also
1930 Details : During a commanded failover stop, we set the disk device
1931 read-only while the server shuts down. We now also set any
1932 external journal device read-only at the same time.
1935 Frequency : when upgrading from 1.4 while trying to change parameters
1937 Description: The wrong (new) MDC name was used when setting parameters for
1938 upgraded MDT's. Also allows changing of OSC (and MDC)
1939 parameters if --writeconf is specified at tunefs upgrade time.
1942 Frequency : when setting specific ost indicies
1944 Description: QOS code breaks on skipped indicies
1945 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1946 created with --index need not be sequential.
1948 Severity : enhancement
1950 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1951 Details : The uninit_groups feature works in conjunction with the kernel
1952 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1953 up the pass1 processing of e2fsck. This is a read-only feature
1954 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1955 mount filesystems that have had this feature enabled.
1957 Severity : enhancement
1959 Description: Improve multi-block allocation algorithm to avoid fragmentation
1960 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1961 allocation locality and avoid filesystem fragmentation.
1966 Description: lfs quota fails with deactivated OSTS
1967 Details : With this patch, three improvements are included:
1968 1. detete the softlimit in mds and osts when use "lfs quota".
1969 2. display the inaccurate data in the output of "lfs quota".
1970 3. try to get quota info when "lfs quota" is executed.
1972 ------------------------------------------------------------------------------
1974 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1976 * Support for kernels:
1977 2.4.21-47.0.1.EL (RHEL 3)
1978 2.6.5-7.283 (SLES 9)
1979 2.6.9-42.0.10.EL (RHEL 4)
1980 2.6.12.6 vanilla (kernel.org)
1981 2.6.16.27-0.9 (SLES 10)
1982 * Recommended e2fsprogs version: 1.39.cfs5
1984 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1988 Frequency : occasional, depends on client load and configuration
1989 Bugzilla : 12181, 12203
1990 Description: data loss for recently-modified files
1992 Details : In some cases it is possible that recently written or created
1993 files may not be written to disk in a timely manner (this should
1994 normally be within 30s unless client IO load is very high).
1995 The problem appears as zero-length files or files that are a
1996 multiple of 1MB in size after a client crash or client eviction
1997 that are missing data at the end of the file.
1999 This problem is more likely to be hit on clients where files are
2000 repeatedly created and unlinked in the same directory, clients
2001 have a large amount of RAM, have many CPUs, the filesystem has
2002 many OSTs, the clients are rebooted frequently, and/or the files
2003 are not accessed by other nodes after being written.
2005 The presence of the problem can be detected by looking at
2006 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2007 smaller than the second (nr_unused) then dirty files will not
2008 be flushed automatically to disk. "sync; sleep 10" should be
2009 run several times on the node before unmounting it to update
2010 Lustre (this is also safe to run on nodes without this problem).
2012 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2013 that can cause this same problem, so customers using that kernel
2014 also need to update the kernel in addition to Lustre. In order
2015 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2017 It is normal that files written just before a client crash (less
2018 than 30s) may not yet have been flushed to disk, even for local
2022 Frequency : frequent on thin XT3 nodes
2024 Description: UUID collision on thin XT3 Linux nodes
2025 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2026 randomly, since we relied on an insufficiently-seeded PRNG.
2031 Description: OSS hangs after "All ost request buffers busy"
2032 Details : A deadlock between quota and journal operations caused OSS
2033 hangs after printing "All ost request buffers busy."
2036 Frequency : always on liblustre builds
2038 Description: Cleanup compiler warnings on liblustre
2041 Frequency : always on liblustre builds on XT3
2043 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2047 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2048 > 32000 subdirectories
2049 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2050 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2051 subdirectory count crosses 32000. This will aid e2fsck to
2052 correctly handle more than 32000 subdirectories.
2055 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2057 Description: System hang on startup
2058 Details : This bug allowed the liblustre (e.g. catamount) client to
2059 return to the app before handling all startup RPCs. This
2060 could leave the node unresponsive to lustre network traffic
2061 and manifested as a server ptllnd timeout.
2063 Severity : enhancement
2065 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2066 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2067 boolean to control whether to print peer debug info when a
2068 client's RPC times out.
2073 Description: make O_SYNC working on 2.6 kernels
2074 Details : 2.6 kernels use different method for mark pages for write,
2075 so need add a code to lustre for O_SYNC work.
2080 Description: Failure to close file and release space on NFS
2081 Details : Put inode details into lock acquired in ll_intent_file_open.
2082 Use mdc_intent_lock in ll_intent_open to properly
2083 detect all kind of errors unhandled by mdc_enqueue.
2088 Description: proc file read during shutdown sometimes raced obd removal,
2090 Details : Add lock to prevent obd access after proc file removal.
2093 Frequency : Only for files larger than 4GB on 32-bit clients.
2095 Description: improperly doing page alignment of locks
2096 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2097 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2102 Description: RPCs being resent when they shouldn't be
2103 Details : Some RPCs that should not be resent are being resent. This
2104 can cause inconsistencies in the RPC state machine. Do not
2105 resend such requests.
2108 Frequency : rare, only with NFS export
2110 Description: Crash on NFS re-export node
2111 Details : under very unusual load conditions an assertion is hit in
2112 ll_intent_file_open()
2115 Frequency : only if OST filesystem is corrupted
2117 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2118 Details : for a short time RPCs with bulk IO are in the replay list,
2119 but replay of bulk IOs is unimplemented. If the OST filesystem
2120 is corrupted due to disk cache incoherency and then replay is
2121 started it is possible to trip an assertion. Avoid putting
2122 committed RPCs into the replay list at all to avoid this issue.
2127 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2128 Details : Large single O_DIRECT read and write calls can fail to allocate
2129 a sufficiently large buffer to process the request. In case of
2130 allocation failure the allocation is retried with a smaller
2131 buffer and broken into smaller requests.
2133 Severity : enhancement
2135 Description: Add -o localflock option to simulate old noflock behaviour.
2136 Details : This will achieve local-only flock/fcntl locks coherentness.
2141 Description: versioning check is incomplete
2142 Details : Checking the version difference of client vs. server, report
2143 error if the gap is too big.
2148 Description: add support PG_writeback bit
2149 Details : add support for PG_writeback bit for Lustre, for more carefull
2150 work with page cache in 2.6 kernel. This also fix some deadlocks
2151 and remove hack for work O_SYNC with 2.6 kernel.
2153 Severity : enhancement
2155 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2156 Details : The uninit_groups feature works in conjunction with the kernel
2157 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2158 up the pass1 processing of e2fsck. This is a read-only feature
2159 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2160 mount filesystems that have had this feature enabled.
2162 Severity : enhancement
2164 Description: Improve multi-block allocation algorithm to avoid fragmentation
2165 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2166 allocation locality and avoid filesystem fragmentation.
2168 ------------------------------------------------------------------------------
2170 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2172 * Support for kernels:
2173 2.6.9-42.0.3.EL (RHEL 4)
2174 2.6.5-7.276 (SLES 9)
2175 2.4.21-47.0.1.EL (RHEL 3)
2176 2.6.12.6 vanilla (kernel.org)
2177 2.6.16.21-0.8 (SLES10)
2178 * Recommended e2fsprogs version: 1.39.cfs2-0
2180 * The backwards-compatible /proc/sys/portals symlink has been removed
2181 in this release. Before upgrading, please ensure that you change
2182 any configuration scripts or /etc/sysctl.conf files that access
2183 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2184 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2185 in advance of the upgrade on any system running Lustre 1.4.6 or
2186 newer, since /proc/sys/lnet was added in that version.
2187 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2191 Frequency : only when quota is used
2193 Description: avoid scanning export list for quota master
2194 Details : Change the algorithms to avoid scanning export list in order
2195 to improve the efficiency.
2198 Frequency : MDS failover only, very rarely
2200 Description: "went back in time" messages on mds failover
2201 Details : The greatest transno may be lost when the current operation
2202 finishes with an error (transno==0) and the client's last_rcvd
2203 record is over-written. Save the greatest transno in the
2204 mds_last_transno for this case.
2207 Frequency : always for specific kernels and striping counts
2209 Description: client may get "Matching packet too big" without ACL support
2210 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2211 when trying to access files in certain configurations. The
2212 clients should in fact be denied when mounting because they do
2213 not understand ACLs.
2216 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2218 Description: many clients connecting with IO in progress causes connect timeouts
2219 Details : Avoid synchronous journal commits to avoid delays caused by many
2220 clients connecting/disconnecting when bulk IO is in progress.
2221 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2222 OST_IO_PORTAL to avoid delays behind potentially many pending
2226 Frequency : occasionally with multiple writers to a single file
2228 Description: shared writes to file may result in wrong size reported by stat()
2229 Details : Allow growing of kms when extent lock is cancelled
2232 Frequency : always with random mmap IO to multi-striped file
2234 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2235 Details : Only if the hole is at the end of OST object so that kms is too
2236 small. Fix is to increase kms accordingly in ll_nopage.
2239 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2241 Description: writes to a missing object would leak memory on the OST
2242 Details : If there is an inconsistency between the MDS and OST filesystems,
2243 such that the MDS references an object that doesn't exist, writes
2244 to that object will leak memory due to incorrect cleanup in the
2245 error handling path, eventually running out of memory on the OST.
2250 Description: Creating too long symlink causes lustre errors
2251 Details : Check symlink and name lengths before sending requests to MDS.
2254 Frequency : only if flock is enabled (not on by default)
2256 Description: posix locks not released on fd closure on 2.6.9+
2257 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2258 kernels, this caused such locks not to be released on fd close and
2259 then assertions on fs unmount about still used locks.
2262 Frequency : MDS failover only, very rarely
2264 Description: clients may get ASSERTION(granted_lock != NULL)
2265 Details : When request was taking a long time, and a client was resending
2266 a getattr by name lock request. The were multiple lock requests
2267 with the same client lock handle and
2268 mds_getattr_name->fixup_handle_for_resent_request found one of the
2269 lock handles but later failed with ASSERTION(granted_lock != NULL).
2274 Description: handle->h_buffer_credits > 0, assertion failure
2275 Details : h_buffer_credits is zero after truncate, causing assertion
2276 failure. This patch extends the transaction or creates a new
2280 Frequency : NFS re-export or patchless client
2281 Bugzilla : 11179, 10796
2282 Description: Crash on NFS re-export node (__d_move)
2283 Details : We do not want to hash the dentry if we don't have a lock.
2284 But if this dentry is later used in d_move, we'd hit uninitialised
2285 list head d_hash, so we just do this to init d_hash field but
2286 leave dentry unhashed.
2289 Frequency : NFS re-export or patchless client
2291 Description: NFS exports has problem with symbolic link
2292 Details : lustre client didn't properly install dentry when re-exported
2293 to NFS or running patchless client.
2296 Frequency : NFS re-export or patchless client
2298 Description: Various nfs/patchless fixes.
2299 Details : fixes reuse disconected alias for lookup process - this fixes
2300 warning "find_exported_dentry: npd != pd",
2301 fix permission error with open files at nfs.
2302 fix apply umask when do revalidate.
2305 Frequency : occasional
2307 Description: Crash on NFS re-export node
2308 Details : calling clear_page() on the wrong pointer triggered oops in
2309 generic_mapping_read().
2312 Frequency : rarely, using O_DIRECT IO
2314 Description: unaligned directio crashes client with LASSERT
2315 Details : check for unaligned buffers before trying any requests.
2318 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2320 Description: stale data returned from RAID cache
2321 Details : If only a small amount of IO is done to the RAID device before
2322 reading it again it is possible to get stale data from the RAID
2323 cache instead of reading it from disk.
2326 Frequency : always for sles10 kernel
2328 Description: sles10 support
2329 Details : ll_follow_link: compile fixes and using of nd_set_link
2330 under newer kernels.
2333 Frequency : depends on arch, kernel and compiler version, always on sles10
2336 Description: recursive or deep enough symlinks cause stack overflow
2337 Details : getting rid of large stack-allocated variable in
2341 Frequency : depends on hardware
2343 Description: lustre write performance loss in the SLES10 kernel
2344 Details : the performance loss is caused by using of write barriers in the
2345 ext3 code. The SLES10 kernel turns barrier support on by
2346 default. The fix is to undo that change for ldiskfs.
2348 ------------------------------------------------------------------------------
2350 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2352 * Support for kernels:
2353 2.6.9-42.0.3EL (RHEL 4)
2354 2.6.5-7.276 (SLES 9)
2355 2.4.21-47.0.1.EL (RHEL 3)
2356 2.6.12.6 vanilla (kernel.org)
2360 Frequency : quota enabled and large files being deleted
2362 Description: releasing more than 4GB of quota at once hangs OST
2363 Details : If a user deletes more than 4GB of files on a single OST it
2364 will cause the OST to spin in an infinite loop. Release
2365 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2370 Description: statfs data retrieved from /proc may be stale or zero
2371 Details : When reading per-device statfs data from /proc, in the
2372 {kbytes,files}_{total,free,avail} files, it may appear
2373 as zero or be out of date.
2376 Frequency : systems with MD RAID1 external journal devices
2378 Description: lconf's call to blkid is confused by RAID1 journal devices
2379 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2380 of returning all block devices that match the journal UUID.
2383 Frequency : always, for aggregate stripe size over 4GB
2385 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2386 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2387 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2390 Frequency : always if "lfs find" used on a local file/directory
2392 Description: "lfs find" segfaults if used on a local file/directory
2393 Details : The case where a directory component was not specified wasn't
2394 handled correctly. Handle this properly.
2397 Frequency : always on ppc64
2399 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2400 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2401 perform endianness conversion.
2404 Frequency : rarely (truncate to non-zero file size after write under load)
2405 Bugzilla : 10730, 10687
2406 Description: Files padded with zeros to next 4K multiple
2407 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2408 it is possible that files that are truncated to a non-zero size
2409 immediately after being written are filled with zero bytes beyond
2410 the truncated size. No file data is lost.
2412 Severity : enhancement
2414 Description: Allow recovery/failover for liblustre clients.
2415 Details : liblustre clients were unaware of failover configurations until
2418 Severity : enhancement
2420 Description: user file locks should fail when not mounting with flock option
2421 Details : Set up an error-returning stub in ll_file_operations.lock field
2422 to prevent incorrect behaviour when client is mounted without
2423 flock option. Also, set up properly f_op->flock field for
2427 Frequency : always on ia64
2429 Description: "lfs df" loops on printing out MDS statfs information
2430 Details : The obd_ioctl_data was not initialized and in some systems
2431 this caused a failure during the ioctl that did not return
2432 an error. Initialize the struct and return an error on failure.
2435 Frequency : SLES 9 only
2437 Description: Error of copying files with lustre special EAs as root
2438 Details : Client side always return success for setxattr call for lustre
2439 special xattr (currently only "trusted.lov").
2442 Frequency : rarely on clusters with both ia64+i386 clients
2444 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2445 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2446 clients are doing concurrent writes to the same file it is
2447 possible that the ia64 clients may LASSERT because the OST
2448 extent locks are not PAGE_SIZE aligned. Ensure that grown
2449 locks are always aligned on the request boundary.
2452 Frequency : specific use, occasional
2454 Description: Overwriting in use executable truncates on-disk binary image
2455 Details : If one node attempts to overwrite an executable in use by
2456 another node, we now correctly return ETXTBSY instead of
2457 truncating the file.
2459 Severity : enhancement
2461 Description: Async OSC create to avoid the blocking unnecessarily.
2462 Details : If a OST has no remain object, system will block on the creating
2463 when need to create a new object on this OST. Now, ways use
2464 pre-created objects when available, instead of blocking on an
2465 empty osc while others are not empty. If we must block, we block
2466 for the shortest possible period of time.
2471 Description: chmod on Lustre root is propagated to other clients
2472 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2473 invalid by the follow_mount time.
2476 Frequency : liblustre clients only
2478 Description: Race in 'instant cancel' lock handling could lead to such locks
2479 never to be granted in case of SMP MDS
2480 Details : Do not destroy not yet granted but cbpending locks in
2484 Frequency : replay/resend of open
2486 Description: non null lock assetion failure in mds_intent_policy
2487 Details : Trying to replay/resend lockless open requests resulted in
2488 mds_open() returning 0 with no lock. Now it sets a flag if
2489 a lock is going to be returned.
2491 Severity : enhancement
2493 Description: Checksum enhancements
2494 Details : New checksum enhancements allow for resending RPCs that failed
2497 Severity : enhancement
2499 Description: Tunables on number of dirty pages in cacche
2500 Details : Allow to set limit on number of dirty pages cached.
2505 Description: client crash on unmount - lock still has references
2506 Details : In some error handling cases it was possible to leak a lock
2507 reference on a client while accessing a file. This was not
2508 harmful to the client during operation, but would cause the
2509 client to crash when the filesystem is unmounted.
2512 Frequency : specific case, rare
2514 Description: ETXTBSY on mds though file not in use by client
2515 Details : ETXTBSY is no longer incorrectly returned when attempting to
2516 chmod or chown a directory that the user previously tried to
2517 execute or a currently-executing binary.
2520 Frequency : extremely rare except on liblustre-based clients
2522 Description: Lustre space not freed when files are deleted
2523 Details : Clean up open-unlinked files after client eviction. Previously
2524 the unlink was skipped and the files remained as orphans.
2529 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2530 Details : In some cases it was possible to send a blocking callback to
2531 a client doing a glimpse, even though that client didn't get
2532 a lock granted. When the glimpse lock is cancelled on the OST
2533 the freed lock is left on the waiting list and corrupted the list.
2536 Frequency : all core dumps
2538 Description: Broke core dumps to lustre
2539 Details : Negative dentry may be unhashed if parent does not have UPDATE
2540 lock, but some callers, e.g. do_coredump, expect dentry to be
2541 hashed after successful create, hash it in ll_create_it.
2543 ------------------------------------------------------------------------------
2545 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2547 * Support for kernels:
2548 2.6.9-42.0.2.EL (RHEL 4)
2549 2.6.5-7.276 (SLES 9)
2550 2.4.21-40.EL (RHEL 3)
2551 2.6.12.6 vanilla (kernel.org)
2555 Frequency : always on RHEL 3
2557 Description: Number of open files grows over time
2558 Details : The number of open files grows over time, whether or not
2559 Lustre is started. This was due to a filp leak introduced
2560 by one of our kernel patches.
2562 ------------------------------------------------------------------------------
2564 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2566 * Support for kernels:
2567 2.6.9-42.EL (RHEL 4)
2568 2.6.5-7.267 (SLES 9)
2569 2.4.21-40.EL (RHEL 3)
2570 2.6.12.6 vanilla (kernel.org)
2575 Bugzilla : 5719, 9635, 9792, 9684
2576 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2577 Details : If a server is under heavy load and cannot reply to new
2578 connection requests before the client resends the (re)connect,
2579 the connection handling code can behave badly if two service
2580 threads are concurrently handing separate (re)connections from
2581 the same client. Add better locking to the connection handling
2582 code, and ensure that only a single connection will be processed
2583 for a given client UUID, even if the lock is dropped.
2585 Severity : enhancement
2587 Description: add TCP zero-copy support to kernel
2588 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2589 sends if the hardware supports scatter-gather and checksumming.
2590 This allows socklnd to do client-write and server-read more
2591 efficiently and reduce CPU utilization from skbuf copying.
2594 Frequency : only if NFS exporting from client
2596 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2597 Details : If setting attributes on a file created under NFS that had
2598 never been opened it would be possible to oops the client
2599 if the file had no objects.
2603 Bugzilla : 9326, 10402, 10897
2604 Description: client crash in ptlrpcd_wake() thread when sending async RPC
2605 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2606 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2607 before ptlrpcd_wake() was called and the request was freed.
2610 Frequency : always for liblustre
2612 Description: liblustre client does MDS+OSTs setattr RPC for each write
2613 Details : When doing a write from a liblustre client, the client
2614 incorrectly issued an RPC to the MDS and each OST the file was
2615 striped over in order to update the timestamps. When writing
2616 with small chunks and many clients this could overwhelm the MDS
2617 with RPCs. In all cases it would slow down the write because
2618 these RPCs are unnecessary.
2620 Severity : enhancement
2622 Description: allow number of MDS service threads to be changed at module load
2623 Details : It is now possible to change the number of MDS service threads
2624 running. Adding "options mds mds_num_threads={N}" to the MDS's
2625 /etc/modprobe.conf will set the number of threads for the next
2626 time Lustre is restarted (assuming the "mds" module is also
2627 reloaded at that time). The default number of threads will
2628 stay the same, 32 for most systems.
2633 Description: OST crash if filesystem is unformatted or corrupt
2634 Details : If an OST is started on a device that has never been formatted
2635 or if the filesystem is corrupt and cannot even mount then the
2636 error handling cleanup routines would dereference a NULL pointer.
2641 Description: NULL pointer deref in llap_from_page.
2642 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2643 incorrect) mapping if the page was truncated/reclaimed while it was
2644 searched for. Check for this condition and skip such pages when
2645 doing readahead. Introduce extra check to llap_from_page() to
2646 verify page->mapping->host is non-NULL (so page is not anonymous).
2649 Frequency : Sometimes when using sys_sendfile
2651 Description: "page not covered by a lock" warnings from ll_readpage
2652 Details : sendfile called ll_readpage without right page locks present.
2653 Now we introduced ll_file_sendfile that does necessary locking
2654 around call to generic_file_sendfile() much like we do in
2658 Frequency : with certain MDS communication failures at client mount time
2660 Description: NULL pointer deref after failed client mount
2661 Details : a client connection request may delayed by the network layer
2662 and not be sent until after the PTLRPC layer has timed out the
2663 request. If the client fails the mount immediately it will try
2664 to clean up before the network times out the request. Add a
2665 reference from the request import to the obd device and delay
2666 the cleanup until the network drops the request.
2669 Frequency : occasionally during client (re)connect
2671 Description: assertion failure during client (re)connect
2672 Details : processing a client connection request may be delayed by the
2673 client or server longer than the client connect timeout. This
2674 causes the client to resend the connection request. If the
2675 original connection request is replied in this interval, the
2676 client may trip an assertion failure in ptlrpc_connect_interpret()
2677 which thought it would be the only running connect process.
2680 Frequency : only with obd_echo servers and clients that are rebooted
2682 Description: kernel BUG accessing uninitialized data structure
2683 Details : When running an obd_echo server it did not start the ping_evictor
2684 thread, and when a client was evicted an uninitialized data
2685 structure was accessed. Start the ping_evictor in the RPC
2686 service startup instead of the OBD startup.
2688 Severity : enhancement
2689 Bugzilla : 10193 (patchless)
2690 Description: Remove dependency on various unexported kernel interfaces.
2691 Details : No longer need reparent_to_init, exit_mm, exit_files,
2692 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2695 Frequency : rare (only users of deprecated and unsupported LDAP config)
2697 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2698 Details : LDAP apparently contains 'lustreName' attributes instead of
2699 'name'. A simple remapping of the name is sufficient.
2702 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2704 Description: waiting_locks_callback trips kernel BUG if client is evicted
2705 Details : Running with the dump_on_timeout debug flag turned on makes
2706 it possible that the waiting_locks_callback() can try to dump
2707 the Lustre kernel debug logs from an interrupt handler. Defer
2708 this log dumping to the expired_lock_main() thread.
2710 Severity : enhancement
2712 Description: Support NFS exporting on 2.6 kernels.
2713 Details : Implement non-rawops metadata methods for NFS server to use without
2714 changing NFS server code.
2717 Frequency : very rare (synthetic metadata workload only)
2719 Description: two racing renames might cause an MDS thread to deadlock
2720 Details : Running the "racer" program may cause one MDS thread to rename
2721 a file from being the source of a rename to being the target of
2722 a rename at exactly the same time that another thread is doing
2723 so, and the second thread has already enqueued these locks after
2724 doing a lookup of the target and is trying to relock them in
2725 order. Ensure that we don't try to re-lock the same resource.
2728 Frequency : only very large systems with liblustre clients
2730 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2731 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2732 all OSTs in parallel. This allows the eviction of stale liblustre
2733 clients to proceed much faster than if they were done in series,
2734 and also offers similar improvements for other set_info RPCs.
2739 Description: excessive CPU usage during initial read phase on client
2740 Details : During the initial read phase on a client, it would agressively
2741 retry readahead on the file, consuming too much CPU and impacting
2742 performance (since 1.4.5.8). Improve the readahead algorithm
2743 to avoid this, and also improve some other common cases (read
2744 of small files in particular, where "small" is files smaller than
2745 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2750 Description: MDS crash when receiving packet with unknown intent.
2751 Details : Do not LBUG in unknown intent case, just return -EFAULT
2753 Severity : enhancement
2754 Bugzilla : 9293, 9385
2755 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2756 Details : Do not serialize getattr (non-intent version) and statfs.
2759 Frequency : occasional, when OST network is overloaded/intermittent
2761 Description: client evicted by OST after bulk IO timeout
2762 Details : If a client sends a bulk IO request (read or write) the OST
2763 may evict the client if it is unresposive to its data GET/PUT
2764 request. This is incorrect if the network is overloaded (takes
2765 too long to transfer the RPC data) or dropped the OST GET/PUT
2766 request. There is no need to evict the client at all, since
2767 the pinger and/or lock callbacks will handle this, and the
2768 client can restart the bulk request.
2771 Frequency : Always when mmapping file with no objects
2773 Description: client crashes when mmapping file with no objects
2774 Details : Check that we actually have objects in a file before doing any
2775 operations on objects in ll_vm_open, ll_vm_close and
2781 Description: Request leak when working with deleted CWD
2782 Details : Introduce advanced request refcount tracking for requests
2783 referenced from lustre intent.
2785 Severity : Enhancement
2787 Description: Cache open file handles on client.
2788 Details : MDS now will return special lock along with openhandle, if
2789 requested and client is allowed to hold openhandle, even if unused,
2790 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2791 closing files for every read/write openration.
2793 Severity : Enhancement
2795 Description: Cache open negative dentries on client when possible.
2796 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2797 negative dentries on lock revocation.
2802 Description: Remounting a client read-only wasn't possible with a zconf mount
2803 Details : It wasn't possible to remount a client read-only with llmount.
2805 Severity : enhancement
2806 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2807 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2808 MPICH is included to add Lustre-specific ADIO interfaces.
2809 This is based closely on the UFS ADIO layer and only differs
2810 in file creation, in order to allow the OST striping to be set.
2811 This is user-contributed code and not supported by CFS.
2816 Description: extended inode attributes (immutable, append-only) work improperly
2817 when 2.4 and 2.6 kernels are used on client/server or vice versa
2818 Details : Introduce kernel-independent values for these flags.
2820 Severity : enhancement
2823 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2824 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2825 granularity. Now, Fractional MB granularity is be supported,
2826 this is very useful for low memory system.
2828 Severity : enhancement
2830 Description: Getattr by fid
2831 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2832 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2833 have locks and avoids getattr rpc requests for it on every stat.
2836 Frequency : Always, for filesystems larger than 2TB
2838 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2839 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2840 filesystems larger than 2TB to be mounted with mballoc enabled.
2843 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2845 Description: filesystem corruption for non-standard kernels and very large OSTs
2846 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2847 filesystems larger than 2TB is mounted then the kernel will
2848 silently corrupt the start of the filesystem. CONFIG_LBD is
2849 enabled for all CFS-supported kernels, but the possibility of
2850 this happening with a modified kernel config exists.
2852 Severity : enhancement
2854 Description: add client O_DIRECT support for 2.6 kernels
2855 Details : It is now possible to do O_DIRECT reads and writes to files
2856 in the Lustre client mountpoint on 2.6 kernel clients.
2858 Severity : enhancement
2860 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2861 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2862 parallel, not waiting for response from every OST before sending
2863 a rpc to the next OST.
2868 Description: setattr vs write race when updating file timestamps
2869 Details : Client processes that update a file timestamp into the past
2870 right after writing to the file (e.g. tar) it is possible that
2871 the updated file modification time can be reset to the current
2872 time due to a race between processing the setattr and write RPC.
2874 Severity : enhancement
2876 Description: Bring 'lfs find' closer in line with regular Linux find.
2877 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2878 -print0 options and obtains all the needed info through the lustre
2881 Severity : enhancement
2883 Description: support up to 1024 configured devices on one node
2884 Details : change obd_dev array from statically allocated to dynamically
2885 allocated structs as they are first used to reduce memory usage
2890 Description: Flush dirty partially truncated pages during truncate
2891 Details : Immediatelly flush partially truncated pages in filter_setattr,
2892 this way we completely avoid having any pages in page cache on OST
2893 and can retire ugly workarounds during writes to flush such pages.
2898 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2899 Details : protect inode from truncation within vfs_unlink() context
2900 just take a reference before calling vfs_unlink() and release it
2901 when parent's i_sem is free.
2904 Frequency : always, if extents are used on OSTs
2906 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2907 Details : OSTs using the extents format would not zero the high 16 bits of
2908 the index physical block number. This is not a problem for any
2909 OST filesystems smaller than 16TB, and no kernels support ext3
2910 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2911 new/modified files) and can be fixed for existing filesystems
2912 with e2fsprogs-1.39-cfs1.
2917 Description: import connection selection may be incorrect if timer wraps
2918 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2919 import connections to be ignored if the 32-bit jiffies counter
2920 wraps. Use a 64-bit jiffies counter.
2923 Frequency : during server recovery
2925 Description: crash after server is denying duplicate export
2926 Details : If clients are resending connect requests to the server, the
2927 server refuses to allow a client to connect multiple times.
2928 Fixed a bug in the handling of this case.
2931 Frequency : very large clusters immediately after boot
2933 Description: LNET request buffers exhausted under heavy short-term load
2934 Details : If a large number of client requests are generated on a service
2935 that has previously never seen so many requests it is possible
2936 that the request buffer growth cannot keep up with the spike in
2937 demand. Instead of dropping incoming requests, they are held in
2938 the LND until the RPC service can accept more requests.
2941 Frequency : Sometimes during replay
2943 Description: Assertion failure in ll_local_open after replay.
2944 Details : If replay happened on an open request reply before we were able
2945 to set replay handler, reply will become not swabbed tripping the
2946 assertion in ll_local_open. Now we set the handler right after
2947 recognising of open request
2950 Frequency : very rare
2952 Description: kernel reports "badness in vsnprintf"
2953 Details : Reading from the "recovery_status" /proc file in small chunks
2954 may cause a negative length in lprocfs_obd_rd_recovery_status()
2955 call to vsnprintf() (which is otherwise harmless). Exit early
2956 if there is no more space in the output buffer.
2958 Severity : enhancement
2960 Description: clear OBD RPC statistics by writing to them
2961 Details : It is now possible to clear the OBD RPC statistics by writing
2962 to the "stats" file.
2967 Description: Client mtime is not the same on different clients after utimes
2968 Details : In some cases, the client was using the utimes() syscall on
2969 a file cached on another node. The clients now validate the
2970 ctime from the MDS + OSTs to determine which one is right.
2975 Description: Inability to activate failout mode
2976 Details : lconf script incorrectly assumed that in python string's numeric
2977 value is used in comparisons.
2980 Frequency : always with multiple stripes per file
2982 Description: Inefficient object allocation for mutli-stripe files
2983 Details : When selecting which OSTs to stripe files over, for files with
2984 a stripe count that divides evenly into the number of OSTs,
2985 the MDS is always picking the same starting OST for each file.
2986 Return the OST selection heuristic to the original design.
2991 Description: mount failures may take full timeout to return an error
2992 Details : Under some heavy load conditions it is possible that a
2993 failed mount can wait for the full obd_timeout interval,
2994 possibly several minutes, before reporting an error.
2995 Instead return an error as soon as the status is known.
2997 ------------------------------------------------------------------------------
2999 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
3001 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3002 INTEROPERATE with older versions automatically. Please read the
3003 user documentation before upgrading any part of a live system.
3004 * WARNING: Lustre networking configuration changes are required with
3005 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3008 * Support for kernels:
3009 2.6.9-22.0.2.EL (RHEL 4)
3010 2.6.5-7.244 (SLES 9)
3011 2.6.12.6 vanilla (kernel.org)
3014 Severity : enhancement
3015 Bugzilla : 7981/8208
3016 Description: Introduced Lustre Networking (LNET)
3017 Details : LNET is new networking infrastructure for Lustre, it includes
3018 a reorganized network configuration mode (see the user
3019 documentation for full details) as well as support for routing
3020 between different network fabrics. Lustre Networking Devices
3021 (LNDs) for the supported network fabrics have also been
3022 created for this new infrastructure.
3024 Severity : enhancement
3025 Description: Introduced Access control lists
3026 Details : clients can set ACLs on files and directories in order to have
3027 more fine-grained permissions than the standard Unix UGO+RWX.
3028 The MDS must be started with the "-o acl" mount option.
3030 Severity : enhancement
3031 Description: Introduced filesystem quotas
3032 Details : Administrators may now establish per-user quotas on the
3035 Severity : enhancement
3037 Description: Configuration change for the XT3
3038 The PTLLND is now used to run Lustre over Portals on the XT3
3039 The configure option(s) --with-cray-portals are no longer used.
3040 Rather --with-portals=<path-to-portals-includes> is used to
3041 enable building on the XT3. In addition to enable XT3 specific
3042 features the option --enable-cray-xt3 must be used.
3047 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3048 Details : A race between allocating a new llcd and re-getting the llcd_lock
3049 allowed another thread to grab newly-allocated llcd.
3051 Severity : enhancement
3053 Description: 2.6 OST async journal commit and locking fix to improve performance
3054 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3055 2.6 kernels are now async as they already were in 2.4 kernels so
3056 that they can commit concurrently with the network bulk transfer.
3057 For block-allocated files the filter allocation semaphore is held
3058 to avoid filesystem fragmentation during allocation. BKL lock
3059 removed for 2.6 xattr operations where it is no longer needed.
3064 Description: lconf incorrectly determined whether two IP networks could talk
3065 Details : In some more complicated routing and multiple-network
3066 configurations, lconf will avoid trying to make a network
3067 connection to a disjoint part of the IP space. It was doing the
3068 math incorrectly for one set of cases.
3073 Description: Fix for potential infinite loop processing records in an llog.
3074 Details : If an llog record is corrupted/zeroed, it is possible to loop
3075 forever in llog_process(). Validate the llog record length
3076 and skip the remainder of the block on error.
3079 Frequency : occasional (liblustre only)
3081 Description: liblustre could not open files whose last component is a symlink
3082 Details : sysio_path_walk() would incorrectly pass the open intent to
3083 intermediate path components.
3086 Frequency : rare (liblustre only with non-standard tuning)
3087 Bugzilla : 7201 (7350)
3088 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3089 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3090 after it has been released, only do this for referenced locks
3092 Severity : enhancement
3094 Description: specifying an (invalid) directory default stripe_size of -1
3095 would reset the directory default striping
3096 Details : stripe_size -1 was used internally to signal directory stripe
3097 removal, now use "all default" to signal dir stripe removal
3098 as a directory striping of "all default" is not useful
3101 Frequency : common for large clusters running liblustre clients
3103 Description: doing an ls when liblustre clients are running is slow
3104 Details : sending a glimpse AST to a liblustre client waits for every AST
3105 to time out, as liblustre clients will not respond. Since they
3106 cannot cache data we refresh the OST lock LVB from disk instead.
3108 Severity : enhancement
3110 Description: doing an ls at the same time as file IO can be slow
3111 Details : enqueue and other "small" requests can be blocked behind many
3112 large IO requests. Create a new OST IO portal for non-IO
3113 requests so they can be processed faster.
3116 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3118 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3119 created file with mode 000
3120 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3121 allow an ftruncate() as a non-root user to a file with mode 000.
3122 Lustre now respects this flag to disable mode checks when
3123 truncating a file owned by the user
3126 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3129 Description: Revoking locks from clients that went dead or catatonic might take
3131 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3132 cancellation of such locks instant on servers without waiting for
3133 any reply from clients. Clients drops these locks when cancel
3134 notification from server is received without replying.
3137 Frequency : liblustre-only, when liblustre client dies or becomes busy
3139 Description: Doing ls on Linux clients can take a long time with active
3141 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3142 granting such locks to it in the first place if possible. Locks
3143 are taken by proxy on the OST during the read or write and
3144 dropped immediately afterward. Add connect flags handling, do
3145 not grant locks to liblustre clients for glimpse ASTs.
3147 Severity : enhancement
3149 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3150 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3151 sequential reads and behaves badly with applications doing random
3152 reads. Improve it by reading ahead at least read region, and
3153 avoiding excessive large RPC for small reads.
3155 Severity : enhancement
3157 Description: Creating more than 1000 files for a single job may cause a load
3158 imbalance on the OSTs if there are also a large number of OSTs.
3159 Details : qos_prep_create() uses an OST index reseed value that is an
3160 even multiple of the number of available OSTs so that if the
3161 reseed happens in the middle of the object allocation it will
3162 still utilize the OSTs as uniformly as possible.
3167 Description: OST or MDS may oops in ping_evictor_main()
3168 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3169 but doesn't restart at beginning of obd_exports_timed list
3172 Severity : enhancement
3174 Description: improve by-nid export eviction on the MDS and OST
3175 Details : allow multiple exports with the same NID to be evicted at one
3176 time without re-searching the exports list.
3179 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3181 Description: MDS may oops in groups_free()
3182 Details : in rare race conditions a newly allocated group_info struct is
3183 freed again, and this can be NULL. The 2.4 compatibility code
3184 for groups_free() checked for a NULL pointer, but 2.6 did not.
3187 Frequency : common for liblustre clients doing little filesystem IO
3188 Bugzilla : 9352, 7313
3189 Description: server may evict liblustre clients accessing contended locks
3190 Details : if a client is granted a lock or receives a completion AST
3191 with a blocking AST already set it would not reply to the AST
3192 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3195 Frequency : lfs setstripe, only systems with more than 160 OSTs
3197 Description: unable to set striping with a starting offset beyond OST 160
3198 Details : llapi_create_file() incorrectly limited the starting stripe
3199 index to the maximum single-file stripe count.
3202 Frequency : LDAP users only
3204 Description: lconf did not handle in-kernel recovery with LDAP properly
3205 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3207 Severity : enhancement
3209 Description: bind OST threads to NUMA nodes to improve performance
3210 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3211 node and do their allocations there to localize memory access
3213 Severity : enhancement
3215 Description: llmount can determine client NID directly from Myrinet (GM)
3216 Details : the client NID code from gmnalnid was moved directly into
3217 llmount, removing the need to use this or specifying the
3218 client NID explicitly when mounting GM clients with zeroconf
3221 Frequency : if client is started with down MDS
3223 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3224 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3225 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3226 ptlrpc_import_delayed_req() because we didn't check if the
3227 request was interrupted, and we also didn't break out of the
3228 event loop if there was no timeout
3233 Description: data loss during non-page-aligned writes to a single file from
3234 both multiple nodes and multiple threads on one node at same time
3235 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3236 inconsistency led to false short-reads, that were cached and later
3237 used by ->prepare_write() to fill in partially written page,
3238 leading to data loss.
3241 Frequency : always, if lconf --abort_recovery used
3243 Description: lconf --abort_recovery fails with 'Operation not supported'
3244 Details : lconf was attempting to abort recovery on the MDT device and not
3247 Severity : enhancement
3249 Description: remove cleanup logs
3250 Details : replace lconf-generated cleanup logs with lustre internal
3251 cleanup routines. Eliminates the need for client-cleanup and
3254 Severity : enhancement
3256 Description: add support for EAs (user and system) on lustre filesystems
3257 Details : it is now possible to store extended attributes in the Lustre
3258 client filesystem, and with the user_xattr mount option it
3259 is possible to allow users to store EAs on their files also
3261 Severity : enhancement
3263 Description: Add possibility (config option) to show minimal available OST free
3265 Details : When compiled with --enable-mindf configure option, statfs(2)
3266 (and so, df) will return least minimal free space available from
3267 all OSTs as amount of free space on FS, instead of summary of
3268 free spaces of all OSTs.
3270 Severity : enhancement
3272 Description: do not expand extent locks acquired on OST-side
3273 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3274 by server: they are not cached anyway.
3277 Frequency : when mmap is used/binaries executed from Lustre
3279 Description: Unmmap pages before throwing them away from read cache.
3280 Details : llap_shrink cache now attempts to unmap pages before discarding
3281 them (if unmapping failed - do not discard). SLES9 kernel has
3282 extra checks that trigger if this unmapping is not done first.
3287 Description: lconf didn't resolve symlinks before checking to see whether a
3288 given mountpoint was already in use
3291 Frequency : when migrating failover services
3292 Bugzilla : 6395, 9514
3293 Description: When migrating a subset of services from a node (e.g. failback
3294 from a failover service node) the remaining services would
3295 time out and evict clients.
3296 Details : lconf --force (implied by --failover) sets the global obd_timeout
3297 to 5 seconds in order to quickly disconnect, but this caused
3298 other RPCs to time out too quickly. Do not change the global
3299 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3301 Severity : enhancement
3302 Frequency : if MDS is started with down OST
3303 Bugzilla : 9439,5706
3304 Description: Allow startup/shutdown of an MDS without depending on the
3305 availability of the OSTs.
3306 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3307 Add appropriate locking and lov-osc refcounts for safe
3308 cleaning. Add osc abort_inflight calls in case the
3309 synchronize never started.
3312 Frequency : occasional (Cray XT3 only)
3314 Description: root not authorized to access files in CRAY_PORTALS environment
3315 Details : The client process capabilities were not honoured on the MDS in
3316 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3317 been accessed by an authorized user then root was able to access
3318 the file on the local client also. The root user capabilities
3319 are now allowed on the MDS, as this environment has secure UID.
3322 Frequency : occasional
3324 Description: ldiskfs "too long searching" message happens too often
3325 Details : A debugging message (otherwise harmless) prints too often on
3326 the OST console. This has been reduced to only happen when
3327 there are fragmentation problems on the filesystem.
3332 Description: Division by zero in statfs when all OSCs are inactive
3333 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3334 lov_statfs divides by value returned by lov_get_stripecnt().
3338 Bugzilla : 9489, 3273
3339 Description: First write from each client to each OST was only 4kB in size,
3340 to initialize client writeback cache, which caused sub-optimal
3341 RPCs and poor layout on disk for the first writen file.
3342 Details : Clients now request an initial cache grant at (re)connect time
3343 and so that they can start streaming writes to the cache right
3344 away and always do full-sized RPCs if there is enough data.
3345 If the OST is rebooted the client also re-establishes its grant
3346 so that client cached writes will be honoured under the grant.
3351 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3352 Details : Now I/O RPCs go to different portal number and (presumably) fast
3353 lock requests (and glimses) and other RPCs get their own service
3354 threads pool that should be able to service those RPCs
3357 Severity : enhancement
3359 Description: Ability to exchange lustre version between client and servers and
3360 issue warnings at client side if client is too old. Also for
3361 liblustre clients there is ability to refuse connection of too old
3363 Details : New 'version' field is added to connect data structure that is
3364 filled with version info. That info is later checked by server and
3368 Frequency : rare, liblustre only.
3369 Bugzilla : 9296, 9581
3370 Description: Two simultaneous writes from liblustre at offset within same page
3371 might proceed at the same time overwriting eachother with stale
3373 Details : I/O lock withing llu_file_prwv was released too early, before data
3374 actually was hitting the wire. Extended lock-holding time until
3375 server acknowledges receiving data.
3378 Frequency : extremely rare. Never observed in practice.
3380 Description: avoid generating lustre_handle cookie of 0.
3381 Details : class_handle_hash() generates handle cookies by incrementing
3382 global counter, and can hit 0 occasionaly (this is unlikely, but
3383 not impossible, because initial value of cookie counter is
3384 selected randonly). Value of 0 is used as a sentinel meaning
3385 "unassigned handle" --- avoid it. Also coalesce two critical
3386 sections in this function into one.
3388 Severity : enhancement
3390 Description: allow liblustre clients to delegate truncate locking to OST
3391 Details : To avoid overhead of locking, liblustre client instructs OST to
3392 take extent lock in ost_punch() on client's behalf. New connection
3393 flag is added to handle backward compatibility.
3395 Severity : enhancement
3396 Bugzilla : 4928, 7341, 9758
3397 Description: allow number of OST service threads to be specified
3398 Details : a module parameter allows the number of OST service threads
3399 to be specified via "options ost ost_num_threads={N}" in the
3400 OSS's /etc/modules.conf or /etc/modprobe.conf.
3404 Bugzilla : 6146, 9635, 9895
3405 Description: servers crash with bad pointer in target_handle_connect()
3406 Details : In rare cases when a client is reconnecting it was possible that
3407 the connection request was the last reference for that export.
3408 We would temporarily drop the export reference and get a new
3409 one, but this may have been the last reference and the export
3410 was just destroyed. Get new reference before dropping old one.
3412 Severity : enhancement
3413 Frequency : if client is started with failover MDS
3415 Description: Allow multiple MDS hostnames in the mount command
3416 Details : Try to read the configuration from all specified MDS
3417 hostnames during a client mount in case the "primary"
3420 Severity : enhancement
3422 Description: Stop sending data to evicted clients as soon as possible.
3423 Details : Check if the client we are about to send or are sending data to
3424 was evicted already. (Check is done every second of waiting,
3425 for which l_wait_event interface was extended to allow checking
3426 of exit condition at specified intervals).
3429 Frequency : rare, normally only when NFS exporting is done from client
3431 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3433 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3436 Frequency : always, if srand() is called before liblustre initialization
3438 Description: Liblustre uses system PRNG disturbing its usage by user application
3439 Details : Introduce internal to lustre fast and high-quality PRNG for
3440 lustre usage and make liblustre and some other places in generic
3441 lustre code to use it.
3443 Severity : enhancement
3444 Bugzilla : 9477, 9557, 9870
3445 Description: Verify that the MDS configuration logs are updated when xml is
3446 Details : Check if the .xml configuration logs are newer than the config
3447 logs stored on the MDS and report an error if this is the case.
3448 Request --write-conf, or allow starting with --old_conf.
3450 Severity : enhancement
3452 Description: Handle symlinks in the path when checking if Lustre is mounted.
3453 Details : Resolve intermediate symlinks when checking if a client has
3454 mounted a filesystem to avoid duplicate client mounts.
3459 Description: lconf can hit an error exception but still return success.
3460 Details : The lconf command catches the Command error exception at the top
3461 level script context and will exit with the associated exit
3462 status, but doesn't ensure that this exit status is non-zero.
3467 Description: failure of ptlrpc thread startup can cause oops
3468 Details : Starting a ptlrpc service thread can fail if there are a large
3469 number of threads or the server memory is very fragmented.
3470 Handle this without oopsing.
3473 Frequency : always, only if liblustre and non-default acceptor port was used
3475 Description: liblustre cannot connect to servers with non-default acceptor port
3476 Details : tcpnal_set_default_params() was not called and was therefore
3477 ignoring the environment varaible TCPNAL_PORT, as well as other
3478 TCPNAL_ environment variables
3483 Description: two objects could be created on the same OST for a single file
3484 Details : If an OST is down, in some cases it was possible to create two
3485 objects on a single OST for a single file. No problems other
3486 than potential performance impact and spurious error messages.
3490 Bugzilla : 5681, 9562
3491 Description: Client may oops in ll_unhash_aliases
3492 Details : Client dcache may become inconsistent in race condition.
3493 In some cases "getcwd" can fail if the current directory is
3499 Description: Inode refcounting problems in NFS export code
3500 Details : link_raw functions used to call d_instantiate without obtaining
3501 extra inode reference first.
3505 Bugzilla : 9942, 9903
3506 Description: Referencing freed requests leading to crash, memleaks with NFS.
3507 Details : We used to require that call to ll_revalidate_it was always
3508 followed by ll_lookup_it. Also with revalidate_special() it is
3509 possible to call ll_revalidate_it() twice for the same dentry
3510 even if first occurence returned success. This fix changes semantic
3511 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3512 reference on a request referred from the intent.
3513 ll_intent_release() then releases such a request.
3516 Frequency : rare, normally benchmark loads only
3518 Description: unlinked inodes were kept in memory on the client
3519 Details : If a client is repeatedly creating and unlinking files it
3520 can accumulate a lot of stale inodes in the inode slab cache.
3521 If there is no other client load running this can cause the
3522 client node to run out of memory. Instead flush old inodes
3523 from client cache that have the same inode number as a new inode.
3526 Frequency : SLES9 2.6.5 kernel and long filenames only
3527 Bugzilla : 9969, 10379
3528 Description: utime reports stale NFS file handle
3529 Details : SLES9 uses out-of-dentry names in some cases, which confused
3530 the lustre dentry revalidation. Change it to always use the
3534 Frequency : rare, unless heavy write-truncate concurrency is continuous
3535 Bugzilla : 4180, 6984, 7171, 9963, 9331
3536 Description: OST becomes very slow and/or deadlocked during object unlink
3537 Details : filter_destroy() was holding onto the parent directory lock
3538 while truncating+unlinking objects. For very large objects this
3539 may block other threads for a long time and slow overall OST
3540 responsiveness. It may also be possible to get a lock ordering
3541 deadlock in this case, or run out of journal credits because of
3542 the combined truncate+unlink. Solution is to do object truncate
3543 first in one transaction without parent lock, and then do the
3544 final unlink in a new transaction with the parent lock. This
3545 reduces the lock hold time dramatically.
3548 Frequency : rare, 2.4 kernels only
3550 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3551 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3552 before unmounting their filesystem in order to avoid deadlock.
3553 The kernel_locked() function in 2.4 kernels only checks whether
3554 the kernel lock is held, not whether it is this process that is
3555 holding it as 2.6 kernels do.
3560 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3561 Details : The client ptlrpc code may be trying to reconnect to a down
3562 server before a previous connection attempt has timed out.
3563 Increase the reconnect interval to be longer than the connection
3564 timeout interval to avoid sending duplicate connections to
3568 Frequency : echo_client brw_test command
3570 Description: fix echo_client to work with OST preallocated code
3571 Details : OST preallocation code (5137) didn't take echo_client IO path
3572 into account: echo_client calls filter methods outside of any
3573 OST thread and, hence, there is no per-thread preallocated
3574 pages and buffers to use. Solution: hijack pga pages for IO. As
3575 a byproduct, this avoids unnecessary data copying.
3579 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3580 Description: Client can oops in mdc_commit_close() after open replay
3581 Details : It was possible for the MDS to return an open request with no
3582 transaction number in mds_finish_transno() if the client was
3583 evicted, but without actually returning an error. Clients
3584 would later try to replay that open and may trip an assertion
3585 Simplify the client close codepath, and always return an error
3586 from the MDS in case the open is not successful.
3589 Frequency : rare, 2.6 OSTs only
3591 Description: OST may deadlock under high load on fragmented files
3592 Details : If there was a heavy load and highly-fragmented OST filesystems
3593 it was possible to have all the OST threads deadlock waiting on
3594 allocation of biovecs, because the biovecs were not released
3595 until the entire RPC IO was completed. Instead, release biovecs
3596 as soon as they are complete to ensure forward IO progress.
3598 Severity : enhancement
3600 Description: Support for specifying external journal device at mount
3601 Details : If an OST or MDS device is formatted with an external journal
3602 device, this device major/minor is stored in the ext3 superblock
3603 and may not be valid for failover. Allow detecting and
3604 specifying the external journal at mount time.
3609 Description: Mounting an MDS with pending unlinked files may cause oops
3610 Details : target_finish_recovery() calls mds_postrecov() which returned
3611 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3612 considers this an error and immediately begins cleaning up the
3613 lov, just after starting the mds_lov process
3615 Severity : enhancement
3617 Description: Implement 'lfs df' to report actual free space on per-OST basis
3618 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3619 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3620 usage of objects; '-h' to report in human readable format.
3622 ------------------------------------------------------------------------------
3624 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3631 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3632 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3633 at startup during orphan unlinking. Instead, skip buddy update
3634 before setup, as it will be regenerated after recovery is complete.
3639 Description: If an OST is inactive, its locks might reference stale inodes.
3640 Details : lov_change_cbdata() must iterate over all namespaces, even if
3641 they are inactive to clear inode references from the lock.
3643 Severity : enhancement
3644 Frequency : occasional, if non-standard max_dirty_mb used
3646 Description: Client will block write RPCs if not enough grant
3647 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3648 then the client will block writes while waiting for another RPC
3649 to complete instead of consuming its dirty limit. With change
3650 we get improved performance when max_dirty_mb is small.
3652 Severity : enhancement
3653 Bugzilla : 3389, 6253
3654 Description: Add support for supplementary groups on the MDS.
3655 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3656 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3657 lookups for user supplementary groups into a cache.
3661 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3662 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3663 for liblustre O_EXCL is enforced by MDS.
3667 Description: readdir never returns NULL in liblustre.
3668 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3669 dentry in struct dirent.
3673 Frequency : liblustre only, depends on application IO pattern
3674 Description: liblustre clients evicted if not contacting servers
3675 Details : Don't put liblustre clients into the ping_evictor list, so
3676 they will not be evicted by the pinger ever.
3678 Severity : enhancement
3680 Description: Add ability to evict clients by NID from MDS.
3681 Details : By echoing "nid:$NID" string into
3682 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3683 $NID would be instantly evicted from this MDS and from all active
3684 OSTs connected to it.
3688 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3689 Details : lookup_it_finish() used to query file size from OSTs that was not
3694 Description: service threads change working directory to that of init
3695 Details : Starting lustre service threads may pin the working directory
3696 of the parent thread, making that filesystem busy. Threads
3697 now change to the working directory of init to avoid this.
3701 Frequency : during shutdown only
3702 Description: shutdown with a failed MDS or OST can cause unmount to hang
3703 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3708 Frequency : 2.6 only
3709 Description: chmod/chown may include an extra supplementary group
3710 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3711 supplementary group and if none is specified this is used.
3714 Bugzilla : 5479 (6816)
3716 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3717 Details : If lookup is in progress on a file that is unlinked we might try
3718 to revalidate the inode and fail in revalidate after lookup is
3719 complete and ll_file_open() enqueues the open again but
3720 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3723 Frequency : always, if lconf --abort_recovery used
3725 Description: lconf --abort_recovery fails with 'Operation not supported'
3726 Details : lconf was attempting to abort recovery on the MDT device and not
3729 ------------------------------------------------------------------------------
3731 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3736 Frequency : rare (only unsupported configurations with a node running as an
3738 Bugzilla : 6514, 5137
3739 Description: Mounting a Lustre file system on a node running as an OST could
3741 Details : OSTs now preallocates memory needed to write out data at
3742 startup, instead of when needed, to avoid having to
3743 allocate memory in possibly low memory situations.
3744 Specifically, if the file system is mounted on on OST,
3745 memory pressure could force it to try to write out data,
3746 which it needed to allocate memory to do. Due to the low
3747 memory, it would be unable to do so and the node would
3748 become unresponsive.
3750 Severity : enhancement
3752 Description: Addition of lconf --service command line option
3753 Details : lconf now accepts a '--service <arg>' option, which is
3754 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3756 Severity : enhancement
3758 Description: Failover mode is now the default for OSTs.
3759 Details : By default, OSTs will now run in failover mode. To return to
3760 the old behaviour, add '--failout' to the lmc line for OSTs.
3762 Severity : enhancement
3764 Description: Health checks are now provided for MDS and OSTs
3765 Details : Additional detailed health check information on MSD and OSTs
3766 is now provided through the procfs health_check value.
3769 Frequency : occasional, depends on IO load
3771 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3772 after numerous create/delete cycles
3773 Details : The ext3 inode allocation policy would not allocate new inodes
3774 very well on the OSTs because there are no new directories
3775 being created. Instead we look for groups with free space if
3776 the parent directories are nearly full.
3781 Description: Network or server problems during mount may cause partially
3782 mounted clients instead of returning an error.
3783 Details : The config llog parsing code may overwrite the error return
3784 code during mount error handling, returning success instead
3790 Description: MDS can fail to allocate large reply buffers
3791 Details : After long uptimes the MDS can fail to allocate large reply
3792 buffers (e.g. zconf client mount config records) due to memory
3793 fragmentation or consumption by the buffer cache. Preallocate
3794 some large reply buffers so that these replies can be sent even
3795 under memory pressure.
3799 Frequency : rare (liblustre)
3800 Description: fsx running with liblustre complained that using truncate() to
3801 extend the file doesn't work. This patch corrects that issue.
3802 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3803 ATTR_SIZE and lsm use in llu_setattr_raw.
3807 Frequency : rare, only 2.6 kernels
3808 Description: Unusual file access patterns on the MDS may result in inode
3809 data being lost in very rare circumstances.
3810 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3811 code in the 2.6 kernel caused the inode and/or EA data not to
3812 be read from disk, causing single-file corruption.
3816 Frequency : rare, only 2.6 filesystems using extents
3817 Description: Heavy concurrent write and delete load may cause data corruption.
3818 Details : It was possible under high-load situations to have an extent
3819 metadata block in the block device cache from a just-unlinked
3820 file overwrite a newly-allocated data block. We now unmap any
3821 metadata buffers that alias just-allocated data blocks.
3825 Frequency : filesystems with default stripe_count larger than 77
3826 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3827 Details : lconf specifies an inode size of 4096 bytes when the default
3828 stripe_count is larger than 77. This conflicts with the default
3829 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3830 this case to avoid pinning too much memory for large EAs.
3832 ------------------------------------------------------------------------------
3834 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3839 Frequency : rare (extremely heavy IO load with hundreds of clients)
3841 Description: Client is evicted, gets IO error writing to file
3842 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3843 caused clients to be evicted for AST timeouts. The fixes in
3844 bug 5192 mean we no longer need to have such short AST timeouts
3845 so ldlm_timeout has been increased.
3848 Frequency : occasional during --force or --failover shutdown under load
3849 Bugzilla : 5949, 4834
3850 Description: Server oops/LBUG if stopped with --force or --failover under load
3851 Details : a collection of import/export refcount and cleanup ordering
3852 issues fixed for safer force cleanup
3855 Frequency : only filesystems larger than 120 OSTs
3856 Bugzilla : 5990, 6223
3857 Description: lfs getstripe would oops on a very large filesystem
3858 Details : lov_getconfig used kfree on vmalloc'd memory
3861 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3862 Bugzilla : 6242, 6243
3863 Description: reading from files that had been truncated to a non-zero size
3864 but never opened returned no data
3865 Details : ll_file_read() reads zeros from no-object files to EOF
3870 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3871 Details : There's little harm in aborting MDS/OSS recovery and letting it
3872 try again, so I removed the LASSERT and return an error instead.
3874 Severity : enhancement
3876 Description: New debugging infrastructure for tracking down data corruption
3877 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3878 (b) cover more of the client-side code path, and (c) try to narrow
3879 down where problems occurred
3883 Bugzilla : 3819, 4364, 4397, 6313
3884 Description: Racing close and eviction MDS could cause assertion in mds_close
3885 Details : It was possible to get multiple mfd references during close and
3886 client eviction, leading to one thread referencing a freed mfd.
3888 Severity: : enhancement
3889 Bugzilla : 3262, 6359
3890 Description: Attempts to reconnect to servers are now more aggressive.
3891 Details : This builds on the enhanced upcall-less recovery that was added
3892 in 1.4.2. When trying to reconnect to servers, clients will
3893 now try each server in the failover group every 10 seconds. By
3894 default, clients would previously try one server every 25 seconds.
3899 Description: After recovery, certain operations trigger a failed
3900 assertion on a client.
3901 Details : Failing over an mds, using lconf -d --failover, while a
3902 client was doing a readdir() call would cause the client to
3903 LBUG after recovery completed and the readdir() was resent.
3905 Severity : enhancement
3907 Description: Default groups are now added by lconf
3908 Details : You can now run lconf --group <servicename> without having to
3909 manually add groups with lmc.
3912 Frequency : occasional
3914 Description: Nodes with an elan id of 0 trigger a failed assertion
3917 Frequency : always when accessing e.g. tty/console device nodes
3919 Description: tty and some other devices nodes cannot be used on lustre
3920 Details : file's private_data field is used by device data and lustre
3921 values in there got lost. New field was added to struct file to
3922 store fs-specific private data.
3925 Frequency : when exporting Lustre via NFS
3927 Description: NFSD failed occasionally when looking up a path component
3928 Details : NFSD is looking up ".." which was broken in ext3 directories
3929 that had grown large enough to become hashed.
3932 Frequency : Clusters with multiple interfaces not on the same subnet
3934 Description: Nodes will repeatedly try to reconnect to an interface which it
3935 cannot reach and report an error to the log.
3936 Details : Extra peer list entries will be created by lconf with some peers
3937 unreachable. lconf now validates the peer before adding it.
3940 Frequency : Only if a default stripe is set on the filesystem root.
3942 Description: Setting a default stripe on the filesystem root prevented the
3943 filesystem from being remounted.
3944 Details : The client was sending extra request flags in the root getattr
3945 request and did not allocate a reply buffer for the dir EA.
3948 Frequency : occasional, higher if lots of files are accessed by one client
3949 Bugzilla : 6159, 6097
3950 Description: Client trips assertion regarding lsm mismatch/magic
3951 Details : While revalidating inodes the VFS looks up inodes with ifind()
3952 and in rare cases can find an inode that is being freed.
3953 The ll_test_inode() code will free the lsm during ifind()
3954 when it finds an existing inode and then the VFS later attaches
3955 this free lsm to a new inode.
3959 Bugzilla : 6422, 7030
3960 Description: MDS deadlock between mkdir and client eviction
3961 Details : Creating a new file via mkdir or mknod (starting a transaction
3962 and getting the ns lock) can deadlock with client eviction
3963 (gets ns lock and trying to finish a synchronous transaction).
3966 Frequency : occasional
3967 Description: While starting a server, the fsfilt_ext3 module could not be
3969 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3970 kernels. Previously, lconf would still use the ext3 name
3971 when trying to load modules. Now, it will correctly use
3972 ext3 on 2.4 and ldiskfs on 2.6.
3974 Severity : enhancement
3975 Description: The default stripe count has been changed to 1
3976 Details : The interpretation of the default stripe count (0, to lfs
3977 or lmc) has been changed to mean striping across a single
3978 OST, rather than all available. For general usage we have
3979 found a stripe count of 1 or 2 works best.
3981 Severity : enhancement
3982 Description: Add support for compiling against Cray portals.
3983 Details : Conditional compiling for some areas that are different
3987 Frequency : occasional
3988 Bugzilla : 6409, 6834
3989 Description: Creating files with an explicit stripe count may lead to
3990 a failed assertion on the MDS
3991 Details : If some OSTs are full or unavailable, creating files may
3992 trigger a failed assertion on the MDS. Now, Lustre will
3993 try to use other servers or return an error to the
3997 Frequency : occasional
3999 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4000 caused lock timeout during readahead (since 1.4.2).
4001 Details : Processes doing readahead might match a lock that hasn't been
4002 granted yet if there are overlapping and conflicting lock
4003 requests. The readahead process waits on ungranted lock
4004 (original lock is CBPENDING), while OST waits for that process
4005 to cancel CBPENDING read lock and eventually evicts client.
4007 Severity : enhancement
4009 Description: Initial enabling of flock support for clients
4010 Details : Implements fcntl advisory locking and file status functions.
4011 This feature is provided as an optional mount flag (default
4012 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4013 locking are implemented yet, and those that are are not guaranteed
4014 to be completely correct in production environments.
4015 mount -t lustre -o [flock|noflock] ...
4018 Frequency : occasional
4020 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4021 assertion in the ext3 JBD (journaling) layer.
4022 Details : The b_committed_data struct is protected by the big kernel lock
4023 in 2.4 kernels, serializing journal_commit_transaction() and
4024 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4025 access to this struct. In 2.6 kernels there is finer grained
4026 locking to improve SMP performance of the JBD layer.
4030 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4032 -----------------------------------------------------------------------------
4034 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4036 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4037 versions of Lustre. Please update all servers and clients to
4038 version 1.4.2 or later at the same time. You must also run
4039 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4040 to update the configuration logs.
4042 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4043 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4044 - swab llog records of type '0' so we get proper header size/idx (5861)
4045 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4046 - fix rename of one directory over another leaking an inode (5953)
4047 - avoid SetPageDirty on 2.6 (5981)
4048 - don't re-add just-being-destroyed locks to the waiting list (5653)
4049 - when creating new directories, inherit the parent's custom
4050 striping settings if present parent (3048)
4051 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4052 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4053 - don't hold client locks on temporary worklist from l_lru (5666)
4054 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4055 - automatically evict dead clients (5921)
4056 - Update file size properly in create+truncate+fstat case (6196)
4057 - Do not unhash mountpoint dentries, do not allow removal of
4059 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4060 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4061 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4062 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4063 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4064 - Correctly handle reads of files with no objects (6243)
4065 - lctl recover will also mark a device active if deactivate used (5933)
4067 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4068 - allow --write-conf on an MDS with different nettype than client (5619)
4069 - don't write config llogs to MDS for mounts not from that MDS (5617)
4070 - lconf should create multiple TCP connections from a client (5201)
4071 - init scripts are now turned off by default; run chkconfig --on
4072 lustre and chkconfig --on lustrefs to use them
4073 - upcalls are no longer needed for clients to recover to failover
4075 - add --abort-recovery option to lconf to abort recovery on device
4077 - add support for an arbitrary number of OSTs (3026)
4078 - Quota support protocol changes.
4079 - forward compatibility changes to wire structs (6007)
4080 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4081 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4082 - improved "lustre status" script
4083 - initialize blocksize for non-regular files (6062)
4084 - added --disable-server and --disable-client configure options (5782)
4085 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4086 - Vanilla 2.4.29 support
4087 - increase maximum number of obd devices to 520 (6242)
4088 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4089 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4090 SLES 9 2.6.5 kernels
4091 - limit stripes per file to 160 (the maximum EA size) (6093)
4093 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4096 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4097 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4098 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4099 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4100 - small ext3 extents cleanups and fixes (5733)
4101 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4102 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4103 - avoid needless client->OST connect, fix handle mismatch (5317)
4104 - fix DLM error path that led to out-of-sync client, long delays (5779)
4105 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4106 - fix several locking issues related to i_size (5492,5624,5654,5672)
4107 - don't move pending lock onto export if it is already evicted (5683)
4108 - fix kernel oops when creating .foo in unlinked directory (5548)
4109 - fix deadlock in obdfilter statistics vs. object create (5811)
4110 - use time_{before,after} to avoid timer jiffies wrap (5882)
4111 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4112 - Do not leak request if server was not able to process it (5154)
4113 - If mds_open unable to find parent dir, make that negative lookup(5154)
4114 - don't create new directories with extent-mapping (5909, 5936)
4116 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4117 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4118 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4119 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4120 - support for RHEL4 kernel on i686 (5773)
4121 - provide error messages when incompatible logs are encountered (5898)
4123 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4124 * version 1.4.0.10 (1.4.1 release candidate 1)
4126 - don't keep a lock reference when lock is not granted (4238)
4127 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4128 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4129 - drop import inflight refcount on signal_completed_replay error (5255)
4130 - unlock page after async write error during send (3677)
4131 - handle missing objects in filter_preprw_read properly (5265)
4132 - no transno return for symlink open, don't save no-trasno open (3440)
4133 - don't try to complete elan receive that already failed (4012)
4134 - free RPC server reply state on error (5406)
4135 - clean up thread from ptlrpc_start_thread() on error (5160)
4136 - readahead could read extra page into cache that wasn't ejected (5388)
4137 - prevent races in class_attach/setup/cleanup/detach (5260)
4138 - don't dereference de->d_inode after l_dput of de (5458)
4139 - use "int" for stripe value returned from lock_to_stripe (5544)
4140 - mballoc allocation and error-checking fixes in 2.6 (5504)
4141 - block device patches to fix I/O request sizes in 2.6 (5482)
4142 - look up hostnames for IB nals (5602)
4143 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4144 - don't start multiple acceptors for the same port (5277)
4145 - fix incorrect LASSERT in mds_getattr_name (5635)
4146 - export a proc file for general "ping" checking (5628)
4147 - fix "lfs check" to not block when the MDS is down (5628)
4149 - service request history (4965)
4150 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4151 - create an "evict_client" /proc entry on OSTs, like the MDS has
4152 - fix mount usage message, return errors per mount(8) (5168)
4153 - change grep [] to grep "[]" in tests so they work in more UMLs
4154 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4155 - remove ancient LOV_MAGIC_V0 EA support (5047)
4156 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4157 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4158 - allow readahead allocations to fail when low on memory (5383)
4159 - mmap locking landed again, after considerable improvement (2828)
4160 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4162 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4165 - send OST transaction number in read/write reply to free req (4966)
4166 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4167 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4168 - the watchdog thread now runs as interruptible (5246)
4169 - flock/lockf fixes (but it's still disabled, pending 5135)
4170 - don't use EXT3 constants in llite code (5094)
4171 - memory shortage at startup could cause assertion (5176)
4173 - reorganization of lov code
4174 - single portals codebase
4176 - add extents/mballoc support (5025)
4177 - direct I/O reads in the obdfilter (4048)
4178 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4180 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4183 - send OST transaction number in read/write reply to free req (4966)
4184 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4185 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4186 - the watchdog thread now runs as interruptible (5246)
4187 - handle missing objects in filter_preprw_read properly (5265)
4188 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4189 - drop import inflight refcount on signal_completed_replay error (5255)
4190 - unlock page after async write error during send (3677)
4191 - return original error code on reconstructed replies (3761)
4192 - no transno return for symlink open, don't save no-trasno open (3440)
4194 - add pid to ldlm debugging output (4922)
4195 - bump the watchdog timeouts -- we can't handle 30sec yet
4196 - extra debugging for orphan dentry/inode bug (5259)
4198 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4201 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4202 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4203 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4204 - fix NULL dereference in /proc/sys/portals/routes (4827)
4205 - allow failed mdc_close() operations to be interrupted (4561)
4206 - stop precreate on OST before MDS would time out on it (4778)
4207 - don't send partial-page writes before EOF from client (4410)
4208 - discard client grant for sub-page writes on large-page clients (4520)
4209 - don't free dentries not owned by NFS code, check generation (4806)
4210 - fix lsm leak if mds_create_objects() fails (4801)
4211 - limit debug_daemon file size, always print CERROR messages (4789)
4212 - use transno after validating reply (3892)
4213 - process timed out requests if import state changes (3754)
4214 - update mtime on OST during writes, return in glimpse (4829)
4215 - add mkfsoptions to LDAP (4679)
4216 - use ->max_readahead method instead of zapping global ra (5039)
4217 - don't interrupt __l_wait_event() during strace
4219 - add software watchdogs to catch hung threads quickly (4941)
4220 - make lustrefs init script start after nfs is mounted
4221 - fix CWARN/ERROR duplication (4930)
4222 - return async write errors to application if possible (2248)
4223 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4224 - print NAL number in %x format (4645)
4225 - update barely-supported suse-2.4.21-171 series (4842)
4226 - support for sles 9 %post scripts
4227 - support for building 2.6 kernel-source packages
4228 - support for sles km_* packages
4230 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4233 - ignore -ENOENT errors in osc_destroy (3639)
4234 - notify osc create thread that OSC is being cleaned up (4600)
4235 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4236 - reconstruct ost_handle() like mds_handle() (4657)
4237 - create a new thread to do import eviction to avoid deadlock (3969)
4238 - let lconf resolve symlinked-to devices (4629)
4239 - don't unlink "objects" from directory with default EA (4554)
4240 - hold socknal file ref over connect in case target is down (4394)
4241 - allow more than 32000 subdirectories in a single directory (3244)
4242 - fix blocks count for O_DIRECT writes (3751)
4243 - OST returns ENOSPC from object create when no space left (4539)
4244 - don't send truncate RPC if file size isn't changing (4410)
4245 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4246 - bind to privileged port in socknal and tcpnal (3689)
4248 - rate limit CERROR/CWARN console message to avoid overload (4519)
4249 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4250 - basic mmap support (3918)
4251 - kernel patch series update from b1_4 (4711)
4253 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4256 - avoid crash during MDS cleanup with OST shut down (2775)
4257 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4258 - don't use bad inodes on the MDS (3744)
4259 - dynamic object preallocation to improve recovery speed (4236)
4260 - don't hold spinlock over lock dumping or change debug flags (4401)
4261 - don't zero obd_dev when it is force cleaned (3651)
4262 - print grants to console if they go negative (4431)
4263 - "lctl deactivate" will stop automatic recovery attempts (3406)
4264 - look for existing locks in ldlm_handle_enqueue() (3764)
4265 - don't resolve lock handle twice in recovery avoiding race (4401)
4266 - revalidate should check working dir is a directory (4134)
4268 - don't always mark "slow" obdfilter messages as errors (4418)
4270 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4273 - don't close LustreDB during write_conf until it is done (3860)
4274 - fix typo in lconf for_each_profile (3821)
4275 - allow dumping logs from multiple threads at one time (3820)
4276 - don't allow multiple threads in OSC recovery (3812)
4277 - fix debug_size parameters (3864)
4278 - fix mds_postrecov to initialize import for llog ctxt (3121)
4279 - replace config semaphore with spinlock (3306)
4280 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4281 - don't allow enqueue to complete on a destroyed export (3822)
4282 - down write_lock before checking llog header bitmap (3825)
4283 - recover from lock replay timeout (3764)
4284 - up llog sem before sending rpc (3652)
4285 - reduce ns lock hold times when setting kms (3267)
4286 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4287 - fix NULL deref and obd_dev leak on setup error (3312)
4288 - replace some LBUG about llog ops with error handling (3841)
4289 - don't match INVALID dentries from d_lookup and spin (3784)
4290 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4291 - fix invalid assertion in ptlrpc_set_wait (3880)
4293 - add libwrap support for the TCP acceptor (3996)
4294 - add /proc/sys/portals/routes for non-root route listing (3994)
4295 - allow setting MDS UUID in .xml (2580)
4296 - print the stack of a process that LBUGs (4228)
4298 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4301 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4302 - make sure to unset replay flag from failed open requests (3440)
4303 - if default stripe count is 0, use OST count for inode size (3636)
4304 - update parent mtime/ctime on client for create/unlink (2611)
4305 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4306 - free recovery state on server during a forced cleanup (3571)
4307 - unregister_reply for resent reqs (3063)
4308 - loop back devices mounting and status check on 2.6 (3563)
4309 - fix resource-creation race that can provoke i_size == 0 (3513)
4310 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4311 - more debugging for page-accounting assertion (3746)
4312 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4313 - avoid deadlock after precreation failure (3758)
4314 - fix race and lock order deadlock in orphan handling (3450, 3750)
4315 - add validity checks when grabbing inodes from l_ast_data (3599)
4317 - add /proc/.../recovery_status to obdfilter (3428)
4318 - lightweight CDEBUG infrastructure, debug daemon (3668)
4319 - change default OSC RPC parameters to be better on small clusters
4320 - turn off OST read cache for files smaller than 32MB
4321 - install man pages and include them in rpms (3100)
4322 - add new init script for (un)mounting lustre filesystems (2593)
4323 - run chkconfig in %post for init scripts (3701)
4324 - drop scimac NAL (unmaintained)
4326 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4329 - clean kiobufs before and after use (3485)
4330 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4331 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4332 - change init script to fail more clearly if not run as root (1528)
4333 - allow clients to reconnect during replay (1742)
4334 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4335 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4336 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4337 - fix cond_resched (3554)
4338 - only evict unfinished clients after recovery (3515)
4339 - allow bulk resend, prevent data loss (3570)
4340 - dynamic ptlrpc request buffer allocation (2102)
4341 - don't allow unlinking open directory if it isn't empty (2904)
4342 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4343 - remove extraneous obd dereference causing LASSERT failure (3334)
4344 - don't use get_cycles() when creating temp. files on the mds (3156)
4345 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4346 - handle EEXIST for set-stripe, set proper directory name (3336)
4348 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4349 - fix ksocknal_fmb_callback() error messages (2918)
4351 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4354 - don't copy lvb into (possibly NULL) reply on error (2983)
4355 - don't deref dentry after dput, don't free lvb on error (2922)
4356 - use the kms to determine writeback rpc length (2947)
4357 - increment oti_logcookies when osc is inactive (2948)
4358 - update client's i_blocks count via lvb messages (2543)
4359 - handle intent open/close of special files properly (1557)
4360 - mount MDS with errors=remount-ro, like obdfilter (2009)
4361 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4362 - don't use cancelling-locks' kms values (2947)
4363 - use highest lock extent for kms, not last one (2925)
4364 - don't dereference ERR_PTR() dentry in error handling path (3107)
4365 - fix thread race in portals_debug_dumplog() (3122)
4366 - create lprocfs device entries at setup instead of at attach (1519)
4367 - common AST error handler, don't evict client on completion race (3145)
4368 - zero nameidata in detach_mnt in 2.6 (3118)
4369 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4370 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4371 - initialize RPC timeout timer earlier for 2.6 (3219)
4372 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4373 - print nal/nid for unknown nid (3258)
4374 - additional checks for oscc recovery before doing precreate (3284)
4375 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4376 - don't crash in mdc_close for bad permissions on open (3285)
4377 - zero i_rdev for non-device files (3147)
4378 - clear page->private before handing to FS, better assertion (3119)
4379 - tune the read pipeline (3236)
4380 - fix incorrect decref of invalidated dentry (2350)
4381 - provide read-ahead stats and refine rpc in flight stats (3328)
4382 - don't hold journal transaction open across create RPC (3313)
4383 - update atime on MDS at close time (3265)
4384 - close LDAP connection when recovering to avoid server load (3315)
4385 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4386 - don't leak open file on MDS after open resend (3325)
4387 - serialize filter_precreate and filter_destroy_precreated (3329)
4388 - loop device shouldn't call sync_dev() for nul device (3092)
4389 - clear page cache after eviction (2766)
4390 - resynchronize MDS->OST in background (2824)
4391 - refuse to mount the same filesystem twice on same mountpoint (3394)
4392 - allow llmount to create routes for mounting behind routers (3320)
4393 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4394 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4395 - fix rare problem with rename on htree directories (3417)
4397 - allow default OST striping configuration per directory (1414)
4398 - fix compilation for qswnal for 2.6 kernels (3125)
4399 - increase maximum number of MDS request buffers for large systems
4400 - change liblustreapi to be useful for external progs like lfsck (3098)
4401 - increase local configuration timeout for slow disks (3353)
4402 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4404 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4407 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4408 - don't overwrite extent policy data in reply if lock was blocked (2901)
4409 - drop filter export grants atomically with removal from device (2663)
4410 - del obd_self_export from work_list in class_disconnect_exports (2908)
4411 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4412 - swab reply message in mdc_close, other PPC fixes (2464)
4413 - fix destroying of named logs (2325)
4414 - overwrite old logs when running lconf --write_conf (2264)
4415 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4416 - fix race in target_handle_connect (2898)
4417 - mds_reint_create() should take same inode create lock (2926)
4418 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4419 - don't close files for self_export to avoid uninitialized obd (2936)
4420 - allow MDS with the same name as client node (2939)
4421 - hold dentry reference for closed log files for unlink (2325)
4422 - reserve space for all logs during transactions (2059)
4423 - don't evict page beyond end of stripe extent (2925)
4424 - don't oops on a deleted current working directory (2399)
4425 - handle hard links to targets without a parent properly (2517)
4426 - don't dereference NULL lock when racing during eviction (2867)
4427 - don't grow lock extents when lots of conflicting locks (2919)
4429 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4432 - account for cache space usage on clients to avoid data loss (974)
4433 - lfsck support in lustre kernel code (2349)
4434 - reduce journal credits needed for BRW writes (2370)
4435 - orphan handling to avoid losing space on client/server crashes
4436 - ptlrpcd can be blocked, stopping ALL progress (2477)
4437 - use lock value blocks to assist in proper KMS, faster stat (1021)
4438 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4439 - recovery for initial connections (2355)
4440 - fixes for mds_cleanup_orphans (1934)
4441 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4442 - block all file creations until orphan recovery completes (1901)
4443 - client remove rq_connection from request struct (2423)
4444 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4445 - recovery timer race (2670)
4446 - mdc_close recovey bug (2532)
4447 - ptlrpc cleanup bug (2710)
4448 - mds timeout on local locks (2588)
4449 - namespace lock held during RPCs (2431)
4450 - handle interrupted sync write properly (2503)
4451 - don't try to handle a message that hasn't been replied to (2699)
4452 - client assert failure during cleanup after abort recovery (2701)
4453 - leak mdc device after failed mount (2712)
4454 - ptlrpc_check_set allows timedout requests to complete (2714)
4455 - wait for inflight reqs when ptlrpcd finishes (2710)
4456 - make sure unregistered services are removed from the srv_list
4457 - reset bulk XID's when resending them (caught by 1138 test)
4458 - unregister_bulk after timeout
4459 - fix lconf error (2694)
4460 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4461 - readahead locks pages, leaves pending causing memory pressure (2673)
4462 - increase OST request buffers to 4096 on large machines (2729)
4463 - fix up permission of existing directories in simple_mkdir (2661)
4464 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4465 - don't assign transno to errored transactions (2742)
4466 - don't delete objects on OST if given a bogus objid from MDS (2751)
4467 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4468 - if rq_no_resend, then timeout request after recovery (2432)
4469 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4470 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4471 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4472 - MDS recovery shouldn't delete objects that it has given out (2730)
4473 - if enqueue arrives after completion, don't clobber LVB (2819)
4474 - don't unlock pages twice when trigger_group_io returns error (2814)
4475 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4476 - don't write pages to disk if there was an error (1450)
4477 - don't ping imports that have recovery disabled (2676)
4478 - take buffered bytes into account when balancing socknal conn (2817)
4479 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4480 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4481 - remove little-endian swabbing of llog records (1987)
4482 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4483 - retry reposting request buffers if they fail (1191)
4484 - grow extent at grant time to avoid granting a revoked lock (2809)
4485 - lock revoke doesn't evict page if covered by a second lock (2765)
4486 - disable VM readahead to avoid reading outside lock extents (2805)
4488 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4489 - updated kernel patches for hp-2.4.20 kernel (2681)
4491 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4494 - fix truncated write corruption (2366)
4495 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4497 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4498 - ensure len(uuid) < 37 in lmc (1171)
4499 - fix ia64 OOPS in llog_test (2255)
4500 - zero end of page at obdfilter for partial page writes (2648)
4501 - don't leave stale dentries around after renames (bug 2428)
4502 - fix timeouts when evicting a client with a single lock held (2642)
4503 - set deadline for the initial HELLO message to drain (2634)
4504 - print out dotted-quad IP addresses in the socknal (2302)
4506 - additional debugging for MDS client eviction problem (2443)
4507 - fix mkfsoptions support for osts (2603, 2604)
4509 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4512 - add series for the vanilla 2.6.0 kernel
4513 - add series for the vanilla 2.4.24 kernel
4514 - add series for a cray x86/64 UL kernel drop
4515 - fix xattr patches for the vanilla 2.4.19 series
4517 - generate true UUIDs in lmc (1171)
4518 - have portals stack dumping break in UML (2466)
4519 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4520 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4521 - wake callback waiting threads on client eviction (2460)
4522 - Add --ptldebug and --subsystem to lmc (1719)
4523 - update assertion to allow safe interrupt allocation
4524 - set rq_no_resend for cancel requests (2432)
4525 - recalculate ptlrpcd timeout after resend (2494)
4526 - call vfs_rmdir when removing pending directories (2368)
4527 - fix renaming a file to itself (2429)
4528 - lmc creates a default one-stripe lov (2454)
4529 - expand procfs space to handle large clusters (2326)
4530 - increase UML stack to avoid overflow
4531 - update lconf's list of debug and subsystem masks
4532 - fix lfs find --obd (2510)
4533 - /proc tunable for disabling filter read caching (2591)
4534 - stop rpm packages from altering slapd.conf (2301)
4535 - disable nagle in the socknal under 0conf (2578)
4536 - choose mds inode size based on stripe count (2572)
4537 - fix kernel-source rpm problems (2516)
4539 - add --disable-doc to avoid pdf generation (2421)
4540 - update documentation, tests, type-os, comments
4541 - avoid format warnings on ia64
4542 - remove the TOE NAL
4543 - tiny code cleanups by removing unused fields
4545 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4548 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4549 - remove the most blatant lies from BUILDING file (2371)
4550 - change default debug level to reasonable production setting
4551 - reduce client side cache size to reduce cache flush time
4552 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4553 - make TCP zerocopy and pinger support enabled by default (2476)
4554 - sync writes completed after process exits caused crashes (2319)
4555 - maintain correct mount count on the MDS (2356)
4556 - backout 1557, because 2316 wasn't really fixed
4557 - better file I/O statistics gathering in /proc
4558 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4559 - another kernel patch to fix zero-copy TCP function export
4560 - don't take duplicate lock when processing re-sent getattr (2420)
4561 - lctl uses obd_self_export instead of creating new conn (2353)
4562 - MDS/OST recovery case which requires object creation asserted (2425)
4563 - move lfs from /usr/sbin to /usr/bin in packages
4564 - fix race between mds_client_add and mds_client_free (2417)
4565 - use kmalloc instead of slabs in portals (2430)
4566 - don't create duplicate records when a failover MDS is present (2442)
4567 - remove unnecessary mount age check (2332)
4568 - don't remove directory inodes from locks prematurely (2451)
4569 - don't break if MDS service name is the same as hostname (2103)
4570 - fix races in client write RPC generation when cache full (2482)
4572 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4575 - remove now-unused request->rq_obd (278)
4576 - if an allocation fails, print out how much memory we've used (1933)
4577 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4578 - add the "configurable stack size" patch to most series files (1256)
4579 - ability to write large log records, for 100+ OST configs (2306)
4580 - fix NULL deref when filter_prep fails (2314)
4581 - fix operator precedence error in filter_sync
4582 - dynamic allocation of socknal TX descriptors (2315)
4583 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4584 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4585 - ensure that CWARN messages in Portals always get to the syslog
4586 - __init/__exit are not for prototype decls (ldlm_init/exit)
4587 - x86-64 compile warning fixes
4588 - fix gateway LMC keyword conflict (2318)
4589 - fix MDS lock inversions in getattr/reint paths (1844)
4590 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4591 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4592 - DLM race condition prevented some lock evictions (2328)
4593 - ENOMEM detection and retry on socknal sends (2230)
4594 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4595 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4596 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4597 - better error messages when a client is rejected during recovery (1505)
4598 - avoid cancelling locks which were never granted, after failure (2330)
4599 - fix i_sem/journal inversion in mds_client_add (2333)
4600 - fix truncate/getattr lock cycle deadlock (2334)
4601 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4602 - fix two rare exit paths which could leak an l_lock() ref (2321)
4603 - fencepost error in MDS/OST orphan recovery (2226)
4604 - make log record alignment 8 bytes (1988)
4605 - lstripe now fails when requested offset > ost_count (2237)
4606 - ensure that all kernel series have a complete list.h (1607)
4607 - fix crashes in special-file operations (2316)
4608 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4610 - allow configurable automake binary, for testing new versions
4611 - small update to the lfs documentation
4613 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4615 * fix negative export reference count in fsfilt_sync (2312)
4617 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4618 * release candidate 0.9.1
4620 - orphans are moved into the PENDING directory for possible recovery
4621 - replayed opens now open by fid for orphan/rename safety (1042)
4622 - last close of an orphan inode generates a transno (683)
4623 - chdir() and mount() now pin the directory entry (1020)
4624 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4625 - discard very old requests without processing them (1502)
4626 - remove some common, well-understood CERRORs (1505)
4627 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4628 - clear "grant" flags in OST replies until OST grant code lands (1644)
4629 - fix read performance by not clobbering i_blksize on client (1598)
4630 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4631 - make LRU size a /proc tunable, clears locks when reduced (707)
4632 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4633 - server locks take references on exports now (1558)
4634 - build fixes for 2.4.20-rh trees (1663)
4635 - return an error from lov_create if all OSCs are inactive (1751)
4636 - fix import levels when a reconnect happens without a timeout (1597)
4637 - exit early from mds_open if we get a lookup error (1749)
4638 - partial page read at EOF wouldn't wait for disk before sending (1642)
4639 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4640 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4641 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4642 - write/truncate lock inversion (1639)
4643 - Don't auto-load obdclass, portals modules during cleanup (1495)
4644 - fix timestamps from jumping to "now" (1763)
4645 - extra journal assertions (1648)
4646 - add an extra multiunlink test (1771)
4647 - fix read_record/write_record API (1776)
4648 - fix leak of offset_extent, possible incorrect i_size later (1772)
4649 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4650 - Debugging for the kqswnal_get_idle_tx problems (1820)
4651 - Allow recovery to be attempted multiple times (1536)
4652 - Write out MDS last_rcvd file after it is first created (1600)
4653 - Fix tx_descriptor leak in failed transmit situations (1827)
4654 - ext3 journaling fixes for assertion failure after IO error (1871)
4655 - class_export_put() on freed export after completion AST error (1896)
4656 - Fix revalidate looping in VFS (1322)
4657 - Don't access a freed export during MDS_REINT timeout (1521)
4658 - Add open-unlink recovery support on the MDS (1673,1764)
4659 - Return an error if no MDS data was read from last_rcvd (1946)
4660 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4661 - Don't setup a disk device that doesn't match exported UUID (317)
4662 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4663 - avoid committing NULL handle in force close
4664 - local.sh is now a one-stripe LOV configuration
4665 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4666 - fix ext3 htree duplicate directory entry corruption (1516)
4667 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4668 - update to new LOV EA format (2097)
4669 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4671 2003-06-15 Phil Schwan <phil@clusterfs.com>
4674 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4675 - per-import/export recovery handling (958, 931, 959)
4676 - multiple last-rcvd slots, for serving multiple FSes (949)
4677 - connections are again shared between multiple imp/exports (963, 964)
4678 - "umount -f" would hang if any requests needed to be sent (393, 978)
4679 - avoid pinning large req buffer by copying for queued messages (989)
4680 - add "uuid" to "lctl device" command to help upcalls (991)
4681 - "open" RPCs with transnos would confuse recovery counters (1037)
4682 - do proper endian conversion of all wire messages (288, 340, 891)
4683 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4684 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4685 - fix ptlrpc_connection leak in target_handle_connect (1174)
4686 - fix import refcounting bug in OST and MDS cleanup (1134)
4687 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4688 - fix very unlikely obd_types race condition (501)
4689 - remove osc_open hack for echo_client (1187)
4690 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4691 - a failure in read_inode2 leads to deadlock (1139)
4692 - cancel ack-locks as soon as transaction is committed (1072)
4693 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4694 - make sure to commitrw after any preprw to avoid deadlock (1162)
4695 - failing to execute a file in a lustre FS would lock inode (1203)
4696 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4697 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4698 - obd_uuid2tgt would walk off the end of the list (1255)
4699 - on IA64 the osc would give portals incorrect bulk size (1258)
4700 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4701 - fix lock inversion caused by new llite matching code (1282)
4702 - limit the number of dirty pages on a client to 10MB (1286)
4703 - timed out locks were not being corrected cancelled (1289)
4704 - fix O_DIRECT above 4GB on IA-32 (1292)
4705 * major user-visible changes
4706 - fail out/fail over policy now controlled by the upcall (993)
4708 - add OBD_PING to check server availability and failure (954)
4709 - lustre messages are now sent in sending host order (288, 340, 891)
4710 - add eadatalen to MDS getattr reply (340)
4711 - OST read replies may contain second buffer, with per-page status (593)
4713 2003-03-11 Phil Schwan <phil@clusterfs.com>
4716 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4717 - failed open()s could cause deadlock; fixed (867, 869)
4718 - stop cancelling OST locks when files are closed (481)
4719 - overlapping XID spaces caused network corruption (851, 853)
4720 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4721 - setattr_raw added, to do single-RPC, server-side setattrs
4722 - lmc/lconf syntax change for OST UUIDs
4723 - fix crashy race condition between ptlrpc_free_req and osc_close
4724 - don't use request in mdc_enqueue if we hit a timeout (889)
4725 - don't set the inode i_size for regular files from the MDS (896)
4726 - handle out of order completion AST (842)
4727 - don't LBUG if a lock request times out after receiving AST (913)
4728 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4729 - if a bad lock AST arrives, send an error instead of dropping entirely
4730 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4731 - fix leak in bulk IO when only partially completed (899, 900, 926)
4732 - fix O_DIRECT for ia64 (55)
4733 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4734 - C-z after timeout could hang a process forever; fixed (977)
4736 - client-side I/O cache (678, 924, 929, 941, 970)
4738 - READPAGE and SETATTRs which don't take server-side locks get
4741 2003-02-11 Phil Schwan <phil@clusterfs.com>
4744 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4745 - Fix fsfilt_extN_readpage() to read a full page of directory
4746 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4747 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4748 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4749 - Most intent operations were removed, in favour of a new RPC mode
4750 that does a single RPC to the server and bypasses most of the VFS
4751 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4752 - Aggressively cancel local locks on DLM servers
4753 - mds_reint_unlink sends EA to the client if it's the last nlink.
4754 client uses that EA to unlink OST objects.
4755 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4756 - recursive symlinks were fixed (439)
4757 - fixed NULL deref in DEBUG_REQ
4758 - filter_update_lastobjid no longer calls sync, which annoyed extN
4759 - fixed multi-client small-writes to a single file problem (445)
4760 - fixed mtime updates during file writes (607)
4761 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4762 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4763 - fixed Linux OST size reporting problem (444, 656)
4764 - OST now updates object mtime with writes or setattr (607, 619)
4765 - client verifies file size before zeroing page past EOF (445)
4766 - OST now writes last allocated objid to disk with allocation (108)
4767 - LOV on echo now works (409)
4769 - mds_reint_unlink sends a new buffer, with the EA included. this
4770 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4771 set if a regular file was being unlinked, and it was the last link
4772 - use PtlGet from the target for bulk writes (315)
4773 - OST now updates object mtime with writes or setattr (607, 619)
4774 - LDLM now has a grant-time callback to revalidate locked items, if
4776 - Many MDS operations were reorganized to combat race conditions
4778 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4779 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4781 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4784 - Fully reactivate OST imports after reconnection (512, others)
4785 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4786 - More graceful error handling for truncating on dead OST (515)
4787 - Don't error out unless we're actually accessing dead stripes (474)
4788 - Fix garbage sizes when stripes are missing (410)
4789 - LRU counters were broken, causing constant lock purge (433, 432)
4790 - garbage on read from stripes with failed OSTs (441)
4791 - mark OSCs as active before reconnecting during recovery (438)
4792 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4793 - lfind did not preserve OST order in output (443)
4794 - symlinks cause hung clients, incorrect data (439)
4795 - stop dereferencing request after dropping refcount (457)
4796 - don't LASSERT(spin_is_locked) on non-SMP (455)
4797 - fixes for many rename() bugs
4798 - fstat didn't correctly synchronize attributes (399)
4799 - server must handle lock cancellation during blocking AST prep (487)
4800 - bulk descriptors were free()d too soon (511)
4801 - fix paths in lconf, which would load incorrect modules (451, 507)
4802 - fix confusing lconf 'host not found' error message (386)
4803 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4804 - fix race condition in mdc_blocking_ast() for inode access (526)
4805 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4806 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4807 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4808 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4809 - fix lstripe in several ways (kernel side) (527)
4810 - fix request leak in ldlm_cli_enqueue (262)
4811 - incorrect OSC was marked inactive after OST failure
4812 - call mds_fs_cleanup before unmounting filesystem (524)
4813 - fix races between taking ns_lock and ldlm_lock_change_resource
4814 - fix races updating LOV export open file list
4815 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4816 - fix recovery NULL deref in ldlm_cli_cancel_unused
4817 - fix some DLM races by using new hash table for lock handles (419)
4818 - permit the client to specify desired inodes, at replay
4819 - duplicate requests when we queue them for replay reintegration
4820 - fix last_rcvd offset calculation
4821 - sync after each recovered transaction, so we always make progress
4822 - never, not always, ERESTART requests without transnos
4823 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4825 - skip replay if the MDS says that the client is already connected
4826 - don't check for a recovery-enabled export to match lctl's UUID
4827 - don't INC_USE_COUNT for phantom exports
4828 - don't crash when cleaning up phantom exports (567)
4829 - don't double-finish or set replay data for errored mdc_open requests
4830 - abort requests when they time out, so we don't get old replies
4831 - send/receive replies for AST messages again
4832 - if the client says that it doesn't have the lock, cancel it on the
4834 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4835 mark it as destroyed, it will all work out when decref is called
4836 - fix module use counts (22, 581)
4838 - ASTs now expect a reply (server cancels lock on error reply)
4840 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4843 - fix many simultaneous client startup (392)
4844 - fix dentry->d_it clobbering
4845 - credentials weren't being shipped for readdir/getattr operations
4846 - remove invalid assertions triggered during some concurrent MD
4848 - proper Lustre versions added (336, 389)
4849 - fix memory leak for create error case (398)
4850 - fix LOV locking bug that would get cli/srv out of sync
4851 - fix echo client over LOV (409)
4852 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4853 - fix double-O_EXCL intent crash (424)
4854 - avoid sending multiple lock CANCELs (352)
4856 - MDS can do multi-client recovery (modulo bugs in new code)
4858 - many updates, edits, cleanups
4860 2002-11-18 Phil Schwan <phil@clusterfs.com>
4863 - fix null d_it dereference (346)
4864 - fix full OST/dbench hang (333)
4865 - fix permission problem with file removal (286)
4866 - fix removal of OSCs from LOV when they fail
4867 - fix NULL deref during bulk timeout (214)
4868 - fix problems related to multiple filesystems on one MDS (241)
4869 - fixed serious subtle metadata locking bugs
4870 - free locks on clients when inodes are removed due to memory
4872 - fix inode pointer in lock data (285)
4873 - partial support for multiple MDS on a single host (241)
4874 - data locks weren't cancelled at clear_inode time (290, 311)
4875 - intent locks could lead to unbounded lock growth (205)
4876 - added a maximum lock count, an LRU list, and a flusher
4877 - fix multiple rename (365)
4878 - properly abstracted the echo client
4879 - OSC locked 1 byte too many; fixed
4880 - rewrote brw callback code:
4881 - fixed recovery bugs related to LOVs (306)
4882 - fixed too-many-pages-in-one-write crash (191)
4883 - fixed (again) crash in sync_io_timeout (214)
4884 - probably fixed callback-related race (385)
4886 - Add capability to MDS protocol
4887 - LDLM cancellations and callbacks on different portals
4889 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4892 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4893 - defer open object destruction to close time (601981, 138)
4894 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4895 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4896 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4897 - don't call obd_setattr in ll_file_release for destroyed objects
4898 * protocol change to lustre_msg: move |version| and add |flags|
4899 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4900 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4901 * added replay of create, unlink, link and rename operations during
4902 MDS failover; recovery should be much more robust now
4903 * remove failed OSCs from LOVs (only lov_create uses this so far)
4904 * the lustre-HOWTO was brought (more) up to date (582544)
4906 2002-10-23 Phil Schwan <phil@clusterfs.com>
4909 - in-use dentries weren't being reused properly (617851)
4910 - prevent multiple LDLM setup (599178)
4911 - fix LOV size calculations for truncate (617853)
4912 - fix client handling of MDS intent errors (POSIX)
4913 - fix permission bug in lovstripe.c test (624321)
4914 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4915 - truncate past end of file could corrupt data
4916 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4917 - a race in recovery could return ETIMEDOUT to apps (623947)
4918 - building outside the source directory was fixed
4919 * the lustre-HOWTO was brought (more) up to date (582544)
4920 * major progress was made on recovery functionality
4922 2002-10-10 Phil Schwan <phil@clusterfs.com>
4925 - recovery deadlock fix
4926 - rm -rf causes LBUG fix (617817)
4927 - file open by multiple tasks fix (618962)
4928 - directory permissions bugs (602707 and 620007)
4929 - journal_stop fixed with locking (611313)
4930 - O_APPEND failures resolved (618273, perhaps 614459)
4931 - lconf PATH fix (619770)
4932 - IA64 build fix (621450)
4933 - RPC buffer sizes scale with amount of memory
4935 2002-10-01 Phil Schwan <phil@clusterfs.com>
4938 - locks would be cancelled without throwing away data pages,
4939 resulting in inconsistent data (605627)
4940 - inode attributes were not always being refreshed (605627, 612449)
4941 - lconf now continues to cleanup after lctl reports an error
4942 - MDS now enforces user permissions (602707)
4943 - lprocfs cleanup fixed, but not yet enabled (614157)
4944 - fixed infinite server hang, should a client not respond to an AST
4945 - avoid going into recovery if user calls readlink() with a buffer
4946 that's too small (613941)
4947 - AST RPCs no longer require replies (614867) -- this may be changed
4948 - don't crash server if client sends an IOV that's too big (611336)
4949 - fixed lock conversion deadlock (611892)
4950 - fixed the following of symlinks (614622)
4951 * recovery: the server can remove locks from a client that dies, other
4952 clients can make progress
4953 * more extN patch fixes
4954 * compile-time configurable ptlrpc buffer allocations
4956 - collaborative read cache document
4957 - Lustre Lite Performance CDR document-in-progress
4959 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4962 - fix typo in patch-2.4.18
4964 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4967 - clear ptlrpc request each time in handle_incoming_request()
4968 - unlink of files now destroys the object on the OST
4970 2002-09-19 Peter Braam <braam@clusterfs.com>
4972 * add hard link support
4973 * change obdfile creation method
4974 * kernel patch changed
4976 2002-09-19 Peter Braam <braam@clusterfs.com>
4979 - stack overflow bug in extN fixed
4981 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4983 * documentation updates
4984 - add man pages for config tools
4985 - update tests/README to describe testing with new config tools
4986 - finish metadata API descriptions
4987 * bug fixes and cleanups
4988 - statfs workaround for 16TB limit
4989 - LOV stripe allocation improved, can stripe on subset of OSTs
4990 - LOV file size/IO offset was wrong for files > 4GB in size
4991 - object EA data was being dropped, caused files to be unreadable
4992 - memory overflow with non-LOV OST caused memory corruption
4993 - fixed regression tests to work with new config tools, obdfilter
4994 - fixed bug when directory size became larger than 1 block
4995 - fixed bug (for single client case) when PWD was deleted
4996 - invalidate local directory pages when doing intent-based ops
4997 - avoid LDLM oops when lock callback contained bad data
4999 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5001 * documentation updates
5002 * bug fixes and cleanups
5003 - configuration tools
5006 - 64-bit compile warnings
5007 - 64-bit internal statfs data
5009 * test_brw on persistent OST devices
5011 * lprocfs (disabled)
5013 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5015 * documentation updates
5016 * bug fixes and cleanups
5017 * configuration tools
5019 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5022 * many small fixes to 0_5_4
5023 * io/network handling
5024 * thinkos in MDS operations
5026 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5029 * crucial basic fixes to 0.5.3
5030 * IOR, Iozone work over Elan
5033 2002-08-07 Phil Schwan <phil@clusterfs.com>
5034 * version 0_5_3, our first alpha
5035 * we use the new Portals iovs
5036 * documentation updates
5037 * bug fixes and cleanups
5038 * small changes in the DLM wire protocol
5040 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5041 * version 0_5_1 with some initial stability,
5042 * locking on MD and file I/O.
5043 * documentation updates
5044 * several bug fixes since 0.5.0
5045 * small changes in wire protocol
5047 2002-07-18 Phil Schwan <phil@clusterfs.com>
5049 * delivered as Lustre Light Alpha
5050 * fixed a crash after handling invalid MDS requests
5051 * fixed directory pages for architectures with non-4k pages sizes
5053 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5054 * release version v0_4_4
5055 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5057 * Fixes a number of interruption problems with OST operations.
5058 * Update documentation for portals header changes
5059 * Move all wire protocol structs/defines to lustre_idl.h
5060 * Fixes symlink length bug.
5061 * Add tcpdump to repository.
5063 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5064 * release version v0_4_3
5065 * Fixes statfs for inodes on extN.
5066 * Fixes bug in runtests which would delete /etc/hosts.
5067 * Use 64-bit object IDs wherever possible (not into VFS though)
5068 Remove ost_get_info, which is unused by lustre, and out of date.
5070 2002-07-03 Peter Braam <braam@clusterfs.com>
5071 * release version v0_4_2 Fixes a lookup error (type not passed)
5072 * move forward to head of Portals
5073 * move forward to latest Lustre kernel
5075 2002-06-25 Peter Braam <braam@clusterfs.com>
5076 * release version v0_4_1. Hopefully stable on single node use.