1 tbd Sun Microsystems, Inc.
5 2.6.9-67.0.7.EL (RHEL 4),
6 2.6.16.54-0.2.5 (SLES 10),
7 2.6.18-53.1.14.el5 (RHEL 5),
8 2.6.22.14 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 * Client support for unpatched kernels:
13 we do not recommend using patchless RHEL4 clients with kernels
14 prior to 2.6.9-55EL (RHEL4U5).
15 * Recommended e2fsprogs version: 1.40.7-sun1
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).
19 * A new quota file format has been introduced in 1.6.5.
20 The format conversion from prior releases is handled transparently,
21 but releases older than 1.4.12/1.6.5 will not understand this new
22 format. The automatic format conversion can be avoided by running
23 the following command on the MDS:
24 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
25 For more information, please refer to bugzilla 13904.
29 Frequency : after recovery
30 Description: precreate to many object's after del orphan.
31 Details : del orphan st in oscc last_id == next_id and this triger growing
32 count of precreated objects. Set flag LOW to skip increase count
33 of precreated objects.
37 Frequency : rare, on clear nid stats
38 Description: ASSERTION(client_stat->nid_exp_ref_count == 0)
39 Details : when clean nid stats sometimes try destroy live entry,
40 and this produce panic in free.
44 Description: Stack overflow during MDS log replay
45 Details : ease stack pressure by using a thread dealing llog_process.
50 Description: MDT cannot be unmounted, reporting "Mount still busy"
51 Details : Mountpoint references were being leaked during open reply
52 reconstruction after an MDS restart. Drop mountpoint reference
53 in reconstruct_open() and free dentry reference also.
57 Description: wait until IO finished before start new when do lock cancel.
58 Details : VM protocol want old IO finished before start new, in this case
59 need wait until PG_writeback is cleared until check dirty flag and
60 call writepages in lock cancel callback.
64 Description: mds_mfd_close() ASSERTION(rc == 0)
65 Details : In mds_mfd_close(), we need protect inode's writecount change
66 within its orphan write semaphore to prevent possible races.
70 Frequency : rare, on shutdown ost
71 Description: don't hit live lock with umount ost.
72 Details : shrink_dcache_parent can be in long loop with destroy dentries,
73 use shrink_dcache_sb instead.
77 Description: don't panic with use echo client
78 Details : echo client pass NULL as client nid pointer and this produce null
83 Description: fix build on ppc32
84 Details : compile code with -m64 flag produce wrong object file for ppc32.
88 Description: MDS LBUG: ASSERTION(!IS_ERR(dchild))
89 Details : In reconstruct_* functions, LASSERTs on both the data supplied by
90 a client, and the data on disk are dangerous and incorrect. Change
91 them with client eviction.
93 -------------------------------------------------------------------------------
96 2008-05-26 Sun Microsystems, Inc.
98 * Support for kernels:
100 2.6.9-67.0.7.EL (RHEL 4),
101 2.6.16.54-0.2.5 (SLES 10),
102 2.6.22.14 vanilla (kernel.org)
103 * Client support for unpatched kernels:
104 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
105 2.6.16 - 2.6.22 vanilla (kernel.org)
106 * Due to problems with nested symlinks and FMODE_EXEC (bug 12652),
107 we do not recommend using patchless RHEL4 clients with kernels
108 prior to 2.6.9-55EL (RHEL4U5).
109 * Recommended e2fsprogs version: 1.40.7-sun1
110 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
111 * RHEL 4 and RHEL 5/SLES 10 clients behaves differently on 'cd' to a
112 removed cwd "./" (refer to Bugzilla 14399).
113 * A new quota file format has been introduced in 1.6.5.
114 The format conversion from prior releases is handled transparently,
115 but releases older than 1.4.12/1.6.5 will not understand this new
116 format. The automatic format conversion can be avoided by running
117 the following command on the MDS before upgrading:
118 'tunefs.lustre --param="mdt.quota_type=ug1" $MDTDEV'.
119 For more information, please refer to bugzilla 13904.
122 Frequency : very rare, if additional xattrs are used on kernels >= 2.6.12
124 Description: MDS may lose file striping (and hence file data) in some cases
125 Details : If there are additional extended attributes stored on the MDS,
126 in particular ACLs, SELinux, or user attributes (if user_xattr
127 is specified for the client mount options) then there is a risk
128 of attribute loss. Additionally, the Lustre file striping
129 needs to be larger than default (e.g. striped over all OSTs),
130 and an additional attribute must be stored initially in the
131 inode and then increase in size enough to be moved to the
132 external attribute block (e.g. ACL growing in size) for file
135 Severity : enhancement
137 Description: add message levels for liblustreapi
142 Description: MDT cannot be unmounted, reporting "Mount still busy"
143 Details : Mountpoint references were being leaked during open reply
144 reconstruction after an MDS restart. Drop mountpoint reference
145 in reconstruct_open() and free dentry reference also.
150 Description: fix for occasional failure case of -ENOSPC in recovery-small tests
151 Details : Move the 'good_osts' check before the 'total_bavail' check. This
152 will result in an -EAGAIN and in the exit call path we call
153 alloc_rr() which will with increasing aggressiveness attempt to
154 aquire precreated objects on the minimum number of required OSCs.
158 Description: Use old size assignment to avoid deadlock
159 Details : This reverts the changes in bugs 2369 and bug 14138 that introduced
160 the scheduling while holding a spinlock. We do not need locking
161 for size in ll_update_inode() because size is only updated from
162 the MDS for directories or files without objects, so there is no
163 other place to do the update, and concurrent access to such inodes
164 are protected by the inode lock.
168 Description: Use __u64 instead of int for valid bits
172 Description: resolve "_IOWR redefined" build error on SLES10
176 Description: dump the memory debugging after all modules are unloaded to
177 suppress false negative in conf_sanity test 39
181 Description: the recovery timer never expires
182 Details : for new client connect request, the recovery timer should not be
183 reset, otherwise recovery timer will never expired, if the old
184 client never come. Only old client connect and first connection
185 req should trigger recovery timer reset.
189 Description: the min numbers of lproc stats are wrong
190 Details : adding a new constant LC_MIN_INIT and use it for initialization
194 Frequency : always with interactive lfs
196 Description: Reinitialize optind to 0 so that interactive lfs works in all cases
199 Frequency : with multiple concurrent readdir processes in same directory
200 Bugzilla : 15406, 15169, 15175
201 Description: misc fixes for directory readahead.
202 Details : prevent previous statahead async RPC callback from processing the
203 current "statahead_info", race condition between async RPC callback
204 add dentry into dentry hash table and "ls" thread revalidate such
205 dentry, statahead his/miss control for hidden items, and so on.
207 Severity : enhancement
209 Description: build kernel-ib packages for OFED 1.3 in our release cycle
213 Description: incore types cleaning in quota code (with respect to 64-bit limits)
214 Details : several u32 variables declarations are replaced with u64 declarations
219 Description: fix SLES kernel versioning
220 Details : the kernel version for our SLES 10 kernel did not include a "-"
221 before the "smp" at the end. while this was not a problem in
222 general, it did mean that software trying to use the kernel
223 version to try to detect a vendor specific kernel would fail.
224 this was most evident by the OFED build scripts.
229 Description: Don't update lov_desc members until making sure they are valid
230 Details : When updating lov_desc members via proc fs, need fix their
231 validities before doing the real update.
234 Frequency : very rare
236 Description: don't put request into delay list while invalidate in flight.
237 Details : ptlrpc_delay_request sometimes put in delay list while invalidate
238 import in flight. this produce timeout for invalidate and sometimes
239 can cause stale data.
241 Severity : enhancement
243 Description: Update kernel to SLES9 2.6.5-7.311.
245 Severity : enhancement
247 Description: Update kernel to RHEL4 2.6.9-67.0.7.
252 Frequency : on PPC only
253 Description: not convert ost objects for directory because it's not exist.
254 Details : ll_dir_getstripe assume dirrectory has ost objects but this wrong.
256 Severity : enhancement
258 Description: Fix warnings with compile liblustre at sles10/rhel5 which have
259 __u64 as usingied long long type.
262 Frequency : rare, on shutdown
264 Description: race process ast vs remove callback
265 Details : removing callback before disconnect import open race with
268 Severity : enhancement
270 Description: Update kernel to SLES9 2.6.5-7.311.
272 Severity : enhancement
274 Description: Files open for execute are not marked busy on SLES10
275 Details : Add FMODE_EXEC to SLES10 SP1 server kernel series.
277 Severity : enhancement
279 Description: Add server support for vanilla-2.6.22.14.
282 Frequency : occasional
284 Description: Avoid lov_create() getting stuck in obd_statfs_rqset()
285 Details : If an OST is down the MDS will hang indefinitely in
286 obd_statfs_rqset() waiting for the statfs data. While for
287 MDS QOS usage of statfs, it should not stuck in waiting.
289 Severity : enhancement
291 Description: Disable adaptive timeouts by default
294 Frequency : on network error
296 Description: panic with double free request if network error
297 Details : mdc_finish_enqueue is finish request if any network error occuring,
298 but it's true only for synchronus enqueue, for async enqueue
299 (via ptlrpcd) this incorrect and ptlrpcd want finish request
303 Frequency : rare, on recovery
305 Description: read procfs can produce deadlock in some situation
306 Details : Holding lprocfs lock with send rpc can produce block for destroy
307 obd objects and this also block reconnect with -EALREADY.
308 This isn't fix all lprocfs bugs - but make it rare.
310 Severity : enhancement
312 Description: Update kernel to RHEL5 2.6.18-53.1.14.el5.
315 Frequency : frequent on X2 node
317 Description: mdc_set_open_replay_data LBUG
318 Details : Set replay data for requests that are eligible for replay.
323 Description: lustre_mgs: operation 101 on unconnected MGS
324 Details : When MGC is disconnected from MGS long enough, MGS will evict the
325 MGC, and late on MGC cannot successfully connect to MGS and a lot
326 of the error messages complaining that MGS is not connected.
329 Frequency : rare, depends on device drivers and load
331 Description: MDS or OSS nodes crash due to stack overflow
332 Details : Code changes in 1.6.4 increased the stack usage of some functions.
333 In some cases, in conjunction with device drivers that use a lot
334 of stack, the MDS (or possibly OSS) service threads could overflow
335 the stack. One change which was identified to consume additional
336 stack has been reworked to avoid the extra stack usage.
338 Severity : enhancement
340 Description: Update to RHEL5 latest kernel-2.6.18-53.1.13.el5.
342 Severity : enhancement
344 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.5.
346 Severity : enhancement
348 Description: Update to RHEL5 latest kernel-2.6.18-53.1.6.el5.
350 Serverity : enhancement
352 Description: Update RHEL4 kernel to 2.6.9-67.0.4.
355 Frequency : rare on shutdown OST
357 Description: Don't allow skipping OSTs if index has been specified.
358 Details : Don't allow skipping OSTs if index has been specified, make locking
359 in internal create lots better.
364 Description: ASSERTION(!PageDirty(page)) failed
365 Details : Wrong check could lead to an assertion failure under specific
371 Description: LBUG in ptlrpc_check_set() bad phase ebc0de00
372 Details : access to bitfield in structure is always rounded to long
373 and this produce problem with not atomic change any bit.
378 Description: Lustre make rpms failed.
379 Details : Remove ldiskfs spec file to avoids rpmbuild be confused when
380 builds Lustre rpms from tarball.
382 Severity : enhancement
384 Description: Update to SLES9 SP4 kernel-2.6.5-7.308.
387 Frequency : rare on shutdown OST
389 Description: If llog cancel was not send before clean_exports phase, this can
390 produce deadlock in llog code.
391 Details : If llog thread has last reference to obd and call class_import_put
392 this produce deadlock because llog_cleanup_commit_master wait when
393 last llog_commit_thread exited, but this never success because was
394 called from llog_commit_thread.
397 Frequency : only if OST index is skipped
399 Description: NULL lov_tgts causing MDS oops
400 Details : more safe checks for NULL lov_tgts for avoid oops.
402 Severity : enhancement
404 Description: Update to RHEL4 latest kernel-2.6.9-67.0.1.EL.
406 Severity : enhancement
408 Description: Update to RHEL5 latest kernel-2.6.18-53.1.4.el5.
413 Description: make mgs_setparam() handle fsname containing dash
414 Details : fsname containing a dash does not work with lctl conf_param
416 Severity : enhancement
418 Description: Update to RHEL4 Update-6 kernel-2.6.9-67.EL.
421 Frequency : rare, in recovery and (or) destroy lovobjid file.
423 Description: rewrite lov objid code.
424 Details : Cleanup for lov objid code, remove scability problems and wrong
425 locking. Fix sending last_id into ost.
427 Severity : enhancement
429 Description: Update to SLES10 SP1 latest kernel-2.6.16.54-0.2.3.
431 Severity : enhancement
433 Description: Update to RHEL5 Update-1 kernel 2.6.18-53.el5.
434 Details : Use d_move_locked instead of __d_move.
437 Frequency : rare, at shutdown
439 Description: access already free / zero obd_namespace.
440 Details : if client_disconnect_export was called without force flag set,
441 and exist connect request in flight, this can produce access to
442 NULL pointer (or already free pointer) when connect_interpret
443 store ocd flags in obd_namespace.
446 Frequency : only at startup
448 Description: not alloc memory with spinlock held.
449 Details : allocation memory with GFP_KERNEL can produce sleep deadlock,
450 if any spinlock held.
455 Description: lfs find does not continue on file error
456 Details : Continue other files processing when a file/dir is absent.
460 Description: Inconsistent usage of lustre_pack_reply()
461 Details : Standardize the usage of lustre_pack_reply() such that it
462 always generate a CERROR on failure.
465 Frequency : very rare
467 Description: Fix replay if there is an un-replied request and open
468 Details : In some cases, older replay request will revert the
469 mcd->mcd_last_xid on MDS which is used to record the client's
472 Severity : enhancement
474 Description: Update to RHEL5 kernel 2.6.18-8.1.15.el5.
476 Severity : enhancement
478 Description: Update to SLES10 SP1 kernel 2.6.16.53-0.16
480 Severity : enhancement
482 Description: Update to SLES9 kernel-2.6.5-7.287.3.
484 Severity : enhancement
486 Description: Update to RHEL4 kernel-2.6.9-55.0.12.EL.
488 Severity : enhancement
490 Description: Build SLES10 patchless client fails
491 Details : The configure was broken by run ./configure with
492 --with-linux-obj=.... argument for patchless client. When the
493 configure use --with-linux-obj, the LINUXINCLUDE= -Iinclude
494 can't search header adequately. Use absolute path such as
495 -I($LINUX)/include instead.
497 Severity : enhancement
499 Description: Lustre Page Accounting
500 Details : New macros for page alloc and free which enable accounting
501 of page allocation of Lustre. Use percpu counters to store memory
505 Frequency : only if debugging is disabled
507 Description: LASSERT_{REQ,REP}SWAB macros are buggy
508 Details : If SWAB_PARANOIA is disabled, the LASSERT_REQSWAB and
509 LASSERT_REPSWAB macros become no-ops, which is incorrect. Drop
510 these macros and replace them with their definitions instead.
515 Description: interrupt oig_wait produce painc on resend.
516 Details : brw_redo_request can be used for resend requests from ptlrpcd and
517 private set, and this produce situation when rq_ptlrpcd_data not
518 copyed to new allocated request and triggered LBUG on assert
519 req->rq_ptlrpcd_data != NULL. But this member used only for wakeup
520 ptlrpcd set if request is changed and can be safety changed to use
523 Severity : enhancement
525 Description: organize the server-side client stats on per-nid basis
526 Details : Change the structure of stats under obdfilter and mds to
534 The "uuid"s file would list the uuids of _active_ exports.
535 And the clear entry is to clear all stats and stale nids.
540 Description: Processes looping in ll_readdir() on Lustre clients finally causing
541 a full node pseudo-hang
542 Details : Concurrent access to the same directory from multiple clients with
543 intensive file creation/removal can cause a client node to spin in
544 ll_readdir(). i_version must be increased every time the lock is
545 cancelled to ensure a revalidate is done.
550 Description: touch file failed when fs is not full
551 Details : OST in recovery should not be discarded by MDS in alloc_qos(),
552 otherwise we can get ENOSP while fs is not full.
555 Frequency : only for Cray XT3
556 Bugzilla : 12829/13455
557 Description: Changing primary group doesn't change the group lustre assigns to
559 Details : When CRAY_XT3 is defined, the fsgid supplied by the client is
560 overridden with the primary group provided by the group upcall,
561 whereas the supplied fsgid can be trusted if it is in the list of
562 supplementary groups returned by the group upcall.
564 Severity : enhancement
566 Description: Root Squash Functionality
567 Details : Implementation of NFS-like root squash capability. Specifically,
568 don't allow someone with root access on a client node to be able
569 to manipulate files owned by root on a server node.
571 Severity : enhancement
573 Description: Slow trucate/writes to huge files at high offsets.
574 Details : Directly associate cached pages to lock that protect those pages,
575 this allows us to quickly find what pages to write and remove
576 once lock callback is received.
581 Description: Too many locks accumulating on client during NFS usage
582 Details : mds_open improperly used accmode to find out access mode to a
583 file. Also mdc_intent_lock logic to find out if we already have
584 lock similar to just received was flawed since introduction of
585 skiplists - locks are now added to the front of the granted
590 Description: Hit ASSERTION(obd->obd_stopping == 1) failed in some setup failed
592 Details : In obd setup failure handler, obd_stopping will not necessarily to
593 be 1, and obd_set_up should also be checked to make sure whether
594 obd is completely setup.
596 Severity : enhancement
598 Description: Allow masking D_WARNING, D_ERROR messages from console
599 Details : Console messages can now be disabled via lnet.printk.
604 Description: User code with malformed file open parameter crashes client node
605 Details : Before packing join_file req, all the related reference should be
606 checked carefully in case some malformed flags cause fake join_file
612 Description: shrink/enlarge qunit size when needed; fix the problem of coarse
613 grain of quota doing harm to quota's accuracy
614 Details : qunit size will be changed when quota limitation is too low/high;
615 record the pending quota write in order to get more accureate
616 quota; delete the patch for bug12588, which is unnecessary when
617 this patch is landed. This bug also contains fixes for bug 14526,
618 14299, 14601 and 13794.
622 Description: LDLM_ENQUEUE races with LDLM_CP_CALLBACK
623 Details : ldlm_completion_ast() assumes that a lock is granted when the req
624 mode is equal to the granted mode. However, it should also check
625 that LDLM_FL_CP_REQD is not set.
629 Description: Heavy nfs access might result in deadlocks
630 Details : After ELC code landed, it is now improper to enqueue any mds
631 locks under och_sem, because enqueue might want to decide to
632 cancel open locks for same inode we are holding och_sem for.
636 Description: 35% write performance drop with ldiskfs2 when quotas are on
637 Details : Enable ext3 journalled quota by default to improve performance
638 when quotas are turned on.
642 Description: Client eviction while running blogbench
643 Details : A lot of unlink operations with concurrent I/O can lead to a
644 deadlock causing evictions. To address the problem, the number of
645 oustanding OST_DESTROY requests is now throttled to
646 max_rpcs_in_flight per OSC and LDLM_FL_DISCARD_DATA blocking
647 callbacks are processed in priority.
650 Frequency : RHEL4 only
652 Description: mkfs is very slow on IA64/RHEL4
653 Details : A performance regression has been discovered in the MPT Fusion
654 driver between versions 3.02.73rh and 3.02.99.00rh. As a
655 consequence, we have downgraded the MPT Fusion driver in the RHEL4
656 kernel from 3.02.99.00 to 3.02.73 until this problem is fixed.
659 Frequency : PPC/PPC64 only
661 Description: conflicts between asm-ppc64/types.h and lustre_types.h
662 Details : fix duplicated definitions between asm-ppc64/types.h and
663 lustre_types.h on PPC.
666 Frequency : PPC/PPC64 only
668 Description: asm-ppc/segment.h does not exist
669 Details : fix compile issue on PPC.
673 Description: data checksumming impacts single node performance
674 Details : add support for several checksum algorithms. Currently, CRC32 and
675 Adler-32 are supported. The checksum type can be changed on the fly
676 through /proc/fs/lustre/osc/*/checksum_type.
680 Description: use adler32 for page checksums
681 Details : when available, use the Adler-32 algorithm instead of CRC32 for
686 Description: better handle error messages in extents code
688 Severity : enhancement
690 Description: SNMP support enhancement
691 Details : Adding total number of sampled request for an MDS node in snmp
694 Severity : enhancement
696 Description: Optimize ldlm waiting list processing for PR extent locks
697 Details : When processing waiting list for read extent lock and meeting read
698 lock that is same or wider to it that is not contended, skip
699 processing rest of the list and immediatelly return current
700 status of conflictness, since we are guaranteed there are no
701 conflicting locks in the rest of the list.
705 Description: Time out and refuse to reconnect
706 Details : When the failover node is the primary node, it is possible
707 to have two identical connections in imp_conn_list. We must
708 compare not conn's pointers but NIDs, otherwise we
709 can defeat connection throttling.
713 Description: Client not clear own cache if answer to reconnect is lost.
714 Details : Client gets evicted from server. Now client also thinks it is
715 disconnected (or gets ENOTCONN on its operation) and decides to
716 reconnect. Server receives reconnect message, but cannot find
717 export. New export is created that is fully valid (new cookie!),
718 but reply is lost and not reported to client. Client reconnects
719 again and gets back a just-created connection, but it is not new
720 so client thinks it was not evicted and does not do recovery.
724 Description: Detect stride IO mode in read-ahead
725 Details : When a client does stride read, read-ahead should detect that and
726 read-ahead pages according to the detected stride pattern.
730 Description: build for x2 fails
731 Details : fix compile issue on Cray systems.
733 Severity : enhancement
735 Description: implement readv/writev APIs(aio_read/aio_writes in newer kernels)
736 Details : This greatly improves speed of NFS writes on 2.6 kernels.
739 Frequency : only on PPC/SLES10
741 Description: "BITS_PER_LONG is not 32 or 64" in linux/idr.h
742 Details : On SLES10/PPC, fs.h includes idr.h which requires BITS_PER_LONG to
743 be defined. Add a hack in mkfs_lustre.c to work around this compile
748 Description: LASSERT on MDS when client holding flock lock dies
749 Details : ldlm pool logic depends on number of granted locks equal to
750 number of released locks which is not true for flock locks, so
751 just exclude such locks from consideration.
755 Description: MDS deadlock with many ll_sync_lov threads and I/O stalled
756 Details : Use fsfilt_sync() for both the whole filesystem sync and
757 individual file sync to eliminate dangerous inode locking
758 with I_LOCK that can lead to a deadlock.
762 Description: Update an obsolete wirecheck.c generator
763 Details : Update wirecheck.c/wirehdr.c and regenerate wiretest.c
767 Description: Client can panic on open sometimes
768 Details : It is possible that we try to free already freed request in
769 ll_file_open in some error cases when we send request from
774 Description: performance in 1.6.3
775 Details : Force q->max_phys_segments to MAX_PHYS_SEGMENTS on SLES10 to be
776 sure that 1MB requests are not fragmented by the block layer.
780 Description: LDLM soft lockups - improvement
781 Details : It is be possible to send the lock handle along with each read
782 or write request because the client is already doing a lock match
783 itself so there isn't any reason the OST should have to re-do that
789 Description: lfs quota fails with deactivated OSTS
790 Details : With this patch, three improvements are included:
791 1. detete the softlimit in mds and osts when use "lfs quota".
792 2. display the inaccurate data in the output of "lfs quota".
793 3. try to get quota info when "lfs quota" is executed.
798 Description: Extent locks not granted with no conflicts sometimes.
799 Details : When race occurs in glimpse handler and nothing is returned,
800 we do not reprocess the queue after lock cancel, and that leads
801 to a stall until next activity on a resource
804 Frequency : failover with quotaon
806 Description: during mds failovers with quota on, OSTs got into deadlock state
807 and causing dumpstack.
808 Details : for every quota slave, at any time, there is only one quota req
809 is sent to quota master for every uid/gid. Before that quota req
810 returns, all the thread relative to the same uid/gid will wait.
811 So if the quota req is lost because mds failovers or any other
812 reasons, this bug will be hit. Now, dqacq_interpret() will handle
813 quota reqs who time out.
815 Severity : enhancement
818 Description: when quota slave checks if quota is enough, there is an unnecessary
820 Details : place this wait on necessary place instead of always waiting.
822 --------------------------------------------------------------------------------
824 2007-12-07 Cluster File Systems, Inc. <info@clusterfs.com>
826 * Support for kernels:
827 2.6.5-7.286 (SLES 9),
828 2.6.9-55.0.9.EL (RHEL 4),
829 2.6.16.53-0.8 (SLES 10),
830 2.6.18-8.1.14.el5 (RHEL 5),
831 2.6.18.8 vanilla (kernel.org)
832 * Client support for unpatched kernels:
833 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
834 2.6.16 - 2.6.22 vanilla (kernel.org)
835 * Due to recently discovered recovery problems, we do not recommend
836 using patchless RHEL 4 clients with this or any earlier release.
837 * Recommended e2fsprogs version: 1.40.2-cfs1
838 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
841 Frequency : occasional
843 Description: MDS hang or stay in waiting lock
844 Details : If client receive lock with CBPENDING flag ldlm need send lock
845 cancel as separate rpc, to avoid situation when cancel request
846 can't processed due all i/o threads stay in wait lock.
849 Frequency : occasional
851 Description: Do not fail import if osc_interpret_create gets -EAGAIN
852 Details : If osc_interpret_create got -EAGAIN it immediately exits and
853 wakeup oscc_waitq. After wakeup oscc_wait_for_objects call
854 oscc_has_objects and see OSC has no objests and call
855 oscc_internal_create to resend create request.
857 Severity : enhancement
859 Description: Update kernel patches for SLES10 2.6.16.53-0.8.
860 Details : Update which_patch & target file for SLES10 latest kernel.
862 Severity : enhancement
864 Description: add --type and --size parameters to lfs find
865 Details : Enhance lfs find by adding filetype and filesize parameters. Also
866 multiple OBDs can now be specified for the --obd option.
868 Severity : enhancement
870 Description: eliminate client locks in face of contention
871 Details : file contention detection and lockless i/o implementation
874 Severity : enhancement
876 Description: Remove client patches from SLES 10 kernel.
877 Details : This causes SLES 10 clients to behave as patchless clients
878 even on a Lustre-patched (server) kernel.
880 Severity : enhancement
882 Description: use i_size_read and i_size_write in 2.6 port
883 Details : replace inode->i_size access with i_size_read/write()
886 Frequency : when removing large files
888 Description: scheduling issue during removal of large Lustre files
889 Details : Don't take the BKL in fsfilt_ext3_setattr() for 2.6 kernels.
890 It causes scheduling issues when removing large files (17TB in the
896 Description: 1.4.11 Can't handle directories with stripe set and extended ACLs
897 Details : Impossible (EPROTO is returned) to access a directory that has a
898 non-default striping and ACLs.
901 Frequency : only on ppc
903 Description: /proc/fs/lustre/devices broken on ppc
904 Details : The patch as applied to 1.6.2 doesn't look correct for all arches.
905 We should make sure the type of 'index' is loff_t and then cast
906 explicitly as needed below. Do not assign an explicitly cast
910 Frequency : only for rhel5
912 Description: Kernel patches update for RHEL5 2.6.18-8.1.10.el5.
913 Details : Modify the target file & which_kernel.
916 Frequency : if the uninit_groups feature is enabled on ldiskfs
918 Description: e2fsck reports "invalid unused inodes count"
919 Details : If a new ldiskfs filesystem is created with the "uninit_groups"
920 feature and only a single inode is created in a group then the
921 "bg_unused_inodes" count is incorrectly updated. Creating a
922 second inode in that group would update it correctly.
927 Description: buffer overruns could theoretically occur
928 Details : llapi_semantic_traverse() modifies the "path" argument by
929 appending values to the end of the origin string, and a buffer
930 overrun may occur. Adding buffer overrun check in liblustreapi.
932 Severity : enhancement
934 Description: Add jbd statistics patch for RHEL5 and 2.6.18-vanilla.
937 Frequency : only if filesystem is inconsistent
939 Description: handle "serious error: objid * already exists" more gracefully
940 Details : If LAST_ID value on disk is smaller than the objects existing in
941 the O/0/d* directories, it indicates disk corruption and causes an
942 LBUG(). If the object is 0-length, then we should use the existing
943 object. This will help to avoid a full fsck in most cases.
945 Severity : enhancement
947 Description: Kernel patches update for RHEL4 2.6.9-55.0.6.
948 Details : Modify vm-tunables-rhel4.patch.
950 Severity : enhancement
952 Description: Kernel config for 2.6.18-vanilla.
953 Details : Modify targets/2.6-vanilla.target.in.
954 Add config file kernel-2.6.18-2.6-vanilla-i686.config.
955 Add config file kernel-2.6.18-2.6-vanilla-i686-smp.config.
956 Add config file kernel-2.6.18-2.6-vanilla-x86_64.config.
957 Add config file kernel-2.6.18-2.6-vanilla-x86_64-smp.config.
960 Frequency : occasional
962 Description: improve handling recoverable errors
963 Details : If request processed with error which can be recoverable on server
964 request should be resend, otherwise page released from cache and
970 Description: Kernel patches update for RHEL5 2.6.18-8.1.14.el5.
971 Details : Modify target file & which_patch.
972 A flaw was found in the IA32 system call emulation provided
973 on AMD64 and Intel 64 platforms. An improperly validated 64-bit
974 value could be stored in the %RAX register, which could trigger an
975 out-of-bounds system call table access. An untrusted local user
976 could exploit this flaw to run code in the kernel
977 (ie a root privilege escalation). (CVE-2007-4573).
981 Description: change order of libsysio includes
982 Details : '#include sysio.h' should always come before '#include xtio.h'
984 Severity : enhancement
986 Description: adapt the lustre_config script to support the upgrade case
987 Details : Add "-u" option for lustre_config script to support upgrading 1.4
988 server targets to 1.6 in parallel.
993 Description: To avoid grant space > avaible space when the disk is almost
994 full. Without this patch you might see the error "grant XXXX >
995 available" or some LBUG about grant, when the disk is almost
997 Details : In filter_check_grant, for non_grant cache write, we should
998 check the left space by if (*left > ungranted + bytes), instead
999 of (*left > ungranted), because only we are sure the left space
1000 is enough for another "bytes", then the ungrant space should be
1001 increase. In client, we should update cl_avail_grant only there is
1002 OBD_MD_FLGRANT in the reply.
1007 Description: Update RHEL 4 kernel to fix local root privilege escalation.
1008 Details : Update to the latest RHEL 4 kernel to fix the vulnerability
1009 described in CVE-2007-4573. This problem could allow untrusted
1010 local users to gain root access.
1013 Frequency : when using O_DIRECT and quotas
1015 Description: Incorrect file ownership on O_DIRECT output files
1016 Details : block usage reported by 'lfs quota' does not take into account
1017 files that have been written with O_DIRECT.
1019 --------------------------------------------------------------------------------
1021 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
1023 * Support for kernels:
1024 2.6.5-7.286 (SLES 9),
1025 2.6.9-55.0.2.EL (RHEL 4),
1026 2.6.16.46-0.14 (SLES 10),
1027 2.6.18-8.1.8.el5 (RHEL 5),
1028 2.6.18.8 vanilla (kernel.org)
1029 * Client support for unpatched kernels:
1030 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1031 2.6.16 - 2.6.21 vanilla (kernel.org)
1032 * Due to recently discovered recovery problems, we do not recommend
1033 using patchless RHEL 4 clients with this or any earlier release.
1034 * Recommended e2fsprogs version: 1.40.2-cfs1
1035 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1039 Description: Fix errors in lfs documentation
1040 Details : Fixes man pages
1042 Severity : enhancement
1044 Description: Adaptive timeouts
1045 Details : RPC timeouts adapt to changing server load and network
1046 conditions to reduce resend attempts and improve recovery time.
1048 Severity : enhancement
1050 Description: llapi_file_create() does not allow some changes
1051 Details : add llapi_file_open() that allows specifying the file creation
1052 mode and open flags, and also returns an open file handle.
1054 Severity : enhancement
1056 Description: df doesn't work properly if diskfs blocksize != 4K
1057 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
1060 Frequency : always on directories with default striping set
1062 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
1063 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
1065 Severity : enhancement
1067 Description: merge and cleanup kernel patches.
1068 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
1070 Severity : enhancement
1072 Description: RedHat Update kernel for RHEL5
1073 Details : Kernel config file for RHEL5.
1075 Severity : enhancement
1077 Description: OSS needs mutliple precreate threads
1078 Details : Add ability to start more than one create thread per OSS.
1081 Frequency : only with quota on the root user
1083 Description: mds_obd_create error creating tmp object
1084 Details : When the user sets quota on root, llog will be affected and can't
1085 create files and write files.
1088 Frequency : Always on ia64 patchless client, and possibly others.
1090 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
1091 Details : This allows the patchless client to be loaded on architectures
1092 without this export.
1097 Description: disorder of journal start and llog_add cause deadlock.
1098 Details : in llog_origin_connect, journal start should happen before
1099 llog_add keep the same order as other functions to avoid
1102 Severity : enhancement
1104 Description: RedHat Update kernel for RHEL5
1105 Details : Modify the kernel config file more closer RHEL5.
1107 Severity : enhancement
1109 Description: Build failure against Centos5 (RHEL5)
1110 Details : Define PAGE_SIZE when it isn't present.
1113 Frequency : occasionally when using NFS
1115 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
1116 Details : with NFS, the anon dentry's parent was set to itself in
1117 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
1118 corresponding dentry other than use rec->ur_name.
1120 Severity : enhancement
1122 Description: enable data checksumming by default
1123 Details : enable checksum by default, allow --disable-checksum
1124 configure option and "-o nochecksum" mount option. Checksums
1125 can also be disabled at runtime via $LPROC/osc/*/checksum_pages.
1128 Frequency : Occasionally with failover
1130 Description: Client eviction due to failover config
1131 Details : after a connection loss, the lustre client should attempt to
1132 reconnect to the last active server first before trying the
1133 other potential connections.
1135 Severity : enhancement
1137 Description: client-side metadata stat-ahead during readdir(directory readahead)
1138 Details : perform client-side metadata stat-ahead when the client detects
1139 readdir and sequential stat of dir entries therein
1144 Description: when mds and osts use different quota unit(32bit and 64bit),
1145 quota will be released repeatly.
1146 Details : void sending multiple quota reqs to mds, which will keep the
1147 status between the reqs.
1150 Frequency : only with liblustre clients on XT3
1152 Description: evictions taking too long
1153 Details : allow llrd to evict clients directly on OSTs
1157 Description: osts not allocated evenly to files
1158 Details : change the condition to increase offset_idx
1160 Severity : enhancement
1162 Description: self-adjustable client's lru lists
1163 Details : use adaptive algorithm for managing client cached locks lru
1164 lists according to current server load, other client's work
1165 pattern, memory activities, etc. Both, server and client
1166 side namespaces provide number of proc tunables for controlling
1171 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
1172 Details : rewrite inherited code (uuid parsing code from ext2 utils and
1173 readdir code from ext3) from scratch preserving functionality.
1177 Description: Only those disconnect error should be returned by rq_status.
1178 Details : In open/enqueue processs, Some errors, which will cause client
1179 disconnected, should be returned by rq_status, while other
1180 errors should still be returned by intent, then mdc or llite will
1183 Severity : enhancement
1185 Description: Tune the kernel for good SCSI performance.
1186 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
1187 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
1191 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
1192 Bugzilla : 13547, 13627
1193 Description: Data corruption for OSTs that are formatted larger than 2TB
1195 Details : When generating the bio request for lustre file writes the
1196 sector number would overflow a temporary variable before being
1197 used for the IO. The data reads correctly from Lustre (which
1198 will overflow in a similar manner) but other file data or
1199 filesystem metadata may be corrupted in some cases.
1203 Description: "lfs find -obd UUID" prints directories
1204 Details : "lfs find -obd UUID" will return all directory names instead
1205 of just file names. It is incorrect because the directories
1206 do not reside on the OSTs.
1209 Bugzilla : 13596, 13614
1210 Description: MDS hang after unclean shutdown of lots of clients
1211 Details : Never resend AST requests.
1215 Frequency : Always, for kernels after 2.6.16
1216 Description: Fix warning idr_remove called for id=.. which is not allocated.
1217 Details : Last kernels save old s_dev before kill super and not allow
1218 to restore from callback - restore it before call kill_anon_super.
1222 Description: Fix error on 'ls .' at the top of the Lustre mount.
1223 Details : Don't revalidate dentry if it is a root dentry.
1225 --------------------------------------------------------------------------------
1227 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
1229 * Support for kernels:
1230 2.6.5-7.286 (SLES 9),
1231 2.6.9-55.0.2.EL (RHEL 4),
1232 2.6.16.46-0.14 (SLES 10),
1233 2.6.18-8.1.8.el5 (RHEL 5),
1234 2.6.18.8 vanilla (kernel.org)
1235 * Client support for unpatched kernels:
1236 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1237 2.6.16 - 2.6.22 vanilla (kernel.org)
1238 * Due to recently discovered recovery problems, we do not recommend
1239 using patchless RHEL 4 clients with this or any earlier release.
1240 * Recommended e2fsprogs version: 1.39.cfs8
1241 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1246 Description: block reactivating mgc import until all deactivates complete
1247 Details : Fix race when failing back MDT/MGS to itself (testing)
1249 Severity : enhancement
1251 Description: lfs setstripe enhancement
1252 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
1255 Frequency : mds/oss recovery
1257 Description: llog ctxt is refrenced after it has been freed.
1258 Details : llog ctxt refcount was added to avoide the race between ctxt free
1259 and llog recovery process. Each llog user must hold ctxt refcount
1260 before it access the llog. And the llog ctxt can only be freed
1261 when its refcount is zero.
1263 Severity : enhancement
1265 Description: randomly memory allocation failure util
1266 Details : Make lustre randomly failed allocating memory for testing purpose.
1269 Frequency : only for SLES10
1271 Description: Update kernel patch for SLES10 SP1
1272 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
1274 Severity : enhancement
1276 Description: omit set fsid for export NFS
1277 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
1278 as FS_REQUIRES_DEV for avoid problems with generate fsid.
1281 Frequency : after network failures
1283 Description: Add sync option to mount_lustre.c
1284 Details : Client loses data written to lustre after a network interruption.
1286 Severity : enhancement
1288 Description: Error message improvement.
1289 Details : Merging of two LCONSOLE_ERROR_MSG into one.
1291 Severity : enhancement
1293 Description: don't use GFP_* in generic Lustre code.
1294 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
1296 Severity : enhancement
1298 Description: obdclass is limited by single OBD_ALLOC(idarray)
1299 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
1301 Severity : enhancement
1303 Description: updated patchess for new RHEL4 kernel
1304 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
1306 Severity : enhancement
1308 Description: warnings with build patchless client with vanila 2.6.19 and up
1309 Details : change old ctl_table style and replace ctl_table/ctl_table_header
1310 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
1312 Severity : enhancement
1314 Description: O_DIRECT bypasses client statistics.
1315 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
1316 read_ahead_stats were updated. Copied stats section from
1317 osc_send_oap_rpc() into async_internal().
1320 Frequency : only for Cray XT3
1322 Description: peer credits not enough on many OST per OSS systems.
1323 Details : Use new lnet way to add credits as we need those for pings and ASTs
1326 Frequency : only with liblustre
1328 Description: Liblustre is not releasing flock locks on file close.
1329 Details : Release flock locks on file close.
1332 Frequency : only for RHEL4
1334 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
1335 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
1336 Update target file and kernel config.
1340 Description: lustre support for RHEL5
1341 Details : Add support for RHEL5.
1346 Description: ASSERTION(export != NULL) failed in target_handle_connect
1347 Details : Assetion hit is result of rare race between disconnect and connet
1348 to same nid. target_handle_connect found old connect cockie and
1349 tried to reconnect, but can't find export for this cockie.
1354 Description: umount blocks forever on error
1355 Details : In result of wrong using obd_no_recov and obd_force flags client
1356 can hand if cancel or some other requests is lost.
1359 Frequency : Only for SLES
1361 Description: sanity_quota fail test_1
1362 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
1363 file, which makes TSTID[2] inunique.
1365 Severity : enhancement
1367 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
1368 Details : Update target/ChangeLog/which_patch .
1370 Severity : enhancement
1372 Description: jbd statistics
1373 Details : Port older jbd statistics patch for sles10
1378 Description: lvbo_init failed for resource with missing objects.
1379 Details : Fix returning error if we do stat for file with missing/corrupted
1380 objects and i_size set to all sum of size of all avaible objects.
1381 if we to truncate/write to missing object - it is recreated.
1386 Description: Oops in read and write path when failing to allocate lock.
1387 Details : Check if lock allocation failed and return error back.
1390 Frequency : When flocks are used.
1392 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
1393 Details : Flock locks might destroy just granted lock if it could be merged
1394 with another existing flock, this is done in completion handler,
1395 so teach ldlm_cli_enquque_fini that this is a valid case for
1401 Description: reply_lock_interpret crash due to race with it and lock cancel.
1402 Details : Do not replay locks that are being cancelled. Do not reference
1403 locks by their address during replay, just by their handle.
1406 Frequency : only with deactivated OSTs
1408 Description: lstripe command fails for valid OST index
1409 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
1410 instead of lov->desc.ld_active_tgt_count.
1412 Severity : enhancement
1414 Description: Build failure against Centos5 (RHEL5)
1415 Details : Use getpagesize() instead of PAGE_SIZE.
1417 --------------------------------------------------------------------------------
1419 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
1421 * Support for kernels:
1422 2.6.5-7.283 (SLES 9),
1423 2.6.9-55.EL (RHEL 4),
1424 2.6.16.46-0.14 (SLES 10),
1425 2.6.18.8 vanilla (kernel.org)
1426 * Client support for unpatched kernels:
1427 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
1428 2.6.16 - 2.6.22 vanilla (kernel.org)
1429 * Due to recently discovered recovery problems, we do not recommend
1430 using patchless RHEL 4 clients with this or any earlier release.
1431 * Recommended e2fsprogs version: 1.39.cfs8
1432 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
1433 * Starting with this release, the ldiskfs backing filesystem required
1434 by Lustre is now in its own package, lustre-ldiskfs. This package
1435 should be installed. It is versioned separately from Lustre and
1436 may be released separately in future.
1441 Description: server LBUG when shutting down
1442 Details : Block umount forever until the mount refcount is zero rather
1443 than giving up after an arbitrary timeout.
1445 Severity : enhancement
1447 Description: add optional extra BUILD_VERSION info
1448 Details : add a new environment variable (namely LUSTRE_VERS) which allows
1449 to override the lustre version.
1452 Frequency : 2.6.18 servers only
1454 Description: ll_kern_mount() doesn't release the module reference
1455 Details : The ldiskfs module reference count never drops down to 0
1456 because ll_kern_mount() doesn't release the module reference.
1461 Description: server LBUG when using old ost_num_threads parameter
1462 Details : Accept the old ost_num_threads parameter but warn that it
1463 is deprecated, and fix an off-by-one error that caused an LBUG.
1468 Description: Transient SCSI error results in persistent IO issue
1469 Details : iobuf->dr_error is not reinitialized to 0 between two
1473 Frequency : sometimes when underlying device returns I/O errors
1475 Description: OSTs not going read-only during write failures
1476 Details : OSTs are not remounted read-only when the journal commit threads
1477 get I/O errors because fsfilt_ext3 calls journal_start/stop()
1478 instead of the ext3 wrappers.
1481 Frequency : only with 10000 clients or more
1483 Description: poor connect scaling with increasing client count
1484 Details : Don't run filter_grant_sanity_check for more than 100 exports
1485 to improve scaling for large numbers of clients.
1488 Frequency : SLES10 only
1490 Description: sanity-quota.sh quotacheck failed: rc = -22
1491 Details : Quotas cannot be enabled on SLES10.
1494 Frequency : liblustre clients only
1496 Description: getdirentries does not give error when run on compute nodes
1497 Details : getdirentries does not fail when the size specified as an argument
1498 is too small to contain at least one entry
1500 Severity : enhancement
1502 Description: Add LNET router traceability for debug purposes
1503 Details : If a checksum failure occurs with a router as part of the
1504 IO path, the NID of the last router that forwarded the bulk data
1505 is printed so it can be identified.
1510 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
1511 Details : Due to a race condition, liblustre clients were occasionally
1512 evicted incorrectly.
1514 Severity : enhancement
1516 Description: lfs setstripe use optional parameters instead of postional
1519 Severity : enhancement
1521 Description: Nanosecond timestamp support for ldiskfs
1522 Details : The on-disk ldiskfs filesystem has added support for nanosecond
1523 resolution timestamps. There is not yet support for this at
1524 the Lustre filesystem level.
1527 Frequency : during server recovery
1529 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
1530 Details : request with rq_no_resend flag not awake l_wait_event if they get a
1534 Frequency : nfs export on patchless client
1536 Description: connectathon hang when test nfs export over patchless client
1537 Details : Disconnected dentry cannot be found with lookup, so we do not need
1538 to unhash it or make it invalid
1541 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
1542 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
1543 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
1548 Description: open req refcounting wrong on reconnect
1549 Details : If reconnect happened between getting open reply from server and
1550 call to mdc_set_replay_data in ll_file_open, we will schedule
1551 replay for unreferenced request that we are about to free.
1552 Subsequent close will crash in variety of ways.
1553 Check that request is still eligible for replay in
1554 mdc_set_replay_data().
1559 Description: disable writes to filesystem when reading health_check file
1560 Details : the default for reading the health_check proc file has changed
1561 to NOT do a journal transaction and write to disk, because this
1562 can cause reads of the /proc file to hang and block HA state
1563 checking on a healthy but otherwise heavily loaded system. It
1564 is possible to return to the previous behaviour during configure
1565 with --enable-health-write.
1567 Severity : enhancement
1569 Description: 64-bit inode version
1570 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
1571 to the inode. This will be required for version-based recovery.
1576 Description: MDS fails to start if a duplicate client export is detected
1577 Details : in some rare cases it was possible for a client to connect to
1578 an MDS multiple times. Upon recovery the MDS would detect this
1579 and fail during startup. Handle this more gracefully.
1581 Severity : enhancement
1583 Description: Add -o localflock option to simulate old noflock behaviour.
1584 Details : This will achieve local-only flock/fcntl locks coherentness.
1589 Description: log_commit_thread vs filter_destroy race leads to crash
1590 Details : Take import reference before releasing llog record semaphore
1595 Description: Wrong request locking in request set processing
1596 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
1597 imp_delayed_list, in this place should be used imp_lock.
1600 Frequency : when reconnecting
1602 Description: Grant leak when OSC reconnect to OST
1603 Details : When osc reconnect ost, OST(filter) should check whether it
1604 should grant more space to client by comparing fed_grant and
1605 cl_avail_grant, and return the granted space to client instead
1606 of "new granted" space, because client will call osc_init_grant
1607 to update the client grant space info.
1610 Frequency : when client reconnects to OST
1612 Description: Grant leak when OSC does a resend and replays bulk write
1613 Details : When osc reconnect to OST, OST(filter) should clear grant info of
1614 bulk write request, because the grant info while be sync between
1615 OSC and OST when reconnect, and we should ignore the grant info
1616 these of resend/replay write req.
1621 Description: Grant space more than avaiable space sometimes.
1622 Details : When then OST is about to be full, if two bulk writing from
1623 different clients came to OST. Accord the avaliable space of the
1624 OST, the first req should be permitted, and the second one
1625 should be denied by ENOSPC. But if the seconde arrived before
1626 the first one is commited. The OST might wrongly permit second
1627 writing, which will cause grant space > avaiable space.
1630 Frequency : when client is evicted
1632 Description: Grant might be wrongly erased when osc is evicted by OST
1633 Details : when the import is evicted by server, it will fork another
1634 thread ptlrpc_invalidate_import_thread to invalidate the
1635 import, where the grant will be set to 0. While the original
1636 thread will update the grant it got when connecting. So if
1637 the former happened latter, the grant will be wrongly errased
1638 because of this race.
1643 Description: Checking Stale with correct fid
1644 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
1645 to check whether it is stale, because sometimes, we want the
1646 enqueue happened anyway, and op_data.fid2 will not be initialized.
1648 Severity : enhancement
1650 Description: update patchless client
1651 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
1654 Frequency : only with 2.4 kernel
1656 Description: random memory corruption
1657 Details : size of struct ll_inode_info is to big for union inode.u and this
1658 can be cause of random memory corruption.
1663 Description: Memory leak in recovery
1664 Details : Lov_mds_md was not free in an error handler in mds_create_object.
1665 It should also check obd_fail before fsfilt_start, otherwise if
1666 fsfilt_start return -EROFS,(failover mds during mds recovery).
1667 then the req will return with repmsg->transno = 0 and rc = EROFS.
1668 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
1669 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
1670 be freed no matter whether fsfilt_commit success or not.
1673 Frequency : only with huge numbers of clients
1675 Description: Prevents from taking the superblock lock in llap_from_page for
1677 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
1678 for llap_from_page call in ll_removepage() prevents from taking
1679 the superblock lock for a soon died page.
1684 Description: Not check open intent error before release open handle
1685 Details : in some rare cases, the open intent error is not checked before
1686 release open handle, which may cause
1687 ASSERTION(open_req->rq_transno != 0), because it tries to release
1688 the failed open handle.
1693 Description: Set cat log bitmap only after create log success.
1694 Details : in some rare cases, the cat log bitmap is set too early. and it
1695 should be set only after create log success.
1699 Description: Accessing a block bevice can re-enable I/O when Lustre is
1700 tearing down a device.
1701 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
1706 Description: service threads may hog cpus when there are a lot of requests
1707 Details : Insert cond_resched to give other threads a chance to use some CPU
1712 Description: the cat log was not initialized in recovery
1713 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
1714 unlink log on mds will not be initialized until mds post
1715 recovery. And also in mds post recovery, the unlink log will
1716 initialization will be done asynchronausly, so there will be race
1717 between add unlink log and unlink log initialization.
1721 Description: brw_stats were being printed incorrectly
1722 Details : brw_stats were being printed as log2 but all of them were not
1723 recorded as log2. Also remove some code duplication arising from
1724 filter_tally_{read,write}.
1728 Frequency : rare, only in recovery.
1729 Description: ASSERTION(req->rq_type != LI_POISON) failed
1730 Details : imp_lock should be held while iterating over imp_sending_list for
1731 prevent destroy request after get timeout in ptlrpc_queue_wait.
1735 Description: replay-single.sh test 52 fails
1736 Details : A lock's skiplist need to be cleanup when it being unlinked
1737 from its resource list.
1742 Description: Short directio read returns full requested size rather than
1744 Details : Direct I/O operations should return actual amount of bytes
1745 transferred rather than requested size.
1747 Severity : enhancement
1749 Description: metadata RPC reduction (e.g. for rm performance)
1750 Details : decrease the amount of synchronous RPC between clients and servers
1751 by canceling conflicing lock before the operation on the client
1752 and packing thier handles into the main operation RPC to server.
1754 Severity : enhancement
1756 Description: add #ifdef HAVE_KERNEL_CONFIG_H
1757 Details : kernels from 2.6.19 not need include linux/config.h, but add
1758 include linux/autoconf.h in commpiler command line.
1760 Severity : enhancement
1762 Description: patchless client support for 2.6.22 kernel
1763 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
1764 removed. In this case we need drop using os depended interface to
1765 kmem_cache and use cfs_mem_cache API.
1770 Description: fix mal-formatted messages
1771 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
1775 Frequency : always in liblustre
1776 Description: wrong IS_ERR implementation in liblustre.h
1777 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
1782 Description: Correct condition for output debug message.
1783 Details : inode i_nlink equal zero is not enough for output message about
1784 disk corruption, i_ctime and i_mode should be also checked.
1788 Frequency : always in patchless client
1789 Description: add configure check for truncate_complete_page
1790 Details : improve checks for exported symbols. This allow run check without
1791 sources, but with Module.symvers shipped with kernel distribution.
1792 add check for truncate_complete_page used by patchless client.
1796 Description: sanity.sh test_77h fails with "test_77h file compare failed"
1797 Details : test_77h uses a file which was messed by other test case.
1801 Description: Not Check whether lov_tgts is NULL in some lov functions
1802 Details : Checking whether lov_tgts is NULL in some functions.
1806 Description: replace obdo_alloc() with OBDO_ALLOC macro
1807 Details : nothing special is done in obdo_alloc() function, and for
1808 debugging purpose, it needs to be replaced with macros.
1812 Description: bad return value and errno from fcntl call
1813 Details : In liblustre API, errno should be a negative value if error
1818 Description: ptlrpc_check_set() LBUG
1819 Details : In case of positive reply from server and failed client bulk
1820 callback after bulk transfer shouldn't LBUG, but process this
1821 request as erroneous.
1823 Severity : enhancement
1825 Description: VFS operations stats tool.
1826 Details : Tool which collects stats by tracking value written in pid,
1827 ppid, gid and uses llstat to generate output to plot graph using
1829 Updated lustre/utils/Makefile.am
1830 Added lustre/utils/ltrack_stats.c
1832 Severity : enhancement
1834 Description: 2.6.18 server support (lustre 1.6.1)
1835 Details : Support for 2.6.18 kernels on the server side.
1840 Description: ASSERTION(imp->imp_conn_current) failed
1841 Details : an assertion failure is hit if a client node boots and attempts to
1842 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
1845 Frequency : only for i686
1847 Description: 1.4.11 RC1 build fails for RHEL 4, i686
1848 Details : Fixed config variable for build.
1853 Description: Updated patchess for new RHEL4 kernel
1854 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
1855 Updated patch jbd-stats-2.6.9.patch
1856 Updated patch qsnet-rhel4-2.6.patch
1857 Updated patch quota-deadlock-on-pagelock-core.patch
1858 Updated patch vfs_intent-2.6-rhel4.patch
1859 Updated patch vfs_races-2.6-rhel4.patch
1860 Updated series file 2.6-rhel4-titech.series
1861 Updated series file 2.6-rhel4.series
1862 Updated kernel config files
1867 Description: lquota slave complains LBUG when reconnecting with mds
1869 Details : quota slave depends qctxt->lqc_import to send its quota request.
1870 This pointer will be invalid if mds did failover or broke its
1871 connect to osts, which leads to LBUG.
1874 Frequency : when qunit size is too small(less than 20M)
1876 Description: write is stopped by improper -EDQUOT
1877 Details : If the master is busy and qunit size is small enough(let's say 1M),
1878 the slave can not get quota from master on time, which will lead
1879 slave to trigger a -EQUOTA to client.
1884 Description: Deadlock during metadata tests
1885 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
1886 called with the per-dentry lock held.
1889 Frequency : SLES9 only
1891 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
1892 Details : set CONFIG_NR_CPUS to 128 instead of 8.
1894 Severity : enhancement
1896 Description: remove fs_prep_san_write operation and related patches
1897 Details : remove the ext3-san-jdike patches which are no longer useful.
1902 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
1903 Details : a disk corruption can cause the mballoc code to assert on a
1904 double free or other extent corruptions. Handle these with
1905 ext3_error() instead of assertions.
1908 Frequency : only with mballoc3 code and deep extent trees
1910 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
1911 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
1912 can cause the filesystem to be remounted read-only.
1917 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
1918 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
1919 few external users of this ioctl number is lfsck's call to
1920 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
1921 Add the old ioctl number to the handler so both old and new
1926 Description: parallel lock callbacks
1927 Details : Instead of sending blocking and completion callbacks as separated
1928 requests, adding them to a set and sending in parallel.
1932 Description: Disable most debugging by default
1933 Details : To improve performance, disable most logging (for debug purposes)
1934 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
1935 default, and HA includes fewer messages.
1939 Frequency : only run on patchless client.
1940 Description: use do_facet on sanity.sh for test handling recoverables errors
1941 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
1945 Description: hash tables for lists of nids, connections and uuids
1946 Details : Hash tables noticeably help when a lot of clients connect to a
1947 server, to faster identify duplicate connections or reconnects,
1948 also to faster find export to evict in manual eviction case.
1952 Description: Sometimes, when the server evict a client, and the client will
1953 not be evicted as soon as possible.
1954 Details : In enqueue req, the error was returned by intent, instead of
1955 rq_status which make ptlrpc layer not detect this error, and
1956 does not evict the client. So enqueue error should be returned
1960 Frequency : only at startup
1962 Description: Delay client connections to MDT until fist MDT->OST connect
1963 Details : If a client tried to create a new file before the MDT had
1964 connected to any OSTs, the create would return EIO. Now
1965 the client will simply block until the MDT connects to the
1966 first OST and the create can succeed.
1969 Frequency : only for SLES9
1971 Description: Routinely utilize latest Quadrics drivers in CFS releases
1972 Details : Update patch qsnet-suse-2.6.patch.
1975 Frequency : only for sles10
1977 Description: Update patches for SLES 10 SP1 kernel.
1978 Details : Update the patch vfs_intent-2.6-sles10.patch.
1983 Description: Routinely utilize latest Quadrics drivers in CFS releases
1984 Details : Update patch qsnet-rhel4-2.6.patch.
1987 Frequency : at statup only
1989 Description: mds_lov_synchronize race leads to various problems
1990 Details : simultaneous MDT->OST connections at startup can cause the
1991 sync to abort, leaving the OSC in a bad state.
1996 Description: Using wrong pointer in osc_brw_prep_request
1997 Details : Access to array[-1] can produce panic if kernel compiled with
1998 CONFIG_PAGE_ALLOC enabled
2000 Severity : enhancement
2002 Description: Async OSC create to avoid the blocking unnecessarily.
2003 Details : If a OST has no remain object, system will block on the creating
2004 when need to create a new object on this OST. Now, ways use
2005 pre-created objects when available, instead of blocking on an
2006 empty osc while others are not empty. If we must block, we block
2007 for the shortest possible period of time.
2011 Frequency : only in recovery
2012 Description: Mark OST as early accessible if his start SYNC.
2013 Details : osc_precreate return flag early accessible if oscc marked as
2014 OSCC_FLAG_SYNC_IN_PROGRESS.
2019 Description: Sometimes precreate code can triger create object on wrong ost
2020 Details : Wrong protected or not not restored variables aftre precreate loop
2021 can produce creation object on wrong ost.
2024 Frequency : oss recovery
2026 Description: llog_commit_thread cleanup should sync with llog_commit_thread
2028 Details : llog_commit_thread_count should be synced between llog_commit
2029 start and cleanup, so new llog_commit thread should not be started
2030 when llog_commit threads being stopped to avoid accessing some
2033 Severity : enhancement
2035 Description: Add printing inode info into message about error in writepage.
2037 --------------------------------------------------------------------------------
2039 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
2044 Frequency : on some architectures
2046 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
2047 Details : Uninitialized flags sometimes cause configuration commands to
2051 Frequency : patchless clients only
2053 Description: missing __iget() symbol export
2054 Details : The __iget() symbol export is missing. To avoid the need for
2055 this on patchless clients the deathrow inode reaper is turned
2056 off, and we depend on the VM to clean up old inodes. This
2057 dependency was during via the fix for bug 12181.
2062 Description: sanity.sh fail: test_52b
2063 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
2064 incorrect inode's flags to client.
2066 --------------------------------------------------------------------------------
2068 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
2070 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
2071 INTEROPERATE with older versions automatically. In many cases a
2072 special upgrade step is needed. Please read the
2073 user documentation before upgrading any part of a 1.4.x system.
2074 * WARNING: Lustre configuration and startup changes are required with
2075 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
2077 * Support for kernels:
2078 2.4.21-47.0.1.EL (RHEL 3),
2079 2.6.5-7.283 (SLES 9),
2080 2.6.9-42.0.10.EL (RHEL 4),
2081 2.6.12.6 vanilla (kernel.org),
2082 2.6.16.27-0.9 (SLES10)
2083 * Client support for unpatched kernels:
2084 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
2085 2.6.16 - 2.6.19 vanilla (kernel.org),
2086 2.6.9-42.0.8EL (RHEL 4)
2087 * Recommended e2fsprogs version: 1.39.cfs6
2088 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2091 Severity : enhancement
2093 Description: Async OSC create to avoid the blocking unnecessarily.
2094 Details : If a OST has no remain object, system will block on the creating
2095 when need to create a new object on this OST. Now, ways use
2096 pre-created objects when available, instead of blocking on an
2097 empty osc while others are not empty. If we must block, we block
2098 for the shortest possible period of time.
2100 Severity : enhancement
2102 Description: MountConf
2103 Details : Lustre configuration is now managed via mkfs and mount
2104 commands instead of lmc and lconf. New obd types (MGS, MGC)
2105 are added for dynamic configuration management. See
2106 https://mail.clusterfs.com/wikis/lustre/MountConf for
2109 Severity : enhancement
2111 Description: dynamic OST addition
2112 Details : OSTs can now be added to a live filesystem
2114 Severity : enhancement
2116 Description: startup order invariance
2117 Details : MDTs and OSTs can be started in any order. Clients only
2118 require the MDT to complete startup.
2120 Severity : enhancement
2122 Description: parallel, asynchronous orphan cleanup
2123 Details : orphan cleanup is now performed in separate threads for each
2124 OST, allowing parallel non-blocking operation.
2126 Severity : enhancement
2128 Description: optimized stripe assignment
2129 Details : stripe assignments are now made based on ost space available,
2130 ost previous usage, and OSS previous usage, in order to try
2131 to optimize storage space and networking resources.
2133 Severity : enhancement
2135 Description: Permanently set tunables
2136 Details : All writable /proc/fs/lustre tunables can now be permanently
2137 set on a per-server basis, at mkfs time or on a live
2140 Severity : enhancement
2142 Description: Lustre message v2
2143 Details : Add lustre message format v2.
2145 Severity : enhancement
2147 Description: client OST exclusion list
2148 Details : Clients can be started with a list of OSTs that should be
2149 declared "inactive" for known non-responsive OSTs.
2153 Description: ENOENT returned for valid filehandle during dbench.
2154 Details : Check if a directory has children when invalidating dentries
2155 associated with an inode during lock cancellation. This fixes
2156 an incorrect ENOENT sometimes seen for valid filehandles during
2157 testing with dbench.
2160 Frequency : SFS test only (otherwise harmless)
2162 Description: SPEC SFS validation failure on NFS v2 over lustre.
2163 Details : Changes the blocksize for regular files to be 2x RPC size,
2164 and not depend on stripe size.
2166 Severity : enhancement
2168 Description: fine-grained SMP locking inside DLM
2169 Details : Improve DLM performance on SMP systems by removing the single
2170 per-namespace lock and replace it with per-resource locks.
2172 Severity : enhancement
2174 Description: don't hold multiple extent locks at one time
2175 Details : To avoid client eviction during large writes, locks are not
2176 held on multiple stripes at one time or for very large writes.
2177 Otherwise, clients can block waiting for a lock on a failed OST
2178 while holding locks on other OSTs and be evicted.
2180 Severity : enhancement
2182 Description: Multiple MD RPCs in flight.
2183 Details : Further unserialise some read-only MDT RPCs - learn about intents.
2184 To avoid overly-overloading MDT, introduce a limit on number of
2185 MDT RPCs in flight for a single client and add /proc controls
2186 to adjust this limit.
2188 Severity : enhancement
2190 Description: client read/write statistics
2191 Details : Add client read/write call usage stats for performance
2192 analysis of user processes.
2193 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
2194 file access. extents_stats shows chunk size distribution.
2195 extents_stats_per_process show chunk size distribution per
2198 Severity : enhancement
2200 Description: per-client statistics on server
2201 Details : Add ldlm and operations statistics for each client in
2202 /proc/fs/lustre/mds|obdfilter/*/exports/
2204 Severity : enhancement
2206 Description: improved MDT statistics
2207 Details : Add detailed MDT operations statistics in
2208 /proc/fs/lustre/mds/*/stats
2210 Severity : enhancement
2212 Description: VFS operations stats
2213 Details : Add client VFS call stats, trackable by pid, ppid, or gid
2214 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
2219 Description: Fix client-side osc byte counters
2220 Details : The osc read/write byte counters in
2221 /proc/fs/lustre/osc/*/stats are now working
2224 Frequency : always as root on SLES
2226 Description: Failure of copying files with lustre special EAs.
2227 Details : Client side always return success for setxattr call for lustre
2228 special xattr (currently only "trusted.lov").
2233 Description: Refcount LNET uuids
2234 Details : The global LNET uuid list grew linearly with every startup;
2235 refcount repeated list entries instead of always adding to
2238 Severity : enhancement
2240 Description: Dynamic service threads
2241 Details : Within a small range, start extra service threads
2242 automatically when the request queue builds up.
2245 Frequency : mixed-endian client/server environments
2247 Description: mixed-endian crashes
2248 Details : The new msg_v2 system had some failures in mixed-endian
2251 Severity : enhancement
2253 Description: Easy OST removal
2254 Details : OSTs can be permanently deactivated with e.g. 'lctl
2255 conf_param lustre-OST0001.osc.active=0'
2257 Severity : enhancement
2259 Description: MGS proc entries
2260 Details : Added basic proc entries for the MGS showing what filesystems
2263 Severity : enhancement
2265 Description: provide MGS failover
2266 Details : Added config lock reacquisition after MGS server failover.
2268 Severity : enhancement
2270 Description: add Linux 2.4 support
2271 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
2275 Description: a large application tries to do I/O to the same resource and dies
2276 in the middle of it.
2277 Details : Check the req->rq_arrival time after the call to
2278 ost_brw_lock_get(), but before we do anything about
2279 processing it & sending the BULK transfer request. This
2280 should help move old stale pending locks off the queue as
2281 quickly as obd_timeout.
2284 Frequency : when an incorrect nid is specified during startup
2286 Description: ptlrpc connect to non-existant node causes kernel crash
2287 Details : LNET can't be re-entered from an event callback, which
2288 happened when we expire a message after the export has been
2289 cleaned up. Instead, hand the zombie cleanup off to another
2292 Severity : enhancement
2294 Description: plain/inodebits lock performance improvement
2295 Details : Grouping plain/inodebits in granted list by their request modes
2296 and bits policy, thus improving the performance of search through
2300 Frequency : only if OST filesystem is corrupted
2302 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2303 Details : for a short time RPCs with bulk IO are in the replay list,
2304 but replay of bulk IOs is unimplemented. If the OST filesystem
2305 is corrupted due to disk cache incoherency and then replay is
2306 started it is possible to trip an assertion. Avoid putting
2307 committed RPCs into the replay list at all to avoid this issue.
2310 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2312 Description: System hang on startup
2313 Details : This bug allowed the liblustre (e.g. catamount) client to
2314 return to the app before handling all startup RPCs. This
2315 could leave the node unresponsive to lustre network traffic
2316 and manifested as a server ptllnd timeout.
2318 Severity : enhancement
2320 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2321 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
2322 boolean to control whether to print peer debug info when a
2323 client's RPC times out.
2326 Frequency : only for kernels with patches from Lustre below 1.4.3
2328 Description: Remove old rdonly API
2329 Details : Remove old rdonly API which unused from at least lustre 1.4.3
2332 Frequency : only for devices with external journals
2334 Description: Set external device read-only also
2335 Details : During a commanded failover stop, we set the disk device
2336 read-only while the server shuts down. We now also set any
2337 external journal device read-only at the same time.
2340 Frequency : when upgrading from 1.4 while trying to change parameters
2342 Description: The wrong (new) MDC name was used when setting parameters for
2343 upgraded MDT's. Also allows changing of OSC (and MDC)
2344 parameters if --writeconf is specified at tunefs upgrade time.
2347 Frequency : when setting specific ost indicies
2349 Description: QOS code breaks on skipped indicies
2350 Details : Add checks for missing OST indicies in the QOS code, so OSTs
2351 created with --index need not be sequential.
2353 Severity : enhancement
2355 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2356 Details : The uninit_groups feature works in conjunction with the kernel
2357 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2358 up the pass1 processing of e2fsck. This is a read-only feature
2359 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2360 mount filesystems that have had this feature enabled.
2362 Severity : enhancement
2364 Description: Improve multi-block allocation algorithm to avoid fragmentation
2365 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2366 allocation locality and avoid filesystem fragmentation.
2368 ------------------------------------------------------------------------------
2370 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
2372 * Support for kernels:
2373 2.4.21-47.0.1.EL (RHEL 3)
2374 2.6.5-7.283 (SLES 9)
2375 2.6.9-42.0.10.EL (RHEL 4)
2376 2.6.12.6 vanilla (kernel.org)
2377 2.6.16.27-0.9 (SLES 10)
2378 * Recommended e2fsprogs version: 1.39.cfs5
2380 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2384 Frequency : occasional, depends on client load and configuration
2385 Bugzilla : 12181, 12203
2386 Description: data loss for recently-modified files
2388 Details : In some cases it is possible that recently written or created
2389 files may not be written to disk in a timely manner (this should
2390 normally be within 30s unless client IO load is very high).
2391 The problem appears as zero-length files or files that are a
2392 multiple of 1MB in size after a client crash or client eviction
2393 that are missing data at the end of the file.
2395 This problem is more likely to be hit on clients where files are
2396 repeatedly created and unlinked in the same directory, clients
2397 have a large amount of RAM, have many CPUs, the filesystem has
2398 many OSTs, the clients are rebooted frequently, and/or the files
2399 are not accessed by other nodes after being written.
2401 The presence of the problem can be detected by looking at
2402 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
2403 smaller than the second (nr_unused) then dirty files will not
2404 be flushed automatically to disk. "sync; sleep 10" should be
2405 run several times on the node before unmounting it to update
2406 Lustre (this is also safe to run on nodes without this problem).
2408 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
2409 that can cause this same problem, so customers using that kernel
2410 also need to update the kernel in addition to Lustre. In order
2411 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
2413 It is normal that files written just before a client crash (less
2414 than 30s) may not yet have been flushed to disk, even for local
2418 Frequency : frequent on thin XT3 nodes
2420 Description: UUID collision on thin XT3 Linux nodes
2421 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
2422 randomly, since we relied on an insufficiently-seeded PRNG.
2427 Description: OSS hangs after "All ost request buffers busy"
2428 Details : A deadlock between quota and journal operations caused OSS
2429 hangs after printing "All ost request buffers busy."
2432 Frequency : always on liblustre builds
2434 Description: Cleanup compiler warnings on liblustre
2437 Frequency : always on liblustre builds on XT3
2439 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
2443 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
2444 > 32000 subdirectories
2445 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
2446 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
2447 subdirectory count crosses 32000. This will aid e2fsck to
2448 correctly handle more than 32000 subdirectories.
2451 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
2453 Description: System hang on startup
2454 Details : This bug allowed the liblustre (e.g. catamount) client to
2455 return to the app before handling all startup RPCs. This
2456 could leave the node unresponsive to lustre network traffic
2457 and manifested as a server ptllnd timeout.
2459 Severity : enhancement
2461 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
2462 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
2463 boolean to control whether to print peer debug info when a
2464 client's RPC times out.
2469 Description: make O_SYNC working on 2.6 kernels
2470 Details : 2.6 kernels use different method for mark pages for write,
2471 so need add a code to lustre for O_SYNC work.
2476 Description: Failure to close file and release space on NFS
2477 Details : Put inode details into lock acquired in ll_intent_file_open.
2478 Use mdc_intent_lock in ll_intent_open to properly
2479 detect all kind of errors unhandled by mdc_enqueue.
2484 Description: proc file read during shutdown sometimes raced obd removal,
2486 Details : Add lock to prevent obd access after proc file removal.
2489 Frequency : Only for files larger than 4GB on 32-bit clients.
2491 Description: improperly doing page alignment of locks
2492 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
2493 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
2498 Description: RPCs being resent when they shouldn't be
2499 Details : Some RPCs that should not be resent are being resent. This
2500 can cause inconsistencies in the RPC state machine. Do not
2501 resend such requests.
2504 Frequency : rare, only with NFS export
2506 Description: Crash on NFS re-export node
2507 Details : under very unusual load conditions an assertion is hit in
2508 ll_intent_file_open()
2511 Frequency : only if OST filesystem is corrupted
2513 Description: client incorrectly hits assertion in ptlrpc_replay_req()
2514 Details : for a short time RPCs with bulk IO are in the replay list,
2515 but replay of bulk IOs is unimplemented. If the OST filesystem
2516 is corrupted due to disk cache incoherency and then replay is
2517 started it is possible to trip an assertion. Avoid putting
2518 committed RPCs into the replay list at all to avoid this issue.
2523 Description: large O_DIRECT requests fail under memory pressure/fragmentation
2524 Details : Large single O_DIRECT read and write calls can fail to allocate
2525 a sufficiently large buffer to process the request. In case of
2526 allocation failure the allocation is retried with a smaller
2527 buffer and broken into smaller requests.
2529 Severity : enhancement
2531 Description: Add -o localflock option to simulate old noflock behaviour.
2532 Details : This will achieve local-only flock/fcntl locks coherentness.
2537 Description: versioning check is incomplete
2538 Details : Checking the version difference of client vs. server, report
2539 error if the gap is too big.
2544 Description: add support PG_writeback bit
2545 Details : add support for PG_writeback bit for Lustre, for more carefull
2546 work with page cache in 2.6 kernel. This also fix some deadlocks
2547 and remove hack for work O_SYNC with 2.6 kernel.
2549 Severity : enhancement
2551 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
2552 Details : The uninit_groups feature works in conjunction with the kernel
2553 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
2554 up the pass1 processing of e2fsck. This is a read-only feature
2555 in ldiskfs2 only, so older kernels and current ldiskfs cannot
2556 mount filesystems that have had this feature enabled.
2558 Severity : enhancement
2560 Description: Improve multi-block allocation algorithm to avoid fragmentation
2561 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
2562 allocation locality and avoid filesystem fragmentation.
2564 ------------------------------------------------------------------------------
2566 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
2568 * Support for kernels:
2569 2.6.9-42.0.3.EL (RHEL 4)
2570 2.6.5-7.276 (SLES 9)
2571 2.4.21-47.0.1.EL (RHEL 3)
2572 2.6.12.6 vanilla (kernel.org)
2573 2.6.16.21-0.8 (SLES10)
2574 * Recommended e2fsprogs version: 1.39.cfs2-0
2576 * The backwards-compatible /proc/sys/portals symlink has been removed
2577 in this release. Before upgrading, please ensure that you change
2578 any configuration scripts or /etc/sysctl.conf files that access
2579 /proc/sys/portals/* or sysctl portals.* to use the corresponding
2580 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
2581 in advance of the upgrade on any system running Lustre 1.4.6 or
2582 newer, since /proc/sys/lnet was added in that version.
2583 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
2587 Frequency : only when quota is used
2589 Description: avoid scanning export list for quota master
2590 Details : Change the algorithms to avoid scanning export list in order
2591 to improve the efficiency.
2594 Frequency : MDS failover only, very rarely
2596 Description: "went back in time" messages on mds failover
2597 Details : The greatest transno may be lost when the current operation
2598 finishes with an error (transno==0) and the client's last_rcvd
2599 record is over-written. Save the greatest transno in the
2600 mds_last_transno for this case.
2603 Frequency : always for specific kernels and striping counts
2605 Description: client may get "Matching packet too big" without ACL support
2606 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
2607 when trying to access files in certain configurations. The
2608 clients should in fact be denied when mounting because they do
2609 not understand ACLs.
2612 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
2614 Description: many clients connecting with IO in progress causes connect timeouts
2615 Details : Avoid synchronous journal commits to avoid delays caused by many
2616 clients connecting/disconnecting when bulk IO is in progress.
2617 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
2618 OST_IO_PORTAL to avoid delays behind potentially many pending
2622 Frequency : occasionally with multiple writers to a single file
2624 Description: shared writes to file may result in wrong size reported by stat()
2625 Details : Allow growing of kms when extent lock is cancelled
2628 Frequency : always with random mmap IO to multi-striped file
2630 Description: mmap write might be lost if we are writing to a 'hole' in stripe
2631 Details : Only if the hole is at the end of OST object so that kms is too
2632 small. Fix is to increase kms accordingly in ll_nopage.
2635 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
2637 Description: writes to a missing object would leak memory on the OST
2638 Details : If there is an inconsistency between the MDS and OST filesystems,
2639 such that the MDS references an object that doesn't exist, writes
2640 to that object will leak memory due to incorrect cleanup in the
2641 error handling path, eventually running out of memory on the OST.
2646 Description: Creating too long symlink causes lustre errors
2647 Details : Check symlink and name lengths before sending requests to MDS.
2650 Frequency : only if flock is enabled (not on by default)
2652 Description: posix locks not released on fd closure on 2.6.9+
2653 Details : We failed to add posix locks to list of inode locks on 2.6.9+
2654 kernels, this caused such locks not to be released on fd close and
2655 then assertions on fs unmount about still used locks.
2658 Frequency : MDS failover only, very rarely
2660 Description: clients may get ASSERTION(granted_lock != NULL)
2661 Details : When request was taking a long time, and a client was resending
2662 a getattr by name lock request. The were multiple lock requests
2663 with the same client lock handle and
2664 mds_getattr_name->fixup_handle_for_resent_request found one of the
2665 lock handles but later failed with ASSERTION(granted_lock != NULL).
2670 Description: handle->h_buffer_credits > 0, assertion failure
2671 Details : h_buffer_credits is zero after truncate, causing assertion
2672 failure. This patch extends the transaction or creates a new
2676 Frequency : NFS re-export or patchless client
2677 Bugzilla : 11179, 10796
2678 Description: Crash on NFS re-export node (__d_move)
2679 Details : We do not want to hash the dentry if we don't have a lock.
2680 But if this dentry is later used in d_move, we'd hit uninitialised
2681 list head d_hash, so we just do this to init d_hash field but
2682 leave dentry unhashed.
2685 Frequency : NFS re-export or patchless client
2687 Description: NFS exports has problem with symbolic link
2688 Details : lustre client didn't properly install dentry when re-exported
2689 to NFS or running patchless client.
2692 Frequency : NFS re-export or patchless client
2694 Description: Various nfs/patchless fixes.
2695 Details : fixes reuse disconected alias for lookup process - this fixes
2696 warning "find_exported_dentry: npd != pd",
2697 fix permission error with open files at nfs.
2698 fix apply umask when do revalidate.
2701 Frequency : occasional
2703 Description: Crash on NFS re-export node
2704 Details : calling clear_page() on the wrong pointer triggered oops in
2705 generic_mapping_read().
2708 Frequency : rarely, using O_DIRECT IO
2710 Description: unaligned directio crashes client with LASSERT
2711 Details : check for unaligned buffers before trying any requests.
2714 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
2716 Description: stale data returned from RAID cache
2717 Details : If only a small amount of IO is done to the RAID device before
2718 reading it again it is possible to get stale data from the RAID
2719 cache instead of reading it from disk.
2722 Frequency : always for sles10 kernel
2724 Description: sles10 support
2725 Details : ll_follow_link: compile fixes and using of nd_set_link
2726 under newer kernels.
2729 Frequency : depends on arch, kernel and compiler version, always on sles10
2732 Description: recursive or deep enough symlinks cause stack overflow
2733 Details : getting rid of large stack-allocated variable in
2737 Frequency : depends on hardware
2739 Description: lustre write performance loss in the SLES10 kernel
2740 Details : the performance loss is caused by using of write barriers in the
2741 ext3 code. The SLES10 kernel turns barrier support on by
2742 default. The fix is to undo that change for ldiskfs.
2744 ------------------------------------------------------------------------------
2746 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
2748 * Support for kernels:
2749 2.6.9-42.0.3EL (RHEL 4)
2750 2.6.5-7.276 (SLES 9)
2751 2.4.21-47.0.1.EL (RHEL 3)
2752 2.6.12.6 vanilla (kernel.org)
2756 Frequency : quota enabled and large files being deleted
2758 Description: releasing more than 4GB of quota at once hangs OST
2759 Details : If a user deletes more than 4GB of files on a single OST it
2760 will cause the OST to spin in an infinite loop. Release
2761 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
2766 Description: statfs data retrieved from /proc may be stale or zero
2767 Details : When reading per-device statfs data from /proc, in the
2768 {kbytes,files}_{total,free,avail} files, it may appear
2769 as zero or be out of date.
2772 Frequency : systems with MD RAID1 external journal devices
2774 Description: lconf's call to blkid is confused by RAID1 journal devices
2775 Details : Use the "blkid -l" flag to locate the MD RAID device instead
2776 of returning all block devices that match the journal UUID.
2779 Frequency : always, for aggregate stripe size over 4GB
2781 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
2782 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
2783 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
2786 Frequency : always if "lfs find" used on a local file/directory
2788 Description: "lfs find" segfaults if used on a local file/directory
2789 Details : The case where a directory component was not specified wasn't
2790 handled correctly. Handle this properly.
2793 Frequency : always on ppc64
2795 Description: the write to an ext3 filesystem mounted with mballoc got stuck
2796 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
2797 perform endianness conversion.
2800 Frequency : rarely (truncate to non-zero file size after write under load)
2801 Bugzilla : 10730, 10687
2802 Description: Files padded with zeros to next 4K multiple
2803 Details : With filesystems mounted using the "extents" option (2.6 kernels)
2804 it is possible that files that are truncated to a non-zero size
2805 immediately after being written are filled with zero bytes beyond
2806 the truncated size. No file data is lost.
2808 Severity : enhancement
2810 Description: Allow recovery/failover for liblustre clients.
2811 Details : liblustre clients were unaware of failover configurations until
2814 Severity : enhancement
2816 Description: user file locks should fail when not mounting with flock option
2817 Details : Set up an error-returning stub in ll_file_operations.lock field
2818 to prevent incorrect behaviour when client is mounted without
2819 flock option. Also, set up properly f_op->flock field for
2823 Frequency : always on ia64
2825 Description: "lfs df" loops on printing out MDS statfs information
2826 Details : The obd_ioctl_data was not initialized and in some systems
2827 this caused a failure during the ioctl that did not return
2828 an error. Initialize the struct and return an error on failure.
2831 Frequency : SLES 9 only
2833 Description: Error of copying files with lustre special EAs as root
2834 Details : Client side always return success for setxattr call for lustre
2835 special xattr (currently only "trusted.lov").
2838 Frequency : rarely on clusters with both ia64+i386 clients
2840 Description: ia64+i686 clients doing shared IO on the same file may LBUG
2841 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
2842 clients are doing concurrent writes to the same file it is
2843 possible that the ia64 clients may LASSERT because the OST
2844 extent locks are not PAGE_SIZE aligned. Ensure that grown
2845 locks are always aligned on the request boundary.
2848 Frequency : specific use, occasional
2850 Description: Overwriting in use executable truncates on-disk binary image
2851 Details : If one node attempts to overwrite an executable in use by
2852 another node, we now correctly return ETXTBSY instead of
2853 truncating the file.
2855 Severity : enhancement
2857 Description: Async OSC create to avoid the blocking unnecessarily.
2858 Details : If a OST has no remain object, system will block on the creating
2859 when need to create a new object on this OST. Now, ways use
2860 pre-created objects when available, instead of blocking on an
2861 empty osc while others are not empty. If we must block, we block
2862 for the shortest possible period of time.
2867 Description: chmod on Lustre root is propagated to other clients
2868 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
2869 invalid by the follow_mount time.
2872 Frequency : liblustre clients only
2874 Description: Race in 'instant cancel' lock handling could lead to such locks
2875 never to be granted in case of SMP MDS
2876 Details : Do not destroy not yet granted but cbpending locks in
2880 Frequency : replay/resend of open
2882 Description: non null lock assetion failure in mds_intent_policy
2883 Details : Trying to replay/resend lockless open requests resulted in
2884 mds_open() returning 0 with no lock. Now it sets a flag if
2885 a lock is going to be returned.
2887 Severity : enhancement
2889 Description: Checksum enhancements
2890 Details : New checksum enhancements allow for resending RPCs that failed
2893 Severity : enhancement
2895 Description: Tunables on number of dirty pages in cacche
2896 Details : Allow to set limit on number of dirty pages cached.
2901 Description: client crash on unmount - lock still has references
2902 Details : In some error handling cases it was possible to leak a lock
2903 reference on a client while accessing a file. This was not
2904 harmful to the client during operation, but would cause the
2905 client to crash when the filesystem is unmounted.
2908 Frequency : specific case, rare
2910 Description: ETXTBSY on mds though file not in use by client
2911 Details : ETXTBSY is no longer incorrectly returned when attempting to
2912 chmod or chown a directory that the user previously tried to
2913 execute or a currently-executing binary.
2916 Frequency : extremely rare except on liblustre-based clients
2918 Description: Lustre space not freed when files are deleted
2919 Details : Clean up open-unlinked files after client eviction. Previously
2920 the unlink was skipped and the files remained as orphans.
2925 Description: OST failure "would be an LBUG" in waiting_locks_callback()
2926 Details : In some cases it was possible to send a blocking callback to
2927 a client doing a glimpse, even though that client didn't get
2928 a lock granted. When the glimpse lock is cancelled on the OST
2929 the freed lock is left on the waiting list and corrupted the list.
2932 Frequency : all core dumps
2934 Description: Broke core dumps to lustre
2935 Details : Negative dentry may be unhashed if parent does not have UPDATE
2936 lock, but some callers, e.g. do_coredump, expect dentry to be
2937 hashed after successful create, hash it in ll_create_it.
2939 ------------------------------------------------------------------------------
2941 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
2943 * Support for kernels:
2944 2.6.9-42.0.2.EL (RHEL 4)
2945 2.6.5-7.276 (SLES 9)
2946 2.4.21-40.EL (RHEL 3)
2947 2.6.12.6 vanilla (kernel.org)
2951 Frequency : always on RHEL 3
2953 Description: Number of open files grows over time
2954 Details : The number of open files grows over time, whether or not
2955 Lustre is started. This was due to a filp leak introduced
2956 by one of our kernel patches.
2958 ------------------------------------------------------------------------------
2960 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
2962 * Support for kernels:
2963 2.6.9-42.EL (RHEL 4)
2964 2.6.5-7.267 (SLES 9)
2965 2.4.21-40.EL (RHEL 3)
2966 2.6.12.6 vanilla (kernel.org)
2971 Bugzilla : 5719, 9635, 9792, 9684
2972 Description: OST (or MDS) trips assertions in (re)connection under heavy load
2973 Details : If a server is under heavy load and cannot reply to new
2974 connection requests before the client resends the (re)connect,
2975 the connection handling code can behave badly if two service
2976 threads are concurrently handing separate (re)connections from
2977 the same client. Add better locking to the connection handling
2978 code, and ensure that only a single connection will be processed
2979 for a given client UUID, even if the lock is dropped.
2981 Severity : enhancement
2983 Description: add TCP zero-copy support to kernel
2984 Details : Add support to the kernel TCP stack to allow zero-copy bulk
2985 sends if the hardware supports scatter-gather and checksumming.
2986 This allows socklnd to do client-write and server-read more
2987 efficiently and reduce CPU utilization from skbuf copying.
2990 Frequency : only if NFS exporting from client
2992 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
2993 Details : If setting attributes on a file created under NFS that had
2994 never been opened it would be possible to oops the client
2995 if the file had no objects.
2999 Bugzilla : 9326, 10402, 10897
3000 Description: client crash in ptlrpcd_wake() thread when sending async RPC
3001 Details : It is possible that ptlrpcd_wake() dereferences a freed async
3002 RPC. In rare cases the ptlrpcd thread alread processed the RPC
3003 before ptlrpcd_wake() was called and the request was freed.
3006 Frequency : always for liblustre
3008 Description: liblustre client does MDS+OSTs setattr RPC for each write
3009 Details : When doing a write from a liblustre client, the client
3010 incorrectly issued an RPC to the MDS and each OST the file was
3011 striped over in order to update the timestamps. When writing
3012 with small chunks and many clients this could overwhelm the MDS
3013 with RPCs. In all cases it would slow down the write because
3014 these RPCs are unnecessary.
3016 Severity : enhancement
3018 Description: allow number of MDS service threads to be changed at module load
3019 Details : It is now possible to change the number of MDS service threads
3020 running. Adding "options mds mds_num_threads={N}" to the MDS's
3021 /etc/modprobe.conf will set the number of threads for the next
3022 time Lustre is restarted (assuming the "mds" module is also
3023 reloaded at that time). The default number of threads will
3024 stay the same, 32 for most systems.
3029 Description: OST crash if filesystem is unformatted or corrupt
3030 Details : If an OST is started on a device that has never been formatted
3031 or if the filesystem is corrupt and cannot even mount then the
3032 error handling cleanup routines would dereference a NULL pointer.
3037 Description: NULL pointer deref in llap_from_page.
3038 Details : get_cache_page_nowait can return a page with NULL (or otherwise
3039 incorrect) mapping if the page was truncated/reclaimed while it was
3040 searched for. Check for this condition and skip such pages when
3041 doing readahead. Introduce extra check to llap_from_page() to
3042 verify page->mapping->host is non-NULL (so page is not anonymous).
3045 Frequency : Sometimes when using sys_sendfile
3047 Description: "page not covered by a lock" warnings from ll_readpage
3048 Details : sendfile called ll_readpage without right page locks present.
3049 Now we introduced ll_file_sendfile that does necessary locking
3050 around call to generic_file_sendfile() much like we do in
3054 Frequency : with certain MDS communication failures at client mount time
3056 Description: NULL pointer deref after failed client mount
3057 Details : a client connection request may delayed by the network layer
3058 and not be sent until after the PTLRPC layer has timed out the
3059 request. If the client fails the mount immediately it will try
3060 to clean up before the network times out the request. Add a
3061 reference from the request import to the obd device and delay
3062 the cleanup until the network drops the request.
3065 Frequency : occasionally during client (re)connect
3067 Description: assertion failure during client (re)connect
3068 Details : processing a client connection request may be delayed by the
3069 client or server longer than the client connect timeout. This
3070 causes the client to resend the connection request. If the
3071 original connection request is replied in this interval, the
3072 client may trip an assertion failure in ptlrpc_connect_interpret()
3073 which thought it would be the only running connect process.
3076 Frequency : only with obd_echo servers and clients that are rebooted
3078 Description: kernel BUG accessing uninitialized data structure
3079 Details : When running an obd_echo server it did not start the ping_evictor
3080 thread, and when a client was evicted an uninitialized data
3081 structure was accessed. Start the ping_evictor in the RPC
3082 service startup instead of the OBD startup.
3084 Severity : enhancement
3085 Bugzilla : 10193 (patchless)
3086 Description: Remove dependency on various unexported kernel interfaces.
3087 Details : No longer need reparent_to_init, exit_mm, exit_files,
3088 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
3091 Frequency : rare (only users of deprecated and unsupported LDAP config)
3093 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
3094 Details : LDAP apparently contains 'lustreName' attributes instead of
3095 'name'. A simple remapping of the name is sufficient.
3098 Frequency : rare (only with non-default dump_on_timeout debug enabled)
3100 Description: waiting_locks_callback trips kernel BUG if client is evicted
3101 Details : Running with the dump_on_timeout debug flag turned on makes
3102 it possible that the waiting_locks_callback() can try to dump
3103 the Lustre kernel debug logs from an interrupt handler. Defer
3104 this log dumping to the expired_lock_main() thread.
3106 Severity : enhancement
3108 Description: Support NFS exporting on 2.6 kernels.
3109 Details : Implement non-rawops metadata methods for NFS server to use without
3110 changing NFS server code.
3113 Frequency : very rare (synthetic metadata workload only)
3115 Description: two racing renames might cause an MDS thread to deadlock
3116 Details : Running the "racer" program may cause one MDS thread to rename
3117 a file from being the source of a rename to being the target of
3118 a rename at exactly the same time that another thread is doing
3119 so, and the second thread has already enqueued these locks after
3120 doing a lookup of the target and is trying to relock them in
3121 order. Ensure that we don't try to re-lock the same resource.
3124 Frequency : only very large systems with liblustre clients
3126 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
3127 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
3128 all OSTs in parallel. This allows the eviction of stale liblustre
3129 clients to proceed much faster than if they were done in series,
3130 and also offers similar improvements for other set_info RPCs.
3135 Description: excessive CPU usage during initial read phase on client
3136 Details : During the initial read phase on a client, it would agressively
3137 retry readahead on the file, consuming too much CPU and impacting
3138 performance (since 1.4.5.8). Improve the readahead algorithm
3139 to avoid this, and also improve some other common cases (read
3140 of small files in particular, where "small" is files smaller than
3141 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
3146 Description: MDS crash when receiving packet with unknown intent.
3147 Details : Do not LBUG in unknown intent case, just return -EFAULT
3149 Severity : enhancement
3150 Bugzilla : 9293, 9385
3151 Description: MDS RPCs are serialised on client. This is unnecessary for some.
3152 Details : Do not serialize getattr (non-intent version) and statfs.
3155 Frequency : occasional, when OST network is overloaded/intermittent
3157 Description: client evicted by OST after bulk IO timeout
3158 Details : If a client sends a bulk IO request (read or write) the OST
3159 may evict the client if it is unresposive to its data GET/PUT
3160 request. This is incorrect if the network is overloaded (takes
3161 too long to transfer the RPC data) or dropped the OST GET/PUT
3162 request. There is no need to evict the client at all, since
3163 the pinger and/or lock callbacks will handle this, and the
3164 client can restart the bulk request.
3167 Frequency : Always when mmapping file with no objects
3169 Description: client crashes when mmapping file with no objects
3170 Details : Check that we actually have objects in a file before doing any
3171 operations on objects in ll_vm_open, ll_vm_close and
3177 Description: Request leak when working with deleted CWD
3178 Details : Introduce advanced request refcount tracking for requests
3179 referenced from lustre intent.
3181 Severity : Enhancement
3183 Description: Cache open file handles on client.
3184 Details : MDS now will return special lock along with openhandle, if
3185 requested and client is allowed to hold openhandle, even if unused,
3186 until such a lock is revoked. Helps NFS a lot, since NFS is opening
3187 closing files for every read/write openration.
3189 Severity : Enhancement
3191 Description: Cache open negative dentries on client when possible.
3192 Details : Guard negative dentries with UPDATE lock on parent dir, drop
3193 negative dentries on lock revocation.
3198 Description: Remounting a client read-only wasn't possible with a zconf mount
3199 Details : It wasn't possible to remount a client read-only with llmount.
3201 Severity : enhancement
3202 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
3203 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
3204 MPICH is included to add Lustre-specific ADIO interfaces.
3205 This is based closely on the UFS ADIO layer and only differs
3206 in file creation, in order to allow the OST striping to be set.
3207 This is user-contributed code and not supported by CFS.
3212 Description: extended inode attributes (immutable, append-only) work improperly
3213 when 2.4 and 2.6 kernels are used on client/server or vice versa
3214 Details : Introduce kernel-independent values for these flags.
3216 Severity : enhancement
3219 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
3220 Details : Many of the /proc/ tunables can only be tuned at a megabyte
3221 granularity. Now, Fractional MB granularity is be supported,
3222 this is very useful for low memory system.
3224 Severity : enhancement
3226 Description: Getattr by fid
3227 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
3228 locks, avoids extra getattr rpc requests to MDS, allows '/' to
3229 have locks and avoids getattr rpc requests for it on every stat.
3232 Frequency : Always, for filesystems larger than 2TB
3234 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
3235 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
3236 filesystems larger than 2TB to be mounted with mballoc enabled.
3239 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
3241 Description: filesystem corruption for non-standard kernels and very large OSTs
3242 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
3243 filesystems larger than 2TB is mounted then the kernel will
3244 silently corrupt the start of the filesystem. CONFIG_LBD is
3245 enabled for all CFS-supported kernels, but the possibility of
3246 this happening with a modified kernel config exists.
3248 Severity : enhancement
3250 Description: add client O_DIRECT support for 2.6 kernels
3251 Details : It is now possible to do O_DIRECT reads and writes to files
3252 in the Lustre client mountpoint on 2.6 kernel clients.
3254 Severity : enhancement
3256 Description: parallel glimpse, setattr, statfs, punch, destroy requests
3257 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
3258 parallel, not waiting for response from every OST before sending
3259 a rpc to the next OST.
3264 Description: setattr vs write race when updating file timestamps
3265 Details : Client processes that update a file timestamp into the past
3266 right after writing to the file (e.g. tar) it is possible that
3267 the updated file modification time can be reset to the current
3268 time due to a race between processing the setattr and write RPC.
3270 Severity : enhancement
3272 Description: Bring 'lfs find' closer in line with regular Linux find.
3273 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
3274 -print0 options and obtains all the needed info through the lustre
3277 Severity : enhancement
3279 Description: support up to 1024 configured devices on one node
3280 Details : change obd_dev array from statically allocated to dynamically
3281 allocated structs as they are first used to reduce memory usage
3286 Description: Flush dirty partially truncated pages during truncate
3287 Details : Immediatelly flush partially truncated pages in filter_setattr,
3288 this way we completely avoid having any pages in page cache on OST
3289 and can retire ugly workarounds during writes to flush such pages.
3294 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
3295 Details : protect inode from truncation within vfs_unlink() context
3296 just take a reference before calling vfs_unlink() and release it
3297 when parent's i_sem is free.
3300 Frequency : always, if extents are used on OSTs
3302 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
3303 Details : OSTs using the extents format would not zero the high 16 bits of
3304 the index physical block number. This is not a problem for any
3305 OST filesystems smaller than 16TB, and no kernels support ext3
3306 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
3307 new/modified files) and can be fixed for existing filesystems
3308 with e2fsprogs-1.39-cfs1.
3313 Description: import connection selection may be incorrect if timer wraps
3314 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
3315 import connections to be ignored if the 32-bit jiffies counter
3316 wraps. Use a 64-bit jiffies counter.
3319 Frequency : during server recovery
3321 Description: crash after server is denying duplicate export
3322 Details : If clients are resending connect requests to the server, the
3323 server refuses to allow a client to connect multiple times.
3324 Fixed a bug in the handling of this case.
3327 Frequency : very large clusters immediately after boot
3329 Description: LNET request buffers exhausted under heavy short-term load
3330 Details : If a large number of client requests are generated on a service
3331 that has previously never seen so many requests it is possible
3332 that the request buffer growth cannot keep up with the spike in
3333 demand. Instead of dropping incoming requests, they are held in
3334 the LND until the RPC service can accept more requests.
3337 Frequency : Sometimes during replay
3339 Description: Assertion failure in ll_local_open after replay.
3340 Details : If replay happened on an open request reply before we were able
3341 to set replay handler, reply will become not swabbed tripping the
3342 assertion in ll_local_open. Now we set the handler right after
3343 recognising of open request
3346 Frequency : very rare
3348 Description: kernel reports "badness in vsnprintf"
3349 Details : Reading from the "recovery_status" /proc file in small chunks
3350 may cause a negative length in lprocfs_obd_rd_recovery_status()
3351 call to vsnprintf() (which is otherwise harmless). Exit early
3352 if there is no more space in the output buffer.
3354 Severity : enhancement
3356 Description: clear OBD RPC statistics by writing to them
3357 Details : It is now possible to clear the OBD RPC statistics by writing
3358 to the "stats" file.
3363 Description: Client mtime is not the same on different clients after utimes
3364 Details : In some cases, the client was using the utimes() syscall on
3365 a file cached on another node. The clients now validate the
3366 ctime from the MDS + OSTs to determine which one is right.
3371 Description: Inability to activate failout mode
3372 Details : lconf script incorrectly assumed that in python string's numeric
3373 value is used in comparisons.
3376 Frequency : always with multiple stripes per file
3378 Description: Inefficient object allocation for mutli-stripe files
3379 Details : When selecting which OSTs to stripe files over, for files with
3380 a stripe count that divides evenly into the number of OSTs,
3381 the MDS is always picking the same starting OST for each file.
3382 Return the OST selection heuristic to the original design.
3387 Description: mount failures may take full timeout to return an error
3388 Details : Under some heavy load conditions it is possible that a
3389 failed mount can wait for the full obd_timeout interval,
3390 possibly several minutes, before reporting an error.
3391 Instead return an error as soon as the status is known.
3393 ------------------------------------------------------------------------------
3395 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
3397 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
3398 INTEROPERATE with older versions automatically. Please read the
3399 user documentation before upgrading any part of a live system.
3400 * WARNING: Lustre networking configuration changes are required with
3401 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
3404 * Support for kernels:
3405 2.6.9-22.0.2.EL (RHEL 4)
3406 2.6.5-7.244 (SLES 9)
3407 2.6.12.6 vanilla (kernel.org)
3410 Severity : enhancement
3411 Bugzilla : 7981/8208
3412 Description: Introduced Lustre Networking (LNET)
3413 Details : LNET is new networking infrastructure for Lustre, it includes
3414 a reorganized network configuration mode (see the user
3415 documentation for full details) as well as support for routing
3416 between different network fabrics. Lustre Networking Devices
3417 (LNDs) for the supported network fabrics have also been
3418 created for this new infrastructure.
3420 Severity : enhancement
3421 Description: Introduced Access control lists
3422 Details : clients can set ACLs on files and directories in order to have
3423 more fine-grained permissions than the standard Unix UGO+RWX.
3424 The MDS must be started with the "-o acl" mount option.
3426 Severity : enhancement
3427 Description: Introduced filesystem quotas
3428 Details : Administrators may now establish per-user quotas on the
3431 Severity : enhancement
3433 Description: Configuration change for the XT3
3434 The PTLLND is now used to run Lustre over Portals on the XT3
3435 The configure option(s) --with-cray-portals are no longer used.
3436 Rather --with-portals=<path-to-portals-includes> is used to
3437 enable building on the XT3. In addition to enable XT3 specific
3438 features the option --enable-cray-xt3 must be used.
3443 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
3444 Details : A race between allocating a new llcd and re-getting the llcd_lock
3445 allowed another thread to grab newly-allocated llcd.
3447 Severity : enhancement
3449 Description: 2.6 OST async journal commit and locking fix to improve performance
3450 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
3451 2.6 kernels are now async as they already were in 2.4 kernels so
3452 that they can commit concurrently with the network bulk transfer.
3453 For block-allocated files the filter allocation semaphore is held
3454 to avoid filesystem fragmentation during allocation. BKL lock
3455 removed for 2.6 xattr operations where it is no longer needed.
3460 Description: lconf incorrectly determined whether two IP networks could talk
3461 Details : In some more complicated routing and multiple-network
3462 configurations, lconf will avoid trying to make a network
3463 connection to a disjoint part of the IP space. It was doing the
3464 math incorrectly for one set of cases.
3469 Description: Fix for potential infinite loop processing records in an llog.
3470 Details : If an llog record is corrupted/zeroed, it is possible to loop
3471 forever in llog_process(). Validate the llog record length
3472 and skip the remainder of the block on error.
3475 Frequency : occasional (liblustre only)
3477 Description: liblustre could not open files whose last component is a symlink
3478 Details : sysio_path_walk() would incorrectly pass the open intent to
3479 intermediate path components.
3482 Frequency : rare (liblustre only with non-standard tuning)
3483 Bugzilla : 7201 (7350)
3484 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
3485 Details : llu_lookup_finish_locks() tries to set lock data on a lock
3486 after it has been released, only do this for referenced locks
3488 Severity : enhancement
3490 Description: specifying an (invalid) directory default stripe_size of -1
3491 would reset the directory default striping
3492 Details : stripe_size -1 was used internally to signal directory stripe
3493 removal, now use "all default" to signal dir stripe removal
3494 as a directory striping of "all default" is not useful
3497 Frequency : common for large clusters running liblustre clients
3499 Description: doing an ls when liblustre clients are running is slow
3500 Details : sending a glimpse AST to a liblustre client waits for every AST
3501 to time out, as liblustre clients will not respond. Since they
3502 cannot cache data we refresh the OST lock LVB from disk instead.
3504 Severity : enhancement
3506 Description: doing an ls at the same time as file IO can be slow
3507 Details : enqueue and other "small" requests can be blocked behind many
3508 large IO requests. Create a new OST IO portal for non-IO
3509 requests so they can be processed faster.
3512 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
3514 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
3515 created file with mode 000
3516 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
3517 allow an ftruncate() as a non-root user to a file with mode 000.
3518 Lustre now respects this flag to disable mode checks when
3519 truncating a file owned by the user
3522 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
3525 Description: Revoking locks from clients that went dead or catatonic might take
3527 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
3528 cancellation of such locks instant on servers without waiting for
3529 any reply from clients. Clients drops these locks when cancel
3530 notification from server is received without replying.
3533 Frequency : liblustre-only, when liblustre client dies or becomes busy
3535 Description: Doing ls on Linux clients can take a long time with active
3537 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
3538 granting such locks to it in the first place if possible. Locks
3539 are taken by proxy on the OST during the read or write and
3540 dropped immediately afterward. Add connect flags handling, do
3541 not grant locks to liblustre clients for glimpse ASTs.
3543 Severity : enhancement
3545 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
3546 Details : Existing read-ahead algorithm is tuned for the case of streamlined
3547 sequential reads and behaves badly with applications doing random
3548 reads. Improve it by reading ahead at least read region, and
3549 avoiding excessive large RPC for small reads.
3551 Severity : enhancement
3553 Description: Creating more than 1000 files for a single job may cause a load
3554 imbalance on the OSTs if there are also a large number of OSTs.
3555 Details : qos_prep_create() uses an OST index reseed value that is an
3556 even multiple of the number of available OSTs so that if the
3557 reseed happens in the middle of the object allocation it will
3558 still utilize the OSTs as uniformly as possible.
3563 Description: OST or MDS may oops in ping_evictor_main()
3564 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
3565 but doesn't restart at beginning of obd_exports_timed list
3568 Severity : enhancement
3570 Description: improve by-nid export eviction on the MDS and OST
3571 Details : allow multiple exports with the same NID to be evicted at one
3572 time without re-searching the exports list.
3575 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
3577 Description: MDS may oops in groups_free()
3578 Details : in rare race conditions a newly allocated group_info struct is
3579 freed again, and this can be NULL. The 2.4 compatibility code
3580 for groups_free() checked for a NULL pointer, but 2.6 did not.
3583 Frequency : common for liblustre clients doing little filesystem IO
3584 Bugzilla : 9352, 7313
3585 Description: server may evict liblustre clients accessing contended locks
3586 Details : if a client is granted a lock or receives a completion AST
3587 with a blocking AST already set it would not reply to the AST
3588 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
3591 Frequency : lfs setstripe, only systems with more than 160 OSTs
3593 Description: unable to set striping with a starting offset beyond OST 160
3594 Details : llapi_create_file() incorrectly limited the starting stripe
3595 index to the maximum single-file stripe count.
3598 Frequency : LDAP users only
3600 Description: lconf did not handle in-kernel recovery with LDAP properly
3601 Details : lconf/LustreDB get_refs() is searching the wrong namespace
3603 Severity : enhancement
3605 Description: bind OST threads to NUMA nodes to improve performance
3606 Details : all OST threads are uniformly bound to CPUs on a single NUMA
3607 node and do their allocations there to localize memory access
3609 Severity : enhancement
3611 Description: llmount can determine client NID directly from Myrinet (GM)
3612 Details : the client NID code from gmnalnid was moved directly into
3613 llmount, removing the need to use this or specifying the
3614 client NID explicitly when mounting GM clients with zeroconf
3617 Frequency : if client is started with down MDS
3619 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
3620 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
3621 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
3622 ptlrpc_import_delayed_req() because we didn't check if the
3623 request was interrupted, and we also didn't break out of the
3624 event loop if there was no timeout
3629 Description: data loss during non-page-aligned writes to a single file from
3630 both multiple nodes and multiple threads on one node at same time
3631 Details : updates to KMS and lsm weren't protected by common lock. Resulting
3632 inconsistency led to false short-reads, that were cached and later
3633 used by ->prepare_write() to fill in partially written page,
3634 leading to data loss.
3637 Frequency : always, if lconf --abort_recovery used
3639 Description: lconf --abort_recovery fails with 'Operation not supported'
3640 Details : lconf was attempting to abort recovery on the MDT device and not
3643 Severity : enhancement
3645 Description: remove cleanup logs
3646 Details : replace lconf-generated cleanup logs with lustre internal
3647 cleanup routines. Eliminates the need for client-cleanup and
3650 Severity : enhancement
3652 Description: add support for EAs (user and system) on lustre filesystems
3653 Details : it is now possible to store extended attributes in the Lustre
3654 client filesystem, and with the user_xattr mount option it
3655 is possible to allow users to store EAs on their files also
3657 Severity : enhancement
3659 Description: Add possibility (config option) to show minimal available OST free
3661 Details : When compiled with --enable-mindf configure option, statfs(2)
3662 (and so, df) will return least minimal free space available from
3663 all OSTs as amount of free space on FS, instead of summary of
3664 free spaces of all OSTs.
3666 Severity : enhancement
3668 Description: do not expand extent locks acquired on OST-side
3669 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
3670 by server: they are not cached anyway.
3673 Frequency : when mmap is used/binaries executed from Lustre
3675 Description: Unmmap pages before throwing them away from read cache.
3676 Details : llap_shrink cache now attempts to unmap pages before discarding
3677 them (if unmapping failed - do not discard). SLES9 kernel has
3678 extra checks that trigger if this unmapping is not done first.
3683 Description: lconf didn't resolve symlinks before checking to see whether a
3684 given mountpoint was already in use
3687 Frequency : when migrating failover services
3688 Bugzilla : 6395, 9514
3689 Description: When migrating a subset of services from a node (e.g. failback
3690 from a failover service node) the remaining services would
3691 time out and evict clients.
3692 Details : lconf --force (implied by --failover) sets the global obd_timeout
3693 to 5 seconds in order to quickly disconnect, but this caused
3694 other RPCs to time out too quickly. Do not change the global
3695 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
3697 Severity : enhancement
3698 Frequency : if MDS is started with down OST
3699 Bugzilla : 9439,5706
3700 Description: Allow startup/shutdown of an MDS without depending on the
3701 availability of the OSTs.
3702 Details : Asynchronously call mds_lov_synchronize during MDS startup.
3703 Add appropriate locking and lov-osc refcounts for safe
3704 cleaning. Add osc abort_inflight calls in case the
3705 synchronize never started.
3708 Frequency : occasional (Cray XT3 only)
3710 Description: root not authorized to access files in CRAY_PORTALS environment
3711 Details : The client process capabilities were not honoured on the MDS in
3712 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
3713 been accessed by an authorized user then root was able to access
3714 the file on the local client also. The root user capabilities
3715 are now allowed on the MDS, as this environment has secure UID.
3718 Frequency : occasional
3720 Description: ldiskfs "too long searching" message happens too often
3721 Details : A debugging message (otherwise harmless) prints too often on
3722 the OST console. This has been reduced to only happen when
3723 there are fragmentation problems on the filesystem.
3728 Description: Division by zero in statfs when all OSCs are inactive
3729 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
3730 lov_statfs divides by value returned by lov_get_stripecnt().
3734 Bugzilla : 9489, 3273
3735 Description: First write from each client to each OST was only 4kB in size,
3736 to initialize client writeback cache, which caused sub-optimal
3737 RPCs and poor layout on disk for the first writen file.
3738 Details : Clients now request an initial cache grant at (re)connect time
3739 and so that they can start streaming writes to the cache right
3740 away and always do full-sized RPCs if there is enough data.
3741 If the OST is rebooted the client also re-establishes its grant
3742 so that client cached writes will be honoured under the grant.
3747 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
3748 Details : Now I/O RPCs go to different portal number and (presumably) fast
3749 lock requests (and glimses) and other RPCs get their own service
3750 threads pool that should be able to service those RPCs
3753 Severity : enhancement
3755 Description: Ability to exchange lustre version between client and servers and
3756 issue warnings at client side if client is too old. Also for
3757 liblustre clients there is ability to refuse connection of too old
3759 Details : New 'version' field is added to connect data structure that is
3760 filled with version info. That info is later checked by server and
3764 Frequency : rare, liblustre only.
3765 Bugzilla : 9296, 9581
3766 Description: Two simultaneous writes from liblustre at offset within same page
3767 might proceed at the same time overwriting eachother with stale
3769 Details : I/O lock withing llu_file_prwv was released too early, before data
3770 actually was hitting the wire. Extended lock-holding time until
3771 server acknowledges receiving data.
3774 Frequency : extremely rare. Never observed in practice.
3776 Description: avoid generating lustre_handle cookie of 0.
3777 Details : class_handle_hash() generates handle cookies by incrementing
3778 global counter, and can hit 0 occasionaly (this is unlikely, but
3779 not impossible, because initial value of cookie counter is
3780 selected randonly). Value of 0 is used as a sentinel meaning
3781 "unassigned handle" --- avoid it. Also coalesce two critical
3782 sections in this function into one.
3784 Severity : enhancement
3786 Description: allow liblustre clients to delegate truncate locking to OST
3787 Details : To avoid overhead of locking, liblustre client instructs OST to
3788 take extent lock in ost_punch() on client's behalf. New connection
3789 flag is added to handle backward compatibility.
3791 Severity : enhancement
3792 Bugzilla : 4928, 7341, 9758
3793 Description: allow number of OST service threads to be specified
3794 Details : a module parameter allows the number of OST service threads
3795 to be specified via "options ost ost_num_threads={N}" in the
3796 OSS's /etc/modules.conf or /etc/modprobe.conf.
3800 Bugzilla : 6146, 9635, 9895
3801 Description: servers crash with bad pointer in target_handle_connect()
3802 Details : In rare cases when a client is reconnecting it was possible that
3803 the connection request was the last reference for that export.
3804 We would temporarily drop the export reference and get a new
3805 one, but this may have been the last reference and the export
3806 was just destroyed. Get new reference before dropping old one.
3808 Severity : enhancement
3809 Frequency : if client is started with failover MDS
3811 Description: Allow multiple MDS hostnames in the mount command
3812 Details : Try to read the configuration from all specified MDS
3813 hostnames during a client mount in case the "primary"
3816 Severity : enhancement
3818 Description: Stop sending data to evicted clients as soon as possible.
3819 Details : Check if the client we are about to send or are sending data to
3820 was evicted already. (Check is done every second of waiting,
3821 for which l_wait_event interface was extended to allow checking
3822 of exit condition at specified intervals).
3825 Frequency : rare, normally only when NFS exporting is done from client
3827 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
3829 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
3832 Frequency : always, if srand() is called before liblustre initialization
3834 Description: Liblustre uses system PRNG disturbing its usage by user application
3835 Details : Introduce internal to lustre fast and high-quality PRNG for
3836 lustre usage and make liblustre and some other places in generic
3837 lustre code to use it.
3839 Severity : enhancement
3840 Bugzilla : 9477, 9557, 9870
3841 Description: Verify that the MDS configuration logs are updated when xml is
3842 Details : Check if the .xml configuration logs are newer than the config
3843 logs stored on the MDS and report an error if this is the case.
3844 Request --write-conf, or allow starting with --old_conf.
3846 Severity : enhancement
3848 Description: Handle symlinks in the path when checking if Lustre is mounted.
3849 Details : Resolve intermediate symlinks when checking if a client has
3850 mounted a filesystem to avoid duplicate client mounts.
3855 Description: lconf can hit an error exception but still return success.
3856 Details : The lconf command catches the Command error exception at the top
3857 level script context and will exit with the associated exit
3858 status, but doesn't ensure that this exit status is non-zero.
3863 Description: failure of ptlrpc thread startup can cause oops
3864 Details : Starting a ptlrpc service thread can fail if there are a large
3865 number of threads or the server memory is very fragmented.
3866 Handle this without oopsing.
3869 Frequency : always, only if liblustre and non-default acceptor port was used
3871 Description: liblustre cannot connect to servers with non-default acceptor port
3872 Details : tcpnal_set_default_params() was not called and was therefore
3873 ignoring the environment varaible TCPNAL_PORT, as well as other
3874 TCPNAL_ environment variables
3879 Description: two objects could be created on the same OST for a single file
3880 Details : If an OST is down, in some cases it was possible to create two
3881 objects on a single OST for a single file. No problems other
3882 than potential performance impact and spurious error messages.
3886 Bugzilla : 5681, 9562
3887 Description: Client may oops in ll_unhash_aliases
3888 Details : Client dcache may become inconsistent in race condition.
3889 In some cases "getcwd" can fail if the current directory is
3895 Description: Inode refcounting problems in NFS export code
3896 Details : link_raw functions used to call d_instantiate without obtaining
3897 extra inode reference first.
3901 Bugzilla : 9942, 9903
3902 Description: Referencing freed requests leading to crash, memleaks with NFS.
3903 Details : We used to require that call to ll_revalidate_it was always
3904 followed by ll_lookup_it. Also with revalidate_special() it is
3905 possible to call ll_revalidate_it() twice for the same dentry
3906 even if first occurence returned success. This fix changes semantic
3907 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
3908 reference on a request referred from the intent.
3909 ll_intent_release() then releases such a request.
3912 Frequency : rare, normally benchmark loads only
3914 Description: unlinked inodes were kept in memory on the client
3915 Details : If a client is repeatedly creating and unlinking files it
3916 can accumulate a lot of stale inodes in the inode slab cache.
3917 If there is no other client load running this can cause the
3918 client node to run out of memory. Instead flush old inodes
3919 from client cache that have the same inode number as a new inode.
3922 Frequency : SLES9 2.6.5 kernel and long filenames only
3923 Bugzilla : 9969, 10379
3924 Description: utime reports stale NFS file handle
3925 Details : SLES9 uses out-of-dentry names in some cases, which confused
3926 the lustre dentry revalidation. Change it to always use the
3930 Frequency : rare, unless heavy write-truncate concurrency is continuous
3931 Bugzilla : 4180, 6984, 7171, 9963, 9331
3932 Description: OST becomes very slow and/or deadlocked during object unlink
3933 Details : filter_destroy() was holding onto the parent directory lock
3934 while truncating+unlinking objects. For very large objects this
3935 may block other threads for a long time and slow overall OST
3936 responsiveness. It may also be possible to get a lock ordering
3937 deadlock in this case, or run out of journal credits because of
3938 the combined truncate+unlink. Solution is to do object truncate
3939 first in one transaction without parent lock, and then do the
3940 final unlink in a new transaction with the parent lock. This
3941 reduces the lock hold time dramatically.
3944 Frequency : rare, 2.4 kernels only
3946 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
3947 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
3948 before unmounting their filesystem in order to avoid deadlock.
3949 The kernel_locked() function in 2.4 kernels only checks whether
3950 the kernel lock is held, not whether it is this process that is
3951 holding it as 2.6 kernels do.
3956 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
3957 Details : The client ptlrpc code may be trying to reconnect to a down
3958 server before a previous connection attempt has timed out.
3959 Increase the reconnect interval to be longer than the connection
3960 timeout interval to avoid sending duplicate connections to
3964 Frequency : echo_client brw_test command
3966 Description: fix echo_client to work with OST preallocated code
3967 Details : OST preallocation code (5137) didn't take echo_client IO path
3968 into account: echo_client calls filter methods outside of any
3969 OST thread and, hence, there is no per-thread preallocated
3970 pages and buffers to use. Solution: hijack pga pages for IO. As
3971 a byproduct, this avoids unnecessary data copying.
3975 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
3976 Description: Client can oops in mdc_commit_close() after open replay
3977 Details : It was possible for the MDS to return an open request with no
3978 transaction number in mds_finish_transno() if the client was
3979 evicted, but without actually returning an error. Clients
3980 would later try to replay that open and may trip an assertion
3981 Simplify the client close codepath, and always return an error
3982 from the MDS in case the open is not successful.
3985 Frequency : rare, 2.6 OSTs only
3987 Description: OST may deadlock under high load on fragmented files
3988 Details : If there was a heavy load and highly-fragmented OST filesystems
3989 it was possible to have all the OST threads deadlock waiting on
3990 allocation of biovecs, because the biovecs were not released
3991 until the entire RPC IO was completed. Instead, release biovecs
3992 as soon as they are complete to ensure forward IO progress.
3994 Severity : enhancement
3996 Description: Support for specifying external journal device at mount
3997 Details : If an OST or MDS device is formatted with an external journal
3998 device, this device major/minor is stored in the ext3 superblock
3999 and may not be valid for failover. Allow detecting and
4000 specifying the external journal at mount time.
4005 Description: Mounting an MDS with pending unlinked files may cause oops
4006 Details : target_finish_recovery() calls mds_postrecov() which returned
4007 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
4008 considers this an error and immediately begins cleaning up the
4009 lov, just after starting the mds_lov process
4011 Severity : enhancement
4013 Description: Implement 'lfs df' to report actual free space on per-OST basis
4014 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
4015 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
4016 usage of objects; '-h' to report in human readable format.
4018 ------------------------------------------------------------------------------
4020 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
4027 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
4028 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
4029 at startup during orphan unlinking. Instead, skip buddy update
4030 before setup, as it will be regenerated after recovery is complete.
4035 Description: If an OST is inactive, its locks might reference stale inodes.
4036 Details : lov_change_cbdata() must iterate over all namespaces, even if
4037 they are inactive to clear inode references from the lock.
4039 Severity : enhancement
4040 Frequency : occasional, if non-standard max_dirty_mb used
4042 Description: Client will block write RPCs if not enough grant
4043 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
4044 then the client will block writes while waiting for another RPC
4045 to complete instead of consuming its dirty limit. With change
4046 we get improved performance when max_dirty_mb is small.
4048 Severity : enhancement
4049 Bugzilla : 3389, 6253
4050 Description: Add support for supplementary groups on the MDS.
4051 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
4052 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
4053 lookups for user supplementary groups into a cache.
4057 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
4058 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
4059 for liblustre O_EXCL is enforced by MDS.
4063 Description: readdir never returns NULL in liblustre.
4064 Details : Corrected llu_iop_getdirentries logic, to return offset of next
4065 dentry in struct dirent.
4069 Frequency : liblustre only, depends on application IO pattern
4070 Description: liblustre clients evicted if not contacting servers
4071 Details : Don't put liblustre clients into the ping_evictor list, so
4072 they will not be evicted by the pinger ever.
4074 Severity : enhancement
4076 Description: Add ability to evict clients by NID from MDS.
4077 Details : By echoing "nid:$NID" string into
4078 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
4079 $NID would be instantly evicted from this MDS and from all active
4080 OSTs connected to it.
4084 Description: Do not query file size twice, somewhat slowing stat(2) calls.
4085 Details : lookup_it_finish() used to query file size from OSTs that was not
4090 Description: service threads change working directory to that of init
4091 Details : Starting lustre service threads may pin the working directory
4092 of the parent thread, making that filesystem busy. Threads
4093 now change to the working directory of init to avoid this.
4097 Frequency : during shutdown only
4098 Description: shutdown with a failed MDS or OST can cause unmount to hang
4099 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
4104 Frequency : 2.6 only
4105 Description: chmod/chown may include an extra supplementary group
4106 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
4107 supplementary group and if none is specified this is used.
4110 Bugzilla : 5479 (6816)
4112 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
4113 Details : If lookup is in progress on a file that is unlinked we might try
4114 to revalidate the inode and fail in revalidate after lookup is
4115 complete and ll_file_open() enqueues the open again but
4116 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
4119 Frequency : always, if lconf --abort_recovery used
4121 Description: lconf --abort_recovery fails with 'Operation not supported'
4122 Details : lconf was attempting to abort recovery on the MDT device and not
4125 ------------------------------------------------------------------------------
4127 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
4132 Frequency : rare (only unsupported configurations with a node running as an
4134 Bugzilla : 6514, 5137
4135 Description: Mounting a Lustre file system on a node running as an OST could
4137 Details : OSTs now preallocates memory needed to write out data at
4138 startup, instead of when needed, to avoid having to
4139 allocate memory in possibly low memory situations.
4140 Specifically, if the file system is mounted on on OST,
4141 memory pressure could force it to try to write out data,
4142 which it needed to allocate memory to do. Due to the low
4143 memory, it would be unable to do so and the node would
4144 become unresponsive.
4146 Severity : enhancement
4148 Description: Addition of lconf --service command line option
4149 Details : lconf now accepts a '--service <arg>' option, which is
4150 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
4152 Severity : enhancement
4154 Description: Failover mode is now the default for OSTs.
4155 Details : By default, OSTs will now run in failover mode. To return to
4156 the old behaviour, add '--failout' to the lmc line for OSTs.
4158 Severity : enhancement
4160 Description: Health checks are now provided for MDS and OSTs
4161 Details : Additional detailed health check information on MSD and OSTs
4162 is now provided through the procfs health_check value.
4165 Frequency : occasional, depends on IO load
4167 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
4168 after numerous create/delete cycles
4169 Details : The ext3 inode allocation policy would not allocate new inodes
4170 very well on the OSTs because there are no new directories
4171 being created. Instead we look for groups with free space if
4172 the parent directories are nearly full.
4177 Description: Network or server problems during mount may cause partially
4178 mounted clients instead of returning an error.
4179 Details : The config llog parsing code may overwrite the error return
4180 code during mount error handling, returning success instead
4186 Description: MDS can fail to allocate large reply buffers
4187 Details : After long uptimes the MDS can fail to allocate large reply
4188 buffers (e.g. zconf client mount config records) due to memory
4189 fragmentation or consumption by the buffer cache. Preallocate
4190 some large reply buffers so that these replies can be sent even
4191 under memory pressure.
4195 Frequency : rare (liblustre)
4196 Description: fsx running with liblustre complained that using truncate() to
4197 extend the file doesn't work. This patch corrects that issue.
4198 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
4199 ATTR_SIZE and lsm use in llu_setattr_raw.
4203 Frequency : rare, only 2.6 kernels
4204 Description: Unusual file access patterns on the MDS may result in inode
4205 data being lost in very rare circumstances.
4206 Details : Bad interaction between the ea-in-inode patch and the "no-read"
4207 code in the 2.6 kernel caused the inode and/or EA data not to
4208 be read from disk, causing single-file corruption.
4212 Frequency : rare, only 2.6 filesystems using extents
4213 Description: Heavy concurrent write and delete load may cause data corruption.
4214 Details : It was possible under high-load situations to have an extent
4215 metadata block in the block device cache from a just-unlinked
4216 file overwrite a newly-allocated data block. We now unmap any
4217 metadata buffers that alias just-allocated data blocks.
4221 Frequency : filesystems with default stripe_count larger than 77
4222 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
4223 Details : lconf specifies an inode size of 4096 bytes when the default
4224 stripe_count is larger than 77. This conflicts with the default
4225 inode density of 1 per 4096 bytes. Allocate smaller inodes in
4226 this case to avoid pinning too much memory for large EAs.
4228 ------------------------------------------------------------------------------
4230 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
4235 Frequency : rare (extremely heavy IO load with hundreds of clients)
4237 Description: Client is evicted, gets IO error writing to file
4238 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
4239 caused clients to be evicted for AST timeouts. The fixes in
4240 bug 5192 mean we no longer need to have such short AST timeouts
4241 so ldlm_timeout has been increased.
4244 Frequency : occasional during --force or --failover shutdown under load
4245 Bugzilla : 5949, 4834
4246 Description: Server oops/LBUG if stopped with --force or --failover under load
4247 Details : a collection of import/export refcount and cleanup ordering
4248 issues fixed for safer force cleanup
4251 Frequency : only filesystems larger than 120 OSTs
4252 Bugzilla : 5990, 6223
4253 Description: lfs getstripe would oops on a very large filesystem
4254 Details : lov_getconfig used kfree on vmalloc'd memory
4257 Frequency : only filesystems exporting via NFS to Solaris 10 clients
4258 Bugzilla : 6242, 6243
4259 Description: reading from files that had been truncated to a non-zero size
4260 but never opened returned no data
4261 Details : ll_file_read() reads zeros from no-object files to EOF
4266 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
4267 Details : There's little harm in aborting MDS/OSS recovery and letting it
4268 try again, so I removed the LASSERT and return an error instead.
4270 Severity : enhancement
4272 Description: New debugging infrastructure for tracking down data corruption
4273 Details : The I/O checksum code was replaced to: (a) control it at runtime,
4274 (b) cover more of the client-side code path, and (c) try to narrow
4275 down where problems occurred
4279 Bugzilla : 3819, 4364, 4397, 6313
4280 Description: Racing close and eviction MDS could cause assertion in mds_close
4281 Details : It was possible to get multiple mfd references during close and
4282 client eviction, leading to one thread referencing a freed mfd.
4284 Severity: : enhancement
4285 Bugzilla : 3262, 6359
4286 Description: Attempts to reconnect to servers are now more aggressive.
4287 Details : This builds on the enhanced upcall-less recovery that was added
4288 in 1.4.2. When trying to reconnect to servers, clients will
4289 now try each server in the failover group every 10 seconds. By
4290 default, clients would previously try one server every 25 seconds.
4295 Description: After recovery, certain operations trigger a failed
4296 assertion on a client.
4297 Details : Failing over an mds, using lconf -d --failover, while a
4298 client was doing a readdir() call would cause the client to
4299 LBUG after recovery completed and the readdir() was resent.
4301 Severity : enhancement
4303 Description: Default groups are now added by lconf
4304 Details : You can now run lconf --group <servicename> without having to
4305 manually add groups with lmc.
4308 Frequency : occasional
4310 Description: Nodes with an elan id of 0 trigger a failed assertion
4313 Frequency : always when accessing e.g. tty/console device nodes
4315 Description: tty and some other devices nodes cannot be used on lustre
4316 Details : file's private_data field is used by device data and lustre
4317 values in there got lost. New field was added to struct file to
4318 store fs-specific private data.
4321 Frequency : when exporting Lustre via NFS
4323 Description: NFSD failed occasionally when looking up a path component
4324 Details : NFSD is looking up ".." which was broken in ext3 directories
4325 that had grown large enough to become hashed.
4328 Frequency : Clusters with multiple interfaces not on the same subnet
4330 Description: Nodes will repeatedly try to reconnect to an interface which it
4331 cannot reach and report an error to the log.
4332 Details : Extra peer list entries will be created by lconf with some peers
4333 unreachable. lconf now validates the peer before adding it.
4336 Frequency : Only if a default stripe is set on the filesystem root.
4338 Description: Setting a default stripe on the filesystem root prevented the
4339 filesystem from being remounted.
4340 Details : The client was sending extra request flags in the root getattr
4341 request and did not allocate a reply buffer for the dir EA.
4344 Frequency : occasional, higher if lots of files are accessed by one client
4345 Bugzilla : 6159, 6097
4346 Description: Client trips assertion regarding lsm mismatch/magic
4347 Details : While revalidating inodes the VFS looks up inodes with ifind()
4348 and in rare cases can find an inode that is being freed.
4349 The ll_test_inode() code will free the lsm during ifind()
4350 when it finds an existing inode and then the VFS later attaches
4351 this free lsm to a new inode.
4355 Bugzilla : 6422, 7030
4356 Description: MDS deadlock between mkdir and client eviction
4357 Details : Creating a new file via mkdir or mknod (starting a transaction
4358 and getting the ns lock) can deadlock with client eviction
4359 (gets ns lock and trying to finish a synchronous transaction).
4362 Frequency : occasional
4363 Description: While starting a server, the fsfilt_ext3 module could not be
4365 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
4366 kernels. Previously, lconf would still use the ext3 name
4367 when trying to load modules. Now, it will correctly use
4368 ext3 on 2.4 and ldiskfs on 2.6.
4370 Severity : enhancement
4371 Description: The default stripe count has been changed to 1
4372 Details : The interpretation of the default stripe count (0, to lfs
4373 or lmc) has been changed to mean striping across a single
4374 OST, rather than all available. For general usage we have
4375 found a stripe count of 1 or 2 works best.
4377 Severity : enhancement
4378 Description: Add support for compiling against Cray portals.
4379 Details : Conditional compiling for some areas that are different
4383 Frequency : occasional
4384 Bugzilla : 6409, 6834
4385 Description: Creating files with an explicit stripe count may lead to
4386 a failed assertion on the MDS
4387 Details : If some OSTs are full or unavailable, creating files may
4388 trigger a failed assertion on the MDS. Now, Lustre will
4389 try to use other servers or return an error to the
4393 Frequency : occasional
4395 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
4396 caused lock timeout during readahead (since 1.4.2).
4397 Details : Processes doing readahead might match a lock that hasn't been
4398 granted yet if there are overlapping and conflicting lock
4399 requests. The readahead process waits on ungranted lock
4400 (original lock is CBPENDING), while OST waits for that process
4401 to cancel CBPENDING read lock and eventually evicts client.
4403 Severity : enhancement
4405 Description: Initial enabling of flock support for clients
4406 Details : Implements fcntl advisory locking and file status functions.
4407 This feature is provided as an optional mount flag (default
4408 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
4409 locking are implemented yet, and those that are are not guaranteed
4410 to be completely correct in production environments.
4411 mount -t lustre -o [flock|noflock] ...
4414 Frequency : occasional
4416 Description: OSTs running 2.4 kernels but with extents enabled might trip an
4417 assertion in the ext3 JBD (journaling) layer.
4418 Details : The b_committed_data struct is protected by the big kernel lock
4419 in 2.4 kernels, serializing journal_commit_transaction() and
4420 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
4421 access to this struct. In 2.6 kernels there is finer grained
4422 locking to improve SMP performance of the JBD layer.
4426 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
4428 -----------------------------------------------------------------------------
4430 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
4432 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
4433 versions of Lustre. Please update all servers and clients to
4434 version 1.4.2 or later at the same time. You must also run
4435 "lconf --write-conf {config}.xml" on the MDS while it is stopped
4436 to update the configuration logs.
4438 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
4439 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
4440 - swab llog records of type '0' so we get proper header size/idx (5861)
4441 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
4442 - fix rename of one directory over another leaking an inode (5953)
4443 - avoid SetPageDirty on 2.6 (5981)
4444 - don't re-add just-being-destroyed locks to the waiting list (5653)
4445 - when creating new directories, inherit the parent's custom
4446 striping settings if present parent (3048)
4447 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
4448 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
4449 - don't hold client locks on temporary worklist from l_lru (5666)
4450 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
4451 - automatically evict dead clients (5921)
4452 - Update file size properly in create+truncate+fstat case (6196)
4453 - Do not unhash mountpoint dentries, do not allow removal of
4455 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
4456 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
4457 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
4458 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
4459 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
4460 - Correctly handle reads of files with no objects (6243)
4461 - lctl recover will also mark a device active if deactivate used (5933)
4463 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
4464 - allow --write-conf on an MDS with different nettype than client (5619)
4465 - don't write config llogs to MDS for mounts not from that MDS (5617)
4466 - lconf should create multiple TCP connections from a client (5201)
4467 - init scripts are now turned off by default; run chkconfig --on
4468 lustre and chkconfig --on lustrefs to use them
4469 - upcalls are no longer needed for clients to recover to failover
4471 - add --abort-recovery option to lconf to abort recovery on device
4473 - add support for an arbitrary number of OSTs (3026)
4474 - Quota support protocol changes.
4475 - forward compatibility changes to wire structs (6007)
4476 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
4477 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
4478 - improved "lustre status" script
4479 - initialize blocksize for non-regular files (6062)
4480 - added --disable-server and --disable-client configure options (5782)
4481 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
4482 - Vanilla 2.4.29 support
4483 - increase maximum number of obd devices to 520 (6242)
4484 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
4485 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
4486 SLES 9 2.6.5 kernels
4487 - limit stripes per file to 160 (the maximum EA size) (6093)
4489 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4492 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
4493 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
4494 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
4495 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
4496 - small ext3 extents cleanups and fixes (5733)
4497 - improved mballoc code, several small races and bugs fixed (5733, 5638)
4498 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
4499 - avoid needless client->OST connect, fix handle mismatch (5317)
4500 - fix DLM error path that led to out-of-sync client, long delays (5779)
4501 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
4502 - fix several locking issues related to i_size (5492,5624,5654,5672)
4503 - don't move pending lock onto export if it is already evicted (5683)
4504 - fix kernel oops when creating .foo in unlinked directory (5548)
4505 - fix deadlock in obdfilter statistics vs. object create (5811)
4506 - use time_{before,after} to avoid timer jiffies wrap (5882)
4507 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
4508 - Do not leak request if server was not able to process it (5154)
4509 - If mds_open unable to find parent dir, make that negative lookup(5154)
4510 - don't create new directories with extent-mapping (5909, 5936)
4512 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
4513 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
4514 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
4515 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
4516 - support for RHEL4 kernel on i686 (5773)
4517 - provide error messages when incompatible logs are encountered (5898)
4519 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
4520 * version 1.4.0.10 (1.4.1 release candidate 1)
4522 - don't keep a lock reference when lock is not granted (4238)
4523 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4524 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
4525 - drop import inflight refcount on signal_completed_replay error (5255)
4526 - unlock page after async write error during send (3677)
4527 - handle missing objects in filter_preprw_read properly (5265)
4528 - no transno return for symlink open, don't save no-trasno open (3440)
4529 - don't try to complete elan receive that already failed (4012)
4530 - free RPC server reply state on error (5406)
4531 - clean up thread from ptlrpc_start_thread() on error (5160)
4532 - readahead could read extra page into cache that wasn't ejected (5388)
4533 - prevent races in class_attach/setup/cleanup/detach (5260)
4534 - don't dereference de->d_inode after l_dput of de (5458)
4535 - use "int" for stripe value returned from lock_to_stripe (5544)
4536 - mballoc allocation and error-checking fixes in 2.6 (5504)
4537 - block device patches to fix I/O request sizes in 2.6 (5482)
4538 - look up hostnames for IB nals (5602)
4539 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
4540 - don't start multiple acceptors for the same port (5277)
4541 - fix incorrect LASSERT in mds_getattr_name (5635)
4542 - export a proc file for general "ping" checking (5628)
4543 - fix "lfs check" to not block when the MDS is down (5628)
4545 - service request history (4965)
4546 - put {ll,lov,osc}_async_page structs in a single slab (4699)
4547 - create an "evict_client" /proc entry on OSTs, like the MDS has
4548 - fix mount usage message, return errors per mount(8) (5168)
4549 - change grep [] to grep "[]" in tests so they work in more UMLs
4550 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
4551 - remove ancient LOV_MAGIC_V0 EA support (5047)
4552 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
4553 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
4554 - allow readahead allocations to fail when low on memory (5383)
4555 - mmap locking landed again, after considerable improvement (2828)
4556 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
4558 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
4561 - send OST transaction number in read/write reply to free req (4966)
4562 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4563 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4564 - the watchdog thread now runs as interruptible (5246)
4565 - flock/lockf fixes (but it's still disabled, pending 5135)
4566 - don't use EXT3 constants in llite code (5094)
4567 - memory shortage at startup could cause assertion (5176)
4569 - reorganization of lov code
4570 - single portals codebase
4572 - add extents/mballoc support (5025)
4573 - direct I/O reads in the obdfilter (4048)
4574 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
4576 tbd Cluster File Systems, Inc. <info@clusterfs.com>
4579 - send OST transaction number in read/write reply to free req (4966)
4580 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
4581 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
4582 - the watchdog thread now runs as interruptible (5246)
4583 - handle missing objects in filter_preprw_read properly (5265)
4584 - unsafe list practices (rarely) led to infinite eviction loop (4908)
4585 - drop import inflight refcount on signal_completed_replay error (5255)
4586 - unlock page after async write error during send (3677)
4587 - return original error code on reconstructed replies (3761)
4588 - no transno return for symlink open, don't save no-trasno open (3440)
4590 - add pid to ldlm debugging output (4922)
4591 - bump the watchdog timeouts -- we can't handle 30sec yet
4592 - extra debugging for orphan dentry/inode bug (5259)
4594 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
4597 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
4598 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
4599 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
4600 - fix NULL dereference in /proc/sys/portals/routes (4827)
4601 - allow failed mdc_close() operations to be interrupted (4561)
4602 - stop precreate on OST before MDS would time out on it (4778)
4603 - don't send partial-page writes before EOF from client (4410)
4604 - discard client grant for sub-page writes on large-page clients (4520)
4605 - don't free dentries not owned by NFS code, check generation (4806)
4606 - fix lsm leak if mds_create_objects() fails (4801)
4607 - limit debug_daemon file size, always print CERROR messages (4789)
4608 - use transno after validating reply (3892)
4609 - process timed out requests if import state changes (3754)
4610 - update mtime on OST during writes, return in glimpse (4829)
4611 - add mkfsoptions to LDAP (4679)
4612 - use ->max_readahead method instead of zapping global ra (5039)
4613 - don't interrupt __l_wait_event() during strace
4615 - add software watchdogs to catch hung threads quickly (4941)
4616 - make lustrefs init script start after nfs is mounted
4617 - fix CWARN/ERROR duplication (4930)
4618 - return async write errors to application if possible (2248)
4619 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
4620 - print NAL number in %x format (4645)
4621 - update barely-supported suse-2.4.21-171 series (4842)
4622 - support for sles 9 %post scripts
4623 - support for building 2.6 kernel-source packages
4624 - support for sles km_* packages
4626 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
4629 - ignore -ENOENT errors in osc_destroy (3639)
4630 - notify osc create thread that OSC is being cleaned up (4600)
4631 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
4632 - reconstruct ost_handle() like mds_handle() (4657)
4633 - create a new thread to do import eviction to avoid deadlock (3969)
4634 - let lconf resolve symlinked-to devices (4629)
4635 - don't unlink "objects" from directory with default EA (4554)
4636 - hold socknal file ref over connect in case target is down (4394)
4637 - allow more than 32000 subdirectories in a single directory (3244)
4638 - fix blocks count for O_DIRECT writes (3751)
4639 - OST returns ENOSPC from object create when no space left (4539)
4640 - don't send truncate RPC if file size isn't changing (4410)
4641 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
4642 - bind to privileged port in socknal and tcpnal (3689)
4644 - rate limit CERROR/CWARN console message to avoid overload (4519)
4645 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
4646 - basic mmap support (3918)
4647 - kernel patch series update from b1_4 (4711)
4649 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
4652 - avoid crash during MDS cleanup with OST shut down (2775)
4653 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
4654 - don't use bad inodes on the MDS (3744)
4655 - dynamic object preallocation to improve recovery speed (4236)
4656 - don't hold spinlock over lock dumping or change debug flags (4401)
4657 - don't zero obd_dev when it is force cleaned (3651)
4658 - print grants to console if they go negative (4431)
4659 - "lctl deactivate" will stop automatic recovery attempts (3406)
4660 - look for existing locks in ldlm_handle_enqueue() (3764)
4661 - don't resolve lock handle twice in recovery avoiding race (4401)
4662 - revalidate should check working dir is a directory (4134)
4664 - don't always mark "slow" obdfilter messages as errors (4418)
4666 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
4669 - don't close LustreDB during write_conf until it is done (3860)
4670 - fix typo in lconf for_each_profile (3821)
4671 - allow dumping logs from multiple threads at one time (3820)
4672 - don't allow multiple threads in OSC recovery (3812)
4673 - fix debug_size parameters (3864)
4674 - fix mds_postrecov to initialize import for llog ctxt (3121)
4675 - replace config semaphore with spinlock (3306)
4676 - be sure to send a reply for a CANCEL rpc with bad export (3863)
4677 - don't allow enqueue to complete on a destroyed export (3822)
4678 - down write_lock before checking llog header bitmap (3825)
4679 - recover from lock replay timeout (3764)
4680 - up llog sem before sending rpc (3652)
4681 - reduce ns lock hold times when setting kms (3267)
4682 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
4683 - fix NULL deref and obd_dev leak on setup error (3312)
4684 - replace some LBUG about llog ops with error handling (3841)
4685 - don't match INVALID dentries from d_lookup and spin (3784)
4686 - hold dcache_lock while marking dentries INVALID and hashing (4255)
4687 - fix invalid assertion in ptlrpc_set_wait (3880)
4689 - add libwrap support for the TCP acceptor (3996)
4690 - add /proc/sys/portals/routes for non-root route listing (3994)
4691 - allow setting MDS UUID in .xml (2580)
4692 - print the stack of a process that LBUGs (4228)
4694 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
4697 - don't cleanup request in ll_file_open() on failed MDS open (3430)
4698 - make sure to unset replay flag from failed open requests (3440)
4699 - if default stripe count is 0, use OST count for inode size (3636)
4700 - update parent mtime/ctime on client for create/unlink (2611)
4701 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
4702 - free recovery state on server during a forced cleanup (3571)
4703 - unregister_reply for resent reqs (3063)
4704 - loop back devices mounting and status check on 2.6 (3563)
4705 - fix resource-creation race that can provoke i_size == 0 (3513)
4706 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
4707 - more debugging for page-accounting assertion (3746)
4708 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
4709 - avoid deadlock after precreation failure (3758)
4710 - fix race and lock order deadlock in orphan handling (3450, 3750)
4711 - add validity checks when grabbing inodes from l_ast_data (3599)
4713 - add /proc/.../recovery_status to obdfilter (3428)
4714 - lightweight CDEBUG infrastructure, debug daemon (3668)
4715 - change default OSC RPC parameters to be better on small clusters
4716 - turn off OST read cache for files smaller than 32MB
4717 - install man pages and include them in rpms (3100)
4718 - add new init script for (un)mounting lustre filesystems (2593)
4719 - run chkconfig in %post for init scripts (3701)
4720 - drop scimac NAL (unmaintained)
4722 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
4725 - clean kiobufs before and after use (3485)
4726 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
4727 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
4728 - change init script to fail more clearly if not run as root (1528)
4729 - allow clients to reconnect during replay (1742)
4730 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
4731 - don't do DNS lookups on NIDs too small for IP addresses (3442)
4732 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
4733 - fix cond_resched (3554)
4734 - only evict unfinished clients after recovery (3515)
4735 - allow bulk resend, prevent data loss (3570)
4736 - dynamic ptlrpc request buffer allocation (2102)
4737 - don't allow unlinking open directory if it isn't empty (2904)
4738 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
4739 - remove extraneous obd dereference causing LASSERT failure (3334)
4740 - don't use get_cycles() when creating temp. files on the mds (3156)
4741 - hold i_sem when setting i_size in ll_extent_lock() (3564)
4742 - handle EEXIST for set-stripe, set proper directory name (3336)
4744 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
4745 - fix ksocknal_fmb_callback() error messages (2918)
4747 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
4750 - don't copy lvb into (possibly NULL) reply on error (2983)
4751 - don't deref dentry after dput, don't free lvb on error (2922)
4752 - use the kms to determine writeback rpc length (2947)
4753 - increment oti_logcookies when osc is inactive (2948)
4754 - update client's i_blocks count via lvb messages (2543)
4755 - handle intent open/close of special files properly (1557)
4756 - mount MDS with errors=remount-ro, like obdfilter (2009)
4757 - initialize lock handle to avoid ASSERT on error cleanup (3057)
4758 - don't use cancelling-locks' kms values (2947)
4759 - use highest lock extent for kms, not last one (2925)
4760 - don't dereference ERR_PTR() dentry in error handling path (3107)
4761 - fix thread race in portals_debug_dumplog() (3122)
4762 - create lprocfs device entries at setup instead of at attach (1519)
4763 - common AST error handler, don't evict client on completion race (3145)
4764 - zero nameidata in detach_mnt in 2.6 (3118)
4765 - verify d_inode after revalidate_special is valid in 2.6 (3116)
4766 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
4767 - initialize RPC timeout timer earlier for 2.6 (3219)
4768 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
4769 - print nal/nid for unknown nid (3258)
4770 - additional checks for oscc recovery before doing precreate (3284)
4771 - fix ll_extent_lock() error return code for 64-bit systems (3043)
4772 - don't crash in mdc_close for bad permissions on open (3285)
4773 - zero i_rdev for non-device files (3147)
4774 - clear page->private before handing to FS, better assertion (3119)
4775 - tune the read pipeline (3236)
4776 - fix incorrect decref of invalidated dentry (2350)
4777 - provide read-ahead stats and refine rpc in flight stats (3328)
4778 - don't hold journal transaction open across create RPC (3313)
4779 - update atime on MDS at close time (3265)
4780 - close LDAP connection when recovering to avoid server load (3315)
4781 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
4782 - don't leak open file on MDS after open resend (3325)
4783 - serialize filter_precreate and filter_destroy_precreated (3329)
4784 - loop device shouldn't call sync_dev() for nul device (3092)
4785 - clear page cache after eviction (2766)
4786 - resynchronize MDS->OST in background (2824)
4787 - refuse to mount the same filesystem twice on same mountpoint (3394)
4788 - allow llmount to create routes for mounting behind routers (3320)
4789 - push lock cancellation to blocking thread for glimpse ASTs (3409)
4790 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
4791 - fix rare problem with rename on htree directories (3417)
4793 - allow default OST striping configuration per directory (1414)
4794 - fix compilation for qswnal for 2.6 kernels (3125)
4795 - increase maximum number of MDS request buffers for large systems
4796 - change liblustreapi to be useful for external progs like lfsck (3098)
4797 - increase local configuration timeout for slow disks (3353)
4798 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
4800 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
4803 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
4804 - don't overwrite extent policy data in reply if lock was blocked (2901)
4805 - drop filter export grants atomically with removal from device (2663)
4806 - del obd_self_export from work_list in class_disconnect_exports (2908)
4807 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
4808 - swab reply message in mdc_close, other PPC fixes (2464)
4809 - fix destroying of named logs (2325)
4810 - overwrite old logs when running lconf --write_conf (2264)
4811 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
4812 - fix race in target_handle_connect (2898)
4813 - mds_reint_create() should take same inode create lock (2926)
4814 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
4815 - don't close files for self_export to avoid uninitialized obd (2936)
4816 - allow MDS with the same name as client node (2939)
4817 - hold dentry reference for closed log files for unlink (2325)
4818 - reserve space for all logs during transactions (2059)
4819 - don't evict page beyond end of stripe extent (2925)
4820 - don't oops on a deleted current working directory (2399)
4821 - handle hard links to targets without a parent properly (2517)
4822 - don't dereference NULL lock when racing during eviction (2867)
4823 - don't grow lock extents when lots of conflicting locks (2919)
4825 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
4828 - account for cache space usage on clients to avoid data loss (974)
4829 - lfsck support in lustre kernel code (2349)
4830 - reduce journal credits needed for BRW writes (2370)
4831 - orphan handling to avoid losing space on client/server crashes
4832 - ptlrpcd can be blocked, stopping ALL progress (2477)
4833 - use lock value blocks to assist in proper KMS, faster stat (1021)
4834 - takes i_sem instead of DLM locks internally on obdfilter (2720)
4835 - recovery for initial connections (2355)
4836 - fixes for mds_cleanup_orphans (1934)
4837 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
4838 - block all file creations until orphan recovery completes (1901)
4839 - client remove rq_connection from request struct (2423)
4840 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
4841 - recovery timer race (2670)
4842 - mdc_close recovey bug (2532)
4843 - ptlrpc cleanup bug (2710)
4844 - mds timeout on local locks (2588)
4845 - namespace lock held during RPCs (2431)
4846 - handle interrupted sync write properly (2503)
4847 - don't try to handle a message that hasn't been replied to (2699)
4848 - client assert failure during cleanup after abort recovery (2701)
4849 - leak mdc device after failed mount (2712)
4850 - ptlrpc_check_set allows timedout requests to complete (2714)
4851 - wait for inflight reqs when ptlrpcd finishes (2710)
4852 - make sure unregistered services are removed from the srv_list
4853 - reset bulk XID's when resending them (caught by 1138 test)
4854 - unregister_bulk after timeout
4855 - fix lconf error (2694)
4856 - handle write after unfinished setstripe, stripe-only getstripe (2388)
4857 - readahead locks pages, leaves pending causing memory pressure (2673)
4858 - increase OST request buffers to 4096 on large machines (2729)
4859 - fix up permission of existing directories in simple_mkdir (2661)
4860 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
4861 - don't assign transno to errored transactions (2742)
4862 - don't delete objects on OST if given a bogus objid from MDS (2751)
4863 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
4864 - if rq_no_resend, then timeout request after recovery (2432)
4865 - fix MDS llog_logid record size, 64-bit array alignment (2733)
4866 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
4867 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
4868 - MDS recovery shouldn't delete objects that it has given out (2730)
4869 - if enqueue arrives after completion, don't clobber LVB (2819)
4870 - don't unlock pages twice when trigger_group_io returns error (2814)
4871 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
4872 - don't write pages to disk if there was an error (1450)
4873 - don't ping imports that have recovery disabled (2676)
4874 - take buffered bytes into account when balancing socknal conn (2817)
4875 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
4876 - reconnect unlink llog connection after MDS reconnects to OST (2816)
4877 - remove little-endian swabbing of llog records (1987)
4878 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
4879 - retry reposting request buffers if they fail (1191)
4880 - grow extent at grant time to avoid granting a revoked lock (2809)
4881 - lock revoke doesn't evict page if covered by a second lock (2765)
4882 - disable VM readahead to avoid reading outside lock extents (2805)
4884 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
4885 - updated kernel patches for hp-2.4.20 kernel (2681)
4887 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
4890 - fix truncated write corruption (2366)
4891 - fix for failed assertion in iopen_connect_dentry (1792,2517)
4893 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
4894 - ensure len(uuid) < 37 in lmc (1171)
4895 - fix ia64 OOPS in llog_test (2255)
4896 - zero end of page at obdfilter for partial page writes (2648)
4897 - don't leave stale dentries around after renames (bug 2428)
4898 - fix timeouts when evicting a client with a single lock held (2642)
4899 - set deadline for the initial HELLO message to drain (2634)
4900 - print out dotted-quad IP addresses in the socknal (2302)
4902 - additional debugging for MDS client eviction problem (2443)
4903 - fix mkfsoptions support for osts (2603, 2604)
4905 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
4908 - add series for the vanilla 2.6.0 kernel
4909 - add series for the vanilla 2.4.24 kernel
4910 - add series for a cray x86/64 UL kernel drop
4911 - fix xattr patches for the vanilla 2.4.19 series
4913 - generate true UUIDs in lmc (1171)
4914 - have portals stack dumping break in UML (2466)
4915 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
4916 - allocate with _NOFS in ldlm to avoid deadlock (1933)
4917 - wake callback waiting threads on client eviction (2460)
4918 - Add --ptldebug and --subsystem to lmc (1719)
4919 - update assertion to allow safe interrupt allocation
4920 - set rq_no_resend for cancel requests (2432)
4921 - recalculate ptlrpcd timeout after resend (2494)
4922 - call vfs_rmdir when removing pending directories (2368)
4923 - fix renaming a file to itself (2429)
4924 - lmc creates a default one-stripe lov (2454)
4925 - expand procfs space to handle large clusters (2326)
4926 - increase UML stack to avoid overflow
4927 - update lconf's list of debug and subsystem masks
4928 - fix lfs find --obd (2510)
4929 - /proc tunable for disabling filter read caching (2591)
4930 - stop rpm packages from altering slapd.conf (2301)
4931 - disable nagle in the socknal under 0conf (2578)
4932 - choose mds inode size based on stripe count (2572)
4933 - fix kernel-source rpm problems (2516)
4935 - add --disable-doc to avoid pdf generation (2421)
4936 - update documentation, tests, type-os, comments
4937 - avoid format warnings on ia64
4938 - remove the TOE NAL
4939 - tiny code cleanups by removing unused fields
4941 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
4944 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
4945 - remove the most blatant lies from BUILDING file (2371)
4946 - change default debug level to reasonable production setting
4947 - reduce client side cache size to reduce cache flush time
4948 - reduce max RPCs in flight to avoid unnecessary file fragmentation
4949 - make TCP zerocopy and pinger support enabled by default (2476)
4950 - sync writes completed after process exits caused crashes (2319)
4951 - maintain correct mount count on the MDS (2356)
4952 - backout 1557, because 2316 wasn't really fixed
4953 - better file I/O statistics gathering in /proc
4954 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
4955 - another kernel patch to fix zero-copy TCP function export
4956 - don't take duplicate lock when processing re-sent getattr (2420)
4957 - lctl uses obd_self_export instead of creating new conn (2353)
4958 - MDS/OST recovery case which requires object creation asserted (2425)
4959 - move lfs from /usr/sbin to /usr/bin in packages
4960 - fix race between mds_client_add and mds_client_free (2417)
4961 - use kmalloc instead of slabs in portals (2430)
4962 - don't create duplicate records when a failover MDS is present (2442)
4963 - remove unnecessary mount age check (2332)
4964 - don't remove directory inodes from locks prematurely (2451)
4965 - don't break if MDS service name is the same as hostname (2103)
4966 - fix races in client write RPC generation when cache full (2482)
4968 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
4971 - remove now-unused request->rq_obd (278)
4972 - if an allocation fails, print out how much memory we've used (1933)
4973 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
4974 - add the "configurable stack size" patch to most series files (1256)
4975 - ability to write large log records, for 100+ OST configs (2306)
4976 - fix NULL deref when filter_prep fails (2314)
4977 - fix operator precedence error in filter_sync
4978 - dynamic allocation of socknal TX descriptors (2315)
4979 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
4980 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
4981 - ensure that CWARN messages in Portals always get to the syslog
4982 - __init/__exit are not for prototype decls (ldlm_init/exit)
4983 - x86-64 compile warning fixes
4984 - fix gateway LMC keyword conflict (2318)
4985 - fix MDS lock inversions in getattr/reint paths (1844)
4986 - fix a rare lock re-ordering bug, which caused deadlock (2322)
4987 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
4988 - DLM race condition prevented some lock evictions (2328)
4989 - ENOMEM detection and retry on socknal sends (2230)
4990 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
4991 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
4992 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
4993 - better error messages when a client is rejected during recovery (1505)
4994 - avoid cancelling locks which were never granted, after failure (2330)
4995 - fix i_sem/journal inversion in mds_client_add (2333)
4996 - fix truncate/getattr lock cycle deadlock (2334)
4997 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
4998 - fix two rare exit paths which could leak an l_lock() ref (2321)
4999 - fencepost error in MDS/OST orphan recovery (2226)
5000 - make log record alignment 8 bytes (1988)
5001 - lstripe now fails when requested offset > ost_count (2237)
5002 - ensure that all kernel series have a complete list.h (1607)
5003 - fix crashes in special-file operations (2316)
5004 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
5006 - allow configurable automake binary, for testing new versions
5007 - small update to the lfs documentation
5009 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
5011 * fix negative export reference count in fsfilt_sync (2312)
5013 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
5014 * release candidate 0.9.1
5016 - orphans are moved into the PENDING directory for possible recovery
5017 - replayed opens now open by fid for orphan/rename safety (1042)
5018 - last close of an orphan inode generates a transno (683)
5019 - chdir() and mount() now pin the directory entry (1020)
5020 - avoid CERROR in normal ll_setattr_raw() error case (1500)
5021 - discard very old requests without processing them (1502)
5022 - remove some common, well-understood CERRORs (1505)
5023 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
5024 - clear "grant" flags in OST replies until OST grant code lands (1644)
5025 - fix read performance by not clobbering i_blksize on client (1598)
5026 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
5027 - make LRU size a /proc tunable, clears locks when reduced (707)
5028 - fix some lprocfs rot that prevented ptlbd from loading (1732)
5029 - server locks take references on exports now (1558)
5030 - build fixes for 2.4.20-rh trees (1663)
5031 - return an error from lov_create if all OSCs are inactive (1751)
5032 - fix import levels when a reconnect happens without a timeout (1597)
5033 - exit early from mds_open if we get a lookup error (1749)
5034 - partial page read at EOF wouldn't wait for disk before sending (1642)
5035 - avoid NULL deref in obdfilter when reading page past EOF (1592)
5036 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
5037 - fix LBUG in ll_it_open_error with rc = -2 (1861)
5038 - write/truncate lock inversion (1639)
5039 - Don't auto-load obdclass, portals modules during cleanup (1495)
5040 - fix timestamps from jumping to "now" (1763)
5041 - extra journal assertions (1648)
5042 - add an extra multiunlink test (1771)
5043 - fix read_record/write_record API (1776)
5044 - fix leak of offset_extent, possible incorrect i_size later (1772)
5045 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
5046 - Debugging for the kqswnal_get_idle_tx problems (1820)
5047 - Allow recovery to be attempted multiple times (1536)
5048 - Write out MDS last_rcvd file after it is first created (1600)
5049 - Fix tx_descriptor leak in failed transmit situations (1827)
5050 - ext3 journaling fixes for assertion failure after IO error (1871)
5051 - class_export_put() on freed export after completion AST error (1896)
5052 - Fix revalidate looping in VFS (1322)
5053 - Don't access a freed export during MDS_REINT timeout (1521)
5054 - Add open-unlink recovery support on the MDS (1673,1764)
5055 - Return an error if no MDS data was read from last_rcvd (1946)
5056 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
5057 - Don't setup a disk device that doesn't match exported UUID (317)
5058 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
5059 - avoid committing NULL handle in force close
5060 - local.sh is now a one-stripe LOV configuration
5061 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
5062 - fix ext3 htree duplicate directory entry corruption (1516)
5063 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
5064 - update to new LOV EA format (2097)
5065 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
5067 2003-06-15 Phil Schwan <phil@clusterfs.com>
5070 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
5071 - per-import/export recovery handling (958, 931, 959)
5072 - multiple last-rcvd slots, for serving multiple FSes (949)
5073 - connections are again shared between multiple imp/exports (963, 964)
5074 - "umount -f" would hang if any requests needed to be sent (393, 978)
5075 - avoid pinning large req buffer by copying for queued messages (989)
5076 - add "uuid" to "lctl device" command to help upcalls (991)
5077 - "open" RPCs with transnos would confuse recovery counters (1037)
5078 - do proper endian conversion of all wire messages (288, 340, 891)
5079 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
5080 - call waiting locks callback from LDLM recovery thread (1127, 1151)
5081 - fix ptlrpc_connection leak in target_handle_connect (1174)
5082 - fix import refcounting bug in OST and MDS cleanup (1134)
5083 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
5084 - fix very unlikely obd_types race condition (501)
5085 - remove osc_open hack for echo_client (1187)
5086 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
5087 - a failure in read_inode2 leads to deadlock (1139)
5088 - cancel ack-locks as soon as transaction is committed (1072)
5089 - fix major leaks and crashes in the bulk I/O path (937, 1057)
5090 - make sure to commitrw after any preprw to avoid deadlock (1162)
5091 - failing to execute a file in a lustre FS would lock inode (1203)
5092 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
5093 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
5094 - obd_uuid2tgt would walk off the end of the list (1255)
5095 - on IA64 the osc would give portals incorrect bulk size (1258)
5096 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
5097 - fix lock inversion caused by new llite matching code (1282)
5098 - limit the number of dirty pages on a client to 10MB (1286)
5099 - timed out locks were not being corrected cancelled (1289)
5100 - fix O_DIRECT above 4GB on IA-32 (1292)
5101 * major user-visible changes
5102 - fail out/fail over policy now controlled by the upcall (993)
5104 - add OBD_PING to check server availability and failure (954)
5105 - lustre messages are now sent in sending host order (288, 340, 891)
5106 - add eadatalen to MDS getattr reply (340)
5107 - OST read replies may contain second buffer, with per-page status (593)
5109 2003-03-11 Phil Schwan <phil@clusterfs.com>
5112 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
5113 - failed open()s could cause deadlock; fixed (867, 869)
5114 - stop cancelling OST locks when files are closed (481)
5115 - overlapping XID spaces caused network corruption (851, 853)
5116 - fix unsafe fsfilt counter arithmetic; change to atomic_t
5117 - setattr_raw added, to do single-RPC, server-side setattrs
5118 - lmc/lconf syntax change for OST UUIDs
5119 - fix crashy race condition between ptlrpc_free_req and osc_close
5120 - don't use request in mdc_enqueue if we hit a timeout (889)
5121 - don't set the inode i_size for regular files from the MDS (896)
5122 - handle out of order completion AST (842)
5123 - don't LBUG if a lock request times out after receiving AST (913)
5124 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
5125 - if a bad lock AST arrives, send an error instead of dropping entirely
5126 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
5127 - fix leak in bulk IO when only partially completed (899, 900, 926)
5128 - fix O_DIRECT for ia64 (55)
5129 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
5130 - C-z after timeout could hang a process forever; fixed (977)
5132 - client-side I/O cache (678, 924, 929, 941, 970)
5134 - READPAGE and SETATTRs which don't take server-side locks get
5137 2003-02-11 Phil Schwan <phil@clusterfs.com>
5140 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
5141 - Fix fsfilt_extN_readpage() to read a full page of directory
5142 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
5143 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
5144 - Fix imbalanced LOV object allocation and out-of-bound access (469)
5145 - Most intent operations were removed, in favour of a new RPC mode
5146 that does a single RPC to the server and bypasses most of the VFS
5147 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
5148 - Aggressively cancel local locks on DLM servers
5149 - mds_reint_unlink sends EA to the client if it's the last nlink.
5150 client uses that EA to unlink OST objects.
5151 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
5152 - recursive symlinks were fixed (439)
5153 - fixed NULL deref in DEBUG_REQ
5154 - filter_update_lastobjid no longer calls sync, which annoyed extN
5155 - fixed multi-client small-writes to a single file problem (445)
5156 - fixed mtime updates during file writes (607)
5157 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
5158 - fixed bug in obd_brw_read/write() (under guise of testing 367)
5159 - fixed Linux OST size reporting problem (444, 656)
5160 - OST now updates object mtime with writes or setattr (607, 619)
5161 - client verifies file size before zeroing page past EOF (445)
5162 - OST now writes last allocated objid to disk with allocation (108)
5163 - LOV on echo now works (409)
5165 - mds_reint_unlink sends a new buffer, with the EA included. this
5166 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
5167 set if a regular file was being unlinked, and it was the last link
5168 - use PtlGet from the target for bulk writes (315)
5169 - OST now updates object mtime with writes or setattr (607, 619)
5170 - LDLM now has a grant-time callback to revalidate locked items, if
5172 - Many MDS operations were reorganized to combat race conditions
5174 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
5175 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
5177 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
5180 - Fully reactivate OST imports after reconnection (512, others)
5181 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
5182 - More graceful error handling for truncating on dead OST (515)
5183 - Don't error out unless we're actually accessing dead stripes (474)
5184 - Fix garbage sizes when stripes are missing (410)
5185 - LRU counters were broken, causing constant lock purge (433, 432)
5186 - garbage on read from stripes with failed OSTs (441)
5187 - mark OSCs as active before reconnecting during recovery (438)
5188 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
5189 - lfind did not preserve OST order in output (443)
5190 - symlinks cause hung clients, incorrect data (439)
5191 - stop dereferencing request after dropping refcount (457)
5192 - don't LASSERT(spin_is_locked) on non-SMP (455)
5193 - fixes for many rename() bugs
5194 - fstat didn't correctly synchronize attributes (399)
5195 - server must handle lock cancellation during blocking AST prep (487)
5196 - bulk descriptors were free()d too soon (511)
5197 - fix paths in lconf, which would load incorrect modules (451, 507)
5198 - fix confusing lconf 'host not found' error message (386)
5199 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
5200 - fix race condition in mdc_blocking_ast() for inode access (526)
5201 - fix lov_unpackmd() unpacking wrong number of stripes (537)
5202 - fix lov_set_osc_active() marking wrong OSC inactive (440)
5203 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
5204 - fix multiple writes of stripe MD to MDS (358, maybe 519)
5205 - fix lstripe in several ways (kernel side) (527)
5206 - fix request leak in ldlm_cli_enqueue (262)
5207 - incorrect OSC was marked inactive after OST failure
5208 - call mds_fs_cleanup before unmounting filesystem (524)
5209 - fix races between taking ns_lock and ldlm_lock_change_resource
5210 - fix races updating LOV export open file list
5211 - fix lov_enqueue error path, avoid decref-ing bad lock handle
5212 - fix recovery NULL deref in ldlm_cli_cancel_unused
5213 - fix some DLM races by using new hash table for lock handles (419)
5214 - permit the client to specify desired inodes, at replay
5215 - duplicate requests when we queue them for replay reintegration
5216 - fix last_rcvd offset calculation
5217 - sync after each recovered transaction, so we always make progress
5218 - never, not always, ERESTART requests without transnos
5219 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
5221 - skip replay if the MDS says that the client is already connected
5222 - don't check for a recovery-enabled export to match lctl's UUID
5223 - don't INC_USE_COUNT for phantom exports
5224 - don't crash when cleaning up phantom exports (567)
5225 - don't double-finish or set replay data for errored mdc_open requests
5226 - abort requests when they time out, so we don't get old replies
5227 - send/receive replies for AST messages again
5228 - if the client says that it doesn't have the lock, cancel it on the
5230 - if we timeout during I/O, don't try to cancel an in-use lock; instead
5231 mark it as destroyed, it will all work out when decref is called
5232 - fix module use counts (22, 581)
5234 - ASTs now expect a reply (server cancels lock on error reply)
5236 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
5239 - fix many simultaneous client startup (392)
5240 - fix dentry->d_it clobbering
5241 - credentials weren't being shipped for readdir/getattr operations
5242 - remove invalid assertions triggered during some concurrent MD
5244 - proper Lustre versions added (336, 389)
5245 - fix memory leak for create error case (398)
5246 - fix LOV locking bug that would get cli/srv out of sync
5247 - fix echo client over LOV (409)
5248 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
5249 - fix double-O_EXCL intent crash (424)
5250 - avoid sending multiple lock CANCELs (352)
5252 - MDS can do multi-client recovery (modulo bugs in new code)
5254 - many updates, edits, cleanups
5256 2002-11-18 Phil Schwan <phil@clusterfs.com>
5259 - fix null d_it dereference (346)
5260 - fix full OST/dbench hang (333)
5261 - fix permission problem with file removal (286)
5262 - fix removal of OSCs from LOV when they fail
5263 - fix NULL deref during bulk timeout (214)
5264 - fix problems related to multiple filesystems on one MDS (241)
5265 - fixed serious subtle metadata locking bugs
5266 - free locks on clients when inodes are removed due to memory
5268 - fix inode pointer in lock data (285)
5269 - partial support for multiple MDS on a single host (241)
5270 - data locks weren't cancelled at clear_inode time (290, 311)
5271 - intent locks could lead to unbounded lock growth (205)
5272 - added a maximum lock count, an LRU list, and a flusher
5273 - fix multiple rename (365)
5274 - properly abstracted the echo client
5275 - OSC locked 1 byte too many; fixed
5276 - rewrote brw callback code:
5277 - fixed recovery bugs related to LOVs (306)
5278 - fixed too-many-pages-in-one-write crash (191)
5279 - fixed (again) crash in sync_io_timeout (214)
5280 - probably fixed callback-related race (385)
5282 - Add capability to MDS protocol
5283 - LDLM cancellations and callbacks on different portals
5285 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
5288 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
5289 - defer open object destruction to close time (601981, 138)
5290 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
5291 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
5292 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
5293 - don't call obd_setattr in ll_file_release for destroyed objects
5294 * protocol change to lustre_msg: move |version| and add |flags|
5295 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
5296 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
5297 * added replay of create, unlink, link and rename operations during
5298 MDS failover; recovery should be much more robust now
5299 * remove failed OSCs from LOVs (only lov_create uses this so far)
5300 * the lustre-HOWTO was brought (more) up to date (582544)
5302 2002-10-23 Phil Schwan <phil@clusterfs.com>
5305 - in-use dentries weren't being reused properly (617851)
5306 - prevent multiple LDLM setup (599178)
5307 - fix LOV size calculations for truncate (617853)
5308 - fix client handling of MDS intent errors (POSIX)
5309 - fix permission bug in lovstripe.c test (624321)
5310 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
5311 - truncate past end of file could corrupt data
5312 - proper cleanup after timeouts, crashes, etc (592524, 550815)
5313 - a race in recovery could return ETIMEDOUT to apps (623947)
5314 - building outside the source directory was fixed
5315 * the lustre-HOWTO was brought (more) up to date (582544)
5316 * major progress was made on recovery functionality
5318 2002-10-10 Phil Schwan <phil@clusterfs.com>
5321 - recovery deadlock fix
5322 - rm -rf causes LBUG fix (617817)
5323 - file open by multiple tasks fix (618962)
5324 - directory permissions bugs (602707 and 620007)
5325 - journal_stop fixed with locking (611313)
5326 - O_APPEND failures resolved (618273, perhaps 614459)
5327 - lconf PATH fix (619770)
5328 - IA64 build fix (621450)
5329 - RPC buffer sizes scale with amount of memory
5331 2002-10-01 Phil Schwan <phil@clusterfs.com>
5334 - locks would be cancelled without throwing away data pages,
5335 resulting in inconsistent data (605627)
5336 - inode attributes were not always being refreshed (605627, 612449)
5337 - lconf now continues to cleanup after lctl reports an error
5338 - MDS now enforces user permissions (602707)
5339 - lprocfs cleanup fixed, but not yet enabled (614157)
5340 - fixed infinite server hang, should a client not respond to an AST
5341 - avoid going into recovery if user calls readlink() with a buffer
5342 that's too small (613941)
5343 - AST RPCs no longer require replies (614867) -- this may be changed
5344 - don't crash server if client sends an IOV that's too big (611336)
5345 - fixed lock conversion deadlock (611892)
5346 - fixed the following of symlinks (614622)
5347 * recovery: the server can remove locks from a client that dies, other
5348 clients can make progress
5349 * more extN patch fixes
5350 * compile-time configurable ptlrpc buffer allocations
5352 - collaborative read cache document
5353 - Lustre Lite Performance CDR document-in-progress
5355 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5358 - fix typo in patch-2.4.18
5360 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
5363 - clear ptlrpc request each time in handle_incoming_request()
5364 - unlink of files now destroys the object on the OST
5366 2002-09-19 Peter Braam <braam@clusterfs.com>
5368 * add hard link support
5369 * change obdfile creation method
5370 * kernel patch changed
5372 2002-09-19 Peter Braam <braam@clusterfs.com>
5375 - stack overflow bug in extN fixed
5377 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
5379 * documentation updates
5380 - add man pages for config tools
5381 - update tests/README to describe testing with new config tools
5382 - finish metadata API descriptions
5383 * bug fixes and cleanups
5384 - statfs workaround for 16TB limit
5385 - LOV stripe allocation improved, can stripe on subset of OSTs
5386 - LOV file size/IO offset was wrong for files > 4GB in size
5387 - object EA data was being dropped, caused files to be unreadable
5388 - memory overflow with non-LOV OST caused memory corruption
5389 - fixed regression tests to work with new config tools, obdfilter
5390 - fixed bug when directory size became larger than 1 block
5391 - fixed bug (for single client case) when PWD was deleted
5392 - invalidate local directory pages when doing intent-based ops
5393 - avoid LDLM oops when lock callback contained bad data
5395 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
5397 * documentation updates
5398 * bug fixes and cleanups
5399 - configuration tools
5402 - 64-bit compile warnings
5403 - 64-bit internal statfs data
5405 * test_brw on persistent OST devices
5407 * lprocfs (disabled)
5409 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
5411 * documentation updates
5412 * bug fixes and cleanups
5413 * configuration tools
5415 2002-08-30 Peter J. Braam <braam@clusterfs.com>
5418 * many small fixes to 0_5_4
5419 * io/network handling
5420 * thinkos in MDS operations
5422 2002-08-24 Peter J. Braam <braam@clusterfs.com>
5425 * crucial basic fixes to 0.5.3
5426 * IOR, Iozone work over Elan
5429 2002-08-07 Phil Schwan <phil@clusterfs.com>
5430 * version 0_5_3, our first alpha
5431 * we use the new Portals iovs
5432 * documentation updates
5433 * bug fixes and cleanups
5434 * small changes in the DLM wire protocol
5436 2002-07-25 Peter J. Braam <braam@clusterfs.com>
5437 * version 0_5_1 with some initial stability,
5438 * locking on MD and file I/O.
5439 * documentation updates
5440 * several bug fixes since 0.5.0
5441 * small changes in wire protocol
5443 2002-07-18 Phil Schwan <phil@clusterfs.com>
5445 * delivered as Lustre Light Alpha
5446 * fixed a crash after handling invalid MDS requests
5447 * fixed directory pages for architectures with non-4k pages sizes
5449 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
5450 * release version v0_4_4
5451 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
5453 * Fixes a number of interruption problems with OST operations.
5454 * Update documentation for portals header changes
5455 * Move all wire protocol structs/defines to lustre_idl.h
5456 * Fixes symlink length bug.
5457 * Add tcpdump to repository.
5459 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
5460 * release version v0_4_3
5461 * Fixes statfs for inodes on extN.
5462 * Fixes bug in runtests which would delete /etc/hosts.
5463 * Use 64-bit object IDs wherever possible (not into VFS though)
5464 Remove ost_get_info, which is unused by lustre, and out of date.
5466 2002-07-03 Peter Braam <braam@clusterfs.com>
5467 * release version v0_4_2 Fixes a lookup error (type not passed)
5468 * move forward to head of Portals
5469 * move forward to latest Lustre kernel
5471 2002-06-25 Peter Braam <braam@clusterfs.com>
5472 * release version v0_4_1. Hopefully stable on single node use.