1 tbd Cluster File Systems, Inc. <info@clusterfs.com>
5 2.6.9-55.0.2.EL (RHEL 4),
6 2.6.16.46-0.14 (SLES 10),
7 2.6.18-8.1.8.el5 (RHEL 5),
8 2.6.18.8 vanilla (kernel.org)
9 * Client support for unpatched kernels:
10 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
11 2.6.16 - 2.6.22 vanilla (kernel.org)
12 * Due to recently discovered recovery problems, we do not recommend
13 using patchless RHEL 4 clients with this or any earlier release.
14 * Recommended e2fsprogs version: 1.40.2-cfs1
15 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
19 Description: Fix errors in lfs documentation
20 Details : Fixes man pages
22 Severity : enhancement
24 Description: Tune the kernel for good SCSI performance.
25 Details : Set the value of /sys/block/{dev}/queue/max_sectors_kb
26 to the value of /sys/block/{dev}/queue/max_hw_sectors_kb
29 --------------------------------------------------------------------------------
31 2007-09-27 Cluster File Systems, Inc. <info@clusterfs.com>
33 * Support for kernels:
35 2.6.9-55.0.2.EL (RHEL 4),
36 2.6.16.46-0.14 (SLES 10),
37 2.6.18-8.1.8.el5 (RHEL 5),
38 2.6.18.8 vanilla (kernel.org)
39 * Client support for unpatched kernels:
40 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
41 2.6.16 - 2.6.22 vanilla (kernel.org)
42 * Due to recently discovered recovery problems, we do not recommend
43 using patchless RHEL 4 clients with this or any earlier release.
44 * Recommended e2fsprogs version: 1.40.2-cfs1
45 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
47 Severity : enhancement
49 Description: Adaptive timeouts
50 Details : RPC timeouts adapt to changing server load and network
51 conditions to reduce resend attempts and improve recovery time.
54 Frequency : Heavily loaded servers
56 Description: obd_health_check_timeout too short
57 Details : set obd_health_check_timeout as 1.5x of obd_timeout
59 Severity : enhancement
61 Description: llapi_file_create() does not allow some changes
62 Details : add llapi_file_open() that allows specifying the file creation
63 mode and open flags, and also returns an open file handle.
65 Severity : enhancement
67 Description: df doesn't work properly if diskfs blocksize != 4K
68 Details : Choose biggest blocksize of OST's as the LOV's blocksize.
71 Frequency : always on directories with default striping set
73 Description: lfs find on -1 stripe looping in lsm_lmm_verify_common()
74 Details : Avoid lov_verify_lmm_common() on directory with -1 stripe count.
76 Severity : enhancement
78 Description: merge and cleanup kernel patches.
79 Details : Remove mnt_lustre_list in vfs_intent-2.6-rhel4.patch.
81 Severity : enhancement
83 Description: Add journal checksum support.(Kernel part)
84 Details : The journal checksum feature adds two new flags i.e
85 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT and
86 JBD2_FEATURE_COMPAT_CHECKSUM. JBD2_FEATURE_CHECKSUM flag
87 indicates that the commit block contains the checksum for
88 the blocks described by the descriptor blocks. Now commit
89 record can be sent to disk without waiting for descriptor
90 blocks to be written to disk. This behavior is controlled
91 using JBD2_FEATURE_ASYNC_COMMIT flag.
93 Severity : enhancement
95 Description: RedHat Update kernel for RHEL5
96 Details : Kernel config file for RHEL5.
98 Severity : enhancement
100 Description: OSS needs mutliple precreate threads
101 Details : Add ability to start more than one create thread per OSS.
104 Frequency : only with quota on the root user
106 Description: mds_obd_create error creating tmp object
107 Details : When the user sets quota on root, llog will be affected and can't
108 create files and write files.
111 Frequency : Always on ia64 patchless client, and possibly others.
113 Description: Add EXPORT_SYMBOL check for node_to_cpumask symbol.
114 Details : This allows the patchless client to be loaded on architectures
120 Description: disorder of journal start and llog_add cause deadlock.
121 Details : in llog_origin_connect, journal start should happen before
122 llog_add keep the same order as other functions to avoid
125 Severity : enhancement
127 Description: RedHat Update kernel for RHEL5
128 Details : Modify the kernel config file more closer RHEL5.
130 Severity : enhancement
132 Description: Build failure against Centos5 (RHEL5)
133 Details : Define PAGE_SIZE when it isn't present.
136 Frequency : occasionally when using NFS
138 Description: "ll_intent_file_open()) lock enqueue: err: -13" with nfs
139 Details : with NFS, the anon dentry's parent was set to itself in
140 d_alloc_anon(), so in MDS, we use rec->ur_fid1 to find the
141 corresponding dentry other than use rec->ur_name.
143 Severity : enhancement
145 Description: enable data checksumming by default
146 Details : enable checksum by default, allow --disable-checksum
147 configure option and "-o nochecksum" mount option. Checksums
148 can also be disabled at runtime via $LPROC/llite/*/checksum_pages.
151 Frequency : Occasionally with failover
153 Description: Client eviction due to failover config
154 Details : after a connection loss, the lustre client should attempt to
155 reconnect to the last active server first before trying the
156 other potential connections.
158 Severity : enhancement
160 Description: eliminate client locks in face of contention
161 Details : file contention detection and lockless i/o implementation
164 Severity : enhancement
166 Description: client-side metadata stat-ahead during readdir(directory readahead)
167 Details : perform client-side metadata stat-ahead when the client detects
168 readdir and sequential stat of dir entries therein
173 Description: when mds and osts use different quota unit(32bit and 64bit),
174 quota will be released repeatly.
175 Details : void sending multiple quota reqs to mds, which will keep the
176 status between the reqs.
179 Frequency : only with liblustre clients on XT3
181 Description: evictions taking too long
182 Details : allow llrd to evict clients directly on OSTs
186 Description: osts not allocated evenly to files
187 Details : change the condition to increase offset_idx
189 Severity : enhancement
191 Description: self-adjustable client's lru lists
192 Details : use adaptive algorithm for managing client cached locks lru
193 lists according to current server load, other client's work
194 pattern, memory activities, etc. Both, server and client
195 side namespaces provide number of proc tunables for controlling
200 Description: rewrite ext2-derived code in llite/dir.c and obdclass/uuid.c
201 Details : rewrite inherited code (uuid parsing code from ext2 utils and
202 readdir code from ext3) from scratch preserving functionality.
206 Description: Only those disconnect error should be returned by rq_status.
207 Details : In open/enqueue processs, Some errors, which will cause client
208 disconnected, should be returned by rq_status, while other
209 errors should still be returned by intent, then mdc or llite will
213 Frequency : Always for filesystems larger than 2TB on 32-bit systems.
215 Description: Data corruption for OSTs that are formatted larger than 2TB
217 Details : When generating the bio request for lustre file writes the
218 sector number would overflow a temporary variable before being
219 used for the IO. The data reads correctly from Lustre (which
220 will overflow in a similar manner) but other file data or
221 filesystem metadata may be corrupted in some cases.
223 --------------------------------------------------------------------------------
225 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
227 * Support for kernels:
228 2.6.5-7.286 (SLES 9),
229 2.6.9-55.0.2.EL (RHEL 4),
230 2.6.16.46-0.14 (SLES 10),
231 2.6.18-8.1.8.el5 (RHEL 5),
232 2.6.18.8 vanilla (kernel.org)
233 * Client support for unpatched kernels:
234 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
235 2.6.16 - 2.6.22 vanilla (kernel.org)
236 * Due to recently discovered recovery problems, we do not recommend
237 using patchless RHEL 4 clients with this or any earlier release.
238 * Recommended e2fsprogs version: 1.39.cfs8
239 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
244 Description: block reactivating mgc import until all deactivates complete
245 Details : Fix race when failing back MDT/MGS to itself (testing)
247 Severity : enhancement
249 Description: lfs setstripe enhancement
250 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
253 Frequency : mds/oss recovery
255 Description: llog ctxt is refrenced after it has been freed.
256 Details : llog ctxt refcount was added to avoide the race between ctxt free
257 and llog recovery process. Each llog user must hold ctxt refcount
258 before it access the llog. And the llog ctxt can only be freed
259 when its refcount is zero.
261 Severity : enhancement
263 Description: randomly memory allocation failure util
264 Details : Make lustre randomly failed allocating memory for testing purpose.
267 Frequency : only for SLES10
269 Description: Update kernel patch for SLES10 SP1
270 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
272 Severity : enhancement
274 Description: omit set fsid for export NFS
275 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
276 as FS_REQUIRES_DEV for avoid problems with generate fsid.
279 Frequency : after network failures
281 Description: Add sync option to mount_lustre.c
282 Details : Client loses data written to lustre after a network interruption.
284 Severity : enhancement
286 Description: Error message improvement.
287 Details : Merging of two LCONSOLE_ERROR_MSG into one.
289 Severity : enhancement
291 Description: don't use GFP_* in generic Lustre code.
292 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
294 Severity : enhancement
296 Description: obdclass is limited by single OBD_ALLOC(idarray)
297 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
299 Severity : enhancement
301 Description: updated patchess for new RHEL4 kernel
302 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
304 Severity : enhancement
306 Description: warnings with build patchless client with vanila 2.6.19 and up
307 Details : change old ctl_table style and replace ctl_table/ctl_table_header
308 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
310 Severity : enhancement
312 Description: O_DIRECT bypasses client statistics.
313 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
314 read_ahead_stats were updated. Copied stats section from
315 osc_send_oap_rpc() into async_internal().
318 Frequency : only for Cray XT3
320 Description: peer credits not enough on many OST per OSS systems.
321 Details : Use new lnet way to add credits as we need those for pings and ASTs
324 Frequency : only with liblustre
326 Description: Liblustre is not releasing flock locks on file close.
327 Details : Release flock locks on file close.
330 Frequency : only for RHEL4
332 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
333 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
334 Update target file and kernel config.
338 Description: lustre support for RHEL5
339 Details : Add support for RHEL5.
344 Description: ASSERTION(export != NULL) failed in target_handle_connect
345 Details : Assetion hit is result of rare race between disconnect and connet
346 to same nid. target_handle_connect found old connect cockie and
347 tried to reconnect, but can't find export for this cockie.
352 Description: umount blocks forever on error
353 Details : In result of wrong using obd_no_recov and obd_force flags client
354 can hand if cancel or some other requests is lost.
357 Frequency : Only for SLES
359 Description: sanity_quota fail test_1
360 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
361 file, which makes TSTID[2] inunique.
363 Severity : enhancement
365 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
366 Details : Update target/ChangeLog/which_patch .
368 Severity : enhancement
370 Description: jbd statistics
371 Details : Port older jbd statistics patch for sles10
376 Description: lvbo_init failed for resource with missing objects.
377 Details : Fix returning error if we do stat for file with missing/corrupted
378 objects and i_size set to all sum of size of all avaible objects.
379 if we to truncate/write to missing object - it is recreated.
384 Description: Oops in read and write path when failing to allocate lock.
385 Details : Check if lock allocation failed and return error back.
388 Frequency : When flocks are used.
390 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
391 Details : Flock locks might destroy just granted lock if it could be merged
392 with another existing flock, this is done in completion handler,
393 so teach ldlm_cli_enquque_fini that this is a valid case for
399 Description: reply_lock_interpret crash due to race with it and lock cancel.
400 Details : Do not replay locks that are being cancelled. Do not reference
401 locks by their address during replay, just by their handle.
404 Frequency : only with deactivated OSTs
406 Description: lstripe command fails for valid OST index
407 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
408 instead of lov->desc.ld_active_tgt_count.
410 Severity : enhancement
412 Description: Build failure against Centos5 (RHEL5)
413 Details : Use getpagesize() instead of PAGE_SIZE.
415 --------------------------------------------------------------------------------
417 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
419 * Support for kernels:
420 2.6.5-7.283 (SLES 9),
421 2.6.9-55.EL (RHEL 4),
422 2.6.16.46-0.14 (SLES 10),
423 2.6.18.8 vanilla (kernel.org)
424 * Client support for unpatched kernels:
425 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
426 2.6.16 - 2.6.22 vanilla (kernel.org)
427 * Due to recently discovered recovery problems, we do not recommend
428 using patchless RHEL 4 clients with this or any earlier release.
429 * Recommended e2fsprogs version: 1.39.cfs8
430 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
431 * Starting with this release, the ldiskfs backing filesystem required
432 by Lustre is now in its own package, lustre-ldiskfs. This package
433 should be installed. It is versioned separately from Lustre and
434 may be released separately in future.
439 Description: server LBUG when shutting down
440 Details : Block umount forever until the mount refcount is zero rather
441 than giving up after an arbitrary timeout.
443 Severity : enhancement
445 Description: add optional extra BUILD_VERSION info
446 Details : add a new environment variable (namely LUSTRE_VERS) which allows
447 to override the lustre version.
450 Frequency : 2.6.18 servers only
452 Description: ll_kern_mount() doesn't release the module reference
453 Details : The ldiskfs module reference count never drops down to 0
454 because ll_kern_mount() doesn't release the module reference.
459 Description: server LBUG when using old ost_num_threads parameter
460 Details : Accept the old ost_num_threads parameter but warn that it
461 is deprecated, and fix an off-by-one error that caused an LBUG.
466 Description: Transient SCSI error results in persistent IO issue
467 Details : iobuf->dr_error is not reinitialized to 0 between two
471 Frequency : sometimes when underlying device returns I/O errors
473 Description: OSTs not going read-only during write failures
474 Details : OSTs are not remounted read-only when the journal commit threads
475 get I/O errors because fsfilt_ext3 calls journal_start/stop()
476 instead of the ext3 wrappers.
479 Frequency : only with 10000 clients or more
481 Description: poor connect scaling with increasing client count
482 Details : Don't run filter_grant_sanity_check for more than 100 exports
483 to improve scaling for large numbers of clients.
486 Frequency : SLES10 only
488 Description: sanity-quota.sh quotacheck failed: rc = -22
489 Details : Quotas cannot be enabled on SLES10.
492 Frequency : liblustre clients only
494 Description: getdirentries does not give error when run on compute nodes
495 Details : getdirentries does not fail when the size specified as an argument
496 is too small to contain at least one entry
498 Severity : enhancement
500 Description: Add LNET router traceability for debug purposes
501 Details : If a checksum failure occurs with a router as part of the
502 IO path, the NID of the last router that forwarded the bulk data
503 is printed so it can be identified.
508 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
509 Details : Due to a race condition, liblustre clients were occasionally
512 Severity : enhancement
514 Description: lfs setstripe use optional parameters instead of postional
517 Severity : enhancement
519 Description: Nanosecond timestamp support for ldiskfs
520 Details : The on-disk ldiskfs filesystem has added support for nanosecond
521 resolution timestamps. There is not yet support for this at
522 the Lustre filesystem level.
525 Frequency : during server recovery
527 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
528 Details : request with rq_no_resend flag not awake l_wait_event if they get a
532 Frequency : nfs export on patchless client
534 Description: connectathon hang when test nfs export over patchless client
535 Details : Disconnected dentry cannot be found with lookup, so we do not need
536 to unhash it or make it invalid
539 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
540 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
541 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
546 Description: open req refcounting wrong on reconnect
547 Details : If reconnect happened between getting open reply from server and
548 call to mdc_set_replay_data in ll_file_open, we will schedule
549 replay for unreferenced request that we are about to free.
550 Subsequent close will crash in variety of ways.
551 Check that request is still eligible for replay in
552 mdc_set_replay_data().
557 Description: disable writes to filesystem when reading health_check file
558 Details : the default for reading the health_check proc file has changed
559 to NOT do a journal transaction and write to disk, because this
560 can cause reads of the /proc file to hang and block HA state
561 checking on a healthy but otherwise heavily loaded system. It
562 is possible to return to the previous behaviour during configure
563 with --enable-health-write.
565 Severity : enhancement
567 Description: 64-bit inode version
568 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
569 to the inode. This will be required for version-based recovery.
574 Description: MDS fails to start if a duplicate client export is detected
575 Details : in some rare cases it was possible for a client to connect to
576 an MDS multiple times. Upon recovery the MDS would detect this
577 and fail during startup. Handle this more gracefully.
579 Severity : enhancement
581 Description: Add -o localflock option to simulate old noflock behaviour.
582 Details : This will achieve local-only flock/fcntl locks coherentness.
587 Description: log_commit_thread vs filter_destroy race leads to crash
588 Details : Take import reference before releasing llog record semaphore
593 Description: Wrong request locking in request set processing
594 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
595 imp_delayed_list, in this place should be used imp_lock.
598 Frequency : when reconnecting
600 Description: Grant leak when OSC reconnect to OST
601 Details : When osc reconnect ost, OST(filter) should check whether it
602 should grant more space to client by comparing fed_grant and
603 cl_avail_grant, and return the granted space to client instead
604 of "new granted" space, because client will call osc_init_grant
605 to update the client grant space info.
608 Frequency : when client reconnects to OST
610 Description: Grant leak when OSC does a resend and replays bulk write
611 Details : When osc reconnect to OST, OST(filter) should clear grant info of
612 bulk write request, because the grant info while be sync between
613 OSC and OST when reconnect, and we should ignore the grant info
614 these of resend/replay write req.
619 Description: Grant space more than avaiable space sometimes.
620 Details : When then OST is about to be full, if two bulk writing from
621 different clients came to OST. Accord the avaliable space of the
622 OST, the first req should be permitted, and the second one
623 should be denied by ENOSPC. But if the seconde arrived before
624 the first one is commited. The OST might wrongly permit second
625 writing, which will cause grant space > avaiable space.
628 Frequency : when client is evicted
630 Description: Grant might be wrongly erased when osc is evicted by OST
631 Details : when the import is evicted by server, it will fork another
632 thread ptlrpc_invalidate_import_thread to invalidate the
633 import, where the grant will be set to 0. While the original
634 thread will update the grant it got when connecting. So if
635 the former happened latter, the grant will be wrongly errased
636 because of this race.
641 Description: Checking Stale with correct fid
642 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
643 to check whether it is stale, because sometimes, we want the
644 enqueue happened anyway, and op_data.fid2 will not be initialized.
646 Severity : enhancement
648 Description: update patchless client
649 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
652 Frequency : only with 2.4 kernel
654 Description: random memory corruption
655 Details : size of struct ll_inode_info is to big for union inode.u and this
656 can be cause of random memory corruption.
661 Description: Memory leak in recovery
662 Details : Lov_mds_md was not free in an error handler in mds_create_object.
663 It should also check obd_fail before fsfilt_start, otherwise if
664 fsfilt_start return -EROFS,(failover mds during mds recovery).
665 then the req will return with repmsg->transno = 0 and rc = EROFS.
666 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
667 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
668 be freed no matter whether fsfilt_commit success or not.
671 Frequency : only with huge numbers of clients
673 Description: Prevents from taking the superblock lock in llap_from_page for
675 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
676 for llap_from_page call in ll_removepage() prevents from taking
677 the superblock lock for a soon died page.
682 Description: Not check open intent error before release open handle
683 Details : in some rare cases, the open intent error is not checked before
684 release open handle, which may cause
685 ASSERTION(open_req->rq_transno != 0), because it tries to release
686 the failed open handle.
691 Description: Set cat log bitmap only after create log success.
692 Details : in some rare cases, the cat log bitmap is set too early. and it
693 should be set only after create log success.
697 Description: Accessing a block bevice can re-enable I/O when Lustre is
698 tearing down a device.
699 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
704 Description: service threads may hog cpus when there are a lot of requests
705 Details : Insert cond_resched to give other threads a chance to use some CPU
710 Description: the cat log was not initialized in recovery
711 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
712 unlink log on mds will not be initialized until mds post
713 recovery. And also in mds post recovery, the unlink log will
714 initialization will be done asynchronausly, so there will be race
715 between add unlink log and unlink log initialization.
719 Description: brw_stats were being printed incorrectly
720 Details : brw_stats were being printed as log2 but all of them were not
721 recorded as log2. Also remove some code duplication arising from
722 filter_tally_{read,write}.
726 Frequency : rare, only in recovery.
727 Description: ASSERTION(req->rq_type != LI_POISON) failed
728 Details : imp_lock should be held while iterating over imp_sending_list for
729 prevent destroy request after get timeout in ptlrpc_queue_wait.
733 Description: replay-single.sh test 52 fails
734 Details : A lock's skiplist need to be cleanup when it being unlinked
735 from its resource list.
740 Description: Short directio read returns full requested size rather than
742 Details : Direct I/O operations should return actual amount of bytes
743 transferred rather than requested size.
745 Severity : enhancement
747 Description: metadata RPC reduction (e.g. for rm performance)
748 Details : decrease the amount of synchronous RPC between clients and servers
749 by canceling conflicing lock before the operation on the client
750 and packing thier handles into the main operation RPC to server.
752 Severity : enhancement
754 Description: add #ifdef HAVE_KERNEL_CONFIG_H
755 Details : kernels from 2.6.19 not need include linux/config.h, but add
756 include linux/autoconf.h in commpiler command line.
758 Severity : enhancement
760 Description: patchless client support for 2.6.22 kernel
761 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
762 removed. In this case we need drop using os depended interface to
763 kmem_cache and use cfs_mem_cache API.
768 Description: fix mal-formatted messages
769 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
773 Frequency : always in liblustre
774 Description: wrong IS_ERR implementation in liblustre.h
775 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
780 Description: Correct condition for output debug message.
781 Details : inode i_nlink equal zero is not enough for output message about
782 disk corruption, i_ctime and i_mode should be also checked.
786 Frequency : always in patchless client
787 Description: add configure check for truncate_complete_page
788 Details : improve checks for exported symbols. This allow run check without
789 sources, but with Module.symvers shipped with kernel distribution.
790 add check for truncate_complete_page used by patchless client.
794 Description: sanity.sh test_77h fails with "test_77h file compare failed"
795 Details : test_77h uses a file which was messed by other test case.
799 Description: Not Check whether lov_tgts is NULL in some lov functions
800 Details : Checking whether lov_tgts is NULL in some functions.
804 Description: replace obdo_alloc() with OBDO_ALLOC macro
805 Details : nothing special is done in obdo_alloc() function, and for
806 debugging purpose, it needs to be replaced with macros.
810 Description: bad return value and errno from fcntl call
811 Details : In liblustre API, errno should be a negative value if error
816 Description: ptlrpc_check_set() LBUG
817 Details : In case of positive reply from server and failed client bulk
818 callback after bulk transfer shouldn't LBUG, but process this
819 request as erroneous.
821 Severity : enhancement
823 Description: VFS operations stats tool.
824 Details : Tool which collects stats by tracking value written in pid,
825 ppid, gid and uses llstat to generate output to plot graph using
827 Updated lustre/utils/Makefile.am
828 Added lustre/utils/ltrack_stats.c
830 Severity : enhancement
832 Description: 2.6.18 server support (lustre 1.6.1)
833 Details : Support for 2.6.18 kernels on the server side.
838 Description: ASSERTION(imp->imp_conn_current) failed
839 Details : an assertion failure is hit if a client node boots and attempts to
840 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
843 Frequency : only for i686
845 Description: 1.4.11 RC1 build fails for RHEL 4, i686
846 Details : Fixed config variable for build.
851 Description: Updated patchess for new RHEL4 kernel
852 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
853 Updated patch jbd-stats-2.6.9.patch
854 Updated patch qsnet-rhel4-2.6.patch
855 Updated patch quota-deadlock-on-pagelock-core.patch
856 Updated patch vfs_intent-2.6-rhel4.patch
857 Updated patch vfs_races-2.6-rhel4.patch
858 Updated series file 2.6-rhel4-titech.series
859 Updated series file 2.6-rhel4.series
860 Updated kernel config files
865 Description: lquota slave complains LBUG when reconnecting with mds
867 Details : quota slave depends qctxt->lqc_import to send its quota request.
868 This pointer will be invalid if mds did failover or broke its
869 connect to osts, which leads to LBUG.
872 Frequency : when qunit size is too small(less than 20M)
874 Description: write is stopped by improper -EDQUOT
875 Details : If the master is busy and qunit size is small enough(let's say 1M),
876 the slave can not get quota from master on time, which will lead
877 slave to trigger a -EQUOTA to client.
882 Description: Deadlock during metadata tests
883 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
884 called with the per-dentry lock held.
887 Frequency : SLES9 only
889 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
890 Details : set CONFIG_NR_CPUS to 128 instead of 8.
892 Severity : enhancement
894 Description: remove fs_prep_san_write operation and related patches
895 Details : remove the ext3-san-jdike patches which are no longer useful.
900 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
901 Details : a disk corruption can cause the mballoc code to assert on a
902 double free or other extent corruptions. Handle these with
903 ext3_error() instead of assertions.
906 Frequency : only with mballoc3 code and deep extent trees
908 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
909 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
910 can cause the filesystem to be remounted read-only.
915 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
916 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
917 few external users of this ioctl number is lfsck's call to
918 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
919 Add the old ioctl number to the handler so both old and new
924 Description: parallel lock callbacks
925 Details : Instead of sending blocking and completion callbacks as separated
926 requests, adding them to a set and sending in parallel.
930 Description: Disable most debugging by default
931 Details : To improve performance, disable most logging (for debug purposes)
932 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
933 default, and HA includes fewer messages.
937 Frequency : only run on patchless client.
938 Description: use do_facet on sanity.sh for test handling recoverables errors
939 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
943 Description: hash tables for lists of nids, connections and uuids
944 Details : Hash tables noticeably help when a lot of clients connect to a
945 server, to faster identify duplicate connections or reconnects,
946 also to faster find export to evict in manual eviction case.
950 Description: Sometimes, when the server evict a client, and the client will
951 not be evicted as soon as possible.
952 Details : In enqueue req, the error was returned by intent, instead of
953 rq_status which make ptlrpc layer not detect this error, and
954 does not evict the client. So enqueue error should be returned
958 Frequency : only at startup
960 Description: Delay client connections to MDT until fist MDT->OST connect
961 Details : If a client tried to create a new file before the MDT had
962 connected to any OSTs, the create would return EIO. Now
963 the client will simply block until the MDT connects to the
964 first OST and the create can succeed.
967 Frequency : only for SLES9
969 Description: Routinely utilize latest Quadrics drivers in CFS releases
970 Details : Update patch qsnet-suse-2.6.patch.
973 Frequency : only for sles10
975 Description: Update patches for SLES 10 SP1 kernel.
976 Details : Update the patch vfs_intent-2.6-sles10.patch.
981 Description: Routinely utilize latest Quadrics drivers in CFS releases
982 Details : Update patch qsnet-rhel4-2.6.patch.
985 Frequency : at statup only
987 Description: mds_lov_synchronize race leads to various problems
988 Details : simultaneous MDT->OST connections at startup can cause the
989 sync to abort, leaving the OSC in a bad state.
994 Description: Using wrong pointer in osc_brw_prep_request
995 Details : Access to array[-1] can produce panic if kernel compiled with
996 CONFIG_PAGE_ALLOC enabled
998 Severity : enhancement
1000 Description: Async OSC create to avoid the blocking unnecessarily.
1001 Details : If a OST has no remain object, system will block on the creating
1002 when need to create a new object on this OST. Now, ways use
1003 pre-created objects when available, instead of blocking on an
1004 empty osc while others are not empty. If we must block, we block
1005 for the shortest possible period of time.
1009 Frequency : only in recovery
1010 Description: Mark OST as early accessible if his start SYNC.
1011 Details : osc_precreate return flag early accessible if oscc marked as
1012 OSCC_FLAG_SYNC_IN_PROGRESS.
1017 Description: Sometimes precreate code can triger create object on wrong ost
1018 Details : Wrong protected or not not restored variables aftre precreate loop
1019 can produce creation object on wrong ost.
1022 Frequency : oss recovery
1024 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1026 Details : llog_commit_thread_count should be synced between llog_commit
1027 start and cleanup, so new llog_commit thread should not be started
1028 when llog_commit threads being stopped to avoid accessing some
1031 Severity : enhancement
1033 Description: Add printing inode info into message about error in writepage.
1035 --------------------------------------------------------------------------------
1037 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1042 Frequency : on some architectures
1044 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1045 Details : Uninitialized flags sometimes cause configuration commands to
1049 Frequency : patchless clients only
1051 Description: missing __iget() symbol export
1052 Details : The __iget() symbol export is missing. To avoid the need for
1053 this on patchless clients the deathrow inode reaper is turned
1054 off, and we depend on the VM to clean up old inodes. This
1055 dependency was during via the fix for bug 12181.
1060 Description: sanity.sh fail: test_52b
1061 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1062 incorrect inode's flags to client.
1064 --------------------------------------------------------------------------------
1066 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1068 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1069 INTEROPERATE with older versions automatically. In many cases a
1070 special upgrade step is needed. Please read the
1071 user documentation before upgrading any part of a 1.4.x system.
1072 * WARNING: Lustre configuration and startup changes are required with
1073 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1075 * Support for kernels:
1076 2.4.21-47.0.1.EL (RHEL 3),
1077 2.6.5-7.283 (SLES 9),
1078 2.6.9-42.0.10.EL (RHEL 4),
1079 2.6.12.6 vanilla (kernel.org),
1080 2.6.16.27-0.9 (SLES10)
1081 * Client support for unpatched kernels:
1082 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1083 2.6.16 - 2.6.19 vanilla (kernel.org),
1084 2.6.9-42.0.8EL (RHEL 4)
1085 * Recommended e2fsprogs version: 1.39.cfs6
1086 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1089 Severity : enhancement
1091 Description: Async OSC create to avoid the blocking unnecessarily.
1092 Details : If a OST has no remain object, system will block on the creating
1093 when need to create a new object on this OST. Now, ways use
1094 pre-created objects when available, instead of blocking on an
1095 empty osc while others are not empty. If we must block, we block
1096 for the shortest possible period of time.
1098 Severity : enhancement
1100 Description: MountConf
1101 Details : Lustre configuration is now managed via mkfs and mount
1102 commands instead of lmc and lconf. New obd types (MGS, MGC)
1103 are added for dynamic configuration management. See
1104 https://mail.clusterfs.com/wikis/lustre/MountConf for
1107 Severity : enhancement
1109 Description: dynamic OST addition
1110 Details : OSTs can now be added to a live filesystem
1112 Severity : enhancement
1114 Description: startup order invariance
1115 Details : MDTs and OSTs can be started in any order. Clients only
1116 require the MDT to complete startup.
1118 Severity : enhancement
1120 Description: parallel, asynchronous orphan cleanup
1121 Details : orphan cleanup is now performed in separate threads for each
1122 OST, allowing parallel non-blocking operation.
1124 Severity : enhancement
1126 Description: optimized stripe assignment
1127 Details : stripe assignments are now made based on ost space available,
1128 ost previous usage, and OSS previous usage, in order to try
1129 to optimize storage space and networking resources.
1131 Severity : enhancement
1133 Description: Permanently set tunables
1134 Details : All writable /proc/fs/lustre tunables can now be permanently
1135 set on a per-server basis, at mkfs time or on a live
1138 Severity : enhancement
1140 Description: Lustre message v2
1141 Details : Add lustre message format v2.
1143 Severity : enhancement
1145 Description: client OST exclusion list
1146 Details : Clients can be started with a list of OSTs that should be
1147 declared "inactive" for known non-responsive OSTs.
1151 Description: ENOENT returned for valid filehandle during dbench.
1152 Details : Check if a directory has children when invalidating dentries
1153 associated with an inode during lock cancellation. This fixes
1154 an incorrect ENOENT sometimes seen for valid filehandles during
1155 testing with dbench.
1158 Frequency : SFS test only (otherwise harmless)
1160 Description: SPEC SFS validation failure on NFS v2 over lustre.
1161 Details : Changes the blocksize for regular files to be 2x RPC size,
1162 and not depend on stripe size.
1164 Severity : enhancement
1166 Description: fine-grained SMP locking inside DLM
1167 Details : Improve DLM performance on SMP systems by removing the single
1168 per-namespace lock and replace it with per-resource locks.
1170 Severity : enhancement
1172 Description: don't hold multiple extent locks at one time
1173 Details : To avoid client eviction during large writes, locks are not
1174 held on multiple stripes at one time or for very large writes.
1175 Otherwise, clients can block waiting for a lock on a failed OST
1176 while holding locks on other OSTs and be evicted.
1178 Severity : enhancement
1180 Description: Multiple MD RPCs in flight.
1181 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1182 To avoid overly-overloading MDT, introduce a limit on number of
1183 MDT RPCs in flight for a single client and add /proc controls
1184 to adjust this limit.
1186 Severity : enhancement
1188 Description: client read/write statistics
1189 Details : Add client read/write call usage stats for performance
1190 analysis of user processes.
1191 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1192 file access. extents_stats shows chunk size distribution.
1193 extents_stats_per_process show chunk size distribution per
1196 Severity : enhancement
1198 Description: per-client statistics on server
1199 Details : Add ldlm and operations statistics for each client in
1200 /proc/fs/lustre/mds|obdfilter/*/exports/
1202 Severity : enhancement
1204 Description: improved MDT statistics
1205 Details : Add detailed MDT operations statistics in
1206 /proc/fs/lustre/mds/*/stats
1208 Severity : enhancement
1210 Description: VFS operations stats
1211 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1212 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1217 Description: Fix client-side osc byte counters
1218 Details : The osc read/write byte counters in
1219 /proc/fs/lustre/osc/*/stats are now working
1222 Frequency : always as root on SLES
1224 Description: Failure of copying files with lustre special EAs.
1225 Details : Client side always return success for setxattr call for lustre
1226 special xattr (currently only "trusted.lov").
1231 Description: Refcount LNET uuids
1232 Details : The global LNET uuid list grew linearly with every startup;
1233 refcount repeated list entries instead of always adding to
1236 Severity : enhancement
1238 Description: Dynamic service threads
1239 Details : Within a small range, start extra service threads
1240 automatically when the request queue builds up.
1243 Frequency : mixed-endian client/server environments
1245 Description: mixed-endian crashes
1246 Details : The new msg_v2 system had some failures in mixed-endian
1249 Severity : enhancement
1251 Description: Easy OST removal
1252 Details : OSTs can be permanently deactivated with e.g. 'lctl
1253 conf_param lustre-OST0001.osc.active=0'
1255 Severity : enhancement
1257 Description: MGS proc entries
1258 Details : Added basic proc entries for the MGS showing what filesystems
1261 Severity : enhancement
1263 Description: provide MGS failover
1264 Details : Added config lock reacquisition after MGS server failover.
1266 Severity : enhancement
1268 Description: add Linux 2.4 support
1269 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1273 Description: a large application tries to do I/O to the same resource and dies
1274 in the middle of it.
1275 Details : Check the req->rq_arrival time after the call to
1276 ost_brw_lock_get(), but before we do anything about
1277 processing it & sending the BULK transfer request. This
1278 should help move old stale pending locks off the queue as
1279 quickly as obd_timeout.
1282 Frequency : when an incorrect nid is specified during startup
1284 Description: ptlrpc connect to non-existant node causes kernel crash
1285 Details : LNET can't be re-entered from an event callback, which
1286 happened when we expire a message after the export has been
1287 cleaned up. Instead, hand the zombie cleanup off to another
1290 Severity : enhancement
1292 Description: plain/inodebits lock performance improvement
1293 Details : Grouping plain/inodebits in granted list by their request modes
1294 and bits policy, thus improving the performance of search through
1298 Frequency : only if OST filesystem is corrupted
1300 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1301 Details : for a short time RPCs with bulk IO are in the replay list,
1302 but replay of bulk IOs is unimplemented. If the OST filesystem
1303 is corrupted due to disk cache incoherency and then replay is
1304 started it is possible to trip an assertion. Avoid putting
1305 committed RPCs into the replay list at all to avoid this issue.
1308 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1310 Description: System hang on startup
1311 Details : This bug allowed the liblustre (e.g. catamount) client to
1312 return to the app before handling all startup RPCs. This
1313 could leave the node unresponsive to lustre network traffic
1314 and manifested as a server ptllnd timeout.
1316 Severity : enhancement
1318 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1319 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1320 boolean to control whether to print peer debug info when a
1321 client's RPC times out.
1324 Frequency : only for kernels with patches from Lustre below 1.4.3
1326 Description: Remove old rdonly API
1327 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1330 Frequency : only for devices with external journals
1332 Description: Set external device read-only also
1333 Details : During a commanded failover stop, we set the disk device
1334 read-only while the server shuts down. We now also set any
1335 external journal device read-only at the same time.
1338 Frequency : when upgrading from 1.4 while trying to change parameters
1340 Description: The wrong (new) MDC name was used when setting parameters for
1341 upgraded MDT's. Also allows changing of OSC (and MDC)
1342 parameters if --writeconf is specified at tunefs upgrade time.
1345 Frequency : when setting specific ost indicies
1347 Description: QOS code breaks on skipped indicies
1348 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1349 created with --index need not be sequential.
1351 Severity : enhancement
1353 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1354 Details : The uninit_groups feature works in conjunction with the kernel
1355 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1356 up the pass1 processing of e2fsck. This is a read-only feature
1357 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1358 mount filesystems that have had this feature enabled.
1360 Severity : enhancement
1362 Description: Improve multi-block allocation algorithm to avoid fragmentation
1363 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1364 allocation locality and avoid filesystem fragmentation.
1366 ------------------------------------------------------------------------------
1368 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1370 * Support for kernels:
1371 2.4.21-47.0.1.EL (RHEL 3)
1372 2.6.5-7.283 (SLES 9)
1373 2.6.9-42.0.10.EL (RHEL 4)
1374 2.6.12.6 vanilla (kernel.org)
1375 2.6.16.27-0.9 (SLES 10)
1376 * Recommended e2fsprogs version: 1.39.cfs5
1378 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1382 Frequency : occasional, depends on client load and configuration
1383 Bugzilla : 12181, 12203
1384 Description: data loss for recently-modified files
1386 Details : In some cases it is possible that recently written or created
1387 files may not be written to disk in a timely manner (this should
1388 normally be within 30s unless client IO load is very high).
1389 The problem appears as zero-length files or files that are a
1390 multiple of 1MB in size after a client crash or client eviction
1391 that are missing data at the end of the file.
1393 This problem is more likely to be hit on clients where files are
1394 repeatedly created and unlinked in the same directory, clients
1395 have a large amount of RAM, have many CPUs, the filesystem has
1396 many OSTs, the clients are rebooted frequently, and/or the files
1397 are not accessed by other nodes after being written.
1399 The presence of the problem can be detected by looking at
1400 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1401 smaller than the second (nr_unused) then dirty files will not
1402 be flushed automatically to disk. "sync; sleep 10" should be
1403 run several times on the node before unmounting it to update
1404 Lustre (this is also safe to run on nodes without this problem).
1406 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1407 that can cause this same problem, so customers using that kernel
1408 also need to update the kernel in addition to Lustre. In order
1409 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1411 It is normal that files written just before a client crash (less
1412 than 30s) may not yet have been flushed to disk, even for local
1416 Frequency : frequent on thin XT3 nodes
1418 Description: UUID collision on thin XT3 Linux nodes
1419 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1420 randomly, since we relied on an insufficiently-seeded PRNG.
1425 Description: OSS hangs after "All ost request buffers busy"
1426 Details : A deadlock between quota and journal operations caused OSS
1427 hangs after printing "All ost request buffers busy."
1430 Frequency : always on liblustre builds
1432 Description: Cleanup compiler warnings on liblustre
1435 Frequency : always on liblustre builds on XT3
1437 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1441 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1442 > 32000 subdirectories
1443 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1444 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1445 subdirectory count crosses 32000. This will aid e2fsck to
1446 correctly handle more than 32000 subdirectories.
1449 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1451 Description: System hang on startup
1452 Details : This bug allowed the liblustre (e.g. catamount) client to
1453 return to the app before handling all startup RPCs. This
1454 could leave the node unresponsive to lustre network traffic
1455 and manifested as a server ptllnd timeout.
1457 Severity : enhancement
1459 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1460 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1461 boolean to control whether to print peer debug info when a
1462 client's RPC times out.
1467 Description: make O_SYNC working on 2.6 kernels
1468 Details : 2.6 kernels use different method for mark pages for write,
1469 so need add a code to lustre for O_SYNC work.
1474 Description: Failure to close file and release space on NFS
1475 Details : Put inode details into lock acquired in ll_intent_file_open.
1476 Use mdc_intent_lock in ll_intent_open to properly
1477 detect all kind of errors unhandled by mdc_enqueue.
1482 Description: proc file read during shutdown sometimes raced obd removal,
1484 Details : Add lock to prevent obd access after proc file removal.
1487 Frequency : Only for files larger than 4GB on 32-bit clients.
1489 Description: improperly doing page alignment of locks
1490 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1491 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1496 Description: RPCs being resent when they shouldn't be
1497 Details : Some RPCs that should not be resent are being resent. This
1498 can cause inconsistencies in the RPC state machine. Do not
1499 resend such requests.
1502 Frequency : rare, only with NFS export
1504 Description: Crash on NFS re-export node
1505 Details : under very unusual load conditions an assertion is hit in
1506 ll_intent_file_open()
1509 Frequency : only if OST filesystem is corrupted
1511 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1512 Details : for a short time RPCs with bulk IO are in the replay list,
1513 but replay of bulk IOs is unimplemented. If the OST filesystem
1514 is corrupted due to disk cache incoherency and then replay is
1515 started it is possible to trip an assertion. Avoid putting
1516 committed RPCs into the replay list at all to avoid this issue.
1521 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1522 Details : Large single O_DIRECT read and write calls can fail to allocate
1523 a sufficiently large buffer to process the request. In case of
1524 allocation failure the allocation is retried with a smaller
1525 buffer and broken into smaller requests.
1527 Severity : enhancement
1529 Description: Add -o localflock option to simulate old noflock behaviour.
1530 Details : This will achieve local-only flock/fcntl locks coherentness.
1535 Description: versioning check is incomplete
1536 Details : Checking the version difference of client vs. server, report
1537 error if the gap is too big.
1542 Description: add support PG_writeback bit
1543 Details : add support for PG_writeback bit for Lustre, for more carefull
1544 work with page cache in 2.6 kernel. This also fix some deadlocks
1545 and remove hack for work O_SYNC with 2.6 kernel.
1547 Severity : enhancement
1549 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1550 Details : The uninit_groups feature works in conjunction with the kernel
1551 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1552 up the pass1 processing of e2fsck. This is a read-only feature
1553 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1554 mount filesystems that have had this feature enabled.
1556 Severity : enhancement
1558 Description: Improve multi-block allocation algorithm to avoid fragmentation
1559 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1560 allocation locality and avoid filesystem fragmentation.
1562 ------------------------------------------------------------------------------
1564 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1566 * Support for kernels:
1567 2.6.9-42.0.3.EL (RHEL 4)
1568 2.6.5-7.276 (SLES 9)
1569 2.4.21-47.0.1.EL (RHEL 3)
1570 2.6.12.6 vanilla (kernel.org)
1571 2.6.16.21-0.8 (SLES10)
1572 * Recommended e2fsprogs version: 1.39.cfs2-0
1574 * The backwards-compatible /proc/sys/portals symlink has been removed
1575 in this release. Before upgrading, please ensure that you change
1576 any configuration scripts or /etc/sysctl.conf files that access
1577 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1578 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1579 in advance of the upgrade on any system running Lustre 1.4.6 or
1580 newer, since /proc/sys/lnet was added in that version.
1581 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1585 Frequency : only when quota is used
1587 Description: avoid scanning export list for quota master
1588 Details : Change the algorithms to avoid scanning export list in order
1589 to improve the efficiency.
1592 Frequency : MDS failover only, very rarely
1594 Description: "went back in time" messages on mds failover
1595 Details : The greatest transno may be lost when the current operation
1596 finishes with an error (transno==0) and the client's last_rcvd
1597 record is over-written. Save the greatest transno in the
1598 mds_last_transno for this case.
1601 Frequency : always for specific kernels and striping counts
1603 Description: client may get "Matching packet too big" without ACL support
1604 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1605 when trying to access files in certain configurations. The
1606 clients should in fact be denied when mounting because they do
1607 not understand ACLs.
1610 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1612 Description: many clients connecting with IO in progress causes connect timeouts
1613 Details : Avoid synchronous journal commits to avoid delays caused by many
1614 clients connecting/disconnecting when bulk IO is in progress.
1615 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1616 OST_IO_PORTAL to avoid delays behind potentially many pending
1620 Frequency : occasionally with multiple writers to a single file
1622 Description: shared writes to file may result in wrong size reported by stat()
1623 Details : Allow growing of kms when extent lock is cancelled
1626 Frequency : always with random mmap IO to multi-striped file
1628 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1629 Details : Only if the hole is at the end of OST object so that kms is too
1630 small. Fix is to increase kms accordingly in ll_nopage.
1633 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1635 Description: writes to a missing object would leak memory on the OST
1636 Details : If there is an inconsistency between the MDS and OST filesystems,
1637 such that the MDS references an object that doesn't exist, writes
1638 to that object will leak memory due to incorrect cleanup in the
1639 error handling path, eventually running out of memory on the OST.
1644 Description: Creating too long symlink causes lustre errors
1645 Details : Check symlink and name lengths before sending requests to MDS.
1648 Frequency : only if flock is enabled (not on by default)
1650 Description: posix locks not released on fd closure on 2.6.9+
1651 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1652 kernels, this caused such locks not to be released on fd close and
1653 then assertions on fs unmount about still used locks.
1656 Frequency : MDS failover only, very rarely
1658 Description: clients may get ASSERTION(granted_lock != NULL)
1659 Details : When request was taking a long time, and a client was resending
1660 a getattr by name lock request. The were multiple lock requests
1661 with the same client lock handle and
1662 mds_getattr_name->fixup_handle_for_resent_request found one of the
1663 lock handles but later failed with ASSERTION(granted_lock != NULL).
1668 Description: handle->h_buffer_credits > 0, assertion failure
1669 Details : h_buffer_credits is zero after truncate, causing assertion
1670 failure. This patch extends the transaction or creates a new
1674 Frequency : NFS re-export or patchless client
1675 Bugzilla : 11179, 10796
1676 Description: Crash on NFS re-export node (__d_move)
1677 Details : We do not want to hash the dentry if we don't have a lock.
1678 But if this dentry is later used in d_move, we'd hit uninitialised
1679 list head d_hash, so we just do this to init d_hash field but
1680 leave dentry unhashed.
1683 Frequency : NFS re-export or patchless client
1685 Description: NFS exports has problem with symbolic link
1686 Details : lustre client didn't properly install dentry when re-exported
1687 to NFS or running patchless client.
1690 Frequency : NFS re-export or patchless client
1692 Description: Various nfs/patchless fixes.
1693 Details : fixes reuse disconected alias for lookup process - this fixes
1694 warning "find_exported_dentry: npd != pd",
1695 fix permission error with open files at nfs.
1696 fix apply umask when do revalidate.
1699 Frequency : occasional
1701 Description: Crash on NFS re-export node
1702 Details : calling clear_page() on the wrong pointer triggered oops in
1703 generic_mapping_read().
1706 Frequency : rarely, using O_DIRECT IO
1708 Description: unaligned directio crashes client with LASSERT
1709 Details : check for unaligned buffers before trying any requests.
1712 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1714 Description: stale data returned from RAID cache
1715 Details : If only a small amount of IO is done to the RAID device before
1716 reading it again it is possible to get stale data from the RAID
1717 cache instead of reading it from disk.
1720 Frequency : always for sles10 kernel
1722 Description: sles10 support
1723 Details : ll_follow_link: compile fixes and using of nd_set_link
1724 under newer kernels.
1727 Frequency : depends on arch, kernel and compiler version, always on sles10
1730 Description: recursive or deep enough symlinks cause stack overflow
1731 Details : getting rid of large stack-allocated variable in
1735 Frequency : depends on hardware
1737 Description: lustre write performance loss in the SLES10 kernel
1738 Details : the performance loss is caused by using of write barriers in the
1739 ext3 code. The SLES10 kernel turns barrier support on by
1740 default. The fix is to undo that change for ldiskfs.
1742 ------------------------------------------------------------------------------
1744 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1746 * Support for kernels:
1747 2.6.9-42.0.3EL (RHEL 4)
1748 2.6.5-7.276 (SLES 9)
1749 2.4.21-47.0.1.EL (RHEL 3)
1750 2.6.12.6 vanilla (kernel.org)
1754 Frequency : quota enabled and large files being deleted
1756 Description: releasing more than 4GB of quota at once hangs OST
1757 Details : If a user deletes more than 4GB of files on a single OST it
1758 will cause the OST to spin in an infinite loop. Release
1759 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1764 Description: statfs data retrieved from /proc may be stale or zero
1765 Details : When reading per-device statfs data from /proc, in the
1766 {kbytes,files}_{total,free,avail} files, it may appear
1767 as zero or be out of date.
1770 Frequency : systems with MD RAID1 external journal devices
1772 Description: lconf's call to blkid is confused by RAID1 journal devices
1773 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1774 of returning all block devices that match the journal UUID.
1777 Frequency : always, for aggregate stripe size over 4GB
1779 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1780 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1781 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1784 Frequency : always if "lfs find" used on a local file/directory
1786 Description: "lfs find" segfaults if used on a local file/directory
1787 Details : The case where a directory component was not specified wasn't
1788 handled correctly. Handle this properly.
1791 Frequency : always on ppc64
1793 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1794 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1795 perform endianness conversion.
1798 Frequency : rarely (truncate to non-zero file size after write under load)
1799 Bugzilla : 10730, 10687
1800 Description: Files padded with zeros to next 4K multiple
1801 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1802 it is possible that files that are truncated to a non-zero size
1803 immediately after being written are filled with zero bytes beyond
1804 the truncated size. No file data is lost.
1806 Severity : enhancement
1808 Description: Allow recovery/failover for liblustre clients.
1809 Details : liblustre clients were unaware of failover configurations until
1812 Severity : enhancement
1814 Description: user file locks should fail when not mounting with flock option
1815 Details : Set up an error-returning stub in ll_file_operations.lock field
1816 to prevent incorrect behaviour when client is mounted without
1817 flock option. Also, set up properly f_op->flock field for
1821 Frequency : always on ia64
1823 Description: "lfs df" loops on printing out MDS statfs information
1824 Details : The obd_ioctl_data was not initialized and in some systems
1825 this caused a failure during the ioctl that did not return
1826 an error. Initialize the struct and return an error on failure.
1829 Frequency : SLES 9 only
1831 Description: Error of copying files with lustre special EAs as root
1832 Details : Client side always return success for setxattr call for lustre
1833 special xattr (currently only "trusted.lov").
1836 Frequency : rarely on clusters with both ia64+i386 clients
1838 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1839 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1840 clients are doing concurrent writes to the same file it is
1841 possible that the ia64 clients may LASSERT because the OST
1842 extent locks are not PAGE_SIZE aligned. Ensure that grown
1843 locks are always aligned on the request boundary.
1846 Frequency : specific use, occasional
1848 Description: Overwriting in use executable truncates on-disk binary image
1849 Details : If one node attempts to overwrite an executable in use by
1850 another node, we now correctly return ETXTBSY instead of
1851 truncating the file.
1853 Severity : enhancement
1855 Description: Async OSC create to avoid the blocking unnecessarily.
1856 Details : If a OST has no remain object, system will block on the creating
1857 when need to create a new object on this OST. Now, ways use
1858 pre-created objects when available, instead of blocking on an
1859 empty osc while others are not empty. If we must block, we block
1860 for the shortest possible period of time.
1865 Description: chmod on Lustre root is propagated to other clients
1866 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1867 invalid by the follow_mount time.
1870 Frequency : liblustre clients only
1872 Description: Race in 'instant cancel' lock handling could lead to such locks
1873 never to be granted in case of SMP MDS
1874 Details : Do not destroy not yet granted but cbpending locks in
1878 Frequency : replay/resend of open
1880 Description: non null lock assetion failure in mds_intent_policy
1881 Details : Trying to replay/resend lockless open requests resulted in
1882 mds_open() returning 0 with no lock. Now it sets a flag if
1883 a lock is going to be returned.
1885 Severity : enhancement
1887 Description: Checksum enhancements
1888 Details : New checksum enhancements allow for resending RPCs that failed
1891 Severity : enhancement
1893 Description: Tunables on number of dirty pages in cacche
1894 Details : Allow to set limit on number of dirty pages cached.
1899 Description: client crash on unmount - lock still has references
1900 Details : In some error handling cases it was possible to leak a lock
1901 reference on a client while accessing a file. This was not
1902 harmful to the client during operation, but would cause the
1903 client to crash when the filesystem is unmounted.
1906 Frequency : specific case, rare
1908 Description: ETXTBSY on mds though file not in use by client
1909 Details : ETXTBSY is no longer incorrectly returned when attempting to
1910 chmod or chown a directory that the user previously tried to
1911 execute or a currently-executing binary.
1914 Frequency : extremely rare except on liblustre-based clients
1916 Description: Lustre space not freed when files are deleted
1917 Details : Clean up open-unlinked files after client eviction. Previously
1918 the unlink was skipped and the files remained as orphans.
1923 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1924 Details : In some cases it was possible to send a blocking callback to
1925 a client doing a glimpse, even though that client didn't get
1926 a lock granted. When the glimpse lock is cancelled on the OST
1927 the freed lock is left on the waiting list and corrupted the list.
1930 Frequency : all core dumps
1932 Description: Broke core dumps to lustre
1933 Details : Negative dentry may be unhashed if parent does not have UPDATE
1934 lock, but some callers, e.g. do_coredump, expect dentry to be
1935 hashed after successful create, hash it in ll_create_it.
1937 ------------------------------------------------------------------------------
1939 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1941 * Support for kernels:
1942 2.6.9-42.0.2.EL (RHEL 4)
1943 2.6.5-7.276 (SLES 9)
1944 2.4.21-40.EL (RHEL 3)
1945 2.6.12.6 vanilla (kernel.org)
1949 Frequency : always on RHEL 3
1951 Description: Number of open files grows over time
1952 Details : The number of open files grows over time, whether or not
1953 Lustre is started. This was due to a filp leak introduced
1954 by one of our kernel patches.
1956 ------------------------------------------------------------------------------
1958 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1960 * Support for kernels:
1961 2.6.9-42.EL (RHEL 4)
1962 2.6.5-7.267 (SLES 9)
1963 2.4.21-40.EL (RHEL 3)
1964 2.6.12.6 vanilla (kernel.org)
1969 Bugzilla : 5719, 9635, 9792, 9684
1970 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1971 Details : If a server is under heavy load and cannot reply to new
1972 connection requests before the client resends the (re)connect,
1973 the connection handling code can behave badly if two service
1974 threads are concurrently handing separate (re)connections from
1975 the same client. Add better locking to the connection handling
1976 code, and ensure that only a single connection will be processed
1977 for a given client UUID, even if the lock is dropped.
1979 Severity : enhancement
1981 Description: add TCP zero-copy support to kernel
1982 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1983 sends if the hardware supports scatter-gather and checksumming.
1984 This allows socklnd to do client-write and server-read more
1985 efficiently and reduce CPU utilization from skbuf copying.
1988 Frequency : only if NFS exporting from client
1990 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1991 Details : If setting attributes on a file created under NFS that had
1992 never been opened it would be possible to oops the client
1993 if the file had no objects.
1997 Bugzilla : 9326, 10402, 10897
1998 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1999 Details : It is possible that ptlrpcd_wake() dereferences a freed async
2000 RPC. In rare cases the ptlrpcd thread alread processed the RPC
2001 before ptlrpcd_wake() was called and the request was freed.
2004 Frequency : always for liblustre
2006 Description: liblustre client does MDS+OSTs setattr RPC for each write
2007 Details : When doing a write from a liblustre client, the client
2008 incorrectly issued an RPC to the MDS and each OST the file was
2009 striped over in order to update the timestamps. When writing
2010 with small chunks and many clients this could overwhelm the MDS
2011 with RPCs. In all cases it would slow down the write because
2012 these RPCs are unnecessary.
2014 Severity : enhancement
2016 Description: allow number of MDS service threads to be changed at module load
2017 Details : It is now possible to change the number of MDS service threads
2018 running. Adding "options mds mds_num_threads={N}" to the MDS's
2019 /etc/modprobe.conf will set the number of threads for the next
2020 time Lustre is restarted (assuming the "mds" module is also
2021 reloaded at that time). The default number of threads will
2022 stay the same, 32 for most systems.
2027 Description: OST crash if filesystem is unformatted or corrupt
2028 Details : If an OST is started on a device that has never been formatted
2029 or if the filesystem is corrupt and cannot even mount then the
2030 error handling cleanup routines would dereference a NULL pointer.
2035 Description: NULL pointer deref in llap_from_page.
2036 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2037 incorrect) mapping if the page was truncated/reclaimed while it was
2038 searched for. Check for this condition and skip such pages when
2039 doing readahead. Introduce extra check to llap_from_page() to
2040 verify page->mapping->host is non-NULL (so page is not anonymous).
2043 Frequency : Sometimes when using sys_sendfile
2045 Description: "page not covered by a lock" warnings from ll_readpage
2046 Details : sendfile called ll_readpage without right page locks present.
2047 Now we introduced ll_file_sendfile that does necessary locking
2048 around call to generic_file_sendfile() much like we do in
2052 Frequency : with certain MDS communication failures at client mount time
2054 Description: NULL pointer deref after failed client mount
2055 Details : a client connection request may delayed by the network layer
2056 and not be sent until after the PTLRPC layer has timed out the
2057 request. If the client fails the mount immediately it will try
2058 to clean up before the network times out the request. Add a
2059 reference from the request import to the obd device and delay
2060 the cleanup until the network drops the request.
2063 Frequency : occasionally during client (re)connect
2065 Description: assertion failure during client (re)connect
2066 Details : processing a client connection request may be delayed by the
2067 client or server longer than the client connect timeout. This
2068 causes the client to resend the connection request. If the
2069 original connection request is replied in this interval, the
2070 client may trip an assertion failure in ptlrpc_connect_interpret()
2071 which thought it would be the only running connect process.
2074 Frequency : only with obd_echo servers and clients that are rebooted
2076 Description: kernel BUG accessing uninitialized data structure
2077 Details : When running an obd_echo server it did not start the ping_evictor
2078 thread, and when a client was evicted an uninitialized data
2079 structure was accessed. Start the ping_evictor in the RPC
2080 service startup instead of the OBD startup.
2082 Severity : enhancement
2083 Bugzilla : 10193 (patchless)
2084 Description: Remove dependency on various unexported kernel interfaces.
2085 Details : No longer need reparent_to_init, exit_mm, exit_files,
2086 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2089 Frequency : rare (only users of deprecated and unsupported LDAP config)
2091 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2092 Details : LDAP apparently contains 'lustreName' attributes instead of
2093 'name'. A simple remapping of the name is sufficient.
2096 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2098 Description: waiting_locks_callback trips kernel BUG if client is evicted
2099 Details : Running with the dump_on_timeout debug flag turned on makes
2100 it possible that the waiting_locks_callback() can try to dump
2101 the Lustre kernel debug logs from an interrupt handler. Defer
2102 this log dumping to the expired_lock_main() thread.
2104 Severity : enhancement
2106 Description: Support NFS exporting on 2.6 kernels.
2107 Details : Implement non-rawops metadata methods for NFS server to use without
2108 changing NFS server code.
2111 Frequency : very rare (synthetic metadata workload only)
2113 Description: two racing renames might cause an MDS thread to deadlock
2114 Details : Running the "racer" program may cause one MDS thread to rename
2115 a file from being the source of a rename to being the target of
2116 a rename at exactly the same time that another thread is doing
2117 so, and the second thread has already enqueued these locks after
2118 doing a lookup of the target and is trying to relock them in
2119 order. Ensure that we don't try to re-lock the same resource.
2122 Frequency : only very large systems with liblustre clients
2124 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2125 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2126 all OSTs in parallel. This allows the eviction of stale liblustre
2127 clients to proceed much faster than if they were done in series,
2128 and also offers similar improvements for other set_info RPCs.
2133 Description: excessive CPU usage during initial read phase on client
2134 Details : During the initial read phase on a client, it would agressively
2135 retry readahead on the file, consuming too much CPU and impacting
2136 performance (since 1.4.5.8). Improve the readahead algorithm
2137 to avoid this, and also improve some other common cases (read
2138 of small files in particular, where "small" is files smaller than
2139 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2144 Description: MDS crash when receiving packet with unknown intent.
2145 Details : Do not LBUG in unknown intent case, just return -EFAULT
2147 Severity : enhancement
2148 Bugzilla : 9293, 9385
2149 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2150 Details : Do not serialize getattr (non-intent version) and statfs.
2153 Frequency : occasional, when OST network is overloaded/intermittent
2155 Description: client evicted by OST after bulk IO timeout
2156 Details : If a client sends a bulk IO request (read or write) the OST
2157 may evict the client if it is unresposive to its data GET/PUT
2158 request. This is incorrect if the network is overloaded (takes
2159 too long to transfer the RPC data) or dropped the OST GET/PUT
2160 request. There is no need to evict the client at all, since
2161 the pinger and/or lock callbacks will handle this, and the
2162 client can restart the bulk request.
2165 Frequency : Always when mmapping file with no objects
2167 Description: client crashes when mmapping file with no objects
2168 Details : Check that we actually have objects in a file before doing any
2169 operations on objects in ll_vm_open, ll_vm_close and
2175 Description: Request leak when working with deleted CWD
2176 Details : Introduce advanced request refcount tracking for requests
2177 referenced from lustre intent.
2179 Severity : Enhancement
2181 Description: Cache open file handles on client.
2182 Details : MDS now will return special lock along with openhandle, if
2183 requested and client is allowed to hold openhandle, even if unused,
2184 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2185 closing files for every read/write openration.
2187 Severity : Enhancement
2189 Description: Cache open negative dentries on client when possible.
2190 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2191 negative dentries on lock revocation.
2196 Description: Remounting a client read-only wasn't possible with a zconf mount
2197 Details : It wasn't possible to remount a client read-only with llmount.
2199 Severity : enhancement
2200 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2201 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2202 MPICH is included to add Lustre-specific ADIO interfaces.
2203 This is based closely on the UFS ADIO layer and only differs
2204 in file creation, in order to allow the OST striping to be set.
2205 This is user-contributed code and not supported by CFS.
2210 Description: extended inode attributes (immutable, append-only) work improperly
2211 when 2.4 and 2.6 kernels are used on client/server or vice versa
2212 Details : Introduce kernel-independent values for these flags.
2214 Severity : enhancement
2217 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2218 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2219 granularity. Now, Fractional MB granularity is be supported,
2220 this is very useful for low memory system.
2222 Severity : enhancement
2224 Description: Getattr by fid
2225 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2226 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2227 have locks and avoids getattr rpc requests for it on every stat.
2230 Frequency : Always, for filesystems larger than 2TB
2232 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2233 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2234 filesystems larger than 2TB to be mounted with mballoc enabled.
2237 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2239 Description: filesystem corruption for non-standard kernels and very large OSTs
2240 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2241 filesystems larger than 2TB is mounted then the kernel will
2242 silently corrupt the start of the filesystem. CONFIG_LBD is
2243 enabled for all CFS-supported kernels, but the possibility of
2244 this happening with a modified kernel config exists.
2246 Severity : enhancement
2248 Description: add client O_DIRECT support for 2.6 kernels
2249 Details : It is now possible to do O_DIRECT reads and writes to files
2250 in the Lustre client mountpoint on 2.6 kernel clients.
2252 Severity : enhancement
2254 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2255 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2256 parallel, not waiting for response from every OST before sending
2257 a rpc to the next OST.
2262 Description: setattr vs write race when updating file timestamps
2263 Details : Client processes that update a file timestamp into the past
2264 right after writing to the file (e.g. tar) it is possible that
2265 the updated file modification time can be reset to the current
2266 time due to a race between processing the setattr and write RPC.
2268 Severity : enhancement
2270 Description: Bring 'lfs find' closer in line with regular Linux find.
2271 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2272 -print0 options and obtains all the needed info through the lustre
2275 Severity : enhancement
2277 Description: support up to 1024 configured devices on one node
2278 Details : change obd_dev array from statically allocated to dynamically
2279 allocated structs as they are first used to reduce memory usage
2284 Description: Flush dirty partially truncated pages during truncate
2285 Details : Immediatelly flush partially truncated pages in filter_setattr,
2286 this way we completely avoid having any pages in page cache on OST
2287 and can retire ugly workarounds during writes to flush such pages.
2292 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2293 Details : protect inode from truncation within vfs_unlink() context
2294 just take a reference before calling vfs_unlink() and release it
2295 when parent's i_sem is free.
2298 Frequency : always, if extents are used on OSTs
2300 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2301 Details : OSTs using the extents format would not zero the high 16 bits of
2302 the index physical block number. This is not a problem for any
2303 OST filesystems smaller than 16TB, and no kernels support ext3
2304 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2305 new/modified files) and can be fixed for existing filesystems
2306 with e2fsprogs-1.39-cfs1.
2311 Description: import connection selection may be incorrect if timer wraps
2312 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2313 import connections to be ignored if the 32-bit jiffies counter
2314 wraps. Use a 64-bit jiffies counter.
2317 Frequency : during server recovery
2319 Description: crash after server is denying duplicate export
2320 Details : If clients are resending connect requests to the server, the
2321 server refuses to allow a client to connect multiple times.
2322 Fixed a bug in the handling of this case.
2325 Frequency : very large clusters immediately after boot
2327 Description: LNET request buffers exhausted under heavy short-term load
2328 Details : If a large number of client requests are generated on a service
2329 that has previously never seen so many requests it is possible
2330 that the request buffer growth cannot keep up with the spike in
2331 demand. Instead of dropping incoming requests, they are held in
2332 the LND until the RPC service can accept more requests.
2335 Frequency : Sometimes during replay
2337 Description: Assertion failure in ll_local_open after replay.
2338 Details : If replay happened on an open request reply before we were able
2339 to set replay handler, reply will become not swabbed tripping the
2340 assertion in ll_local_open. Now we set the handler right after
2341 recognising of open request
2344 Frequency : very rare
2346 Description: kernel reports "badness in vsnprintf"
2347 Details : Reading from the "recovery_status" /proc file in small chunks
2348 may cause a negative length in lprocfs_obd_rd_recovery_status()
2349 call to vsnprintf() (which is otherwise harmless). Exit early
2350 if there is no more space in the output buffer.
2352 Severity : enhancement
2354 Description: clear OBD RPC statistics by writing to them
2355 Details : It is now possible to clear the OBD RPC statistics by writing
2356 to the "stats" file.
2361 Description: Client mtime is not the same on different clients after utimes
2362 Details : In some cases, the client was using the utimes() syscall on
2363 a file cached on another node. The clients now validate the
2364 ctime from the MDS + OSTs to determine which one is right.
2369 Description: Inability to activate failout mode
2370 Details : lconf script incorrectly assumed that in python string's numeric
2371 value is used in comparisons.
2374 Frequency : always with multiple stripes per file
2376 Description: Inefficient object allocation for mutli-stripe files
2377 Details : When selecting which OSTs to stripe files over, for files with
2378 a stripe count that divides evenly into the number of OSTs,
2379 the MDS is always picking the same starting OST for each file.
2380 Return the OST selection heuristic to the original design.
2385 Description: mount failures may take full timeout to return an error
2386 Details : Under some heavy load conditions it is possible that a
2387 failed mount can wait for the full obd_timeout interval,
2388 possibly several minutes, before reporting an error.
2389 Instead return an error as soon as the status is known.
2391 ------------------------------------------------------------------------------
2393 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2395 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2396 INTEROPERATE with older versions automatically. Please read the
2397 user documentation before upgrading any part of a live system.
2398 * WARNING: Lustre networking configuration changes are required with
2399 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2402 * Support for kernels:
2403 2.6.9-22.0.2.EL (RHEL 4)
2404 2.6.5-7.244 (SLES 9)
2405 2.6.12.6 vanilla (kernel.org)
2408 Severity : enhancement
2409 Bugzilla : 7981/8208
2410 Description: Introduced Lustre Networking (LNET)
2411 Details : LNET is new networking infrastructure for Lustre, it includes
2412 a reorganized network configuration mode (see the user
2413 documentation for full details) as well as support for routing
2414 between different network fabrics. Lustre Networking Devices
2415 (LNDs) for the supported network fabrics have also been
2416 created for this new infrastructure.
2418 Severity : enhancement
2419 Description: Introduced Access control lists
2420 Details : clients can set ACLs on files and directories in order to have
2421 more fine-grained permissions than the standard Unix UGO+RWX.
2422 The MDS must be started with the "-o acl" mount option.
2424 Severity : enhancement
2425 Description: Introduced filesystem quotas
2426 Details : Administrators may now establish per-user quotas on the
2429 Severity : enhancement
2431 Description: Configuration change for the XT3
2432 The PTLLND is now used to run Lustre over Portals on the XT3
2433 The configure option(s) --with-cray-portals are no longer used.
2434 Rather --with-portals=<path-to-portals-includes> is used to
2435 enable building on the XT3. In addition to enable XT3 specific
2436 features the option --enable-cray-xt3 must be used.
2441 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2442 Details : A race between allocating a new llcd and re-getting the llcd_lock
2443 allowed another thread to grab newly-allocated llcd.
2445 Severity : enhancement
2447 Description: 2.6 OST async journal commit and locking fix to improve performance
2448 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2449 2.6 kernels are now async as they already were in 2.4 kernels so
2450 that they can commit concurrently with the network bulk transfer.
2451 For block-allocated files the filter allocation semaphore is held
2452 to avoid filesystem fragmentation during allocation. BKL lock
2453 removed for 2.6 xattr operations where it is no longer needed.
2458 Description: lconf incorrectly determined whether two IP networks could talk
2459 Details : In some more complicated routing and multiple-network
2460 configurations, lconf will avoid trying to make a network
2461 connection to a disjoint part of the IP space. It was doing the
2462 math incorrectly for one set of cases.
2467 Description: Fix for potential infinite loop processing records in an llog.
2468 Details : If an llog record is corrupted/zeroed, it is possible to loop
2469 forever in llog_process(). Validate the llog record length
2470 and skip the remainder of the block on error.
2473 Frequency : occasional (liblustre only)
2475 Description: liblustre could not open files whose last component is a symlink
2476 Details : sysio_path_walk() would incorrectly pass the open intent to
2477 intermediate path components.
2480 Frequency : rare (liblustre only with non-standard tuning)
2481 Bugzilla : 7201 (7350)
2482 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2483 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2484 after it has been released, only do this for referenced locks
2486 Severity : enhancement
2488 Description: specifying an (invalid) directory default stripe_size of -1
2489 would reset the directory default striping
2490 Details : stripe_size -1 was used internally to signal directory stripe
2491 removal, now use "all default" to signal dir stripe removal
2492 as a directory striping of "all default" is not useful
2495 Frequency : common for large clusters running liblustre clients
2497 Description: doing an ls when liblustre clients are running is slow
2498 Details : sending a glimpse AST to a liblustre client waits for every AST
2499 to time out, as liblustre clients will not respond. Since they
2500 cannot cache data we refresh the OST lock LVB from disk instead.
2502 Severity : enhancement
2504 Description: doing an ls at the same time as file IO can be slow
2505 Details : enqueue and other "small" requests can be blocked behind many
2506 large IO requests. Create a new OST IO portal for non-IO
2507 requests so they can be processed faster.
2510 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2512 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2513 created file with mode 000
2514 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2515 allow an ftruncate() as a non-root user to a file with mode 000.
2516 Lustre now respects this flag to disable mode checks when
2517 truncating a file owned by the user
2520 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2523 Description: Revoking locks from clients that went dead or catatonic might take
2525 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2526 cancellation of such locks instant on servers without waiting for
2527 any reply from clients. Clients drops these locks when cancel
2528 notification from server is received without replying.
2531 Frequency : liblustre-only, when liblustre client dies or becomes busy
2533 Description: Doing ls on Linux clients can take a long time with active
2535 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2536 granting such locks to it in the first place if possible. Locks
2537 are taken by proxy on the OST during the read or write and
2538 dropped immediately afterward. Add connect flags handling, do
2539 not grant locks to liblustre clients for glimpse ASTs.
2541 Severity : enhancement
2543 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2544 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2545 sequential reads and behaves badly with applications doing random
2546 reads. Improve it by reading ahead at least read region, and
2547 avoiding excessive large RPC for small reads.
2549 Severity : enhancement
2551 Description: Creating more than 1000 files for a single job may cause a load
2552 imbalance on the OSTs if there are also a large number of OSTs.
2553 Details : qos_prep_create() uses an OST index reseed value that is an
2554 even multiple of the number of available OSTs so that if the
2555 reseed happens in the middle of the object allocation it will
2556 still utilize the OSTs as uniformly as possible.
2561 Description: OST or MDS may oops in ping_evictor_main()
2562 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2563 but doesn't restart at beginning of obd_exports_timed list
2566 Severity : enhancement
2568 Description: improve by-nid export eviction on the MDS and OST
2569 Details : allow multiple exports with the same NID to be evicted at one
2570 time without re-searching the exports list.
2573 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2575 Description: MDS may oops in groups_free()
2576 Details : in rare race conditions a newly allocated group_info struct is
2577 freed again, and this can be NULL. The 2.4 compatibility code
2578 for groups_free() checked for a NULL pointer, but 2.6 did not.
2581 Frequency : common for liblustre clients doing little filesystem IO
2582 Bugzilla : 9352, 7313
2583 Description: server may evict liblustre clients accessing contended locks
2584 Details : if a client is granted a lock or receives a completion AST
2585 with a blocking AST already set it would not reply to the AST
2586 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2589 Frequency : lfs setstripe, only systems with more than 160 OSTs
2591 Description: unable to set striping with a starting offset beyond OST 160
2592 Details : llapi_create_file() incorrectly limited the starting stripe
2593 index to the maximum single-file stripe count.
2596 Frequency : LDAP users only
2598 Description: lconf did not handle in-kernel recovery with LDAP properly
2599 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2601 Severity : enhancement
2603 Description: bind OST threads to NUMA nodes to improve performance
2604 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2605 node and do their allocations there to localize memory access
2607 Severity : enhancement
2609 Description: llmount can determine client NID directly from Myrinet (GM)
2610 Details : the client NID code from gmnalnid was moved directly into
2611 llmount, removing the need to use this or specifying the
2612 client NID explicitly when mounting GM clients with zeroconf
2615 Frequency : if client is started with down MDS
2617 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2618 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2619 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2620 ptlrpc_import_delayed_req() because we didn't check if the
2621 request was interrupted, and we also didn't break out of the
2622 event loop if there was no timeout
2627 Description: data loss during non-page-aligned writes to a single file from
2628 both multiple nodes and multiple threads on one node at same time
2629 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2630 inconsistency led to false short-reads, that were cached and later
2631 used by ->prepare_write() to fill in partially written page,
2632 leading to data loss.
2635 Frequency : always, if lconf --abort_recovery used
2637 Description: lconf --abort_recovery fails with 'Operation not supported'
2638 Details : lconf was attempting to abort recovery on the MDT device and not
2641 Severity : enhancement
2643 Description: remove cleanup logs
2644 Details : replace lconf-generated cleanup logs with lustre internal
2645 cleanup routines. Eliminates the need for client-cleanup and
2648 Severity : enhancement
2650 Description: add support for EAs (user and system) on lustre filesystems
2651 Details : it is now possible to store extended attributes in the Lustre
2652 client filesystem, and with the user_xattr mount option it
2653 is possible to allow users to store EAs on their files also
2655 Severity : enhancement
2657 Description: Add possibility (config option) to show minimal available OST free
2659 Details : When compiled with --enable-mindf configure option, statfs(2)
2660 (and so, df) will return least minimal free space available from
2661 all OSTs as amount of free space on FS, instead of summary of
2662 free spaces of all OSTs.
2664 Severity : enhancement
2666 Description: do not expand extent locks acquired on OST-side
2667 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2668 by server: they are not cached anyway.
2671 Frequency : when mmap is used/binaries executed from Lustre
2673 Description: Unmmap pages before throwing them away from read cache.
2674 Details : llap_shrink cache now attempts to unmap pages before discarding
2675 them (if unmapping failed - do not discard). SLES9 kernel has
2676 extra checks that trigger if this unmapping is not done first.
2681 Description: lconf didn't resolve symlinks before checking to see whether a
2682 given mountpoint was already in use
2685 Frequency : when migrating failover services
2686 Bugzilla : 6395, 9514
2687 Description: When migrating a subset of services from a node (e.g. failback
2688 from a failover service node) the remaining services would
2689 time out and evict clients.
2690 Details : lconf --force (implied by --failover) sets the global obd_timeout
2691 to 5 seconds in order to quickly disconnect, but this caused
2692 other RPCs to time out too quickly. Do not change the global
2693 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2695 Severity : enhancement
2696 Frequency : if MDS is started with down OST
2697 Bugzilla : 9439,5706
2698 Description: Allow startup/shutdown of an MDS without depending on the
2699 availability of the OSTs.
2700 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2701 Add appropriate locking and lov-osc refcounts for safe
2702 cleaning. Add osc abort_inflight calls in case the
2703 synchronize never started.
2706 Frequency : occasional (Cray XT3 only)
2708 Description: root not authorized to access files in CRAY_PORTALS environment
2709 Details : The client process capabilities were not honoured on the MDS in
2710 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2711 been accessed by an authorized user then root was able to access
2712 the file on the local client also. The root user capabilities
2713 are now allowed on the MDS, as this environment has secure UID.
2716 Frequency : occasional
2718 Description: ldiskfs "too long searching" message happens too often
2719 Details : A debugging message (otherwise harmless) prints too often on
2720 the OST console. This has been reduced to only happen when
2721 there are fragmentation problems on the filesystem.
2726 Description: Division by zero in statfs when all OSCs are inactive
2727 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2728 lov_statfs divides by value returned by lov_get_stripecnt().
2732 Bugzilla : 9489, 3273
2733 Description: First write from each client to each OST was only 4kB in size,
2734 to initialize client writeback cache, which caused sub-optimal
2735 RPCs and poor layout on disk for the first writen file.
2736 Details : Clients now request an initial cache grant at (re)connect time
2737 and so that they can start streaming writes to the cache right
2738 away and always do full-sized RPCs if there is enough data.
2739 If the OST is rebooted the client also re-establishes its grant
2740 so that client cached writes will be honoured under the grant.
2745 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2746 Details : Now I/O RPCs go to different portal number and (presumably) fast
2747 lock requests (and glimses) and other RPCs get their own service
2748 threads pool that should be able to service those RPCs
2751 Severity : enhancement
2753 Description: Ability to exchange lustre version between client and servers and
2754 issue warnings at client side if client is too old. Also for
2755 liblustre clients there is ability to refuse connection of too old
2757 Details : New 'version' field is added to connect data structure that is
2758 filled with version info. That info is later checked by server and
2762 Frequency : rare, liblustre only.
2763 Bugzilla : 9296, 9581
2764 Description: Two simultaneous writes from liblustre at offset within same page
2765 might proceed at the same time overwriting eachother with stale
2767 Details : I/O lock withing llu_file_prwv was released too early, before data
2768 actually was hitting the wire. Extended lock-holding time until
2769 server acknowledges receiving data.
2772 Frequency : extremely rare. Never observed in practice.
2774 Description: avoid generating lustre_handle cookie of 0.
2775 Details : class_handle_hash() generates handle cookies by incrementing
2776 global counter, and can hit 0 occasionaly (this is unlikely, but
2777 not impossible, because initial value of cookie counter is
2778 selected randonly). Value of 0 is used as a sentinel meaning
2779 "unassigned handle" --- avoid it. Also coalesce two critical
2780 sections in this function into one.
2782 Severity : enhancement
2784 Description: allow liblustre clients to delegate truncate locking to OST
2785 Details : To avoid overhead of locking, liblustre client instructs OST to
2786 take extent lock in ost_punch() on client's behalf. New connection
2787 flag is added to handle backward compatibility.
2789 Severity : enhancement
2790 Bugzilla : 4928, 7341, 9758
2791 Description: allow number of OST service threads to be specified
2792 Details : a module parameter allows the number of OST service threads
2793 to be specified via "options ost ost_num_threads={N}" in the
2794 OSS's /etc/modules.conf or /etc/modprobe.conf.
2798 Bugzilla : 6146, 9635, 9895
2799 Description: servers crash with bad pointer in target_handle_connect()
2800 Details : In rare cases when a client is reconnecting it was possible that
2801 the connection request was the last reference for that export.
2802 We would temporarily drop the export reference and get a new
2803 one, but this may have been the last reference and the export
2804 was just destroyed. Get new reference before dropping old one.
2806 Severity : enhancement
2807 Frequency : if client is started with failover MDS
2809 Description: Allow multiple MDS hostnames in the mount command
2810 Details : Try to read the configuration from all specified MDS
2811 hostnames during a client mount in case the "primary"
2814 Severity : enhancement
2816 Description: Stop sending data to evicted clients as soon as possible.
2817 Details : Check if the client we are about to send or are sending data to
2818 was evicted already. (Check is done every second of waiting,
2819 for which l_wait_event interface was extended to allow checking
2820 of exit condition at specified intervals).
2823 Frequency : rare, normally only when NFS exporting is done from client
2825 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2827 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2830 Frequency : always, if srand() is called before liblustre initialization
2832 Description: Liblustre uses system PRNG disturbing its usage by user application
2833 Details : Introduce internal to lustre fast and high-quality PRNG for
2834 lustre usage and make liblustre and some other places in generic
2835 lustre code to use it.
2837 Severity : enhancement
2838 Bugzilla : 9477, 9557, 9870
2839 Description: Verify that the MDS configuration logs are updated when xml is
2840 Details : Check if the .xml configuration logs are newer than the config
2841 logs stored on the MDS and report an error if this is the case.
2842 Request --write-conf, or allow starting with --old_conf.
2844 Severity : enhancement
2846 Description: Handle symlinks in the path when checking if Lustre is mounted.
2847 Details : Resolve intermediate symlinks when checking if a client has
2848 mounted a filesystem to avoid duplicate client mounts.
2853 Description: lconf can hit an error exception but still return success.
2854 Details : The lconf command catches the Command error exception at the top
2855 level script context and will exit with the associated exit
2856 status, but doesn't ensure that this exit status is non-zero.
2861 Description: failure of ptlrpc thread startup can cause oops
2862 Details : Starting a ptlrpc service thread can fail if there are a large
2863 number of threads or the server memory is very fragmented.
2864 Handle this without oopsing.
2867 Frequency : always, only if liblustre and non-default acceptor port was used
2869 Description: liblustre cannot connect to servers with non-default acceptor port
2870 Details : tcpnal_set_default_params() was not called and was therefore
2871 ignoring the environment varaible TCPNAL_PORT, as well as other
2872 TCPNAL_ environment variables
2877 Description: two objects could be created on the same OST for a single file
2878 Details : If an OST is down, in some cases it was possible to create two
2879 objects on a single OST for a single file. No problems other
2880 than potential performance impact and spurious error messages.
2884 Bugzilla : 5681, 9562
2885 Description: Client may oops in ll_unhash_aliases
2886 Details : Client dcache may become inconsistent in race condition.
2887 In some cases "getcwd" can fail if the current directory is
2893 Description: Inode refcounting problems in NFS export code
2894 Details : link_raw functions used to call d_instantiate without obtaining
2895 extra inode reference first.
2899 Bugzilla : 9942, 9903
2900 Description: Referencing freed requests leading to crash, memleaks with NFS.
2901 Details : We used to require that call to ll_revalidate_it was always
2902 followed by ll_lookup_it. Also with revalidate_special() it is
2903 possible to call ll_revalidate_it() twice for the same dentry
2904 even if first occurence returned success. This fix changes semantic
2905 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2906 reference on a request referred from the intent.
2907 ll_intent_release() then releases such a request.
2910 Frequency : rare, normally benchmark loads only
2912 Description: unlinked inodes were kept in memory on the client
2913 Details : If a client is repeatedly creating and unlinking files it
2914 can accumulate a lot of stale inodes in the inode slab cache.
2915 If there is no other client load running this can cause the
2916 client node to run out of memory. Instead flush old inodes
2917 from client cache that have the same inode number as a new inode.
2920 Frequency : SLES9 2.6.5 kernel and long filenames only
2921 Bugzilla : 9969, 10379
2922 Description: utime reports stale NFS file handle
2923 Details : SLES9 uses out-of-dentry names in some cases, which confused
2924 the lustre dentry revalidation. Change it to always use the
2928 Frequency : rare, unless heavy write-truncate concurrency is continuous
2929 Bugzilla : 4180, 6984, 7171, 9963, 9331
2930 Description: OST becomes very slow and/or deadlocked during object unlink
2931 Details : filter_destroy() was holding onto the parent directory lock
2932 while truncating+unlinking objects. For very large objects this
2933 may block other threads for a long time and slow overall OST
2934 responsiveness. It may also be possible to get a lock ordering
2935 deadlock in this case, or run out of journal credits because of
2936 the combined truncate+unlink. Solution is to do object truncate
2937 first in one transaction without parent lock, and then do the
2938 final unlink in a new transaction with the parent lock. This
2939 reduces the lock hold time dramatically.
2942 Frequency : rare, 2.4 kernels only
2944 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2945 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2946 before unmounting their filesystem in order to avoid deadlock.
2947 The kernel_locked() function in 2.4 kernels only checks whether
2948 the kernel lock is held, not whether it is this process that is
2949 holding it as 2.6 kernels do.
2954 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2955 Details : The client ptlrpc code may be trying to reconnect to a down
2956 server before a previous connection attempt has timed out.
2957 Increase the reconnect interval to be longer than the connection
2958 timeout interval to avoid sending duplicate connections to
2962 Frequency : echo_client brw_test command
2964 Description: fix echo_client to work with OST preallocated code
2965 Details : OST preallocation code (5137) didn't take echo_client IO path
2966 into account: echo_client calls filter methods outside of any
2967 OST thread and, hence, there is no per-thread preallocated
2968 pages and buffers to use. Solution: hijack pga pages for IO. As
2969 a byproduct, this avoids unnecessary data copying.
2973 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2974 Description: Client can oops in mdc_commit_close() after open replay
2975 Details : It was possible for the MDS to return an open request with no
2976 transaction number in mds_finish_transno() if the client was
2977 evicted, but without actually returning an error. Clients
2978 would later try to replay that open and may trip an assertion
2979 Simplify the client close codepath, and always return an error
2980 from the MDS in case the open is not successful.
2983 Frequency : rare, 2.6 OSTs only
2985 Description: OST may deadlock under high load on fragmented files
2986 Details : If there was a heavy load and highly-fragmented OST filesystems
2987 it was possible to have all the OST threads deadlock waiting on
2988 allocation of biovecs, because the biovecs were not released
2989 until the entire RPC IO was completed. Instead, release biovecs
2990 as soon as they are complete to ensure forward IO progress.
2992 Severity : enhancement
2994 Description: Support for specifying external journal device at mount
2995 Details : If an OST or MDS device is formatted with an external journal
2996 device, this device major/minor is stored in the ext3 superblock
2997 and may not be valid for failover. Allow detecting and
2998 specifying the external journal at mount time.
3003 Description: Mounting an MDS with pending unlinked files may cause oops
3004 Details : target_finish_recovery() calls mds_postrecov() which returned
3005 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
3006 considers this an error and immediately begins cleaning up the
3007 lov, just after starting the mds_lov process
3009 Severity : enhancement
3011 Description: Implement 'lfs df' to report actual free space on per-OST basis
3012 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3013 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3014 usage of objects; '-h' to report in human readable format.
3016 ------------------------------------------------------------------------------
3018 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3025 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3026 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3027 at startup during orphan unlinking. Instead, skip buddy update
3028 before setup, as it will be regenerated after recovery is complete.
3033 Description: If an OST is inactive, its locks might reference stale inodes.
3034 Details : lov_change_cbdata() must iterate over all namespaces, even if
3035 they are inactive to clear inode references from the lock.
3037 Severity : enhancement
3038 Frequency : occasional, if non-standard max_dirty_mb used
3040 Description: Client will block write RPCs if not enough grant
3041 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3042 then the client will block writes while waiting for another RPC
3043 to complete instead of consuming its dirty limit. With change
3044 we get improved performance when max_dirty_mb is small.
3046 Severity : enhancement
3047 Bugzilla : 3389, 6253
3048 Description: Add support for supplementary groups on the MDS.
3049 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3050 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3051 lookups for user supplementary groups into a cache.
3055 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3056 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3057 for liblustre O_EXCL is enforced by MDS.
3061 Description: readdir never returns NULL in liblustre.
3062 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3063 dentry in struct dirent.
3067 Frequency : liblustre only, depends on application IO pattern
3068 Description: liblustre clients evicted if not contacting servers
3069 Details : Don't put liblustre clients into the ping_evictor list, so
3070 they will not be evicted by the pinger ever.
3072 Severity : enhancement
3074 Description: Add ability to evict clients by NID from MDS.
3075 Details : By echoing "nid:$NID" string into
3076 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3077 $NID would be instantly evicted from this MDS and from all active
3078 OSTs connected to it.
3082 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3083 Details : lookup_it_finish() used to query file size from OSTs that was not
3088 Description: service threads change working directory to that of init
3089 Details : Starting lustre service threads may pin the working directory
3090 of the parent thread, making that filesystem busy. Threads
3091 now change to the working directory of init to avoid this.
3095 Frequency : during shutdown only
3096 Description: shutdown with a failed MDS or OST can cause unmount to hang
3097 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3102 Frequency : 2.6 only
3103 Description: chmod/chown may include an extra supplementary group
3104 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3105 supplementary group and if none is specified this is used.
3108 Bugzilla : 5479 (6816)
3110 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3111 Details : If lookup is in progress on a file that is unlinked we might try
3112 to revalidate the inode and fail in revalidate after lookup is
3113 complete and ll_file_open() enqueues the open again but
3114 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3117 Frequency : always, if lconf --abort_recovery used
3119 Description: lconf --abort_recovery fails with 'Operation not supported'
3120 Details : lconf was attempting to abort recovery on the MDT device and not
3123 ------------------------------------------------------------------------------
3125 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3130 Frequency : rare (only unsupported configurations with a node running as an
3132 Bugzilla : 6514, 5137
3133 Description: Mounting a Lustre file system on a node running as an OST could
3135 Details : OSTs now preallocates memory needed to write out data at
3136 startup, instead of when needed, to avoid having to
3137 allocate memory in possibly low memory situations.
3138 Specifically, if the file system is mounted on on OST,
3139 memory pressure could force it to try to write out data,
3140 which it needed to allocate memory to do. Due to the low
3141 memory, it would be unable to do so and the node would
3142 become unresponsive.
3144 Severity : enhancement
3146 Description: Addition of lconf --service command line option
3147 Details : lconf now accepts a '--service <arg>' option, which is
3148 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3150 Severity : enhancement
3152 Description: Failover mode is now the default for OSTs.
3153 Details : By default, OSTs will now run in failover mode. To return to
3154 the old behaviour, add '--failout' to the lmc line for OSTs.
3156 Severity : enhancement
3158 Description: Health checks are now provided for MDS and OSTs
3159 Details : Additional detailed health check information on MSD and OSTs
3160 is now provided through the procfs health_check value.
3163 Frequency : occasional, depends on IO load
3165 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3166 after numerous create/delete cycles
3167 Details : The ext3 inode allocation policy would not allocate new inodes
3168 very well on the OSTs because there are no new directories
3169 being created. Instead we look for groups with free space if
3170 the parent directories are nearly full.
3175 Description: Network or server problems during mount may cause partially
3176 mounted clients instead of returning an error.
3177 Details : The config llog parsing code may overwrite the error return
3178 code during mount error handling, returning success instead
3184 Description: MDS can fail to allocate large reply buffers
3185 Details : After long uptimes the MDS can fail to allocate large reply
3186 buffers (e.g. zconf client mount config records) due to memory
3187 fragmentation or consumption by the buffer cache. Preallocate
3188 some large reply buffers so that these replies can be sent even
3189 under memory pressure.
3193 Frequency : rare (liblustre)
3194 Description: fsx running with liblustre complained that using truncate() to
3195 extend the file doesn't work. This patch corrects that issue.
3196 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3197 ATTR_SIZE and lsm use in llu_setattr_raw.
3201 Frequency : rare, only 2.6 kernels
3202 Description: Unusual file access patterns on the MDS may result in inode
3203 data being lost in very rare circumstances.
3204 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3205 code in the 2.6 kernel caused the inode and/or EA data not to
3206 be read from disk, causing single-file corruption.
3210 Frequency : rare, only 2.6 filesystems using extents
3211 Description: Heavy concurrent write and delete load may cause data corruption.
3212 Details : It was possible under high-load situations to have an extent
3213 metadata block in the block device cache from a just-unlinked
3214 file overwrite a newly-allocated data block. We now unmap any
3215 metadata buffers that alias just-allocated data blocks.
3219 Frequency : filesystems with default stripe_count larger than 77
3220 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3221 Details : lconf specifies an inode size of 4096 bytes when the default
3222 stripe_count is larger than 77. This conflicts with the default
3223 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3224 this case to avoid pinning too much memory for large EAs.
3226 ------------------------------------------------------------------------------
3228 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3233 Frequency : rare (extremely heavy IO load with hundreds of clients)
3235 Description: Client is evicted, gets IO error writing to file
3236 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3237 caused clients to be evicted for AST timeouts. The fixes in
3238 bug 5192 mean we no longer need to have such short AST timeouts
3239 so ldlm_timeout has been increased.
3242 Frequency : occasional during --force or --failover shutdown under load
3243 Bugzilla : 5949, 4834
3244 Description: Server oops/LBUG if stopped with --force or --failover under load
3245 Details : a collection of import/export refcount and cleanup ordering
3246 issues fixed for safer force cleanup
3249 Frequency : only filesystems larger than 120 OSTs
3250 Bugzilla : 5990, 6223
3251 Description: lfs getstripe would oops on a very large filesystem
3252 Details : lov_getconfig used kfree on vmalloc'd memory
3255 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3256 Bugzilla : 6242, 6243
3257 Description: reading from files that had been truncated to a non-zero size
3258 but never opened returned no data
3259 Details : ll_file_read() reads zeros from no-object files to EOF
3264 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3265 Details : There's little harm in aborting MDS/OSS recovery and letting it
3266 try again, so I removed the LASSERT and return an error instead.
3268 Severity : enhancement
3270 Description: New debugging infrastructure for tracking down data corruption
3271 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3272 (b) cover more of the client-side code path, and (c) try to narrow
3273 down where problems occurred
3277 Bugzilla : 3819, 4364, 4397, 6313
3278 Description: Racing close and eviction MDS could cause assertion in mds_close
3279 Details : It was possible to get multiple mfd references during close and
3280 client eviction, leading to one thread referencing a freed mfd.
3282 Severity: : enhancement
3283 Bugzilla : 3262, 6359
3284 Description: Attempts to reconnect to servers are now more aggressive.
3285 Details : This builds on the enhanced upcall-less recovery that was added
3286 in 1.4.2. When trying to reconnect to servers, clients will
3287 now try each server in the failover group every 10 seconds. By
3288 default, clients would previously try one server every 25 seconds.
3293 Description: After recovery, certain operations trigger a failed
3294 assertion on a client.
3295 Details : Failing over an mds, using lconf -d --failover, while a
3296 client was doing a readdir() call would cause the client to
3297 LBUG after recovery completed and the readdir() was resent.
3299 Severity : enhancement
3301 Description: Default groups are now added by lconf
3302 Details : You can now run lconf --group <servicename> without having to
3303 manually add groups with lmc.
3306 Frequency : occasional
3308 Description: Nodes with an elan id of 0 trigger a failed assertion
3311 Frequency : always when accessing e.g. tty/console device nodes
3313 Description: tty and some other devices nodes cannot be used on lustre
3314 Details : file's private_data field is used by device data and lustre
3315 values in there got lost. New field was added to struct file to
3316 store fs-specific private data.
3319 Frequency : when exporting Lustre via NFS
3321 Description: NFSD failed occasionally when looking up a path component
3322 Details : NFSD is looking up ".." which was broken in ext3 directories
3323 that had grown large enough to become hashed.
3326 Frequency : Clusters with multiple interfaces not on the same subnet
3328 Description: Nodes will repeatedly try to reconnect to an interface which it
3329 cannot reach and report an error to the log.
3330 Details : Extra peer list entries will be created by lconf with some peers
3331 unreachable. lconf now validates the peer before adding it.
3334 Frequency : Only if a default stripe is set on the filesystem root.
3336 Description: Setting a default stripe on the filesystem root prevented the
3337 filesystem from being remounted.
3338 Details : The client was sending extra request flags in the root getattr
3339 request and did not allocate a reply buffer for the dir EA.
3342 Frequency : occasional, higher if lots of files are accessed by one client
3343 Bugzilla : 6159, 6097
3344 Description: Client trips assertion regarding lsm mismatch/magic
3345 Details : While revalidating inodes the VFS looks up inodes with ifind()
3346 and in rare cases can find an inode that is being freed.
3347 The ll_test_inode() code will free the lsm during ifind()
3348 when it finds an existing inode and then the VFS later attaches
3349 this free lsm to a new inode.
3353 Bugzilla : 6422, 7030
3354 Description: MDS deadlock between mkdir and client eviction
3355 Details : Creating a new file via mkdir or mknod (starting a transaction
3356 and getting the ns lock) can deadlock with client eviction
3357 (gets ns lock and trying to finish a synchronous transaction).
3360 Frequency : occasional
3361 Description: While starting a server, the fsfilt_ext3 module could not be
3363 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3364 kernels. Previously, lconf would still use the ext3 name
3365 when trying to load modules. Now, it will correctly use
3366 ext3 on 2.4 and ldiskfs on 2.6.
3368 Severity : enhancement
3369 Description: The default stripe count has been changed to 1
3370 Details : The interpretation of the default stripe count (0, to lfs
3371 or lmc) has been changed to mean striping across a single
3372 OST, rather than all available. For general usage we have
3373 found a stripe count of 1 or 2 works best.
3375 Severity : enhancement
3376 Description: Add support for compiling against Cray portals.
3377 Details : Conditional compiling for some areas that are different
3381 Frequency : occasional
3382 Bugzilla : 6409, 6834
3383 Description: Creating files with an explicit stripe count may lead to
3384 a failed assertion on the MDS
3385 Details : If some OSTs are full or unavailable, creating files may
3386 trigger a failed assertion on the MDS. Now, Lustre will
3387 try to use other servers or return an error to the
3391 Frequency : occasional
3393 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3394 caused lock timeout during readahead (since 1.4.2).
3395 Details : Processes doing readahead might match a lock that hasn't been
3396 granted yet if there are overlapping and conflicting lock
3397 requests. The readahead process waits on ungranted lock
3398 (original lock is CBPENDING), while OST waits for that process
3399 to cancel CBPENDING read lock and eventually evicts client.
3401 Severity : enhancement
3403 Description: Initial enabling of flock support for clients
3404 Details : Implements fcntl advisory locking and file status functions.
3405 This feature is provided as an optional mount flag (default
3406 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3407 locking are implemented yet, and those that are are not guaranteed
3408 to be completely correct in production environments.
3409 mount -t lustre -o [flock|noflock] ...
3412 Frequency : occasional
3414 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3415 assertion in the ext3 JBD (journaling) layer.
3416 Details : The b_committed_data struct is protected by the big kernel lock
3417 in 2.4 kernels, serializing journal_commit_transaction() and
3418 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3419 access to this struct. In 2.6 kernels there is finer grained
3420 locking to improve SMP performance of the JBD layer.
3424 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3426 -----------------------------------------------------------------------------
3428 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3430 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3431 versions of Lustre. Please update all servers and clients to
3432 version 1.4.2 or later at the same time. You must also run
3433 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3434 to update the configuration logs.
3436 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3437 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3438 - swab llog records of type '0' so we get proper header size/idx (5861)
3439 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3440 - fix rename of one directory over another leaking an inode (5953)
3441 - avoid SetPageDirty on 2.6 (5981)
3442 - don't re-add just-being-destroyed locks to the waiting list (5653)
3443 - when creating new directories, inherit the parent's custom
3444 striping settings if present parent (3048)
3445 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3446 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3447 - don't hold client locks on temporary worklist from l_lru (5666)
3448 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3449 - automatically evict dead clients (5921)
3450 - Update file size properly in create+truncate+fstat case (6196)
3451 - Do not unhash mountpoint dentries, do not allow removal of
3453 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3454 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3455 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3456 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3457 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3458 - Correctly handle reads of files with no objects (6243)
3459 - lctl recover will also mark a device active if deactivate used (5933)
3461 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3462 - allow --write-conf on an MDS with different nettype than client (5619)
3463 - don't write config llogs to MDS for mounts not from that MDS (5617)
3464 - lconf should create multiple TCP connections from a client (5201)
3465 - init scripts are now turned off by default; run chkconfig --on
3466 lustre and chkconfig --on lustrefs to use them
3467 - upcalls are no longer needed for clients to recover to failover
3469 - add --abort-recovery option to lconf to abort recovery on device
3471 - add support for an arbitrary number of OSTs (3026)
3472 - Quota support protocol changes.
3473 - forward compatibility changes to wire structs (6007)
3474 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3475 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3476 - improved "lustre status" script
3477 - initialize blocksize for non-regular files (6062)
3478 - added --disable-server and --disable-client configure options (5782)
3479 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3480 - Vanilla 2.4.29 support
3481 - increase maximum number of obd devices to 520 (6242)
3482 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3483 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3484 SLES 9 2.6.5 kernels
3485 - limit stripes per file to 160 (the maximum EA size) (6093)
3487 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3490 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3491 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3492 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3493 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3494 - small ext3 extents cleanups and fixes (5733)
3495 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3496 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3497 - avoid needless client->OST connect, fix handle mismatch (5317)
3498 - fix DLM error path that led to out-of-sync client, long delays (5779)
3499 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3500 - fix several locking issues related to i_size (5492,5624,5654,5672)
3501 - don't move pending lock onto export if it is already evicted (5683)
3502 - fix kernel oops when creating .foo in unlinked directory (5548)
3503 - fix deadlock in obdfilter statistics vs. object create (5811)
3504 - use time_{before,after} to avoid timer jiffies wrap (5882)
3505 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3506 - Do not leak request if server was not able to process it (5154)
3507 - If mds_open unable to find parent dir, make that negative lookup(5154)
3508 - don't create new directories with extent-mapping (5909, 5936)
3510 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3511 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3512 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3513 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3514 - support for RHEL4 kernel on i686 (5773)
3515 - provide error messages when incompatible logs are encountered (5898)
3517 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3518 * version 1.4.0.10 (1.4.1 release candidate 1)
3520 - don't keep a lock reference when lock is not granted (4238)
3521 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3522 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3523 - drop import inflight refcount on signal_completed_replay error (5255)
3524 - unlock page after async write error during send (3677)
3525 - handle missing objects in filter_preprw_read properly (5265)
3526 - no transno return for symlink open, don't save no-trasno open (3440)
3527 - don't try to complete elan receive that already failed (4012)
3528 - free RPC server reply state on error (5406)
3529 - clean up thread from ptlrpc_start_thread() on error (5160)
3530 - readahead could read extra page into cache that wasn't ejected (5388)
3531 - prevent races in class_attach/setup/cleanup/detach (5260)
3532 - don't dereference de->d_inode after l_dput of de (5458)
3533 - use "int" for stripe value returned from lock_to_stripe (5544)
3534 - mballoc allocation and error-checking fixes in 2.6 (5504)
3535 - block device patches to fix I/O request sizes in 2.6 (5482)
3536 - look up hostnames for IB nals (5602)
3537 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3538 - don't start multiple acceptors for the same port (5277)
3539 - fix incorrect LASSERT in mds_getattr_name (5635)
3540 - export a proc file for general "ping" checking (5628)
3541 - fix "lfs check" to not block when the MDS is down (5628)
3543 - service request history (4965)
3544 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3545 - create an "evict_client" /proc entry on OSTs, like the MDS has
3546 - fix mount usage message, return errors per mount(8) (5168)
3547 - change grep [] to grep "[]" in tests so they work in more UMLs
3548 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3549 - remove ancient LOV_MAGIC_V0 EA support (5047)
3550 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3551 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3552 - allow readahead allocations to fail when low on memory (5383)
3553 - mmap locking landed again, after considerable improvement (2828)
3554 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3556 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3559 - send OST transaction number in read/write reply to free req (4966)
3560 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3561 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3562 - the watchdog thread now runs as interruptible (5246)
3563 - flock/lockf fixes (but it's still disabled, pending 5135)
3564 - don't use EXT3 constants in llite code (5094)
3565 - memory shortage at startup could cause assertion (5176)
3567 - reorganization of lov code
3568 - single portals codebase
3570 - add extents/mballoc support (5025)
3571 - direct I/O reads in the obdfilter (4048)
3572 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3574 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3577 - send OST transaction number in read/write reply to free req (4966)
3578 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3579 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3580 - the watchdog thread now runs as interruptible (5246)
3581 - handle missing objects in filter_preprw_read properly (5265)
3582 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3583 - drop import inflight refcount on signal_completed_replay error (5255)
3584 - unlock page after async write error during send (3677)
3585 - return original error code on reconstructed replies (3761)
3586 - no transno return for symlink open, don't save no-trasno open (3440)
3588 - add pid to ldlm debugging output (4922)
3589 - bump the watchdog timeouts -- we can't handle 30sec yet
3590 - extra debugging for orphan dentry/inode bug (5259)
3592 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3595 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3596 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3597 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3598 - fix NULL dereference in /proc/sys/portals/routes (4827)
3599 - allow failed mdc_close() operations to be interrupted (4561)
3600 - stop precreate on OST before MDS would time out on it (4778)
3601 - don't send partial-page writes before EOF from client (4410)
3602 - discard client grant for sub-page writes on large-page clients (4520)
3603 - don't free dentries not owned by NFS code, check generation (4806)
3604 - fix lsm leak if mds_create_objects() fails (4801)
3605 - limit debug_daemon file size, always print CERROR messages (4789)
3606 - use transno after validating reply (3892)
3607 - process timed out requests if import state changes (3754)
3608 - update mtime on OST during writes, return in glimpse (4829)
3609 - add mkfsoptions to LDAP (4679)
3610 - use ->max_readahead method instead of zapping global ra (5039)
3611 - don't interrupt __l_wait_event() during strace
3613 - add software watchdogs to catch hung threads quickly (4941)
3614 - make lustrefs init script start after nfs is mounted
3615 - fix CWARN/ERROR duplication (4930)
3616 - return async write errors to application if possible (2248)
3617 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3618 - print NAL number in %x format (4645)
3619 - update barely-supported suse-2.4.21-171 series (4842)
3620 - support for sles 9 %post scripts
3621 - support for building 2.6 kernel-source packages
3622 - support for sles km_* packages
3624 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3627 - ignore -ENOENT errors in osc_destroy (3639)
3628 - notify osc create thread that OSC is being cleaned up (4600)
3629 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3630 - reconstruct ost_handle() like mds_handle() (4657)
3631 - create a new thread to do import eviction to avoid deadlock (3969)
3632 - let lconf resolve symlinked-to devices (4629)
3633 - don't unlink "objects" from directory with default EA (4554)
3634 - hold socknal file ref over connect in case target is down (4394)
3635 - allow more than 32000 subdirectories in a single directory (3244)
3636 - fix blocks count for O_DIRECT writes (3751)
3637 - OST returns ENOSPC from object create when no space left (4539)
3638 - don't send truncate RPC if file size isn't changing (4410)
3639 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3640 - bind to privileged port in socknal and tcpnal (3689)
3642 - rate limit CERROR/CWARN console message to avoid overload (4519)
3643 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3644 - basic mmap support (3918)
3645 - kernel patch series update from b1_4 (4711)
3647 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3650 - avoid crash during MDS cleanup with OST shut down (2775)
3651 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3652 - don't use bad inodes on the MDS (3744)
3653 - dynamic object preallocation to improve recovery speed (4236)
3654 - don't hold spinlock over lock dumping or change debug flags (4401)
3655 - don't zero obd_dev when it is force cleaned (3651)
3656 - print grants to console if they go negative (4431)
3657 - "lctl deactivate" will stop automatic recovery attempts (3406)
3658 - look for existing locks in ldlm_handle_enqueue() (3764)
3659 - don't resolve lock handle twice in recovery avoiding race (4401)
3660 - revalidate should check working dir is a directory (4134)
3662 - don't always mark "slow" obdfilter messages as errors (4418)
3664 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3667 - don't close LustreDB during write_conf until it is done (3860)
3668 - fix typo in lconf for_each_profile (3821)
3669 - allow dumping logs from multiple threads at one time (3820)
3670 - don't allow multiple threads in OSC recovery (3812)
3671 - fix debug_size parameters (3864)
3672 - fix mds_postrecov to initialize import for llog ctxt (3121)
3673 - replace config semaphore with spinlock (3306)
3674 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3675 - don't allow enqueue to complete on a destroyed export (3822)
3676 - down write_lock before checking llog header bitmap (3825)
3677 - recover from lock replay timeout (3764)
3678 - up llog sem before sending rpc (3652)
3679 - reduce ns lock hold times when setting kms (3267)
3680 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3681 - fix NULL deref and obd_dev leak on setup error (3312)
3682 - replace some LBUG about llog ops with error handling (3841)
3683 - don't match INVALID dentries from d_lookup and spin (3784)
3684 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3685 - fix invalid assertion in ptlrpc_set_wait (3880)
3687 - add libwrap support for the TCP acceptor (3996)
3688 - add /proc/sys/portals/routes for non-root route listing (3994)
3689 - allow setting MDS UUID in .xml (2580)
3690 - print the stack of a process that LBUGs (4228)
3692 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3695 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3696 - make sure to unset replay flag from failed open requests (3440)
3697 - if default stripe count is 0, use OST count for inode size (3636)
3698 - update parent mtime/ctime on client for create/unlink (2611)
3699 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3700 - free recovery state on server during a forced cleanup (3571)
3701 - unregister_reply for resent reqs (3063)
3702 - loop back devices mounting and status check on 2.6 (3563)
3703 - fix resource-creation race that can provoke i_size == 0 (3513)
3704 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3705 - more debugging for page-accounting assertion (3746)
3706 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3707 - avoid deadlock after precreation failure (3758)
3708 - fix race and lock order deadlock in orphan handling (3450, 3750)
3709 - add validity checks when grabbing inodes from l_ast_data (3599)
3711 - add /proc/.../recovery_status to obdfilter (3428)
3712 - lightweight CDEBUG infrastructure, debug daemon (3668)
3713 - change default OSC RPC parameters to be better on small clusters
3714 - turn off OST read cache for files smaller than 32MB
3715 - install man pages and include them in rpms (3100)
3716 - add new init script for (un)mounting lustre filesystems (2593)
3717 - run chkconfig in %post for init scripts (3701)
3718 - drop scimac NAL (unmaintained)
3720 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3723 - clean kiobufs before and after use (3485)
3724 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3725 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3726 - change init script to fail more clearly if not run as root (1528)
3727 - allow clients to reconnect during replay (1742)
3728 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3729 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3730 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3731 - fix cond_resched (3554)
3732 - only evict unfinished clients after recovery (3515)
3733 - allow bulk resend, prevent data loss (3570)
3734 - dynamic ptlrpc request buffer allocation (2102)
3735 - don't allow unlinking open directory if it isn't empty (2904)
3736 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3737 - remove extraneous obd dereference causing LASSERT failure (3334)
3738 - don't use get_cycles() when creating temp. files on the mds (3156)
3739 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3740 - handle EEXIST for set-stripe, set proper directory name (3336)
3742 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3743 - fix ksocknal_fmb_callback() error messages (2918)
3745 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3748 - don't copy lvb into (possibly NULL) reply on error (2983)
3749 - don't deref dentry after dput, don't free lvb on error (2922)
3750 - use the kms to determine writeback rpc length (2947)
3751 - increment oti_logcookies when osc is inactive (2948)
3752 - update client's i_blocks count via lvb messages (2543)
3753 - handle intent open/close of special files properly (1557)
3754 - mount MDS with errors=remount-ro, like obdfilter (2009)
3755 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3756 - don't use cancelling-locks' kms values (2947)
3757 - use highest lock extent for kms, not last one (2925)
3758 - don't dereference ERR_PTR() dentry in error handling path (3107)
3759 - fix thread race in portals_debug_dumplog() (3122)
3760 - create lprocfs device entries at setup instead of at attach (1519)
3761 - common AST error handler, don't evict client on completion race (3145)
3762 - zero nameidata in detach_mnt in 2.6 (3118)
3763 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3764 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3765 - initialize RPC timeout timer earlier for 2.6 (3219)
3766 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3767 - print nal/nid for unknown nid (3258)
3768 - additional checks for oscc recovery before doing precreate (3284)
3769 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3770 - don't crash in mdc_close for bad permissions on open (3285)
3771 - zero i_rdev for non-device files (3147)
3772 - clear page->private before handing to FS, better assertion (3119)
3773 - tune the read pipeline (3236)
3774 - fix incorrect decref of invalidated dentry (2350)
3775 - provide read-ahead stats and refine rpc in flight stats (3328)
3776 - don't hold journal transaction open across create RPC (3313)
3777 - update atime on MDS at close time (3265)
3778 - close LDAP connection when recovering to avoid server load (3315)
3779 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3780 - don't leak open file on MDS after open resend (3325)
3781 - serialize filter_precreate and filter_destroy_precreated (3329)
3782 - loop device shouldn't call sync_dev() for nul device (3092)
3783 - clear page cache after eviction (2766)
3784 - resynchronize MDS->OST in background (2824)
3785 - refuse to mount the same filesystem twice on same mountpoint (3394)
3786 - allow llmount to create routes for mounting behind routers (3320)
3787 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3788 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3789 - fix rare problem with rename on htree directories (3417)
3791 - allow default OST striping configuration per directory (1414)
3792 - fix compilation for qswnal for 2.6 kernels (3125)
3793 - increase maximum number of MDS request buffers for large systems
3794 - change liblustreapi to be useful for external progs like lfsck (3098)
3795 - increase local configuration timeout for slow disks (3353)
3796 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3798 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3801 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3802 - don't overwrite extent policy data in reply if lock was blocked (2901)
3803 - drop filter export grants atomically with removal from device (2663)
3804 - del obd_self_export from work_list in class_disconnect_exports (2908)
3805 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3806 - swab reply message in mdc_close, other PPC fixes (2464)
3807 - fix destroying of named logs (2325)
3808 - overwrite old logs when running lconf --write_conf (2264)
3809 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3810 - fix race in target_handle_connect (2898)
3811 - mds_reint_create() should take same inode create lock (2926)
3812 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3813 - don't close files for self_export to avoid uninitialized obd (2936)
3814 - allow MDS with the same name as client node (2939)
3815 - hold dentry reference for closed log files for unlink (2325)
3816 - reserve space for all logs during transactions (2059)
3817 - don't evict page beyond end of stripe extent (2925)
3818 - don't oops on a deleted current working directory (2399)
3819 - handle hard links to targets without a parent properly (2517)
3820 - don't dereference NULL lock when racing during eviction (2867)
3821 - don't grow lock extents when lots of conflicting locks (2919)
3823 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3826 - account for cache space usage on clients to avoid data loss (974)
3827 - lfsck support in lustre kernel code (2349)
3828 - reduce journal credits needed for BRW writes (2370)
3829 - orphan handling to avoid losing space on client/server crashes
3830 - ptlrpcd can be blocked, stopping ALL progress (2477)
3831 - use lock value blocks to assist in proper KMS, faster stat (1021)
3832 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3833 - recovery for initial connections (2355)
3834 - fixes for mds_cleanup_orphans (1934)
3835 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3836 - block all file creations until orphan recovery completes (1901)
3837 - client remove rq_connection from request struct (2423)
3838 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3839 - recovery timer race (2670)
3840 - mdc_close recovey bug (2532)
3841 - ptlrpc cleanup bug (2710)
3842 - mds timeout on local locks (2588)
3843 - namespace lock held during RPCs (2431)
3844 - handle interrupted sync write properly (2503)
3845 - don't try to handle a message that hasn't been replied to (2699)
3846 - client assert failure during cleanup after abort recovery (2701)
3847 - leak mdc device after failed mount (2712)
3848 - ptlrpc_check_set allows timedout requests to complete (2714)
3849 - wait for inflight reqs when ptlrpcd finishes (2710)
3850 - make sure unregistered services are removed from the srv_list
3851 - reset bulk XID's when resending them (caught by 1138 test)
3852 - unregister_bulk after timeout
3853 - fix lconf error (2694)
3854 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3855 - readahead locks pages, leaves pending causing memory pressure (2673)
3856 - increase OST request buffers to 4096 on large machines (2729)
3857 - fix up permission of existing directories in simple_mkdir (2661)
3858 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3859 - don't assign transno to errored transactions (2742)
3860 - don't delete objects on OST if given a bogus objid from MDS (2751)
3861 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3862 - if rq_no_resend, then timeout request after recovery (2432)
3863 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3864 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3865 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3866 - MDS recovery shouldn't delete objects that it has given out (2730)
3867 - if enqueue arrives after completion, don't clobber LVB (2819)
3868 - don't unlock pages twice when trigger_group_io returns error (2814)
3869 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3870 - don't write pages to disk if there was an error (1450)
3871 - don't ping imports that have recovery disabled (2676)
3872 - take buffered bytes into account when balancing socknal conn (2817)
3873 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3874 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3875 - remove little-endian swabbing of llog records (1987)
3876 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3877 - retry reposting request buffers if they fail (1191)
3878 - grow extent at grant time to avoid granting a revoked lock (2809)
3879 - lock revoke doesn't evict page if covered by a second lock (2765)
3880 - disable VM readahead to avoid reading outside lock extents (2805)
3882 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3883 - updated kernel patches for hp-2.4.20 kernel (2681)
3885 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3888 - fix truncated write corruption (2366)
3889 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3891 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3892 - ensure len(uuid) < 37 in lmc (1171)
3893 - fix ia64 OOPS in llog_test (2255)
3894 - zero end of page at obdfilter for partial page writes (2648)
3895 - don't leave stale dentries around after renames (bug 2428)
3896 - fix timeouts when evicting a client with a single lock held (2642)
3897 - set deadline for the initial HELLO message to drain (2634)
3898 - print out dotted-quad IP addresses in the socknal (2302)
3900 - additional debugging for MDS client eviction problem (2443)
3901 - fix mkfsoptions support for osts (2603, 2604)
3903 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3906 - add series for the vanilla 2.6.0 kernel
3907 - add series for the vanilla 2.4.24 kernel
3908 - add series for a cray x86/64 UL kernel drop
3909 - fix xattr patches for the vanilla 2.4.19 series
3911 - generate true UUIDs in lmc (1171)
3912 - have portals stack dumping break in UML (2466)
3913 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3914 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3915 - wake callback waiting threads on client eviction (2460)
3916 - Add --ptldebug and --subsystem to lmc (1719)
3917 - update assertion to allow safe interrupt allocation
3918 - set rq_no_resend for cancel requests (2432)
3919 - recalculate ptlrpcd timeout after resend (2494)
3920 - call vfs_rmdir when removing pending directories (2368)
3921 - fix renaming a file to itself (2429)
3922 - lmc creates a default one-stripe lov (2454)
3923 - expand procfs space to handle large clusters (2326)
3924 - increase UML stack to avoid overflow
3925 - update lconf's list of debug and subsystem masks
3926 - fix lfs find --obd (2510)
3927 - /proc tunable for disabling filter read caching (2591)
3928 - stop rpm packages from altering slapd.conf (2301)
3929 - disable nagle in the socknal under 0conf (2578)
3930 - choose mds inode size based on stripe count (2572)
3931 - fix kernel-source rpm problems (2516)
3933 - add --disable-doc to avoid pdf generation (2421)
3934 - update documentation, tests, type-os, comments
3935 - avoid format warnings on ia64
3936 - remove the TOE NAL
3937 - tiny code cleanups by removing unused fields
3939 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3942 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3943 - remove the most blatant lies from BUILDING file (2371)
3944 - change default debug level to reasonable production setting
3945 - reduce client side cache size to reduce cache flush time
3946 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3947 - make TCP zerocopy and pinger support enabled by default (2476)
3948 - sync writes completed after process exits caused crashes (2319)
3949 - maintain correct mount count on the MDS (2356)
3950 - backout 1557, because 2316 wasn't really fixed
3951 - better file I/O statistics gathering in /proc
3952 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3953 - another kernel patch to fix zero-copy TCP function export
3954 - don't take duplicate lock when processing re-sent getattr (2420)
3955 - lctl uses obd_self_export instead of creating new conn (2353)
3956 - MDS/OST recovery case which requires object creation asserted (2425)
3957 - move lfs from /usr/sbin to /usr/bin in packages
3958 - fix race between mds_client_add and mds_client_free (2417)
3959 - use kmalloc instead of slabs in portals (2430)
3960 - don't create duplicate records when a failover MDS is present (2442)
3961 - remove unnecessary mount age check (2332)
3962 - don't remove directory inodes from locks prematurely (2451)
3963 - don't break if MDS service name is the same as hostname (2103)
3964 - fix races in client write RPC generation when cache full (2482)
3966 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3969 - remove now-unused request->rq_obd (278)
3970 - if an allocation fails, print out how much memory we've used (1933)
3971 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3972 - add the "configurable stack size" patch to most series files (1256)
3973 - ability to write large log records, for 100+ OST configs (2306)
3974 - fix NULL deref when filter_prep fails (2314)
3975 - fix operator precedence error in filter_sync
3976 - dynamic allocation of socknal TX descriptors (2315)
3977 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3978 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3979 - ensure that CWARN messages in Portals always get to the syslog
3980 - __init/__exit are not for prototype decls (ldlm_init/exit)
3981 - x86-64 compile warning fixes
3982 - fix gateway LMC keyword conflict (2318)
3983 - fix MDS lock inversions in getattr/reint paths (1844)
3984 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3985 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3986 - DLM race condition prevented some lock evictions (2328)
3987 - ENOMEM detection and retry on socknal sends (2230)
3988 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3989 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3990 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3991 - better error messages when a client is rejected during recovery (1505)
3992 - avoid cancelling locks which were never granted, after failure (2330)
3993 - fix i_sem/journal inversion in mds_client_add (2333)
3994 - fix truncate/getattr lock cycle deadlock (2334)
3995 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3996 - fix two rare exit paths which could leak an l_lock() ref (2321)
3997 - fencepost error in MDS/OST orphan recovery (2226)
3998 - make log record alignment 8 bytes (1988)
3999 - lstripe now fails when requested offset > ost_count (2237)
4000 - ensure that all kernel series have a complete list.h (1607)
4001 - fix crashes in special-file operations (2316)
4002 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
4004 - allow configurable automake binary, for testing new versions
4005 - small update to the lfs documentation
4007 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
4009 * fix negative export reference count in fsfilt_sync (2312)
4011 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4012 * release candidate 0.9.1
4014 - orphans are moved into the PENDING directory for possible recovery
4015 - replayed opens now open by fid for orphan/rename safety (1042)
4016 - last close of an orphan inode generates a transno (683)
4017 - chdir() and mount() now pin the directory entry (1020)
4018 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4019 - discard very old requests without processing them (1502)
4020 - remove some common, well-understood CERRORs (1505)
4021 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4022 - clear "grant" flags in OST replies until OST grant code lands (1644)
4023 - fix read performance by not clobbering i_blksize on client (1598)
4024 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4025 - make LRU size a /proc tunable, clears locks when reduced (707)
4026 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4027 - server locks take references on exports now (1558)
4028 - build fixes for 2.4.20-rh trees (1663)
4029 - return an error from lov_create if all OSCs are inactive (1751)
4030 - fix import levels when a reconnect happens without a timeout (1597)
4031 - exit early from mds_open if we get a lookup error (1749)
4032 - partial page read at EOF wouldn't wait for disk before sending (1642)
4033 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4034 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4035 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4036 - write/truncate lock inversion (1639)
4037 - Don't auto-load obdclass, portals modules during cleanup (1495)
4038 - fix timestamps from jumping to "now" (1763)
4039 - extra journal assertions (1648)
4040 - add an extra multiunlink test (1771)
4041 - fix read_record/write_record API (1776)
4042 - fix leak of offset_extent, possible incorrect i_size later (1772)
4043 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4044 - Debugging for the kqswnal_get_idle_tx problems (1820)
4045 - Allow recovery to be attempted multiple times (1536)
4046 - Write out MDS last_rcvd file after it is first created (1600)
4047 - Fix tx_descriptor leak in failed transmit situations (1827)
4048 - ext3 journaling fixes for assertion failure after IO error (1871)
4049 - class_export_put() on freed export after completion AST error (1896)
4050 - Fix revalidate looping in VFS (1322)
4051 - Don't access a freed export during MDS_REINT timeout (1521)
4052 - Add open-unlink recovery support on the MDS (1673,1764)
4053 - Return an error if no MDS data was read from last_rcvd (1946)
4054 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4055 - Don't setup a disk device that doesn't match exported UUID (317)
4056 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4057 - avoid committing NULL handle in force close
4058 - local.sh is now a one-stripe LOV configuration
4059 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4060 - fix ext3 htree duplicate directory entry corruption (1516)
4061 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4062 - update to new LOV EA format (2097)
4063 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4065 2003-06-15 Phil Schwan <phil@clusterfs.com>
4068 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4069 - per-import/export recovery handling (958, 931, 959)
4070 - multiple last-rcvd slots, for serving multiple FSes (949)
4071 - connections are again shared between multiple imp/exports (963, 964)
4072 - "umount -f" would hang if any requests needed to be sent (393, 978)
4073 - avoid pinning large req buffer by copying for queued messages (989)
4074 - add "uuid" to "lctl device" command to help upcalls (991)
4075 - "open" RPCs with transnos would confuse recovery counters (1037)
4076 - do proper endian conversion of all wire messages (288, 340, 891)
4077 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4078 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4079 - fix ptlrpc_connection leak in target_handle_connect (1174)
4080 - fix import refcounting bug in OST and MDS cleanup (1134)
4081 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4082 - fix very unlikely obd_types race condition (501)
4083 - remove osc_open hack for echo_client (1187)
4084 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4085 - a failure in read_inode2 leads to deadlock (1139)
4086 - cancel ack-locks as soon as transaction is committed (1072)
4087 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4088 - make sure to commitrw after any preprw to avoid deadlock (1162)
4089 - failing to execute a file in a lustre FS would lock inode (1203)
4090 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4091 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4092 - obd_uuid2tgt would walk off the end of the list (1255)
4093 - on IA64 the osc would give portals incorrect bulk size (1258)
4094 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4095 - fix lock inversion caused by new llite matching code (1282)
4096 - limit the number of dirty pages on a client to 10MB (1286)
4097 - timed out locks were not being corrected cancelled (1289)
4098 - fix O_DIRECT above 4GB on IA-32 (1292)
4099 * major user-visible changes
4100 - fail out/fail over policy now controlled by the upcall (993)
4102 - add OBD_PING to check server availability and failure (954)
4103 - lustre messages are now sent in sending host order (288, 340, 891)
4104 - add eadatalen to MDS getattr reply (340)
4105 - OST read replies may contain second buffer, with per-page status (593)
4107 2003-03-11 Phil Schwan <phil@clusterfs.com>
4110 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4111 - failed open()s could cause deadlock; fixed (867, 869)
4112 - stop cancelling OST locks when files are closed (481)
4113 - overlapping XID spaces caused network corruption (851, 853)
4114 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4115 - setattr_raw added, to do single-RPC, server-side setattrs
4116 - lmc/lconf syntax change for OST UUIDs
4117 - fix crashy race condition between ptlrpc_free_req and osc_close
4118 - don't use request in mdc_enqueue if we hit a timeout (889)
4119 - don't set the inode i_size for regular files from the MDS (896)
4120 - handle out of order completion AST (842)
4121 - don't LBUG if a lock request times out after receiving AST (913)
4122 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4123 - if a bad lock AST arrives, send an error instead of dropping entirely
4124 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4125 - fix leak in bulk IO when only partially completed (899, 900, 926)
4126 - fix O_DIRECT for ia64 (55)
4127 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4128 - C-z after timeout could hang a process forever; fixed (977)
4130 - client-side I/O cache (678, 924, 929, 941, 970)
4132 - READPAGE and SETATTRs which don't take server-side locks get
4135 2003-02-11 Phil Schwan <phil@clusterfs.com>
4138 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4139 - Fix fsfilt_extN_readpage() to read a full page of directory
4140 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4141 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4142 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4143 - Most intent operations were removed, in favour of a new RPC mode
4144 that does a single RPC to the server and bypasses most of the VFS
4145 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4146 - Aggressively cancel local locks on DLM servers
4147 - mds_reint_unlink sends EA to the client if it's the last nlink.
4148 client uses that EA to unlink OST objects.
4149 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4150 - recursive symlinks were fixed (439)
4151 - fixed NULL deref in DEBUG_REQ
4152 - filter_update_lastobjid no longer calls sync, which annoyed extN
4153 - fixed multi-client small-writes to a single file problem (445)
4154 - fixed mtime updates during file writes (607)
4155 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4156 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4157 - fixed Linux OST size reporting problem (444, 656)
4158 - OST now updates object mtime with writes or setattr (607, 619)
4159 - client verifies file size before zeroing page past EOF (445)
4160 - OST now writes last allocated objid to disk with allocation (108)
4161 - LOV on echo now works (409)
4163 - mds_reint_unlink sends a new buffer, with the EA included. this
4164 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4165 set if a regular file was being unlinked, and it was the last link
4166 - use PtlGet from the target for bulk writes (315)
4167 - OST now updates object mtime with writes or setattr (607, 619)
4168 - LDLM now has a grant-time callback to revalidate locked items, if
4170 - Many MDS operations were reorganized to combat race conditions
4172 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4173 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4175 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4178 - Fully reactivate OST imports after reconnection (512, others)
4179 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4180 - More graceful error handling for truncating on dead OST (515)
4181 - Don't error out unless we're actually accessing dead stripes (474)
4182 - Fix garbage sizes when stripes are missing (410)
4183 - LRU counters were broken, causing constant lock purge (433, 432)
4184 - garbage on read from stripes with failed OSTs (441)
4185 - mark OSCs as active before reconnecting during recovery (438)
4186 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4187 - lfind did not preserve OST order in output (443)
4188 - symlinks cause hung clients, incorrect data (439)
4189 - stop dereferencing request after dropping refcount (457)
4190 - don't LASSERT(spin_is_locked) on non-SMP (455)
4191 - fixes for many rename() bugs
4192 - fstat didn't correctly synchronize attributes (399)
4193 - server must handle lock cancellation during blocking AST prep (487)
4194 - bulk descriptors were free()d too soon (511)
4195 - fix paths in lconf, which would load incorrect modules (451, 507)
4196 - fix confusing lconf 'host not found' error message (386)
4197 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4198 - fix race condition in mdc_blocking_ast() for inode access (526)
4199 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4200 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4201 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4202 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4203 - fix lstripe in several ways (kernel side) (527)
4204 - fix request leak in ldlm_cli_enqueue (262)
4205 - incorrect OSC was marked inactive after OST failure
4206 - call mds_fs_cleanup before unmounting filesystem (524)
4207 - fix races between taking ns_lock and ldlm_lock_change_resource
4208 - fix races updating LOV export open file list
4209 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4210 - fix recovery NULL deref in ldlm_cli_cancel_unused
4211 - fix some DLM races by using new hash table for lock handles (419)
4212 - permit the client to specify desired inodes, at replay
4213 - duplicate requests when we queue them for replay reintegration
4214 - fix last_rcvd offset calculation
4215 - sync after each recovered transaction, so we always make progress
4216 - never, not always, ERESTART requests without transnos
4217 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4219 - skip replay if the MDS says that the client is already connected
4220 - don't check for a recovery-enabled export to match lctl's UUID
4221 - don't INC_USE_COUNT for phantom exports
4222 - don't crash when cleaning up phantom exports (567)
4223 - don't double-finish or set replay data for errored mdc_open requests
4224 - abort requests when they time out, so we don't get old replies
4225 - send/receive replies for AST messages again
4226 - if the client says that it doesn't have the lock, cancel it on the
4228 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4229 mark it as destroyed, it will all work out when decref is called
4230 - fix module use counts (22, 581)
4232 - ASTs now expect a reply (server cancels lock on error reply)
4234 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4237 - fix many simultaneous client startup (392)
4238 - fix dentry->d_it clobbering
4239 - credentials weren't being shipped for readdir/getattr operations
4240 - remove invalid assertions triggered during some concurrent MD
4242 - proper Lustre versions added (336, 389)
4243 - fix memory leak for create error case (398)
4244 - fix LOV locking bug that would get cli/srv out of sync
4245 - fix echo client over LOV (409)
4246 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4247 - fix double-O_EXCL intent crash (424)
4248 - avoid sending multiple lock CANCELs (352)
4250 - MDS can do multi-client recovery (modulo bugs in new code)
4252 - many updates, edits, cleanups
4254 2002-11-18 Phil Schwan <phil@clusterfs.com>
4257 - fix null d_it dereference (346)
4258 - fix full OST/dbench hang (333)
4259 - fix permission problem with file removal (286)
4260 - fix removal of OSCs from LOV when they fail
4261 - fix NULL deref during bulk timeout (214)
4262 - fix problems related to multiple filesystems on one MDS (241)
4263 - fixed serious subtle metadata locking bugs
4264 - free locks on clients when inodes are removed due to memory
4266 - fix inode pointer in lock data (285)
4267 - partial support for multiple MDS on a single host (241)
4268 - data locks weren't cancelled at clear_inode time (290, 311)
4269 - intent locks could lead to unbounded lock growth (205)
4270 - added a maximum lock count, an LRU list, and a flusher
4271 - fix multiple rename (365)
4272 - properly abstracted the echo client
4273 - OSC locked 1 byte too many; fixed
4274 - rewrote brw callback code:
4275 - fixed recovery bugs related to LOVs (306)
4276 - fixed too-many-pages-in-one-write crash (191)
4277 - fixed (again) crash in sync_io_timeout (214)
4278 - probably fixed callback-related race (385)
4280 - Add capability to MDS protocol
4281 - LDLM cancellations and callbacks on different portals
4283 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4286 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4287 - defer open object destruction to close time (601981, 138)
4288 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4289 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4290 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4291 - don't call obd_setattr in ll_file_release for destroyed objects
4292 * protocol change to lustre_msg: move |version| and add |flags|
4293 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4294 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4295 * added replay of create, unlink, link and rename operations during
4296 MDS failover; recovery should be much more robust now
4297 * remove failed OSCs from LOVs (only lov_create uses this so far)
4298 * the lustre-HOWTO was brought (more) up to date (582544)
4300 2002-10-23 Phil Schwan <phil@clusterfs.com>
4303 - in-use dentries weren't being reused properly (617851)
4304 - prevent multiple LDLM setup (599178)
4305 - fix LOV size calculations for truncate (617853)
4306 - fix client handling of MDS intent errors (POSIX)
4307 - fix permission bug in lovstripe.c test (624321)
4308 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4309 - truncate past end of file could corrupt data
4310 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4311 - a race in recovery could return ETIMEDOUT to apps (623947)
4312 - building outside the source directory was fixed
4313 * the lustre-HOWTO was brought (more) up to date (582544)
4314 * major progress was made on recovery functionality
4316 2002-10-10 Phil Schwan <phil@clusterfs.com>
4319 - recovery deadlock fix
4320 - rm -rf causes LBUG fix (617817)
4321 - file open by multiple tasks fix (618962)
4322 - directory permissions bugs (602707 and 620007)
4323 - journal_stop fixed with locking (611313)
4324 - O_APPEND failures resolved (618273, perhaps 614459)
4325 - lconf PATH fix (619770)
4326 - IA64 build fix (621450)
4327 - RPC buffer sizes scale with amount of memory
4329 2002-10-01 Phil Schwan <phil@clusterfs.com>
4332 - locks would be cancelled without throwing away data pages,
4333 resulting in inconsistent data (605627)
4334 - inode attributes were not always being refreshed (605627, 612449)
4335 - lconf now continues to cleanup after lctl reports an error
4336 - MDS now enforces user permissions (602707)
4337 - lprocfs cleanup fixed, but not yet enabled (614157)
4338 - fixed infinite server hang, should a client not respond to an AST
4339 - avoid going into recovery if user calls readlink() with a buffer
4340 that's too small (613941)
4341 - AST RPCs no longer require replies (614867) -- this may be changed
4342 - don't crash server if client sends an IOV that's too big (611336)
4343 - fixed lock conversion deadlock (611892)
4344 - fixed the following of symlinks (614622)
4345 * recovery: the server can remove locks from a client that dies, other
4346 clients can make progress
4347 * more extN patch fixes
4348 * compile-time configurable ptlrpc buffer allocations
4350 - collaborative read cache document
4351 - Lustre Lite Performance CDR document-in-progress
4353 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4356 - fix typo in patch-2.4.18
4358 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4361 - clear ptlrpc request each time in handle_incoming_request()
4362 - unlink of files now destroys the object on the OST
4364 2002-09-19 Peter Braam <braam@clusterfs.com>
4366 * add hard link support
4367 * change obdfile creation method
4368 * kernel patch changed
4370 2002-09-19 Peter Braam <braam@clusterfs.com>
4373 - stack overflow bug in extN fixed
4375 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4377 * documentation updates
4378 - add man pages for config tools
4379 - update tests/README to describe testing with new config tools
4380 - finish metadata API descriptions
4381 * bug fixes and cleanups
4382 - statfs workaround for 16TB limit
4383 - LOV stripe allocation improved, can stripe on subset of OSTs
4384 - LOV file size/IO offset was wrong for files > 4GB in size
4385 - object EA data was being dropped, caused files to be unreadable
4386 - memory overflow with non-LOV OST caused memory corruption
4387 - fixed regression tests to work with new config tools, obdfilter
4388 - fixed bug when directory size became larger than 1 block
4389 - fixed bug (for single client case) when PWD was deleted
4390 - invalidate local directory pages when doing intent-based ops
4391 - avoid LDLM oops when lock callback contained bad data
4393 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4395 * documentation updates
4396 * bug fixes and cleanups
4397 - configuration tools
4400 - 64-bit compile warnings
4401 - 64-bit internal statfs data
4403 * test_brw on persistent OST devices
4405 * lprocfs (disabled)
4407 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4409 * documentation updates
4410 * bug fixes and cleanups
4411 * configuration tools
4413 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4416 * many small fixes to 0_5_4
4417 * io/network handling
4418 * thinkos in MDS operations
4420 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4423 * crucial basic fixes to 0.5.3
4424 * IOR, Iozone work over Elan
4427 2002-08-07 Phil Schwan <phil@clusterfs.com>
4428 * version 0_5_3, our first alpha
4429 * we use the new Portals iovs
4430 * documentation updates
4431 * bug fixes and cleanups
4432 * small changes in the DLM wire protocol
4434 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4435 * version 0_5_1 with some initial stability,
4436 * locking on MD and file I/O.
4437 * documentation updates
4438 * several bug fixes since 0.5.0
4439 * small changes in wire protocol
4441 2002-07-18 Phil Schwan <phil@clusterfs.com>
4443 * delivered as Lustre Light Alpha
4444 * fixed a crash after handling invalid MDS requests
4445 * fixed directory pages for architectures with non-4k pages sizes
4447 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4448 * release version v0_4_4
4449 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4451 * Fixes a number of interruption problems with OST operations.
4452 * Update documentation for portals header changes
4453 * Move all wire protocol structs/defines to lustre_idl.h
4454 * Fixes symlink length bug.
4455 * Add tcpdump to repository.
4457 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4458 * release version v0_4_3
4459 * Fixes statfs for inodes on extN.
4460 * Fixes bug in runtests which would delete /etc/hosts.
4461 * Use 64-bit object IDs wherever possible (not into VFS though)
4462 Remove ost_get_info, which is unused by lustre, and out of date.
4464 2002-07-03 Peter Braam <braam@clusterfs.com>
4465 * release version v0_4_2 Fixes a lookup error (type not passed)
4466 * move forward to head of Portals
4467 * move forward to latest Lustre kernel
4469 2002-06-25 Peter Braam <braam@clusterfs.com>
4470 * release version v0_4_1. Hopefully stable on single node use.