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
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
212 --------------------------------------------------------------------------------
214 2007-08-27 Cluster File Systems, Inc. <info@clusterfs.com>
216 * Support for kernels:
217 2.6.5-7.286 (SLES 9),
218 2.6.9-55.0.2.EL (RHEL 4),
219 2.6.16.46-0.14 (SLES 10),
220 2.6.18-8.1.8.el5 (RHEL 5),
221 2.6.18.8 vanilla (kernel.org)
222 * Client support for unpatched kernels:
223 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
224 2.6.16 - 2.6.22 vanilla (kernel.org)
225 * Due to recently discovered recovery problems, we do not recommend
226 using patchless RHEL 4 clients with this or any earlier release.
227 * Recommended e2fsprogs version: 1.39.cfs8
228 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
233 Description: block reactivating mgc import until all deactivates complete
234 Details : Fix race when failing back MDT/MGS to itself (testing)
236 Severity : enhancement
238 Description: lfs setstripe enhancement
239 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
242 Frequency : mds/oss recovery
244 Description: llog ctxt is refrenced after it has been freed.
245 Details : llog ctxt refcount was added to avoide the race between ctxt free
246 and llog recovery process. Each llog user must hold ctxt refcount
247 before it access the llog. And the llog ctxt can only be freed
248 when its refcount is zero.
250 Severity : enhancement
252 Description: randomly memory allocation failure util
253 Details : Make lustre randomly failed allocating memory for testing purpose.
256 Frequency : only for SLES10
258 Description: Update kernel patch for SLES10 SP1
259 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
261 Severity : enhancement
263 Description: omit set fsid for export NFS
264 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
265 as FS_REQUIRES_DEV for avoid problems with generate fsid.
268 Frequency : after network failures
270 Description: Add sync option to mount_lustre.c
271 Details : Client loses data written to lustre after a network interruption.
273 Severity : enhancement
275 Description: Error message improvement.
276 Details : Merging of two LCONSOLE_ERROR_MSG into one.
278 Severity : enhancement
280 Description: don't use GFP_* in generic Lustre code.
281 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
283 Severity : enhancement
285 Description: obdclass is limited by single OBD_ALLOC(idarray)
286 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
288 Severity : enhancement
290 Description: updated patchess for new RHEL4 kernel
291 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
293 Severity : enhancement
295 Description: warnings with build patchless client with vanila 2.6.19 and up
296 Details : change old ctl_table style and replace ctl_table/ctl_table_header
297 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
299 Severity : enhancement
301 Description: O_DIRECT bypasses client statistics.
302 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
303 read_ahead_stats were updated. Copied stats section from
304 osc_send_oap_rpc() into async_internal().
307 Frequency : only for Cray XT3
309 Description: peer credits not enough on many OST per OSS systems.
310 Details : Use new lnet way to add credits as we need those for pings and ASTs
313 Frequency : only with liblustre
315 Description: Liblustre is not releasing flock locks on file close.
316 Details : Release flock locks on file close.
319 Frequency : only for RHEL4
321 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
322 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
323 Update target file and kernel config.
327 Description: lustre support for RHEL5
328 Details : Add support for RHEL5.
333 Description: ASSERTION(export != NULL) failed in target_handle_connect
334 Details : Assetion hit is result of rare race between disconnect and connet
335 to same nid. target_handle_connect found old connect cockie and
336 tried to reconnect, but can't find export for this cockie.
341 Description: umount blocks forever on error
342 Details : In result of wrong using obd_no_recov and obd_force flags client
343 can hand if cancel or some other requests is lost.
346 Frequency : Only for SLES
348 Description: sanity_quota fail test_1
349 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
350 file, which makes TSTID[2] inunique.
352 Severity : enhancement
354 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
355 Details : Update target/ChangeLog/which_patch .
357 Severity : enhancement
359 Description: jbd statistics
360 Details : Port older jbd statistics patch for sles10
365 Description: lvbo_init failed for resource with missing objects.
366 Details : Fix returning error if we do stat for file with missing/corrupted
367 objects and i_size set to all sum of size of all avaible objects.
368 if we to truncate/write to missing object - it is recreated.
373 Description: Oops in read and write path when failing to allocate lock.
374 Details : Check if lock allocation failed and return error back.
377 Frequency : When flocks are used.
379 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
380 Details : Flock locks might destroy just granted lock if it could be merged
381 with another existing flock, this is done in completion handler,
382 so teach ldlm_cli_enquque_fini that this is a valid case for
388 Description: reply_lock_interpret crash due to race with it and lock cancel.
389 Details : Do not replay locks that are being cancelled. Do not reference
390 locks by their address during replay, just by their handle.
393 Frequency : only with deactivated OSTs
395 Description: lstripe command fails for valid OST index
396 Details : The stripe offset is compared to 'lov->desc.ld_tgt_count'
397 instead of lov->desc.ld_active_tgt_count.
399 Severity : enhancement
401 Description: Build failure against Centos5 (RHEL5)
402 Details : Use getpagesize() instead of PAGE_SIZE.
404 --------------------------------------------------------------------------------
406 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
408 * Support for kernels:
409 2.6.5-7.283 (SLES 9),
410 2.6.9-55.EL (RHEL 4),
411 2.6.16.46-0.14 (SLES 10),
412 2.6.18.8 vanilla (kernel.org)
413 * Client support for unpatched kernels:
414 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
415 2.6.16 - 2.6.22 vanilla (kernel.org)
416 * Due to recently discovered recovery problems, we do not recommend
417 using patchless RHEL 4 clients with this or any earlier release.
418 * Recommended e2fsprogs version: 1.39.cfs8
419 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
420 * Starting with this release, the ldiskfs backing filesystem required
421 by Lustre is now in its own package, lustre-ldiskfs. This package
422 should be installed. It is versioned separately from Lustre and
423 may be released separately in future.
428 Description: server LBUG when shutting down
429 Details : Block umount forever until the mount refcount is zero rather
430 than giving up after an arbitrary timeout.
432 Severity : enhancement
434 Description: add optional extra BUILD_VERSION info
435 Details : add a new environment variable (namely LUSTRE_VERS) which allows
436 to override the lustre version.
439 Frequency : 2.6.18 servers only
441 Description: ll_kern_mount() doesn't release the module reference
442 Details : The ldiskfs module reference count never drops down to 0
443 because ll_kern_mount() doesn't release the module reference.
448 Description: server LBUG when using old ost_num_threads parameter
449 Details : Accept the old ost_num_threads parameter but warn that it
450 is deprecated, and fix an off-by-one error that caused an LBUG.
455 Description: Transient SCSI error results in persistent IO issue
456 Details : iobuf->dr_error is not reinitialized to 0 between two
460 Frequency : sometimes when underlying device returns I/O errors
462 Description: OSTs not going read-only during write failures
463 Details : OSTs are not remounted read-only when the journal commit threads
464 get I/O errors because fsfilt_ext3 calls journal_start/stop()
465 instead of the ext3 wrappers.
468 Frequency : only with 10000 clients or more
470 Description: poor connect scaling with increasing client count
471 Details : Don't run filter_grant_sanity_check for more than 100 exports
472 to improve scaling for large numbers of clients.
475 Frequency : SLES10 only
477 Description: sanity-quota.sh quotacheck failed: rc = -22
478 Details : Quotas cannot be enabled on SLES10.
481 Frequency : liblustre clients only
483 Description: getdirentries does not give error when run on compute nodes
484 Details : getdirentries does not fail when the size specified as an argument
485 is too small to contain at least one entry
487 Severity : enhancement
489 Description: Add LNET router traceability for debug purposes
490 Details : If a checksum failure occurs with a router as part of the
491 IO path, the NID of the last router that forwarded the bulk data
492 is printed so it can be identified.
497 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
498 Details : Due to a race condition, liblustre clients were occasionally
501 Severity : enhancement
503 Description: lfs setstripe use optional parameters instead of postional
506 Severity : enhancement
508 Description: Nanosecond timestamp support for ldiskfs
509 Details : The on-disk ldiskfs filesystem has added support for nanosecond
510 resolution timestamps. There is not yet support for this at
511 the Lustre filesystem level.
514 Frequency : during server recovery
516 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
517 Details : request with rq_no_resend flag not awake l_wait_event if they get a
521 Frequency : nfs export on patchless client
523 Description: connectathon hang when test nfs export over patchless client
524 Details : Disconnected dentry cannot be found with lookup, so we do not need
525 to unhash it or make it invalid
528 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
529 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
530 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
535 Description: open req refcounting wrong on reconnect
536 Details : If reconnect happened between getting open reply from server and
537 call to mdc_set_replay_data in ll_file_open, we will schedule
538 replay for unreferenced request that we are about to free.
539 Subsequent close will crash in variety of ways.
540 Check that request is still eligible for replay in
541 mdc_set_replay_data().
546 Description: disable writes to filesystem when reading health_check file
547 Details : the default for reading the health_check proc file has changed
548 to NOT do a journal transaction and write to disk, because this
549 can cause reads of the /proc file to hang and block HA state
550 checking on a healthy but otherwise heavily loaded system. It
551 is possible to return to the previous behaviour during configure
552 with --enable-health-write.
554 Severity : enhancement
556 Description: 64-bit inode version
557 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
558 to the inode. This will be required for version-based recovery.
563 Description: MDS fails to start if a duplicate client export is detected
564 Details : in some rare cases it was possible for a client to connect to
565 an MDS multiple times. Upon recovery the MDS would detect this
566 and fail during startup. Handle this more gracefully.
568 Severity : enhancement
570 Description: Add -o localflock option to simulate old noflock behaviour.
571 Details : This will achieve local-only flock/fcntl locks coherentness.
576 Description: log_commit_thread vs filter_destroy race leads to crash
577 Details : Take import reference before releasing llog record semaphore
582 Description: Wrong request locking in request set processing
583 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
584 imp_delayed_list, in this place should be used imp_lock.
587 Frequency : when reconnecting
589 Description: Grant leak when OSC reconnect to OST
590 Details : When osc reconnect ost, OST(filter) should check whether it
591 should grant more space to client by comparing fed_grant and
592 cl_avail_grant, and return the granted space to client instead
593 of "new granted" space, because client will call osc_init_grant
594 to update the client grant space info.
597 Frequency : when client reconnects to OST
599 Description: Grant leak when OSC does a resend and replays bulk write
600 Details : When osc reconnect to OST, OST(filter) should clear grant info of
601 bulk write request, because the grant info while be sync between
602 OSC and OST when reconnect, and we should ignore the grant info
603 these of resend/replay write req.
608 Description: Grant space more than avaiable space sometimes.
609 Details : When then OST is about to be full, if two bulk writing from
610 different clients came to OST. Accord the avaliable space of the
611 OST, the first req should be permitted, and the second one
612 should be denied by ENOSPC. But if the seconde arrived before
613 the first one is commited. The OST might wrongly permit second
614 writing, which will cause grant space > avaiable space.
617 Frequency : when client is evicted
619 Description: Grant might be wrongly erased when osc is evicted by OST
620 Details : when the import is evicted by server, it will fork another
621 thread ptlrpc_invalidate_import_thread to invalidate the
622 import, where the grant will be set to 0. While the original
623 thread will update the grant it got when connecting. So if
624 the former happened latter, the grant will be wrongly errased
625 because of this race.
630 Description: Checking Stale with correct fid
631 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
632 to check whether it is stale, because sometimes, we want the
633 enqueue happened anyway, and op_data.fid2 will not be initialized.
635 Severity : enhancement
637 Description: update patchless client
638 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
641 Frequency : only with 2.4 kernel
643 Description: random memory corruption
644 Details : size of struct ll_inode_info is to big for union inode.u and this
645 can be cause of random memory corruption.
650 Description: Memory leak in recovery
651 Details : Lov_mds_md was not free in an error handler in mds_create_object.
652 It should also check obd_fail before fsfilt_start, otherwise if
653 fsfilt_start return -EROFS,(failover mds during mds recovery).
654 then the req will return with repmsg->transno = 0 and rc = EROFS.
655 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
656 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
657 be freed no matter whether fsfilt_commit success or not.
660 Frequency : only with huge numbers of clients
662 Description: Prevents from taking the superblock lock in llap_from_page for
664 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
665 for llap_from_page call in ll_removepage() prevents from taking
666 the superblock lock for a soon died page.
671 Description: Not check open intent error before release open handle
672 Details : in some rare cases, the open intent error is not checked before
673 release open handle, which may cause
674 ASSERTION(open_req->rq_transno != 0), because it tries to release
675 the failed open handle.
680 Description: Set cat log bitmap only after create log success.
681 Details : in some rare cases, the cat log bitmap is set too early. and it
682 should be set only after create log success.
686 Description: Accessing a block bevice can re-enable I/O when Lustre is
687 tearing down a device.
688 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
693 Description: service threads may hog cpus when there are a lot of requests
694 Details : Insert cond_resched to give other threads a chance to use some CPU
699 Description: the cat log was not initialized in recovery
700 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
701 unlink log on mds will not be initialized until mds post
702 recovery. And also in mds post recovery, the unlink log will
703 initialization will be done asynchronausly, so there will be race
704 between add unlink log and unlink log initialization.
708 Description: brw_stats were being printed incorrectly
709 Details : brw_stats were being printed as log2 but all of them were not
710 recorded as log2. Also remove some code duplication arising from
711 filter_tally_{read,write}.
715 Frequency : rare, only in recovery.
716 Description: ASSERTION(req->rq_type != LI_POISON) failed
717 Details : imp_lock should be held while iterating over imp_sending_list for
718 prevent destroy request after get timeout in ptlrpc_queue_wait.
722 Description: replay-single.sh test 52 fails
723 Details : A lock's skiplist need to be cleanup when it being unlinked
724 from its resource list.
729 Description: Short directio read returns full requested size rather than
731 Details : Direct I/O operations should return actual amount of bytes
732 transferred rather than requested size.
734 Severity : enhancement
736 Description: metadata RPC reduction (e.g. for rm performance)
737 Details : decrease the amount of synchronous RPC between clients and servers
738 by canceling conflicing lock before the operation on the client
739 and packing thier handles into the main operation RPC to server.
741 Severity : enhancement
743 Description: add #ifdef HAVE_KERNEL_CONFIG_H
744 Details : kernels from 2.6.19 not need include linux/config.h, but add
745 include linux/autoconf.h in commpiler command line.
747 Severity : enhancement
749 Description: patchless client support for 2.6.22 kernel
750 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
751 removed. In this case we need drop using os depended interface to
752 kmem_cache and use cfs_mem_cache API.
757 Description: fix mal-formatted messages
758 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
762 Frequency : always in liblustre
763 Description: wrong IS_ERR implementation in liblustre.h
764 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
769 Description: Correct condition for output debug message.
770 Details : inode i_nlink equal zero is not enough for output message about
771 disk corruption, i_ctime and i_mode should be also checked.
775 Frequency : always in patchless client
776 Description: add configure check for truncate_complete_page
777 Details : improve checks for exported symbols. This allow run check without
778 sources, but with Module.symvers shipped with kernel distribution.
779 add check for truncate_complete_page used by patchless client.
783 Description: sanity.sh test_77h fails with "test_77h file compare failed"
784 Details : test_77h uses a file which was messed by other test case.
788 Description: Not Check whether lov_tgts is NULL in some lov functions
789 Details : Checking whether lov_tgts is NULL in some functions.
793 Description: replace obdo_alloc() with OBDO_ALLOC macro
794 Details : nothing special is done in obdo_alloc() function, and for
795 debugging purpose, it needs to be replaced with macros.
799 Description: bad return value and errno from fcntl call
800 Details : In liblustre API, errno should be a negative value if error
805 Description: ptlrpc_check_set() LBUG
806 Details : In case of positive reply from server and failed client bulk
807 callback after bulk transfer shouldn't LBUG, but process this
808 request as erroneous.
810 Severity : enhancement
812 Description: VFS operations stats tool.
813 Details : Tool which collects stats by tracking value written in pid,
814 ppid, gid and uses llstat to generate output to plot graph using
816 Updated lustre/utils/Makefile.am
817 Added lustre/utils/ltrack_stats.c
819 Severity : enhancement
821 Description: 2.6.18 server support (lustre 1.6.1)
822 Details : Support for 2.6.18 kernels on the server side.
827 Description: ASSERTION(imp->imp_conn_current) failed
828 Details : an assertion failure is hit if a client node boots and attempts to
829 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
832 Frequency : only for i686
834 Description: 1.4.11 RC1 build fails for RHEL 4, i686
835 Details : Fixed config variable for build.
840 Description: Updated patchess for new RHEL4 kernel
841 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
842 Updated patch jbd-stats-2.6.9.patch
843 Updated patch qsnet-rhel4-2.6.patch
844 Updated patch quota-deadlock-on-pagelock-core.patch
845 Updated patch vfs_intent-2.6-rhel4.patch
846 Updated patch vfs_races-2.6-rhel4.patch
847 Updated series file 2.6-rhel4-titech.series
848 Updated series file 2.6-rhel4.series
849 Updated kernel config files
854 Description: lquota slave complains LBUG when reconnecting with mds
856 Details : quota slave depends qctxt->lqc_import to send its quota request.
857 This pointer will be invalid if mds did failover or broke its
858 connect to osts, which leads to LBUG.
861 Frequency : when qunit size is too small(less than 20M)
863 Description: write is stopped by improper -EDQUOT
864 Details : If the master is busy and qunit size is small enough(let's say 1M),
865 the slave can not get quota from master on time, which will lead
866 slave to trigger a -EQUOTA to client.
871 Description: Deadlock during metadata tests
872 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
873 called with the per-dentry lock held.
876 Frequency : SLES9 only
878 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
879 Details : set CONFIG_NR_CPUS to 128 instead of 8.
881 Severity : enhancement
883 Description: remove fs_prep_san_write operation and related patches
884 Details : remove the ext3-san-jdike patches which are no longer useful.
889 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
890 Details : a disk corruption can cause the mballoc code to assert on a
891 double free or other extent corruptions. Handle these with
892 ext3_error() instead of assertions.
895 Frequency : only with mballoc3 code and deep extent trees
897 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
898 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
899 can cause the filesystem to be remounted read-only.
904 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
905 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
906 few external users of this ioctl number is lfsck's call to
907 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
908 Add the old ioctl number to the handler so both old and new
913 Description: parallel lock callbacks
914 Details : Instead of sending blocking and completion callbacks as separated
915 requests, adding them to a set and sending in parallel.
919 Description: Disable most debugging by default
920 Details : To improve performance, disable most logging (for debug purposes)
921 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
922 default, and HA includes fewer messages.
926 Frequency : only run on patchless client.
927 Description: use do_facet on sanity.sh for test handling recoverables errors
928 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
932 Description: hash tables for lists of nids, connections and uuids
933 Details : Hash tables noticeably help when a lot of clients connect to a
934 server, to faster identify duplicate connections or reconnects,
935 also to faster find export to evict in manual eviction case.
939 Description: Sometimes, when the server evict a client, and the client will
940 not be evicted as soon as possible.
941 Details : In enqueue req, the error was returned by intent, instead of
942 rq_status which make ptlrpc layer not detect this error, and
943 does not evict the client. So enqueue error should be returned
947 Frequency : only at startup
949 Description: Delay client connections to MDT until fist MDT->OST connect
950 Details : If a client tried to create a new file before the MDT had
951 connected to any OSTs, the create would return EIO. Now
952 the client will simply block until the MDT connects to the
953 first OST and the create can succeed.
956 Frequency : only for SLES9
958 Description: Routinely utilize latest Quadrics drivers in CFS releases
959 Details : Update patch qsnet-suse-2.6.patch.
962 Frequency : only for sles10
964 Description: Update patches for SLES 10 SP1 kernel.
965 Details : Update the patch vfs_intent-2.6-sles10.patch.
970 Description: Routinely utilize latest Quadrics drivers in CFS releases
971 Details : Update patch qsnet-rhel4-2.6.patch.
974 Frequency : at statup only
976 Description: mds_lov_synchronize race leads to various problems
977 Details : simultaneous MDT->OST connections at startup can cause the
978 sync to abort, leaving the OSC in a bad state.
983 Description: Using wrong pointer in osc_brw_prep_request
984 Details : Access to array[-1] can produce panic if kernel compiled with
985 CONFIG_PAGE_ALLOC enabled
987 Severity : enhancement
989 Description: Async OSC create to avoid the blocking unnecessarily.
990 Details : If a OST has no remain object, system will block on the creating
991 when need to create a new object on this OST. Now, ways use
992 pre-created objects when available, instead of blocking on an
993 empty osc while others are not empty. If we must block, we block
994 for the shortest possible period of time.
998 Frequency : only in recovery
999 Description: Mark OST as early accessible if his start SYNC.
1000 Details : osc_precreate return flag early accessible if oscc marked as
1001 OSCC_FLAG_SYNC_IN_PROGRESS.
1006 Description: Sometimes precreate code can triger create object on wrong ost
1007 Details : Wrong protected or not not restored variables aftre precreate loop
1008 can produce creation object on wrong ost.
1011 Frequency : oss recovery
1013 Description: llog_commit_thread cleanup should sync with llog_commit_thread
1015 Details : llog_commit_thread_count should be synced between llog_commit
1016 start and cleanup, so new llog_commit thread should not be started
1017 when llog_commit threads being stopped to avoid accessing some
1020 Severity : enhancement
1022 Description: Add printing inode info into message about error in writepage.
1024 --------------------------------------------------------------------------------
1026 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
1031 Frequency : on some architectures
1033 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
1034 Details : Uninitialized flags sometimes cause configuration commands to
1038 Frequency : patchless clients only
1040 Description: missing __iget() symbol export
1041 Details : The __iget() symbol export is missing. To avoid the need for
1042 this on patchless clients the deathrow inode reaper is turned
1043 off, and we depend on the VM to clean up old inodes. This
1044 dependency was during via the fix for bug 12181.
1049 Description: sanity.sh fail: test_52b
1050 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
1051 incorrect inode's flags to client.
1053 --------------------------------------------------------------------------------
1055 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
1057 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
1058 INTEROPERATE with older versions automatically. In many cases a
1059 special upgrade step is needed. Please read the
1060 user documentation before upgrading any part of a 1.4.x system.
1061 * WARNING: Lustre configuration and startup changes are required with
1062 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
1064 * Support for kernels:
1065 2.4.21-47.0.1.EL (RHEL 3),
1066 2.6.5-7.283 (SLES 9),
1067 2.6.9-42.0.10.EL (RHEL 4),
1068 2.6.12.6 vanilla (kernel.org),
1069 2.6.16.27-0.9 (SLES10)
1070 * Client support for unpatched kernels:
1071 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
1072 2.6.16 - 2.6.19 vanilla (kernel.org),
1073 2.6.9-42.0.8EL (RHEL 4)
1074 * Recommended e2fsprogs version: 1.39.cfs6
1075 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1078 Severity : enhancement
1080 Description: Async OSC create to avoid the blocking unnecessarily.
1081 Details : If a OST has no remain object, system will block on the creating
1082 when need to create a new object on this OST. Now, ways use
1083 pre-created objects when available, instead of blocking on an
1084 empty osc while others are not empty. If we must block, we block
1085 for the shortest possible period of time.
1087 Severity : enhancement
1089 Description: MountConf
1090 Details : Lustre configuration is now managed via mkfs and mount
1091 commands instead of lmc and lconf. New obd types (MGS, MGC)
1092 are added for dynamic configuration management. See
1093 https://mail.clusterfs.com/wikis/lustre/MountConf for
1096 Severity : enhancement
1098 Description: dynamic OST addition
1099 Details : OSTs can now be added to a live filesystem
1101 Severity : enhancement
1103 Description: startup order invariance
1104 Details : MDTs and OSTs can be started in any order. Clients only
1105 require the MDT to complete startup.
1107 Severity : enhancement
1109 Description: parallel, asynchronous orphan cleanup
1110 Details : orphan cleanup is now performed in separate threads for each
1111 OST, allowing parallel non-blocking operation.
1113 Severity : enhancement
1115 Description: optimized stripe assignment
1116 Details : stripe assignments are now made based on ost space available,
1117 ost previous usage, and OSS previous usage, in order to try
1118 to optimize storage space and networking resources.
1120 Severity : enhancement
1122 Description: Permanently set tunables
1123 Details : All writable /proc/fs/lustre tunables can now be permanently
1124 set on a per-server basis, at mkfs time or on a live
1127 Severity : enhancement
1129 Description: Lustre message v2
1130 Details : Add lustre message format v2.
1132 Severity : enhancement
1134 Description: client OST exclusion list
1135 Details : Clients can be started with a list of OSTs that should be
1136 declared "inactive" for known non-responsive OSTs.
1140 Description: ENOENT returned for valid filehandle during dbench.
1141 Details : Check if a directory has children when invalidating dentries
1142 associated with an inode during lock cancellation. This fixes
1143 an incorrect ENOENT sometimes seen for valid filehandles during
1144 testing with dbench.
1147 Frequency : SFS test only (otherwise harmless)
1149 Description: SPEC SFS validation failure on NFS v2 over lustre.
1150 Details : Changes the blocksize for regular files to be 2x RPC size,
1151 and not depend on stripe size.
1153 Severity : enhancement
1155 Description: fine-grained SMP locking inside DLM
1156 Details : Improve DLM performance on SMP systems by removing the single
1157 per-namespace lock and replace it with per-resource locks.
1159 Severity : enhancement
1161 Description: don't hold multiple extent locks at one time
1162 Details : To avoid client eviction during large writes, locks are not
1163 held on multiple stripes at one time or for very large writes.
1164 Otherwise, clients can block waiting for a lock on a failed OST
1165 while holding locks on other OSTs and be evicted.
1167 Severity : enhancement
1169 Description: Multiple MD RPCs in flight.
1170 Details : Further unserialise some read-only MDT RPCs - learn about intents.
1171 To avoid overly-overloading MDT, introduce a limit on number of
1172 MDT RPCs in flight for a single client and add /proc controls
1173 to adjust this limit.
1175 Severity : enhancement
1177 Description: client read/write statistics
1178 Details : Add client read/write call usage stats for performance
1179 analysis of user processes.
1180 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
1181 file access. extents_stats shows chunk size distribution.
1182 extents_stats_per_process show chunk size distribution per
1185 Severity : enhancement
1187 Description: per-client statistics on server
1188 Details : Add ldlm and operations statistics for each client in
1189 /proc/fs/lustre/mds|obdfilter/*/exports/
1191 Severity : enhancement
1193 Description: improved MDT statistics
1194 Details : Add detailed MDT operations statistics in
1195 /proc/fs/lustre/mds/*/stats
1197 Severity : enhancement
1199 Description: VFS operations stats
1200 Details : Add client VFS call stats, trackable by pid, ppid, or gid
1201 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1206 Description: Fix client-side osc byte counters
1207 Details : The osc read/write byte counters in
1208 /proc/fs/lustre/osc/*/stats are now working
1211 Frequency : always as root on SLES
1213 Description: Failure of copying files with lustre special EAs.
1214 Details : Client side always return success for setxattr call for lustre
1215 special xattr (currently only "trusted.lov").
1220 Description: Refcount LNET uuids
1221 Details : The global LNET uuid list grew linearly with every startup;
1222 refcount repeated list entries instead of always adding to
1225 Severity : enhancement
1227 Description: Dynamic service threads
1228 Details : Within a small range, start extra service threads
1229 automatically when the request queue builds up.
1232 Frequency : mixed-endian client/server environments
1234 Description: mixed-endian crashes
1235 Details : The new msg_v2 system had some failures in mixed-endian
1238 Severity : enhancement
1240 Description: Easy OST removal
1241 Details : OSTs can be permanently deactivated with e.g. 'lctl
1242 conf_param lustre-OST0001.osc.active=0'
1244 Severity : enhancement
1246 Description: MGS proc entries
1247 Details : Added basic proc entries for the MGS showing what filesystems
1250 Severity : enhancement
1252 Description: provide MGS failover
1253 Details : Added config lock reacquisition after MGS server failover.
1255 Severity : enhancement
1257 Description: add Linux 2.4 support
1258 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1262 Description: a large application tries to do I/O to the same resource and dies
1263 in the middle of it.
1264 Details : Check the req->rq_arrival time after the call to
1265 ost_brw_lock_get(), but before we do anything about
1266 processing it & sending the BULK transfer request. This
1267 should help move old stale pending locks off the queue as
1268 quickly as obd_timeout.
1271 Frequency : when an incorrect nid is specified during startup
1273 Description: ptlrpc connect to non-existant node causes kernel crash
1274 Details : LNET can't be re-entered from an event callback, which
1275 happened when we expire a message after the export has been
1276 cleaned up. Instead, hand the zombie cleanup off to another
1279 Severity : enhancement
1281 Description: plain/inodebits lock performance improvement
1282 Details : Grouping plain/inodebits in granted list by their request modes
1283 and bits policy, thus improving the performance of search through
1287 Frequency : only if OST filesystem is corrupted
1289 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1290 Details : for a short time RPCs with bulk IO are in the replay list,
1291 but replay of bulk IOs is unimplemented. If the OST filesystem
1292 is corrupted due to disk cache incoherency and then replay is
1293 started it is possible to trip an assertion. Avoid putting
1294 committed RPCs into the replay list at all to avoid this issue.
1297 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1299 Description: System hang on startup
1300 Details : This bug allowed the liblustre (e.g. catamount) client to
1301 return to the app before handling all startup RPCs. This
1302 could leave the node unresponsive to lustre network traffic
1303 and manifested as a server ptllnd timeout.
1305 Severity : enhancement
1307 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1308 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1309 boolean to control whether to print peer debug info when a
1310 client's RPC times out.
1313 Frequency : only for kernels with patches from Lustre below 1.4.3
1315 Description: Remove old rdonly API
1316 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1319 Frequency : only for devices with external journals
1321 Description: Set external device read-only also
1322 Details : During a commanded failover stop, we set the disk device
1323 read-only while the server shuts down. We now also set any
1324 external journal device read-only at the same time.
1327 Frequency : when upgrading from 1.4 while trying to change parameters
1329 Description: The wrong (new) MDC name was used when setting parameters for
1330 upgraded MDT's. Also allows changing of OSC (and MDC)
1331 parameters if --writeconf is specified at tunefs upgrade time.
1334 Frequency : when setting specific ost indicies
1336 Description: QOS code breaks on skipped indicies
1337 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1338 created with --index need not be sequential.
1340 Severity : enhancement
1342 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1343 Details : The uninit_groups feature works in conjunction with the kernel
1344 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1345 up the pass1 processing of e2fsck. This is a read-only feature
1346 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1347 mount filesystems that have had this feature enabled.
1349 Severity : enhancement
1351 Description: Improve multi-block allocation algorithm to avoid fragmentation
1352 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1353 allocation locality and avoid filesystem fragmentation.
1355 ------------------------------------------------------------------------------
1357 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1359 * Support for kernels:
1360 2.4.21-47.0.1.EL (RHEL 3)
1361 2.6.5-7.283 (SLES 9)
1362 2.6.9-42.0.10.EL (RHEL 4)
1363 2.6.12.6 vanilla (kernel.org)
1364 2.6.16.27-0.9 (SLES 10)
1365 * Recommended e2fsprogs version: 1.39.cfs5
1367 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1371 Frequency : occasional, depends on client load and configuration
1372 Bugzilla : 12181, 12203
1373 Description: data loss for recently-modified files
1375 Details : In some cases it is possible that recently written or created
1376 files may not be written to disk in a timely manner (this should
1377 normally be within 30s unless client IO load is very high).
1378 The problem appears as zero-length files or files that are a
1379 multiple of 1MB in size after a client crash or client eviction
1380 that are missing data at the end of the file.
1382 This problem is more likely to be hit on clients where files are
1383 repeatedly created and unlinked in the same directory, clients
1384 have a large amount of RAM, have many CPUs, the filesystem has
1385 many OSTs, the clients are rebooted frequently, and/or the files
1386 are not accessed by other nodes after being written.
1388 The presence of the problem can be detected by looking at
1389 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1390 smaller than the second (nr_unused) then dirty files will not
1391 be flushed automatically to disk. "sync; sleep 10" should be
1392 run several times on the node before unmounting it to update
1393 Lustre (this is also safe to run on nodes without this problem).
1395 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1396 that can cause this same problem, so customers using that kernel
1397 also need to update the kernel in addition to Lustre. In order
1398 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1400 It is normal that files written just before a client crash (less
1401 than 30s) may not yet have been flushed to disk, even for local
1405 Frequency : frequent on thin XT3 nodes
1407 Description: UUID collision on thin XT3 Linux nodes
1408 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1409 randomly, since we relied on an insufficiently-seeded PRNG.
1414 Description: OSS hangs after "All ost request buffers busy"
1415 Details : A deadlock between quota and journal operations caused OSS
1416 hangs after printing "All ost request buffers busy."
1419 Frequency : always on liblustre builds
1421 Description: Cleanup compiler warnings on liblustre
1424 Frequency : always on liblustre builds on XT3
1426 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1430 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1431 > 32000 subdirectories
1432 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1433 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1434 subdirectory count crosses 32000. This will aid e2fsck to
1435 correctly handle more than 32000 subdirectories.
1438 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1440 Description: System hang on startup
1441 Details : This bug allowed the liblustre (e.g. catamount) client to
1442 return to the app before handling all startup RPCs. This
1443 could leave the node unresponsive to lustre network traffic
1444 and manifested as a server ptllnd timeout.
1446 Severity : enhancement
1448 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1449 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1450 boolean to control whether to print peer debug info when a
1451 client's RPC times out.
1456 Description: make O_SYNC working on 2.6 kernels
1457 Details : 2.6 kernels use different method for mark pages for write,
1458 so need add a code to lustre for O_SYNC work.
1463 Description: Failure to close file and release space on NFS
1464 Details : Put inode details into lock acquired in ll_intent_file_open.
1465 Use mdc_intent_lock in ll_intent_open to properly
1466 detect all kind of errors unhandled by mdc_enqueue.
1471 Description: proc file read during shutdown sometimes raced obd removal,
1473 Details : Add lock to prevent obd access after proc file removal.
1476 Frequency : Only for files larger than 4GB on 32-bit clients.
1478 Description: improperly doing page alignment of locks
1479 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1480 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1485 Description: RPCs being resent when they shouldn't be
1486 Details : Some RPCs that should not be resent are being resent. This
1487 can cause inconsistencies in the RPC state machine. Do not
1488 resend such requests.
1491 Frequency : rare, only with NFS export
1493 Description: Crash on NFS re-export node
1494 Details : under very unusual load conditions an assertion is hit in
1495 ll_intent_file_open()
1498 Frequency : only if OST filesystem is corrupted
1500 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1501 Details : for a short time RPCs with bulk IO are in the replay list,
1502 but replay of bulk IOs is unimplemented. If the OST filesystem
1503 is corrupted due to disk cache incoherency and then replay is
1504 started it is possible to trip an assertion. Avoid putting
1505 committed RPCs into the replay list at all to avoid this issue.
1510 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1511 Details : Large single O_DIRECT read and write calls can fail to allocate
1512 a sufficiently large buffer to process the request. In case of
1513 allocation failure the allocation is retried with a smaller
1514 buffer and broken into smaller requests.
1516 Severity : enhancement
1518 Description: Add -o localflock option to simulate old noflock behaviour.
1519 Details : This will achieve local-only flock/fcntl locks coherentness.
1524 Description: versioning check is incomplete
1525 Details : Checking the version difference of client vs. server, report
1526 error if the gap is too big.
1531 Description: add support PG_writeback bit
1532 Details : add support for PG_writeback bit for Lustre, for more carefull
1533 work with page cache in 2.6 kernel. This also fix some deadlocks
1534 and remove hack for work O_SYNC with 2.6 kernel.
1536 Severity : enhancement
1538 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1539 Details : The uninit_groups feature works in conjunction with the kernel
1540 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1541 up the pass1 processing of e2fsck. This is a read-only feature
1542 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1543 mount filesystems that have had this feature enabled.
1545 Severity : enhancement
1547 Description: Improve multi-block allocation algorithm to avoid fragmentation
1548 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1549 allocation locality and avoid filesystem fragmentation.
1551 ------------------------------------------------------------------------------
1553 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1555 * Support for kernels:
1556 2.6.9-42.0.3.EL (RHEL 4)
1557 2.6.5-7.276 (SLES 9)
1558 2.4.21-47.0.1.EL (RHEL 3)
1559 2.6.12.6 vanilla (kernel.org)
1560 2.6.16.21-0.8 (SLES10)
1561 * Recommended e2fsprogs version: 1.39.cfs2-0
1563 * The backwards-compatible /proc/sys/portals symlink has been removed
1564 in this release. Before upgrading, please ensure that you change
1565 any configuration scripts or /etc/sysctl.conf files that access
1566 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1567 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1568 in advance of the upgrade on any system running Lustre 1.4.6 or
1569 newer, since /proc/sys/lnet was added in that version.
1570 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1574 Frequency : only when quota is used
1576 Description: avoid scanning export list for quota master
1577 Details : Change the algorithms to avoid scanning export list in order
1578 to improve the efficiency.
1581 Frequency : MDS failover only, very rarely
1583 Description: "went back in time" messages on mds failover
1584 Details : The greatest transno may be lost when the current operation
1585 finishes with an error (transno==0) and the client's last_rcvd
1586 record is over-written. Save the greatest transno in the
1587 mds_last_transno for this case.
1590 Frequency : always for specific kernels and striping counts
1592 Description: client may get "Matching packet too big" without ACL support
1593 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1594 when trying to access files in certain configurations. The
1595 clients should in fact be denied when mounting because they do
1596 not understand ACLs.
1599 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1601 Description: many clients connecting with IO in progress causes connect timeouts
1602 Details : Avoid synchronous journal commits to avoid delays caused by many
1603 clients connecting/disconnecting when bulk IO is in progress.
1604 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1605 OST_IO_PORTAL to avoid delays behind potentially many pending
1609 Frequency : occasionally with multiple writers to a single file
1611 Description: shared writes to file may result in wrong size reported by stat()
1612 Details : Allow growing of kms when extent lock is cancelled
1615 Frequency : always with random mmap IO to multi-striped file
1617 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1618 Details : Only if the hole is at the end of OST object so that kms is too
1619 small. Fix is to increase kms accordingly in ll_nopage.
1622 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1624 Description: writes to a missing object would leak memory on the OST
1625 Details : If there is an inconsistency between the MDS and OST filesystems,
1626 such that the MDS references an object that doesn't exist, writes
1627 to that object will leak memory due to incorrect cleanup in the
1628 error handling path, eventually running out of memory on the OST.
1633 Description: Creating too long symlink causes lustre errors
1634 Details : Check symlink and name lengths before sending requests to MDS.
1637 Frequency : only if flock is enabled (not on by default)
1639 Description: posix locks not released on fd closure on 2.6.9+
1640 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1641 kernels, this caused such locks not to be released on fd close and
1642 then assertions on fs unmount about still used locks.
1645 Frequency : MDS failover only, very rarely
1647 Description: clients may get ASSERTION(granted_lock != NULL)
1648 Details : When request was taking a long time, and a client was resending
1649 a getattr by name lock request. The were multiple lock requests
1650 with the same client lock handle and
1651 mds_getattr_name->fixup_handle_for_resent_request found one of the
1652 lock handles but later failed with ASSERTION(granted_lock != NULL).
1657 Description: handle->h_buffer_credits > 0, assertion failure
1658 Details : h_buffer_credits is zero after truncate, causing assertion
1659 failure. This patch extends the transaction or creates a new
1663 Frequency : NFS re-export or patchless client
1664 Bugzilla : 11179, 10796
1665 Description: Crash on NFS re-export node (__d_move)
1666 Details : We do not want to hash the dentry if we don't have a lock.
1667 But if this dentry is later used in d_move, we'd hit uninitialised
1668 list head d_hash, so we just do this to init d_hash field but
1669 leave dentry unhashed.
1672 Frequency : NFS re-export or patchless client
1674 Description: NFS exports has problem with symbolic link
1675 Details : lustre client didn't properly install dentry when re-exported
1676 to NFS or running patchless client.
1679 Frequency : NFS re-export or patchless client
1681 Description: Various nfs/patchless fixes.
1682 Details : fixes reuse disconected alias for lookup process - this fixes
1683 warning "find_exported_dentry: npd != pd",
1684 fix permission error with open files at nfs.
1685 fix apply umask when do revalidate.
1688 Frequency : occasional
1690 Description: Crash on NFS re-export node
1691 Details : calling clear_page() on the wrong pointer triggered oops in
1692 generic_mapping_read().
1695 Frequency : rarely, using O_DIRECT IO
1697 Description: unaligned directio crashes client with LASSERT
1698 Details : check for unaligned buffers before trying any requests.
1701 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1703 Description: stale data returned from RAID cache
1704 Details : If only a small amount of IO is done to the RAID device before
1705 reading it again it is possible to get stale data from the RAID
1706 cache instead of reading it from disk.
1709 Frequency : always for sles10 kernel
1711 Description: sles10 support
1712 Details : ll_follow_link: compile fixes and using of nd_set_link
1713 under newer kernels.
1716 Frequency : depends on arch, kernel and compiler version, always on sles10
1719 Description: recursive or deep enough symlinks cause stack overflow
1720 Details : getting rid of large stack-allocated variable in
1724 Frequency : depends on hardware
1726 Description: lustre write performance loss in the SLES10 kernel
1727 Details : the performance loss is caused by using of write barriers in the
1728 ext3 code. The SLES10 kernel turns barrier support on by
1729 default. The fix is to undo that change for ldiskfs.
1731 ------------------------------------------------------------------------------
1733 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1735 * Support for kernels:
1736 2.6.9-42.0.3EL (RHEL 4)
1737 2.6.5-7.276 (SLES 9)
1738 2.4.21-47.0.1.EL (RHEL 3)
1739 2.6.12.6 vanilla (kernel.org)
1743 Frequency : quota enabled and large files being deleted
1745 Description: releasing more than 4GB of quota at once hangs OST
1746 Details : If a user deletes more than 4GB of files on a single OST it
1747 will cause the OST to spin in an infinite loop. Release
1748 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1753 Description: statfs data retrieved from /proc may be stale or zero
1754 Details : When reading per-device statfs data from /proc, in the
1755 {kbytes,files}_{total,free,avail} files, it may appear
1756 as zero or be out of date.
1759 Frequency : systems with MD RAID1 external journal devices
1761 Description: lconf's call to blkid is confused by RAID1 journal devices
1762 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1763 of returning all block devices that match the journal UUID.
1766 Frequency : always, for aggregate stripe size over 4GB
1768 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1769 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1770 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1773 Frequency : always if "lfs find" used on a local file/directory
1775 Description: "lfs find" segfaults if used on a local file/directory
1776 Details : The case where a directory component was not specified wasn't
1777 handled correctly. Handle this properly.
1780 Frequency : always on ppc64
1782 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1783 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1784 perform endianness conversion.
1787 Frequency : rarely (truncate to non-zero file size after write under load)
1788 Bugzilla : 10730, 10687
1789 Description: Files padded with zeros to next 4K multiple
1790 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1791 it is possible that files that are truncated to a non-zero size
1792 immediately after being written are filled with zero bytes beyond
1793 the truncated size. No file data is lost.
1795 Severity : enhancement
1797 Description: Allow recovery/failover for liblustre clients.
1798 Details : liblustre clients were unaware of failover configurations until
1801 Severity : enhancement
1803 Description: user file locks should fail when not mounting with flock option
1804 Details : Set up an error-returning stub in ll_file_operations.lock field
1805 to prevent incorrect behaviour when client is mounted without
1806 flock option. Also, set up properly f_op->flock field for
1810 Frequency : always on ia64
1812 Description: "lfs df" loops on printing out MDS statfs information
1813 Details : The obd_ioctl_data was not initialized and in some systems
1814 this caused a failure during the ioctl that did not return
1815 an error. Initialize the struct and return an error on failure.
1818 Frequency : SLES 9 only
1820 Description: Error of copying files with lustre special EAs as root
1821 Details : Client side always return success for setxattr call for lustre
1822 special xattr (currently only "trusted.lov").
1825 Frequency : rarely on clusters with both ia64+i386 clients
1827 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1828 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1829 clients are doing concurrent writes to the same file it is
1830 possible that the ia64 clients may LASSERT because the OST
1831 extent locks are not PAGE_SIZE aligned. Ensure that grown
1832 locks are always aligned on the request boundary.
1835 Frequency : specific use, occasional
1837 Description: Overwriting in use executable truncates on-disk binary image
1838 Details : If one node attempts to overwrite an executable in use by
1839 another node, we now correctly return ETXTBSY instead of
1840 truncating the file.
1842 Severity : enhancement
1844 Description: Async OSC create to avoid the blocking unnecessarily.
1845 Details : If a OST has no remain object, system will block on the creating
1846 when need to create a new object on this OST. Now, ways use
1847 pre-created objects when available, instead of blocking on an
1848 empty osc while others are not empty. If we must block, we block
1849 for the shortest possible period of time.
1854 Description: chmod on Lustre root is propagated to other clients
1855 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1856 invalid by the follow_mount time.
1859 Frequency : liblustre clients only
1861 Description: Race in 'instant cancel' lock handling could lead to such locks
1862 never to be granted in case of SMP MDS
1863 Details : Do not destroy not yet granted but cbpending locks in
1867 Frequency : replay/resend of open
1869 Description: non null lock assetion failure in mds_intent_policy
1870 Details : Trying to replay/resend lockless open requests resulted in
1871 mds_open() returning 0 with no lock. Now it sets a flag if
1872 a lock is going to be returned.
1874 Severity : enhancement
1876 Description: Checksum enhancements
1877 Details : New checksum enhancements allow for resending RPCs that failed
1880 Severity : enhancement
1882 Description: Tunables on number of dirty pages in cacche
1883 Details : Allow to set limit on number of dirty pages cached.
1888 Description: client crash on unmount - lock still has references
1889 Details : In some error handling cases it was possible to leak a lock
1890 reference on a client while accessing a file. This was not
1891 harmful to the client during operation, but would cause the
1892 client to crash when the filesystem is unmounted.
1895 Frequency : specific case, rare
1897 Description: ETXTBSY on mds though file not in use by client
1898 Details : ETXTBSY is no longer incorrectly returned when attempting to
1899 chmod or chown a directory that the user previously tried to
1900 execute or a currently-executing binary.
1903 Frequency : extremely rare except on liblustre-based clients
1905 Description: Lustre space not freed when files are deleted
1906 Details : Clean up open-unlinked files after client eviction. Previously
1907 the unlink was skipped and the files remained as orphans.
1912 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1913 Details : In some cases it was possible to send a blocking callback to
1914 a client doing a glimpse, even though that client didn't get
1915 a lock granted. When the glimpse lock is cancelled on the OST
1916 the freed lock is left on the waiting list and corrupted the list.
1919 Frequency : all core dumps
1921 Description: Broke core dumps to lustre
1922 Details : Negative dentry may be unhashed if parent does not have UPDATE
1923 lock, but some callers, e.g. do_coredump, expect dentry to be
1924 hashed after successful create, hash it in ll_create_it.
1926 ------------------------------------------------------------------------------
1928 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1930 * Support for kernels:
1931 2.6.9-42.0.2.EL (RHEL 4)
1932 2.6.5-7.276 (SLES 9)
1933 2.4.21-40.EL (RHEL 3)
1934 2.6.12.6 vanilla (kernel.org)
1938 Frequency : always on RHEL 3
1940 Description: Number of open files grows over time
1941 Details : The number of open files grows over time, whether or not
1942 Lustre is started. This was due to a filp leak introduced
1943 by one of our kernel patches.
1945 ------------------------------------------------------------------------------
1947 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1949 * Support for kernels:
1950 2.6.9-42.EL (RHEL 4)
1951 2.6.5-7.267 (SLES 9)
1952 2.4.21-40.EL (RHEL 3)
1953 2.6.12.6 vanilla (kernel.org)
1958 Bugzilla : 5719, 9635, 9792, 9684
1959 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1960 Details : If a server is under heavy load and cannot reply to new
1961 connection requests before the client resends the (re)connect,
1962 the connection handling code can behave badly if two service
1963 threads are concurrently handing separate (re)connections from
1964 the same client. Add better locking to the connection handling
1965 code, and ensure that only a single connection will be processed
1966 for a given client UUID, even if the lock is dropped.
1968 Severity : enhancement
1970 Description: add TCP zero-copy support to kernel
1971 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1972 sends if the hardware supports scatter-gather and checksumming.
1973 This allows socklnd to do client-write and server-read more
1974 efficiently and reduce CPU utilization from skbuf copying.
1977 Frequency : only if NFS exporting from client
1979 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1980 Details : If setting attributes on a file created under NFS that had
1981 never been opened it would be possible to oops the client
1982 if the file had no objects.
1986 Bugzilla : 9326, 10402, 10897
1987 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1988 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1989 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1990 before ptlrpcd_wake() was called and the request was freed.
1993 Frequency : always for liblustre
1995 Description: liblustre client does MDS+OSTs setattr RPC for each write
1996 Details : When doing a write from a liblustre client, the client
1997 incorrectly issued an RPC to the MDS and each OST the file was
1998 striped over in order to update the timestamps. When writing
1999 with small chunks and many clients this could overwhelm the MDS
2000 with RPCs. In all cases it would slow down the write because
2001 these RPCs are unnecessary.
2003 Severity : enhancement
2005 Description: allow number of MDS service threads to be changed at module load
2006 Details : It is now possible to change the number of MDS service threads
2007 running. Adding "options mds mds_num_threads={N}" to the MDS's
2008 /etc/modprobe.conf will set the number of threads for the next
2009 time Lustre is restarted (assuming the "mds" module is also
2010 reloaded at that time). The default number of threads will
2011 stay the same, 32 for most systems.
2016 Description: OST crash if filesystem is unformatted or corrupt
2017 Details : If an OST is started on a device that has never been formatted
2018 or if the filesystem is corrupt and cannot even mount then the
2019 error handling cleanup routines would dereference a NULL pointer.
2024 Description: NULL pointer deref in llap_from_page.
2025 Details : get_cache_page_nowait can return a page with NULL (or otherwise
2026 incorrect) mapping if the page was truncated/reclaimed while it was
2027 searched for. Check for this condition and skip such pages when
2028 doing readahead. Introduce extra check to llap_from_page() to
2029 verify page->mapping->host is non-NULL (so page is not anonymous).
2032 Frequency : Sometimes when using sys_sendfile
2034 Description: "page not covered by a lock" warnings from ll_readpage
2035 Details : sendfile called ll_readpage without right page locks present.
2036 Now we introduced ll_file_sendfile that does necessary locking
2037 around call to generic_file_sendfile() much like we do in
2041 Frequency : with certain MDS communication failures at client mount time
2043 Description: NULL pointer deref after failed client mount
2044 Details : a client connection request may delayed by the network layer
2045 and not be sent until after the PTLRPC layer has timed out the
2046 request. If the client fails the mount immediately it will try
2047 to clean up before the network times out the request. Add a
2048 reference from the request import to the obd device and delay
2049 the cleanup until the network drops the request.
2052 Frequency : occasionally during client (re)connect
2054 Description: assertion failure during client (re)connect
2055 Details : processing a client connection request may be delayed by the
2056 client or server longer than the client connect timeout. This
2057 causes the client to resend the connection request. If the
2058 original connection request is replied in this interval, the
2059 client may trip an assertion failure in ptlrpc_connect_interpret()
2060 which thought it would be the only running connect process.
2063 Frequency : only with obd_echo servers and clients that are rebooted
2065 Description: kernel BUG accessing uninitialized data structure
2066 Details : When running an obd_echo server it did not start the ping_evictor
2067 thread, and when a client was evicted an uninitialized data
2068 structure was accessed. Start the ping_evictor in the RPC
2069 service startup instead of the OBD startup.
2071 Severity : enhancement
2072 Bugzilla : 10193 (patchless)
2073 Description: Remove dependency on various unexported kernel interfaces.
2074 Details : No longer need reparent_to_init, exit_mm, exit_files,
2075 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
2078 Frequency : rare (only users of deprecated and unsupported LDAP config)
2080 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
2081 Details : LDAP apparently contains 'lustreName' attributes instead of
2082 'name'. A simple remapping of the name is sufficient.
2085 Frequency : rare (only with non-default dump_on_timeout debug enabled)
2087 Description: waiting_locks_callback trips kernel BUG if client is evicted
2088 Details : Running with the dump_on_timeout debug flag turned on makes
2089 it possible that the waiting_locks_callback() can try to dump
2090 the Lustre kernel debug logs from an interrupt handler. Defer
2091 this log dumping to the expired_lock_main() thread.
2093 Severity : enhancement
2095 Description: Support NFS exporting on 2.6 kernels.
2096 Details : Implement non-rawops metadata methods for NFS server to use without
2097 changing NFS server code.
2100 Frequency : very rare (synthetic metadata workload only)
2102 Description: two racing renames might cause an MDS thread to deadlock
2103 Details : Running the "racer" program may cause one MDS thread to rename
2104 a file from being the source of a rename to being the target of
2105 a rename at exactly the same time that another thread is doing
2106 so, and the second thread has already enqueued these locks after
2107 doing a lookup of the target and is trying to relock them in
2108 order. Ensure that we don't try to re-lock the same resource.
2111 Frequency : only very large systems with liblustre clients
2113 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
2114 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
2115 all OSTs in parallel. This allows the eviction of stale liblustre
2116 clients to proceed much faster than if they were done in series,
2117 and also offers similar improvements for other set_info RPCs.
2122 Description: excessive CPU usage during initial read phase on client
2123 Details : During the initial read phase on a client, it would agressively
2124 retry readahead on the file, consuming too much CPU and impacting
2125 performance (since 1.4.5.8). Improve the readahead algorithm
2126 to avoid this, and also improve some other common cases (read
2127 of small files in particular, where "small" is files smaller than
2128 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
2133 Description: MDS crash when receiving packet with unknown intent.
2134 Details : Do not LBUG in unknown intent case, just return -EFAULT
2136 Severity : enhancement
2137 Bugzilla : 9293, 9385
2138 Description: MDS RPCs are serialised on client. This is unnecessary for some.
2139 Details : Do not serialize getattr (non-intent version) and statfs.
2142 Frequency : occasional, when OST network is overloaded/intermittent
2144 Description: client evicted by OST after bulk IO timeout
2145 Details : If a client sends a bulk IO request (read or write) the OST
2146 may evict the client if it is unresposive to its data GET/PUT
2147 request. This is incorrect if the network is overloaded (takes
2148 too long to transfer the RPC data) or dropped the OST GET/PUT
2149 request. There is no need to evict the client at all, since
2150 the pinger and/or lock callbacks will handle this, and the
2151 client can restart the bulk request.
2154 Frequency : Always when mmapping file with no objects
2156 Description: client crashes when mmapping file with no objects
2157 Details : Check that we actually have objects in a file before doing any
2158 operations on objects in ll_vm_open, ll_vm_close and
2164 Description: Request leak when working with deleted CWD
2165 Details : Introduce advanced request refcount tracking for requests
2166 referenced from lustre intent.
2168 Severity : Enhancement
2170 Description: Cache open file handles on client.
2171 Details : MDS now will return special lock along with openhandle, if
2172 requested and client is allowed to hold openhandle, even if unused,
2173 until such a lock is revoked. Helps NFS a lot, since NFS is opening
2174 closing files for every read/write openration.
2176 Severity : Enhancement
2178 Description: Cache open negative dentries on client when possible.
2179 Details : Guard negative dentries with UPDATE lock on parent dir, drop
2180 negative dentries on lock revocation.
2185 Description: Remounting a client read-only wasn't possible with a zconf mount
2186 Details : It wasn't possible to remount a client read-only with llmount.
2188 Severity : enhancement
2189 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
2190 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
2191 MPICH is included to add Lustre-specific ADIO interfaces.
2192 This is based closely on the UFS ADIO layer and only differs
2193 in file creation, in order to allow the OST striping to be set.
2194 This is user-contributed code and not supported by CFS.
2199 Description: extended inode attributes (immutable, append-only) work improperly
2200 when 2.4 and 2.6 kernels are used on client/server or vice versa
2201 Details : Introduce kernel-independent values for these flags.
2203 Severity : enhancement
2206 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2207 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2208 granularity. Now, Fractional MB granularity is be supported,
2209 this is very useful for low memory system.
2211 Severity : enhancement
2213 Description: Getattr by fid
2214 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2215 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2216 have locks and avoids getattr rpc requests for it on every stat.
2219 Frequency : Always, for filesystems larger than 2TB
2221 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2222 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2223 filesystems larger than 2TB to be mounted with mballoc enabled.
2226 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2228 Description: filesystem corruption for non-standard kernels and very large OSTs
2229 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2230 filesystems larger than 2TB is mounted then the kernel will
2231 silently corrupt the start of the filesystem. CONFIG_LBD is
2232 enabled for all CFS-supported kernels, but the possibility of
2233 this happening with a modified kernel config exists.
2235 Severity : enhancement
2237 Description: add client O_DIRECT support for 2.6 kernels
2238 Details : It is now possible to do O_DIRECT reads and writes to files
2239 in the Lustre client mountpoint on 2.6 kernel clients.
2241 Severity : enhancement
2243 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2244 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2245 parallel, not waiting for response from every OST before sending
2246 a rpc to the next OST.
2251 Description: setattr vs write race when updating file timestamps
2252 Details : Client processes that update a file timestamp into the past
2253 right after writing to the file (e.g. tar) it is possible that
2254 the updated file modification time can be reset to the current
2255 time due to a race between processing the setattr and write RPC.
2257 Severity : enhancement
2259 Description: Bring 'lfs find' closer in line with regular Linux find.
2260 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2261 -print0 options and obtains all the needed info through the lustre
2264 Severity : enhancement
2266 Description: support up to 1024 configured devices on one node
2267 Details : change obd_dev array from statically allocated to dynamically
2268 allocated structs as they are first used to reduce memory usage
2273 Description: Flush dirty partially truncated pages during truncate
2274 Details : Immediatelly flush partially truncated pages in filter_setattr,
2275 this way we completely avoid having any pages in page cache on OST
2276 and can retire ugly workarounds during writes to flush such pages.
2281 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2282 Details : protect inode from truncation within vfs_unlink() context
2283 just take a reference before calling vfs_unlink() and release it
2284 when parent's i_sem is free.
2287 Frequency : always, if extents are used on OSTs
2289 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2290 Details : OSTs using the extents format would not zero the high 16 bits of
2291 the index physical block number. This is not a problem for any
2292 OST filesystems smaller than 16TB, and no kernels support ext3
2293 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2294 new/modified files) and can be fixed for existing filesystems
2295 with e2fsprogs-1.39-cfs1.
2300 Description: import connection selection may be incorrect if timer wraps
2301 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2302 import connections to be ignored if the 32-bit jiffies counter
2303 wraps. Use a 64-bit jiffies counter.
2306 Frequency : during server recovery
2308 Description: crash after server is denying duplicate export
2309 Details : If clients are resending connect requests to the server, the
2310 server refuses to allow a client to connect multiple times.
2311 Fixed a bug in the handling of this case.
2314 Frequency : very large clusters immediately after boot
2316 Description: LNET request buffers exhausted under heavy short-term load
2317 Details : If a large number of client requests are generated on a service
2318 that has previously never seen so many requests it is possible
2319 that the request buffer growth cannot keep up with the spike in
2320 demand. Instead of dropping incoming requests, they are held in
2321 the LND until the RPC service can accept more requests.
2324 Frequency : Sometimes during replay
2326 Description: Assertion failure in ll_local_open after replay.
2327 Details : If replay happened on an open request reply before we were able
2328 to set replay handler, reply will become not swabbed tripping the
2329 assertion in ll_local_open. Now we set the handler right after
2330 recognising of open request
2333 Frequency : very rare
2335 Description: kernel reports "badness in vsnprintf"
2336 Details : Reading from the "recovery_status" /proc file in small chunks
2337 may cause a negative length in lprocfs_obd_rd_recovery_status()
2338 call to vsnprintf() (which is otherwise harmless). Exit early
2339 if there is no more space in the output buffer.
2341 Severity : enhancement
2343 Description: clear OBD RPC statistics by writing to them
2344 Details : It is now possible to clear the OBD RPC statistics by writing
2345 to the "stats" file.
2350 Description: Client mtime is not the same on different clients after utimes
2351 Details : In some cases, the client was using the utimes() syscall on
2352 a file cached on another node. The clients now validate the
2353 ctime from the MDS + OSTs to determine which one is right.
2358 Description: Inability to activate failout mode
2359 Details : lconf script incorrectly assumed that in python string's numeric
2360 value is used in comparisons.
2363 Frequency : always with multiple stripes per file
2365 Description: Inefficient object allocation for mutli-stripe files
2366 Details : When selecting which OSTs to stripe files over, for files with
2367 a stripe count that divides evenly into the number of OSTs,
2368 the MDS is always picking the same starting OST for each file.
2369 Return the OST selection heuristic to the original design.
2374 Description: mount failures may take full timeout to return an error
2375 Details : Under some heavy load conditions it is possible that a
2376 failed mount can wait for the full obd_timeout interval,
2377 possibly several minutes, before reporting an error.
2378 Instead return an error as soon as the status is known.
2380 ------------------------------------------------------------------------------
2382 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2384 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2385 INTEROPERATE with older versions automatically. Please read the
2386 user documentation before upgrading any part of a live system.
2387 * WARNING: Lustre networking configuration changes are required with
2388 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2391 * Support for kernels:
2392 2.6.9-22.0.2.EL (RHEL 4)
2393 2.6.5-7.244 (SLES 9)
2394 2.6.12.6 vanilla (kernel.org)
2397 Severity : enhancement
2398 Bugzilla : 7981/8208
2399 Description: Introduced Lustre Networking (LNET)
2400 Details : LNET is new networking infrastructure for Lustre, it includes
2401 a reorganized network configuration mode (see the user
2402 documentation for full details) as well as support for routing
2403 between different network fabrics. Lustre Networking Devices
2404 (LNDs) for the supported network fabrics have also been
2405 created for this new infrastructure.
2407 Severity : enhancement
2408 Description: Introduced Access control lists
2409 Details : clients can set ACLs on files and directories in order to have
2410 more fine-grained permissions than the standard Unix UGO+RWX.
2411 The MDS must be started with the "-o acl" mount option.
2413 Severity : enhancement
2414 Description: Introduced filesystem quotas
2415 Details : Administrators may now establish per-user quotas on the
2418 Severity : enhancement
2420 Description: Configuration change for the XT3
2421 The PTLLND is now used to run Lustre over Portals on the XT3
2422 The configure option(s) --with-cray-portals are no longer used.
2423 Rather --with-portals=<path-to-portals-includes> is used to
2424 enable building on the XT3. In addition to enable XT3 specific
2425 features the option --enable-cray-xt3 must be used.
2430 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2431 Details : A race between allocating a new llcd and re-getting the llcd_lock
2432 allowed another thread to grab newly-allocated llcd.
2434 Severity : enhancement
2436 Description: 2.6 OST async journal commit and locking fix to improve performance
2437 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2438 2.6 kernels are now async as they already were in 2.4 kernels so
2439 that they can commit concurrently with the network bulk transfer.
2440 For block-allocated files the filter allocation semaphore is held
2441 to avoid filesystem fragmentation during allocation. BKL lock
2442 removed for 2.6 xattr operations where it is no longer needed.
2447 Description: lconf incorrectly determined whether two IP networks could talk
2448 Details : In some more complicated routing and multiple-network
2449 configurations, lconf will avoid trying to make a network
2450 connection to a disjoint part of the IP space. It was doing the
2451 math incorrectly for one set of cases.
2456 Description: Fix for potential infinite loop processing records in an llog.
2457 Details : If an llog record is corrupted/zeroed, it is possible to loop
2458 forever in llog_process(). Validate the llog record length
2459 and skip the remainder of the block on error.
2462 Frequency : occasional (liblustre only)
2464 Description: liblustre could not open files whose last component is a symlink
2465 Details : sysio_path_walk() would incorrectly pass the open intent to
2466 intermediate path components.
2469 Frequency : rare (liblustre only with non-standard tuning)
2470 Bugzilla : 7201 (7350)
2471 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2472 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2473 after it has been released, only do this for referenced locks
2475 Severity : enhancement
2477 Description: specifying an (invalid) directory default stripe_size of -1
2478 would reset the directory default striping
2479 Details : stripe_size -1 was used internally to signal directory stripe
2480 removal, now use "all default" to signal dir stripe removal
2481 as a directory striping of "all default" is not useful
2484 Frequency : common for large clusters running liblustre clients
2486 Description: doing an ls when liblustre clients are running is slow
2487 Details : sending a glimpse AST to a liblustre client waits for every AST
2488 to time out, as liblustre clients will not respond. Since they
2489 cannot cache data we refresh the OST lock LVB from disk instead.
2491 Severity : enhancement
2493 Description: doing an ls at the same time as file IO can be slow
2494 Details : enqueue and other "small" requests can be blocked behind many
2495 large IO requests. Create a new OST IO portal for non-IO
2496 requests so they can be processed faster.
2499 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2501 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2502 created file with mode 000
2503 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2504 allow an ftruncate() as a non-root user to a file with mode 000.
2505 Lustre now respects this flag to disable mode checks when
2506 truncating a file owned by the user
2509 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2512 Description: Revoking locks from clients that went dead or catatonic might take
2514 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2515 cancellation of such locks instant on servers without waiting for
2516 any reply from clients. Clients drops these locks when cancel
2517 notification from server is received without replying.
2520 Frequency : liblustre-only, when liblustre client dies or becomes busy
2522 Description: Doing ls on Linux clients can take a long time with active
2524 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2525 granting such locks to it in the first place if possible. Locks
2526 are taken by proxy on the OST during the read or write and
2527 dropped immediately afterward. Add connect flags handling, do
2528 not grant locks to liblustre clients for glimpse ASTs.
2530 Severity : enhancement
2532 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2533 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2534 sequential reads and behaves badly with applications doing random
2535 reads. Improve it by reading ahead at least read region, and
2536 avoiding excessive large RPC for small reads.
2538 Severity : enhancement
2540 Description: Creating more than 1000 files for a single job may cause a load
2541 imbalance on the OSTs if there are also a large number of OSTs.
2542 Details : qos_prep_create() uses an OST index reseed value that is an
2543 even multiple of the number of available OSTs so that if the
2544 reseed happens in the middle of the object allocation it will
2545 still utilize the OSTs as uniformly as possible.
2550 Description: OST or MDS may oops in ping_evictor_main()
2551 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2552 but doesn't restart at beginning of obd_exports_timed list
2555 Severity : enhancement
2557 Description: improve by-nid export eviction on the MDS and OST
2558 Details : allow multiple exports with the same NID to be evicted at one
2559 time without re-searching the exports list.
2562 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2564 Description: MDS may oops in groups_free()
2565 Details : in rare race conditions a newly allocated group_info struct is
2566 freed again, and this can be NULL. The 2.4 compatibility code
2567 for groups_free() checked for a NULL pointer, but 2.6 did not.
2570 Frequency : common for liblustre clients doing little filesystem IO
2571 Bugzilla : 9352, 7313
2572 Description: server may evict liblustre clients accessing contended locks
2573 Details : if a client is granted a lock or receives a completion AST
2574 with a blocking AST already set it would not reply to the AST
2575 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2578 Frequency : lfs setstripe, only systems with more than 160 OSTs
2580 Description: unable to set striping with a starting offset beyond OST 160
2581 Details : llapi_create_file() incorrectly limited the starting stripe
2582 index to the maximum single-file stripe count.
2585 Frequency : LDAP users only
2587 Description: lconf did not handle in-kernel recovery with LDAP properly
2588 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2590 Severity : enhancement
2592 Description: bind OST threads to NUMA nodes to improve performance
2593 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2594 node and do their allocations there to localize memory access
2596 Severity : enhancement
2598 Description: llmount can determine client NID directly from Myrinet (GM)
2599 Details : the client NID code from gmnalnid was moved directly into
2600 llmount, removing the need to use this or specifying the
2601 client NID explicitly when mounting GM clients with zeroconf
2604 Frequency : if client is started with down MDS
2606 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2607 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2608 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2609 ptlrpc_import_delayed_req() because we didn't check if the
2610 request was interrupted, and we also didn't break out of the
2611 event loop if there was no timeout
2616 Description: data loss during non-page-aligned writes to a single file from
2617 both multiple nodes and multiple threads on one node at same time
2618 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2619 inconsistency led to false short-reads, that were cached and later
2620 used by ->prepare_write() to fill in partially written page,
2621 leading to data loss.
2624 Frequency : always, if lconf --abort_recovery used
2626 Description: lconf --abort_recovery fails with 'Operation not supported'
2627 Details : lconf was attempting to abort recovery on the MDT device and not
2630 Severity : enhancement
2632 Description: remove cleanup logs
2633 Details : replace lconf-generated cleanup logs with lustre internal
2634 cleanup routines. Eliminates the need for client-cleanup and
2637 Severity : enhancement
2639 Description: add support for EAs (user and system) on lustre filesystems
2640 Details : it is now possible to store extended attributes in the Lustre
2641 client filesystem, and with the user_xattr mount option it
2642 is possible to allow users to store EAs on their files also
2644 Severity : enhancement
2646 Description: Add possibility (config option) to show minimal available OST free
2648 Details : When compiled with --enable-mindf configure option, statfs(2)
2649 (and so, df) will return least minimal free space available from
2650 all OSTs as amount of free space on FS, instead of summary of
2651 free spaces of all OSTs.
2653 Severity : enhancement
2655 Description: do not expand extent locks acquired on OST-side
2656 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2657 by server: they are not cached anyway.
2660 Frequency : when mmap is used/binaries executed from Lustre
2662 Description: Unmmap pages before throwing them away from read cache.
2663 Details : llap_shrink cache now attempts to unmap pages before discarding
2664 them (if unmapping failed - do not discard). SLES9 kernel has
2665 extra checks that trigger if this unmapping is not done first.
2670 Description: lconf didn't resolve symlinks before checking to see whether a
2671 given mountpoint was already in use
2674 Frequency : when migrating failover services
2675 Bugzilla : 6395, 9514
2676 Description: When migrating a subset of services from a node (e.g. failback
2677 from a failover service node) the remaining services would
2678 time out and evict clients.
2679 Details : lconf --force (implied by --failover) sets the global obd_timeout
2680 to 5 seconds in order to quickly disconnect, but this caused
2681 other RPCs to time out too quickly. Do not change the global
2682 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2684 Severity : enhancement
2685 Frequency : if MDS is started with down OST
2686 Bugzilla : 9439,5706
2687 Description: Allow startup/shutdown of an MDS without depending on the
2688 availability of the OSTs.
2689 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2690 Add appropriate locking and lov-osc refcounts for safe
2691 cleaning. Add osc abort_inflight calls in case the
2692 synchronize never started.
2695 Frequency : occasional (Cray XT3 only)
2697 Description: root not authorized to access files in CRAY_PORTALS environment
2698 Details : The client process capabilities were not honoured on the MDS in
2699 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2700 been accessed by an authorized user then root was able to access
2701 the file on the local client also. The root user capabilities
2702 are now allowed on the MDS, as this environment has secure UID.
2705 Frequency : occasional
2707 Description: ldiskfs "too long searching" message happens too often
2708 Details : A debugging message (otherwise harmless) prints too often on
2709 the OST console. This has been reduced to only happen when
2710 there are fragmentation problems on the filesystem.
2715 Description: Division by zero in statfs when all OSCs are inactive
2716 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2717 lov_statfs divides by value returned by lov_get_stripecnt().
2721 Bugzilla : 9489, 3273
2722 Description: First write from each client to each OST was only 4kB in size,
2723 to initialize client writeback cache, which caused sub-optimal
2724 RPCs and poor layout on disk for the first writen file.
2725 Details : Clients now request an initial cache grant at (re)connect time
2726 and so that they can start streaming writes to the cache right
2727 away and always do full-sized RPCs if there is enough data.
2728 If the OST is rebooted the client also re-establishes its grant
2729 so that client cached writes will be honoured under the grant.
2734 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2735 Details : Now I/O RPCs go to different portal number and (presumably) fast
2736 lock requests (and glimses) and other RPCs get their own service
2737 threads pool that should be able to service those RPCs
2740 Severity : enhancement
2742 Description: Ability to exchange lustre version between client and servers and
2743 issue warnings at client side if client is too old. Also for
2744 liblustre clients there is ability to refuse connection of too old
2746 Details : New 'version' field is added to connect data structure that is
2747 filled with version info. That info is later checked by server and
2751 Frequency : rare, liblustre only.
2752 Bugzilla : 9296, 9581
2753 Description: Two simultaneous writes from liblustre at offset within same page
2754 might proceed at the same time overwriting eachother with stale
2756 Details : I/O lock withing llu_file_prwv was released too early, before data
2757 actually was hitting the wire. Extended lock-holding time until
2758 server acknowledges receiving data.
2761 Frequency : extremely rare. Never observed in practice.
2763 Description: avoid generating lustre_handle cookie of 0.
2764 Details : class_handle_hash() generates handle cookies by incrementing
2765 global counter, and can hit 0 occasionaly (this is unlikely, but
2766 not impossible, because initial value of cookie counter is
2767 selected randonly). Value of 0 is used as a sentinel meaning
2768 "unassigned handle" --- avoid it. Also coalesce two critical
2769 sections in this function into one.
2771 Severity : enhancement
2773 Description: allow liblustre clients to delegate truncate locking to OST
2774 Details : To avoid overhead of locking, liblustre client instructs OST to
2775 take extent lock in ost_punch() on client's behalf. New connection
2776 flag is added to handle backward compatibility.
2778 Severity : enhancement
2779 Bugzilla : 4928, 7341, 9758
2780 Description: allow number of OST service threads to be specified
2781 Details : a module parameter allows the number of OST service threads
2782 to be specified via "options ost ost_num_threads={N}" in the
2783 OSS's /etc/modules.conf or /etc/modprobe.conf.
2787 Bugzilla : 6146, 9635, 9895
2788 Description: servers crash with bad pointer in target_handle_connect()
2789 Details : In rare cases when a client is reconnecting it was possible that
2790 the connection request was the last reference for that export.
2791 We would temporarily drop the export reference and get a new
2792 one, but this may have been the last reference and the export
2793 was just destroyed. Get new reference before dropping old one.
2795 Severity : enhancement
2796 Frequency : if client is started with failover MDS
2798 Description: Allow multiple MDS hostnames in the mount command
2799 Details : Try to read the configuration from all specified MDS
2800 hostnames during a client mount in case the "primary"
2803 Severity : enhancement
2805 Description: Stop sending data to evicted clients as soon as possible.
2806 Details : Check if the client we are about to send or are sending data to
2807 was evicted already. (Check is done every second of waiting,
2808 for which l_wait_event interface was extended to allow checking
2809 of exit condition at specified intervals).
2812 Frequency : rare, normally only when NFS exporting is done from client
2814 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2816 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2819 Frequency : always, if srand() is called before liblustre initialization
2821 Description: Liblustre uses system PRNG disturbing its usage by user application
2822 Details : Introduce internal to lustre fast and high-quality PRNG for
2823 lustre usage and make liblustre and some other places in generic
2824 lustre code to use it.
2826 Severity : enhancement
2827 Bugzilla : 9477, 9557, 9870
2828 Description: Verify that the MDS configuration logs are updated when xml is
2829 Details : Check if the .xml configuration logs are newer than the config
2830 logs stored on the MDS and report an error if this is the case.
2831 Request --write-conf, or allow starting with --old_conf.
2833 Severity : enhancement
2835 Description: Handle symlinks in the path when checking if Lustre is mounted.
2836 Details : Resolve intermediate symlinks when checking if a client has
2837 mounted a filesystem to avoid duplicate client mounts.
2842 Description: lconf can hit an error exception but still return success.
2843 Details : The lconf command catches the Command error exception at the top
2844 level script context and will exit with the associated exit
2845 status, but doesn't ensure that this exit status is non-zero.
2850 Description: failure of ptlrpc thread startup can cause oops
2851 Details : Starting a ptlrpc service thread can fail if there are a large
2852 number of threads or the server memory is very fragmented.
2853 Handle this without oopsing.
2856 Frequency : always, only if liblustre and non-default acceptor port was used
2858 Description: liblustre cannot connect to servers with non-default acceptor port
2859 Details : tcpnal_set_default_params() was not called and was therefore
2860 ignoring the environment varaible TCPNAL_PORT, as well as other
2861 TCPNAL_ environment variables
2866 Description: two objects could be created on the same OST for a single file
2867 Details : If an OST is down, in some cases it was possible to create two
2868 objects on a single OST for a single file. No problems other
2869 than potential performance impact and spurious error messages.
2873 Bugzilla : 5681, 9562
2874 Description: Client may oops in ll_unhash_aliases
2875 Details : Client dcache may become inconsistent in race condition.
2876 In some cases "getcwd" can fail if the current directory is
2882 Description: Inode refcounting problems in NFS export code
2883 Details : link_raw functions used to call d_instantiate without obtaining
2884 extra inode reference first.
2888 Bugzilla : 9942, 9903
2889 Description: Referencing freed requests leading to crash, memleaks with NFS.
2890 Details : We used to require that call to ll_revalidate_it was always
2891 followed by ll_lookup_it. Also with revalidate_special() it is
2892 possible to call ll_revalidate_it() twice for the same dentry
2893 even if first occurence returned success. This fix changes semantic
2894 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2895 reference on a request referred from the intent.
2896 ll_intent_release() then releases such a request.
2899 Frequency : rare, normally benchmark loads only
2901 Description: unlinked inodes were kept in memory on the client
2902 Details : If a client is repeatedly creating and unlinking files it
2903 can accumulate a lot of stale inodes in the inode slab cache.
2904 If there is no other client load running this can cause the
2905 client node to run out of memory. Instead flush old inodes
2906 from client cache that have the same inode number as a new inode.
2909 Frequency : SLES9 2.6.5 kernel and long filenames only
2910 Bugzilla : 9969, 10379
2911 Description: utime reports stale NFS file handle
2912 Details : SLES9 uses out-of-dentry names in some cases, which confused
2913 the lustre dentry revalidation. Change it to always use the
2917 Frequency : rare, unless heavy write-truncate concurrency is continuous
2918 Bugzilla : 4180, 6984, 7171, 9963, 9331
2919 Description: OST becomes very slow and/or deadlocked during object unlink
2920 Details : filter_destroy() was holding onto the parent directory lock
2921 while truncating+unlinking objects. For very large objects this
2922 may block other threads for a long time and slow overall OST
2923 responsiveness. It may also be possible to get a lock ordering
2924 deadlock in this case, or run out of journal credits because of
2925 the combined truncate+unlink. Solution is to do object truncate
2926 first in one transaction without parent lock, and then do the
2927 final unlink in a new transaction with the parent lock. This
2928 reduces the lock hold time dramatically.
2931 Frequency : rare, 2.4 kernels only
2933 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2934 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2935 before unmounting their filesystem in order to avoid deadlock.
2936 The kernel_locked() function in 2.4 kernels only checks whether
2937 the kernel lock is held, not whether it is this process that is
2938 holding it as 2.6 kernels do.
2943 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2944 Details : The client ptlrpc code may be trying to reconnect to a down
2945 server before a previous connection attempt has timed out.
2946 Increase the reconnect interval to be longer than the connection
2947 timeout interval to avoid sending duplicate connections to
2951 Frequency : echo_client brw_test command
2953 Description: fix echo_client to work with OST preallocated code
2954 Details : OST preallocation code (5137) didn't take echo_client IO path
2955 into account: echo_client calls filter methods outside of any
2956 OST thread and, hence, there is no per-thread preallocated
2957 pages and buffers to use. Solution: hijack pga pages for IO. As
2958 a byproduct, this avoids unnecessary data copying.
2962 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2963 Description: Client can oops in mdc_commit_close() after open replay
2964 Details : It was possible for the MDS to return an open request with no
2965 transaction number in mds_finish_transno() if the client was
2966 evicted, but without actually returning an error. Clients
2967 would later try to replay that open and may trip an assertion
2968 Simplify the client close codepath, and always return an error
2969 from the MDS in case the open is not successful.
2972 Frequency : rare, 2.6 OSTs only
2974 Description: OST may deadlock under high load on fragmented files
2975 Details : If there was a heavy load and highly-fragmented OST filesystems
2976 it was possible to have all the OST threads deadlock waiting on
2977 allocation of biovecs, because the biovecs were not released
2978 until the entire RPC IO was completed. Instead, release biovecs
2979 as soon as they are complete to ensure forward IO progress.
2981 Severity : enhancement
2983 Description: Support for specifying external journal device at mount
2984 Details : If an OST or MDS device is formatted with an external journal
2985 device, this device major/minor is stored in the ext3 superblock
2986 and may not be valid for failover. Allow detecting and
2987 specifying the external journal at mount time.
2992 Description: Mounting an MDS with pending unlinked files may cause oops
2993 Details : target_finish_recovery() calls mds_postrecov() which returned
2994 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2995 considers this an error and immediately begins cleaning up the
2996 lov, just after starting the mds_lov process
2998 Severity : enhancement
3000 Description: Implement 'lfs df' to report actual free space on per-OST basis
3001 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
3002 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
3003 usage of objects; '-h' to report in human readable format.
3005 ------------------------------------------------------------------------------
3007 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
3014 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
3015 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
3016 at startup during orphan unlinking. Instead, skip buddy update
3017 before setup, as it will be regenerated after recovery is complete.
3022 Description: If an OST is inactive, its locks might reference stale inodes.
3023 Details : lov_change_cbdata() must iterate over all namespaces, even if
3024 they are inactive to clear inode references from the lock.
3026 Severity : enhancement
3027 Frequency : occasional, if non-standard max_dirty_mb used
3029 Description: Client will block write RPCs if not enough grant
3030 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
3031 then the client will block writes while waiting for another RPC
3032 to complete instead of consuming its dirty limit. With change
3033 we get improved performance when max_dirty_mb is small.
3035 Severity : enhancement
3036 Bugzilla : 3389, 6253
3037 Description: Add support for supplementary groups on the MDS.
3038 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
3039 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
3040 lookups for user supplementary groups into a cache.
3044 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
3045 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
3046 for liblustre O_EXCL is enforced by MDS.
3050 Description: readdir never returns NULL in liblustre.
3051 Details : Corrected llu_iop_getdirentries logic, to return offset of next
3052 dentry in struct dirent.
3056 Frequency : liblustre only, depends on application IO pattern
3057 Description: liblustre clients evicted if not contacting servers
3058 Details : Don't put liblustre clients into the ping_evictor list, so
3059 they will not be evicted by the pinger ever.
3061 Severity : enhancement
3063 Description: Add ability to evict clients by NID from MDS.
3064 Details : By echoing "nid:$NID" string into
3065 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
3066 $NID would be instantly evicted from this MDS and from all active
3067 OSTs connected to it.
3071 Description: Do not query file size twice, somewhat slowing stat(2) calls.
3072 Details : lookup_it_finish() used to query file size from OSTs that was not
3077 Description: service threads change working directory to that of init
3078 Details : Starting lustre service threads may pin the working directory
3079 of the parent thread, making that filesystem busy. Threads
3080 now change to the working directory of init to avoid this.
3084 Frequency : during shutdown only
3085 Description: shutdown with a failed MDS or OST can cause unmount to hang
3086 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
3091 Frequency : 2.6 only
3092 Description: chmod/chown may include an extra supplementary group
3093 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
3094 supplementary group and if none is specified this is used.
3097 Bugzilla : 5479 (6816)
3099 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
3100 Details : If lookup is in progress on a file that is unlinked we might try
3101 to revalidate the inode and fail in revalidate after lookup is
3102 complete and ll_file_open() enqueues the open again but
3103 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
3106 Frequency : always, if lconf --abort_recovery used
3108 Description: lconf --abort_recovery fails with 'Operation not supported'
3109 Details : lconf was attempting to abort recovery on the MDT device and not
3112 ------------------------------------------------------------------------------
3114 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
3119 Frequency : rare (only unsupported configurations with a node running as an
3121 Bugzilla : 6514, 5137
3122 Description: Mounting a Lustre file system on a node running as an OST could
3124 Details : OSTs now preallocates memory needed to write out data at
3125 startup, instead of when needed, to avoid having to
3126 allocate memory in possibly low memory situations.
3127 Specifically, if the file system is mounted on on OST,
3128 memory pressure could force it to try to write out data,
3129 which it needed to allocate memory to do. Due to the low
3130 memory, it would be unable to do so and the node would
3131 become unresponsive.
3133 Severity : enhancement
3135 Description: Addition of lconf --service command line option
3136 Details : lconf now accepts a '--service <arg>' option, which is
3137 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
3139 Severity : enhancement
3141 Description: Failover mode is now the default for OSTs.
3142 Details : By default, OSTs will now run in failover mode. To return to
3143 the old behaviour, add '--failout' to the lmc line for OSTs.
3145 Severity : enhancement
3147 Description: Health checks are now provided for MDS and OSTs
3148 Details : Additional detailed health check information on MSD and OSTs
3149 is now provided through the procfs health_check value.
3152 Frequency : occasional, depends on IO load
3154 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
3155 after numerous create/delete cycles
3156 Details : The ext3 inode allocation policy would not allocate new inodes
3157 very well on the OSTs because there are no new directories
3158 being created. Instead we look for groups with free space if
3159 the parent directories are nearly full.
3164 Description: Network or server problems during mount may cause partially
3165 mounted clients instead of returning an error.
3166 Details : The config llog parsing code may overwrite the error return
3167 code during mount error handling, returning success instead
3173 Description: MDS can fail to allocate large reply buffers
3174 Details : After long uptimes the MDS can fail to allocate large reply
3175 buffers (e.g. zconf client mount config records) due to memory
3176 fragmentation or consumption by the buffer cache. Preallocate
3177 some large reply buffers so that these replies can be sent even
3178 under memory pressure.
3182 Frequency : rare (liblustre)
3183 Description: fsx running with liblustre complained that using truncate() to
3184 extend the file doesn't work. This patch corrects that issue.
3185 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
3186 ATTR_SIZE and lsm use in llu_setattr_raw.
3190 Frequency : rare, only 2.6 kernels
3191 Description: Unusual file access patterns on the MDS may result in inode
3192 data being lost in very rare circumstances.
3193 Details : Bad interaction between the ea-in-inode patch and the "no-read"
3194 code in the 2.6 kernel caused the inode and/or EA data not to
3195 be read from disk, causing single-file corruption.
3199 Frequency : rare, only 2.6 filesystems using extents
3200 Description: Heavy concurrent write and delete load may cause data corruption.
3201 Details : It was possible under high-load situations to have an extent
3202 metadata block in the block device cache from a just-unlinked
3203 file overwrite a newly-allocated data block. We now unmap any
3204 metadata buffers that alias just-allocated data blocks.
3208 Frequency : filesystems with default stripe_count larger than 77
3209 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3210 Details : lconf specifies an inode size of 4096 bytes when the default
3211 stripe_count is larger than 77. This conflicts with the default
3212 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3213 this case to avoid pinning too much memory for large EAs.
3215 ------------------------------------------------------------------------------
3217 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3222 Frequency : rare (extremely heavy IO load with hundreds of clients)
3224 Description: Client is evicted, gets IO error writing to file
3225 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3226 caused clients to be evicted for AST timeouts. The fixes in
3227 bug 5192 mean we no longer need to have such short AST timeouts
3228 so ldlm_timeout has been increased.
3231 Frequency : occasional during --force or --failover shutdown under load
3232 Bugzilla : 5949, 4834
3233 Description: Server oops/LBUG if stopped with --force or --failover under load
3234 Details : a collection of import/export refcount and cleanup ordering
3235 issues fixed for safer force cleanup
3238 Frequency : only filesystems larger than 120 OSTs
3239 Bugzilla : 5990, 6223
3240 Description: lfs getstripe would oops on a very large filesystem
3241 Details : lov_getconfig used kfree on vmalloc'd memory
3244 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3245 Bugzilla : 6242, 6243
3246 Description: reading from files that had been truncated to a non-zero size
3247 but never opened returned no data
3248 Details : ll_file_read() reads zeros from no-object files to EOF
3253 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3254 Details : There's little harm in aborting MDS/OSS recovery and letting it
3255 try again, so I removed the LASSERT and return an error instead.
3257 Severity : enhancement
3259 Description: New debugging infrastructure for tracking down data corruption
3260 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3261 (b) cover more of the client-side code path, and (c) try to narrow
3262 down where problems occurred
3266 Bugzilla : 3819, 4364, 4397, 6313
3267 Description: Racing close and eviction MDS could cause assertion in mds_close
3268 Details : It was possible to get multiple mfd references during close and
3269 client eviction, leading to one thread referencing a freed mfd.
3271 Severity: : enhancement
3272 Bugzilla : 3262, 6359
3273 Description: Attempts to reconnect to servers are now more aggressive.
3274 Details : This builds on the enhanced upcall-less recovery that was added
3275 in 1.4.2. When trying to reconnect to servers, clients will
3276 now try each server in the failover group every 10 seconds. By
3277 default, clients would previously try one server every 25 seconds.
3282 Description: After recovery, certain operations trigger a failed
3283 assertion on a client.
3284 Details : Failing over an mds, using lconf -d --failover, while a
3285 client was doing a readdir() call would cause the client to
3286 LBUG after recovery completed and the readdir() was resent.
3288 Severity : enhancement
3290 Description: Default groups are now added by lconf
3291 Details : You can now run lconf --group <servicename> without having to
3292 manually add groups with lmc.
3295 Frequency : occasional
3297 Description: Nodes with an elan id of 0 trigger a failed assertion
3300 Frequency : always when accessing e.g. tty/console device nodes
3302 Description: tty and some other devices nodes cannot be used on lustre
3303 Details : file's private_data field is used by device data and lustre
3304 values in there got lost. New field was added to struct file to
3305 store fs-specific private data.
3308 Frequency : when exporting Lustre via NFS
3310 Description: NFSD failed occasionally when looking up a path component
3311 Details : NFSD is looking up ".." which was broken in ext3 directories
3312 that had grown large enough to become hashed.
3315 Frequency : Clusters with multiple interfaces not on the same subnet
3317 Description: Nodes will repeatedly try to reconnect to an interface which it
3318 cannot reach and report an error to the log.
3319 Details : Extra peer list entries will be created by lconf with some peers
3320 unreachable. lconf now validates the peer before adding it.
3323 Frequency : Only if a default stripe is set on the filesystem root.
3325 Description: Setting a default stripe on the filesystem root prevented the
3326 filesystem from being remounted.
3327 Details : The client was sending extra request flags in the root getattr
3328 request and did not allocate a reply buffer for the dir EA.
3331 Frequency : occasional, higher if lots of files are accessed by one client
3332 Bugzilla : 6159, 6097
3333 Description: Client trips assertion regarding lsm mismatch/magic
3334 Details : While revalidating inodes the VFS looks up inodes with ifind()
3335 and in rare cases can find an inode that is being freed.
3336 The ll_test_inode() code will free the lsm during ifind()
3337 when it finds an existing inode and then the VFS later attaches
3338 this free lsm to a new inode.
3342 Bugzilla : 6422, 7030
3343 Description: MDS deadlock between mkdir and client eviction
3344 Details : Creating a new file via mkdir or mknod (starting a transaction
3345 and getting the ns lock) can deadlock with client eviction
3346 (gets ns lock and trying to finish a synchronous transaction).
3349 Frequency : occasional
3350 Description: While starting a server, the fsfilt_ext3 module could not be
3352 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3353 kernels. Previously, lconf would still use the ext3 name
3354 when trying to load modules. Now, it will correctly use
3355 ext3 on 2.4 and ldiskfs on 2.6.
3357 Severity : enhancement
3358 Description: The default stripe count has been changed to 1
3359 Details : The interpretation of the default stripe count (0, to lfs
3360 or lmc) has been changed to mean striping across a single
3361 OST, rather than all available. For general usage we have
3362 found a stripe count of 1 or 2 works best.
3364 Severity : enhancement
3365 Description: Add support for compiling against Cray portals.
3366 Details : Conditional compiling for some areas that are different
3370 Frequency : occasional
3371 Bugzilla : 6409, 6834
3372 Description: Creating files with an explicit stripe count may lead to
3373 a failed assertion on the MDS
3374 Details : If some OSTs are full or unavailable, creating files may
3375 trigger a failed assertion on the MDS. Now, Lustre will
3376 try to use other servers or return an error to the
3380 Frequency : occasional
3382 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3383 caused lock timeout during readahead (since 1.4.2).
3384 Details : Processes doing readahead might match a lock that hasn't been
3385 granted yet if there are overlapping and conflicting lock
3386 requests. The readahead process waits on ungranted lock
3387 (original lock is CBPENDING), while OST waits for that process
3388 to cancel CBPENDING read lock and eventually evicts client.
3390 Severity : enhancement
3392 Description: Initial enabling of flock support for clients
3393 Details : Implements fcntl advisory locking and file status functions.
3394 This feature is provided as an optional mount flag (default
3395 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3396 locking are implemented yet, and those that are are not guaranteed
3397 to be completely correct in production environments.
3398 mount -t lustre -o [flock|noflock] ...
3401 Frequency : occasional
3403 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3404 assertion in the ext3 JBD (journaling) layer.
3405 Details : The b_committed_data struct is protected by the big kernel lock
3406 in 2.4 kernels, serializing journal_commit_transaction() and
3407 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3408 access to this struct. In 2.6 kernels there is finer grained
3409 locking to improve SMP performance of the JBD layer.
3413 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3415 -----------------------------------------------------------------------------
3417 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3419 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3420 versions of Lustre. Please update all servers and clients to
3421 version 1.4.2 or later at the same time. You must also run
3422 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3423 to update the configuration logs.
3425 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3426 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3427 - swab llog records of type '0' so we get proper header size/idx (5861)
3428 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3429 - fix rename of one directory over another leaking an inode (5953)
3430 - avoid SetPageDirty on 2.6 (5981)
3431 - don't re-add just-being-destroyed locks to the waiting list (5653)
3432 - when creating new directories, inherit the parent's custom
3433 striping settings if present parent (3048)
3434 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3435 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3436 - don't hold client locks on temporary worklist from l_lru (5666)
3437 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3438 - automatically evict dead clients (5921)
3439 - Update file size properly in create+truncate+fstat case (6196)
3440 - Do not unhash mountpoint dentries, do not allow removal of
3442 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3443 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3444 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3445 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3446 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3447 - Correctly handle reads of files with no objects (6243)
3448 - lctl recover will also mark a device active if deactivate used (5933)
3450 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3451 - allow --write-conf on an MDS with different nettype than client (5619)
3452 - don't write config llogs to MDS for mounts not from that MDS (5617)
3453 - lconf should create multiple TCP connections from a client (5201)
3454 - init scripts are now turned off by default; run chkconfig --on
3455 lustre and chkconfig --on lustrefs to use them
3456 - upcalls are no longer needed for clients to recover to failover
3458 - add --abort-recovery option to lconf to abort recovery on device
3460 - add support for an arbitrary number of OSTs (3026)
3461 - Quota support protocol changes.
3462 - forward compatibility changes to wire structs (6007)
3463 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3464 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3465 - improved "lustre status" script
3466 - initialize blocksize for non-regular files (6062)
3467 - added --disable-server and --disable-client configure options (5782)
3468 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3469 - Vanilla 2.4.29 support
3470 - increase maximum number of obd devices to 520 (6242)
3471 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3472 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3473 SLES 9 2.6.5 kernels
3474 - limit stripes per file to 160 (the maximum EA size) (6093)
3476 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3479 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3480 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3481 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3482 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3483 - small ext3 extents cleanups and fixes (5733)
3484 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3485 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3486 - avoid needless client->OST connect, fix handle mismatch (5317)
3487 - fix DLM error path that led to out-of-sync client, long delays (5779)
3488 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3489 - fix several locking issues related to i_size (5492,5624,5654,5672)
3490 - don't move pending lock onto export if it is already evicted (5683)
3491 - fix kernel oops when creating .foo in unlinked directory (5548)
3492 - fix deadlock in obdfilter statistics vs. object create (5811)
3493 - use time_{before,after} to avoid timer jiffies wrap (5882)
3494 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3495 - Do not leak request if server was not able to process it (5154)
3496 - If mds_open unable to find parent dir, make that negative lookup(5154)
3497 - don't create new directories with extent-mapping (5909, 5936)
3499 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3500 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3501 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3502 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3503 - support for RHEL4 kernel on i686 (5773)
3504 - provide error messages when incompatible logs are encountered (5898)
3506 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3507 * version 1.4.0.10 (1.4.1 release candidate 1)
3509 - don't keep a lock reference when lock is not granted (4238)
3510 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3511 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3512 - drop import inflight refcount on signal_completed_replay error (5255)
3513 - unlock page after async write error during send (3677)
3514 - handle missing objects in filter_preprw_read properly (5265)
3515 - no transno return for symlink open, don't save no-trasno open (3440)
3516 - don't try to complete elan receive that already failed (4012)
3517 - free RPC server reply state on error (5406)
3518 - clean up thread from ptlrpc_start_thread() on error (5160)
3519 - readahead could read extra page into cache that wasn't ejected (5388)
3520 - prevent races in class_attach/setup/cleanup/detach (5260)
3521 - don't dereference de->d_inode after l_dput of de (5458)
3522 - use "int" for stripe value returned from lock_to_stripe (5544)
3523 - mballoc allocation and error-checking fixes in 2.6 (5504)
3524 - block device patches to fix I/O request sizes in 2.6 (5482)
3525 - look up hostnames for IB nals (5602)
3526 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3527 - don't start multiple acceptors for the same port (5277)
3528 - fix incorrect LASSERT in mds_getattr_name (5635)
3529 - export a proc file for general "ping" checking (5628)
3530 - fix "lfs check" to not block when the MDS is down (5628)
3532 - service request history (4965)
3533 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3534 - create an "evict_client" /proc entry on OSTs, like the MDS has
3535 - fix mount usage message, return errors per mount(8) (5168)
3536 - change grep [] to grep "[]" in tests so they work in more UMLs
3537 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3538 - remove ancient LOV_MAGIC_V0 EA support (5047)
3539 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3540 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3541 - allow readahead allocations to fail when low on memory (5383)
3542 - mmap locking landed again, after considerable improvement (2828)
3543 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3545 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3548 - send OST transaction number in read/write reply to free req (4966)
3549 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3550 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3551 - the watchdog thread now runs as interruptible (5246)
3552 - flock/lockf fixes (but it's still disabled, pending 5135)
3553 - don't use EXT3 constants in llite code (5094)
3554 - memory shortage at startup could cause assertion (5176)
3556 - reorganization of lov code
3557 - single portals codebase
3559 - add extents/mballoc support (5025)
3560 - direct I/O reads in the obdfilter (4048)
3561 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3563 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3566 - send OST transaction number in read/write reply to free req (4966)
3567 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3568 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3569 - the watchdog thread now runs as interruptible (5246)
3570 - handle missing objects in filter_preprw_read properly (5265)
3571 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3572 - drop import inflight refcount on signal_completed_replay error (5255)
3573 - unlock page after async write error during send (3677)
3574 - return original error code on reconstructed replies (3761)
3575 - no transno return for symlink open, don't save no-trasno open (3440)
3577 - add pid to ldlm debugging output (4922)
3578 - bump the watchdog timeouts -- we can't handle 30sec yet
3579 - extra debugging for orphan dentry/inode bug (5259)
3581 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3584 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3585 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3586 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3587 - fix NULL dereference in /proc/sys/portals/routes (4827)
3588 - allow failed mdc_close() operations to be interrupted (4561)
3589 - stop precreate on OST before MDS would time out on it (4778)
3590 - don't send partial-page writes before EOF from client (4410)
3591 - discard client grant for sub-page writes on large-page clients (4520)
3592 - don't free dentries not owned by NFS code, check generation (4806)
3593 - fix lsm leak if mds_create_objects() fails (4801)
3594 - limit debug_daemon file size, always print CERROR messages (4789)
3595 - use transno after validating reply (3892)
3596 - process timed out requests if import state changes (3754)
3597 - update mtime on OST during writes, return in glimpse (4829)
3598 - add mkfsoptions to LDAP (4679)
3599 - use ->max_readahead method instead of zapping global ra (5039)
3600 - don't interrupt __l_wait_event() during strace
3602 - add software watchdogs to catch hung threads quickly (4941)
3603 - make lustrefs init script start after nfs is mounted
3604 - fix CWARN/ERROR duplication (4930)
3605 - return async write errors to application if possible (2248)
3606 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3607 - print NAL number in %x format (4645)
3608 - update barely-supported suse-2.4.21-171 series (4842)
3609 - support for sles 9 %post scripts
3610 - support for building 2.6 kernel-source packages
3611 - support for sles km_* packages
3613 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3616 - ignore -ENOENT errors in osc_destroy (3639)
3617 - notify osc create thread that OSC is being cleaned up (4600)
3618 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3619 - reconstruct ost_handle() like mds_handle() (4657)
3620 - create a new thread to do import eviction to avoid deadlock (3969)
3621 - let lconf resolve symlinked-to devices (4629)
3622 - don't unlink "objects" from directory with default EA (4554)
3623 - hold socknal file ref over connect in case target is down (4394)
3624 - allow more than 32000 subdirectories in a single directory (3244)
3625 - fix blocks count for O_DIRECT writes (3751)
3626 - OST returns ENOSPC from object create when no space left (4539)
3627 - don't send truncate RPC if file size isn't changing (4410)
3628 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3629 - bind to privileged port in socknal and tcpnal (3689)
3631 - rate limit CERROR/CWARN console message to avoid overload (4519)
3632 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3633 - basic mmap support (3918)
3634 - kernel patch series update from b1_4 (4711)
3636 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3639 - avoid crash during MDS cleanup with OST shut down (2775)
3640 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3641 - don't use bad inodes on the MDS (3744)
3642 - dynamic object preallocation to improve recovery speed (4236)
3643 - don't hold spinlock over lock dumping or change debug flags (4401)
3644 - don't zero obd_dev when it is force cleaned (3651)
3645 - print grants to console if they go negative (4431)
3646 - "lctl deactivate" will stop automatic recovery attempts (3406)
3647 - look for existing locks in ldlm_handle_enqueue() (3764)
3648 - don't resolve lock handle twice in recovery avoiding race (4401)
3649 - revalidate should check working dir is a directory (4134)
3651 - don't always mark "slow" obdfilter messages as errors (4418)
3653 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3656 - don't close LustreDB during write_conf until it is done (3860)
3657 - fix typo in lconf for_each_profile (3821)
3658 - allow dumping logs from multiple threads at one time (3820)
3659 - don't allow multiple threads in OSC recovery (3812)
3660 - fix debug_size parameters (3864)
3661 - fix mds_postrecov to initialize import for llog ctxt (3121)
3662 - replace config semaphore with spinlock (3306)
3663 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3664 - don't allow enqueue to complete on a destroyed export (3822)
3665 - down write_lock before checking llog header bitmap (3825)
3666 - recover from lock replay timeout (3764)
3667 - up llog sem before sending rpc (3652)
3668 - reduce ns lock hold times when setting kms (3267)
3669 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3670 - fix NULL deref and obd_dev leak on setup error (3312)
3671 - replace some LBUG about llog ops with error handling (3841)
3672 - don't match INVALID dentries from d_lookup and spin (3784)
3673 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3674 - fix invalid assertion in ptlrpc_set_wait (3880)
3676 - add libwrap support for the TCP acceptor (3996)
3677 - add /proc/sys/portals/routes for non-root route listing (3994)
3678 - allow setting MDS UUID in .xml (2580)
3679 - print the stack of a process that LBUGs (4228)
3681 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3684 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3685 - make sure to unset replay flag from failed open requests (3440)
3686 - if default stripe count is 0, use OST count for inode size (3636)
3687 - update parent mtime/ctime on client for create/unlink (2611)
3688 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3689 - free recovery state on server during a forced cleanup (3571)
3690 - unregister_reply for resent reqs (3063)
3691 - loop back devices mounting and status check on 2.6 (3563)
3692 - fix resource-creation race that can provoke i_size == 0 (3513)
3693 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3694 - more debugging for page-accounting assertion (3746)
3695 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3696 - avoid deadlock after precreation failure (3758)
3697 - fix race and lock order deadlock in orphan handling (3450, 3750)
3698 - add validity checks when grabbing inodes from l_ast_data (3599)
3700 - add /proc/.../recovery_status to obdfilter (3428)
3701 - lightweight CDEBUG infrastructure, debug daemon (3668)
3702 - change default OSC RPC parameters to be better on small clusters
3703 - turn off OST read cache for files smaller than 32MB
3704 - install man pages and include them in rpms (3100)
3705 - add new init script for (un)mounting lustre filesystems (2593)
3706 - run chkconfig in %post for init scripts (3701)
3707 - drop scimac NAL (unmaintained)
3709 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3712 - clean kiobufs before and after use (3485)
3713 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3714 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3715 - change init script to fail more clearly if not run as root (1528)
3716 - allow clients to reconnect during replay (1742)
3717 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3718 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3719 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3720 - fix cond_resched (3554)
3721 - only evict unfinished clients after recovery (3515)
3722 - allow bulk resend, prevent data loss (3570)
3723 - dynamic ptlrpc request buffer allocation (2102)
3724 - don't allow unlinking open directory if it isn't empty (2904)
3725 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3726 - remove extraneous obd dereference causing LASSERT failure (3334)
3727 - don't use get_cycles() when creating temp. files on the mds (3156)
3728 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3729 - handle EEXIST for set-stripe, set proper directory name (3336)
3731 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3732 - fix ksocknal_fmb_callback() error messages (2918)
3734 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3737 - don't copy lvb into (possibly NULL) reply on error (2983)
3738 - don't deref dentry after dput, don't free lvb on error (2922)
3739 - use the kms to determine writeback rpc length (2947)
3740 - increment oti_logcookies when osc is inactive (2948)
3741 - update client's i_blocks count via lvb messages (2543)
3742 - handle intent open/close of special files properly (1557)
3743 - mount MDS with errors=remount-ro, like obdfilter (2009)
3744 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3745 - don't use cancelling-locks' kms values (2947)
3746 - use highest lock extent for kms, not last one (2925)
3747 - don't dereference ERR_PTR() dentry in error handling path (3107)
3748 - fix thread race in portals_debug_dumplog() (3122)
3749 - create lprocfs device entries at setup instead of at attach (1519)
3750 - common AST error handler, don't evict client on completion race (3145)
3751 - zero nameidata in detach_mnt in 2.6 (3118)
3752 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3753 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3754 - initialize RPC timeout timer earlier for 2.6 (3219)
3755 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3756 - print nal/nid for unknown nid (3258)
3757 - additional checks for oscc recovery before doing precreate (3284)
3758 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3759 - don't crash in mdc_close for bad permissions on open (3285)
3760 - zero i_rdev for non-device files (3147)
3761 - clear page->private before handing to FS, better assertion (3119)
3762 - tune the read pipeline (3236)
3763 - fix incorrect decref of invalidated dentry (2350)
3764 - provide read-ahead stats and refine rpc in flight stats (3328)
3765 - don't hold journal transaction open across create RPC (3313)
3766 - update atime on MDS at close time (3265)
3767 - close LDAP connection when recovering to avoid server load (3315)
3768 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3769 - don't leak open file on MDS after open resend (3325)
3770 - serialize filter_precreate and filter_destroy_precreated (3329)
3771 - loop device shouldn't call sync_dev() for nul device (3092)
3772 - clear page cache after eviction (2766)
3773 - resynchronize MDS->OST in background (2824)
3774 - refuse to mount the same filesystem twice on same mountpoint (3394)
3775 - allow llmount to create routes for mounting behind routers (3320)
3776 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3777 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3778 - fix rare problem with rename on htree directories (3417)
3780 - allow default OST striping configuration per directory (1414)
3781 - fix compilation for qswnal for 2.6 kernels (3125)
3782 - increase maximum number of MDS request buffers for large systems
3783 - change liblustreapi to be useful for external progs like lfsck (3098)
3784 - increase local configuration timeout for slow disks (3353)
3785 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3787 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3790 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3791 - don't overwrite extent policy data in reply if lock was blocked (2901)
3792 - drop filter export grants atomically with removal from device (2663)
3793 - del obd_self_export from work_list in class_disconnect_exports (2908)
3794 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3795 - swab reply message in mdc_close, other PPC fixes (2464)
3796 - fix destroying of named logs (2325)
3797 - overwrite old logs when running lconf --write_conf (2264)
3798 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3799 - fix race in target_handle_connect (2898)
3800 - mds_reint_create() should take same inode create lock (2926)
3801 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3802 - don't close files for self_export to avoid uninitialized obd (2936)
3803 - allow MDS with the same name as client node (2939)
3804 - hold dentry reference for closed log files for unlink (2325)
3805 - reserve space for all logs during transactions (2059)
3806 - don't evict page beyond end of stripe extent (2925)
3807 - don't oops on a deleted current working directory (2399)
3808 - handle hard links to targets without a parent properly (2517)
3809 - don't dereference NULL lock when racing during eviction (2867)
3810 - don't grow lock extents when lots of conflicting locks (2919)
3812 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3815 - account for cache space usage on clients to avoid data loss (974)
3816 - lfsck support in lustre kernel code (2349)
3817 - reduce journal credits needed for BRW writes (2370)
3818 - orphan handling to avoid losing space on client/server crashes
3819 - ptlrpcd can be blocked, stopping ALL progress (2477)
3820 - use lock value blocks to assist in proper KMS, faster stat (1021)
3821 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3822 - recovery for initial connections (2355)
3823 - fixes for mds_cleanup_orphans (1934)
3824 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3825 - block all file creations until orphan recovery completes (1901)
3826 - client remove rq_connection from request struct (2423)
3827 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3828 - recovery timer race (2670)
3829 - mdc_close recovey bug (2532)
3830 - ptlrpc cleanup bug (2710)
3831 - mds timeout on local locks (2588)
3832 - namespace lock held during RPCs (2431)
3833 - handle interrupted sync write properly (2503)
3834 - don't try to handle a message that hasn't been replied to (2699)
3835 - client assert failure during cleanup after abort recovery (2701)
3836 - leak mdc device after failed mount (2712)
3837 - ptlrpc_check_set allows timedout requests to complete (2714)
3838 - wait for inflight reqs when ptlrpcd finishes (2710)
3839 - make sure unregistered services are removed from the srv_list
3840 - reset bulk XID's when resending them (caught by 1138 test)
3841 - unregister_bulk after timeout
3842 - fix lconf error (2694)
3843 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3844 - readahead locks pages, leaves pending causing memory pressure (2673)
3845 - increase OST request buffers to 4096 on large machines (2729)
3846 - fix up permission of existing directories in simple_mkdir (2661)
3847 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3848 - don't assign transno to errored transactions (2742)
3849 - don't delete objects on OST if given a bogus objid from MDS (2751)
3850 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3851 - if rq_no_resend, then timeout request after recovery (2432)
3852 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3853 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3854 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3855 - MDS recovery shouldn't delete objects that it has given out (2730)
3856 - if enqueue arrives after completion, don't clobber LVB (2819)
3857 - don't unlock pages twice when trigger_group_io returns error (2814)
3858 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3859 - don't write pages to disk if there was an error (1450)
3860 - don't ping imports that have recovery disabled (2676)
3861 - take buffered bytes into account when balancing socknal conn (2817)
3862 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3863 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3864 - remove little-endian swabbing of llog records (1987)
3865 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3866 - retry reposting request buffers if they fail (1191)
3867 - grow extent at grant time to avoid granting a revoked lock (2809)
3868 - lock revoke doesn't evict page if covered by a second lock (2765)
3869 - disable VM readahead to avoid reading outside lock extents (2805)
3871 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3872 - updated kernel patches for hp-2.4.20 kernel (2681)
3874 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3877 - fix truncated write corruption (2366)
3878 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3880 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3881 - ensure len(uuid) < 37 in lmc (1171)
3882 - fix ia64 OOPS in llog_test (2255)
3883 - zero end of page at obdfilter for partial page writes (2648)
3884 - don't leave stale dentries around after renames (bug 2428)
3885 - fix timeouts when evicting a client with a single lock held (2642)
3886 - set deadline for the initial HELLO message to drain (2634)
3887 - print out dotted-quad IP addresses in the socknal (2302)
3889 - additional debugging for MDS client eviction problem (2443)
3890 - fix mkfsoptions support for osts (2603, 2604)
3892 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3895 - add series for the vanilla 2.6.0 kernel
3896 - add series for the vanilla 2.4.24 kernel
3897 - add series for a cray x86/64 UL kernel drop
3898 - fix xattr patches for the vanilla 2.4.19 series
3900 - generate true UUIDs in lmc (1171)
3901 - have portals stack dumping break in UML (2466)
3902 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3903 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3904 - wake callback waiting threads on client eviction (2460)
3905 - Add --ptldebug and --subsystem to lmc (1719)
3906 - update assertion to allow safe interrupt allocation
3907 - set rq_no_resend for cancel requests (2432)
3908 - recalculate ptlrpcd timeout after resend (2494)
3909 - call vfs_rmdir when removing pending directories (2368)
3910 - fix renaming a file to itself (2429)
3911 - lmc creates a default one-stripe lov (2454)
3912 - expand procfs space to handle large clusters (2326)
3913 - increase UML stack to avoid overflow
3914 - update lconf's list of debug and subsystem masks
3915 - fix lfs find --obd (2510)
3916 - /proc tunable for disabling filter read caching (2591)
3917 - stop rpm packages from altering slapd.conf (2301)
3918 - disable nagle in the socknal under 0conf (2578)
3919 - choose mds inode size based on stripe count (2572)
3920 - fix kernel-source rpm problems (2516)
3922 - add --disable-doc to avoid pdf generation (2421)
3923 - update documentation, tests, type-os, comments
3924 - avoid format warnings on ia64
3925 - remove the TOE NAL
3926 - tiny code cleanups by removing unused fields
3928 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3931 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3932 - remove the most blatant lies from BUILDING file (2371)
3933 - change default debug level to reasonable production setting
3934 - reduce client side cache size to reduce cache flush time
3935 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3936 - make TCP zerocopy and pinger support enabled by default (2476)
3937 - sync writes completed after process exits caused crashes (2319)
3938 - maintain correct mount count on the MDS (2356)
3939 - backout 1557, because 2316 wasn't really fixed
3940 - better file I/O statistics gathering in /proc
3941 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3942 - another kernel patch to fix zero-copy TCP function export
3943 - don't take duplicate lock when processing re-sent getattr (2420)
3944 - lctl uses obd_self_export instead of creating new conn (2353)
3945 - MDS/OST recovery case which requires object creation asserted (2425)
3946 - move lfs from /usr/sbin to /usr/bin in packages
3947 - fix race between mds_client_add and mds_client_free (2417)
3948 - use kmalloc instead of slabs in portals (2430)
3949 - don't create duplicate records when a failover MDS is present (2442)
3950 - remove unnecessary mount age check (2332)
3951 - don't remove directory inodes from locks prematurely (2451)
3952 - don't break if MDS service name is the same as hostname (2103)
3953 - fix races in client write RPC generation when cache full (2482)
3955 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3958 - remove now-unused request->rq_obd (278)
3959 - if an allocation fails, print out how much memory we've used (1933)
3960 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3961 - add the "configurable stack size" patch to most series files (1256)
3962 - ability to write large log records, for 100+ OST configs (2306)
3963 - fix NULL deref when filter_prep fails (2314)
3964 - fix operator precedence error in filter_sync
3965 - dynamic allocation of socknal TX descriptors (2315)
3966 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3967 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3968 - ensure that CWARN messages in Portals always get to the syslog
3969 - __init/__exit are not for prototype decls (ldlm_init/exit)
3970 - x86-64 compile warning fixes
3971 - fix gateway LMC keyword conflict (2318)
3972 - fix MDS lock inversions in getattr/reint paths (1844)
3973 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3974 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3975 - DLM race condition prevented some lock evictions (2328)
3976 - ENOMEM detection and retry on socknal sends (2230)
3977 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3978 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3979 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3980 - better error messages when a client is rejected during recovery (1505)
3981 - avoid cancelling locks which were never granted, after failure (2330)
3982 - fix i_sem/journal inversion in mds_client_add (2333)
3983 - fix truncate/getattr lock cycle deadlock (2334)
3984 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3985 - fix two rare exit paths which could leak an l_lock() ref (2321)
3986 - fencepost error in MDS/OST orphan recovery (2226)
3987 - make log record alignment 8 bytes (1988)
3988 - lstripe now fails when requested offset > ost_count (2237)
3989 - ensure that all kernel series have a complete list.h (1607)
3990 - fix crashes in special-file operations (2316)
3991 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3993 - allow configurable automake binary, for testing new versions
3994 - small update to the lfs documentation
3996 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3998 * fix negative export reference count in fsfilt_sync (2312)
4000 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
4001 * release candidate 0.9.1
4003 - orphans are moved into the PENDING directory for possible recovery
4004 - replayed opens now open by fid for orphan/rename safety (1042)
4005 - last close of an orphan inode generates a transno (683)
4006 - chdir() and mount() now pin the directory entry (1020)
4007 - avoid CERROR in normal ll_setattr_raw() error case (1500)
4008 - discard very old requests without processing them (1502)
4009 - remove some common, well-understood CERRORs (1505)
4010 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
4011 - clear "grant" flags in OST replies until OST grant code lands (1644)
4012 - fix read performance by not clobbering i_blksize on client (1598)
4013 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
4014 - make LRU size a /proc tunable, clears locks when reduced (707)
4015 - fix some lprocfs rot that prevented ptlbd from loading (1732)
4016 - server locks take references on exports now (1558)
4017 - build fixes for 2.4.20-rh trees (1663)
4018 - return an error from lov_create if all OSCs are inactive (1751)
4019 - fix import levels when a reconnect happens without a timeout (1597)
4020 - exit early from mds_open if we get a lookup error (1749)
4021 - partial page read at EOF wouldn't wait for disk before sending (1642)
4022 - avoid NULL deref in obdfilter when reading page past EOF (1592)
4023 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
4024 - fix LBUG in ll_it_open_error with rc = -2 (1861)
4025 - write/truncate lock inversion (1639)
4026 - Don't auto-load obdclass, portals modules during cleanup (1495)
4027 - fix timestamps from jumping to "now" (1763)
4028 - extra journal assertions (1648)
4029 - add an extra multiunlink test (1771)
4030 - fix read_record/write_record API (1776)
4031 - fix leak of offset_extent, possible incorrect i_size later (1772)
4032 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
4033 - Debugging for the kqswnal_get_idle_tx problems (1820)
4034 - Allow recovery to be attempted multiple times (1536)
4035 - Write out MDS last_rcvd file after it is first created (1600)
4036 - Fix tx_descriptor leak in failed transmit situations (1827)
4037 - ext3 journaling fixes for assertion failure after IO error (1871)
4038 - class_export_put() on freed export after completion AST error (1896)
4039 - Fix revalidate looping in VFS (1322)
4040 - Don't access a freed export during MDS_REINT timeout (1521)
4041 - Add open-unlink recovery support on the MDS (1673,1764)
4042 - Return an error if no MDS data was read from last_rcvd (1946)
4043 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
4044 - Don't setup a disk device that doesn't match exported UUID (317)
4045 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
4046 - avoid committing NULL handle in force close
4047 - local.sh is now a one-stripe LOV configuration
4048 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
4049 - fix ext3 htree duplicate directory entry corruption (1516)
4050 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
4051 - update to new LOV EA format (2097)
4052 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
4054 2003-06-15 Phil Schwan <phil@clusterfs.com>
4057 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
4058 - per-import/export recovery handling (958, 931, 959)
4059 - multiple last-rcvd slots, for serving multiple FSes (949)
4060 - connections are again shared between multiple imp/exports (963, 964)
4061 - "umount -f" would hang if any requests needed to be sent (393, 978)
4062 - avoid pinning large req buffer by copying for queued messages (989)
4063 - add "uuid" to "lctl device" command to help upcalls (991)
4064 - "open" RPCs with transnos would confuse recovery counters (1037)
4065 - do proper endian conversion of all wire messages (288, 340, 891)
4066 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
4067 - call waiting locks callback from LDLM recovery thread (1127, 1151)
4068 - fix ptlrpc_connection leak in target_handle_connect (1174)
4069 - fix import refcounting bug in OST and MDS cleanup (1134)
4070 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
4071 - fix very unlikely obd_types race condition (501)
4072 - remove osc_open hack for echo_client (1187)
4073 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
4074 - a failure in read_inode2 leads to deadlock (1139)
4075 - cancel ack-locks as soon as transaction is committed (1072)
4076 - fix major leaks and crashes in the bulk I/O path (937, 1057)
4077 - make sure to commitrw after any preprw to avoid deadlock (1162)
4078 - failing to execute a file in a lustre FS would lock inode (1203)
4079 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
4080 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
4081 - obd_uuid2tgt would walk off the end of the list (1255)
4082 - on IA64 the osc would give portals incorrect bulk size (1258)
4083 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
4084 - fix lock inversion caused by new llite matching code (1282)
4085 - limit the number of dirty pages on a client to 10MB (1286)
4086 - timed out locks were not being corrected cancelled (1289)
4087 - fix O_DIRECT above 4GB on IA-32 (1292)
4088 * major user-visible changes
4089 - fail out/fail over policy now controlled by the upcall (993)
4091 - add OBD_PING to check server availability and failure (954)
4092 - lustre messages are now sent in sending host order (288, 340, 891)
4093 - add eadatalen to MDS getattr reply (340)
4094 - OST read replies may contain second buffer, with per-page status (593)
4096 2003-03-11 Phil Schwan <phil@clusterfs.com>
4099 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
4100 - failed open()s could cause deadlock; fixed (867, 869)
4101 - stop cancelling OST locks when files are closed (481)
4102 - overlapping XID spaces caused network corruption (851, 853)
4103 - fix unsafe fsfilt counter arithmetic; change to atomic_t
4104 - setattr_raw added, to do single-RPC, server-side setattrs
4105 - lmc/lconf syntax change for OST UUIDs
4106 - fix crashy race condition between ptlrpc_free_req and osc_close
4107 - don't use request in mdc_enqueue if we hit a timeout (889)
4108 - don't set the inode i_size for regular files from the MDS (896)
4109 - handle out of order completion AST (842)
4110 - don't LBUG if a lock request times out after receiving AST (913)
4111 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
4112 - if a bad lock AST arrives, send an error instead of dropping entirely
4113 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
4114 - fix leak in bulk IO when only partially completed (899, 900, 926)
4115 - fix O_DIRECT for ia64 (55)
4116 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
4117 - C-z after timeout could hang a process forever; fixed (977)
4119 - client-side I/O cache (678, 924, 929, 941, 970)
4121 - READPAGE and SETATTRs which don't take server-side locks get
4124 2003-02-11 Phil Schwan <phil@clusterfs.com>
4127 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
4128 - Fix fsfilt_extN_readpage() to read a full page of directory
4129 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
4130 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
4131 - Fix imbalanced LOV object allocation and out-of-bound access (469)
4132 - Most intent operations were removed, in favour of a new RPC mode
4133 that does a single RPC to the server and bypasses most of the VFS
4134 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
4135 - Aggressively cancel local locks on DLM servers
4136 - mds_reint_unlink sends EA to the client if it's the last nlink.
4137 client uses that EA to unlink OST objects.
4138 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
4139 - recursive symlinks were fixed (439)
4140 - fixed NULL deref in DEBUG_REQ
4141 - filter_update_lastobjid no longer calls sync, which annoyed extN
4142 - fixed multi-client small-writes to a single file problem (445)
4143 - fixed mtime updates during file writes (607)
4144 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
4145 - fixed bug in obd_brw_read/write() (under guise of testing 367)
4146 - fixed Linux OST size reporting problem (444, 656)
4147 - OST now updates object mtime with writes or setattr (607, 619)
4148 - client verifies file size before zeroing page past EOF (445)
4149 - OST now writes last allocated objid to disk with allocation (108)
4150 - LOV on echo now works (409)
4152 - mds_reint_unlink sends a new buffer, with the EA included. this
4153 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
4154 set if a regular file was being unlinked, and it was the last link
4155 - use PtlGet from the target for bulk writes (315)
4156 - OST now updates object mtime with writes or setattr (607, 619)
4157 - LDLM now has a grant-time callback to revalidate locked items, if
4159 - Many MDS operations were reorganized to combat race conditions
4161 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
4162 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
4164 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
4167 - Fully reactivate OST imports after reconnection (512, others)
4168 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
4169 - More graceful error handling for truncating on dead OST (515)
4170 - Don't error out unless we're actually accessing dead stripes (474)
4171 - Fix garbage sizes when stripes are missing (410)
4172 - LRU counters were broken, causing constant lock purge (433, 432)
4173 - garbage on read from stripes with failed OSTs (441)
4174 - mark OSCs as active before reconnecting during recovery (438)
4175 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
4176 - lfind did not preserve OST order in output (443)
4177 - symlinks cause hung clients, incorrect data (439)
4178 - stop dereferencing request after dropping refcount (457)
4179 - don't LASSERT(spin_is_locked) on non-SMP (455)
4180 - fixes for many rename() bugs
4181 - fstat didn't correctly synchronize attributes (399)
4182 - server must handle lock cancellation during blocking AST prep (487)
4183 - bulk descriptors were free()d too soon (511)
4184 - fix paths in lconf, which would load incorrect modules (451, 507)
4185 - fix confusing lconf 'host not found' error message (386)
4186 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
4187 - fix race condition in mdc_blocking_ast() for inode access (526)
4188 - fix lov_unpackmd() unpacking wrong number of stripes (537)
4189 - fix lov_set_osc_active() marking wrong OSC inactive (440)
4190 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
4191 - fix multiple writes of stripe MD to MDS (358, maybe 519)
4192 - fix lstripe in several ways (kernel side) (527)
4193 - fix request leak in ldlm_cli_enqueue (262)
4194 - incorrect OSC was marked inactive after OST failure
4195 - call mds_fs_cleanup before unmounting filesystem (524)
4196 - fix races between taking ns_lock and ldlm_lock_change_resource
4197 - fix races updating LOV export open file list
4198 - fix lov_enqueue error path, avoid decref-ing bad lock handle
4199 - fix recovery NULL deref in ldlm_cli_cancel_unused
4200 - fix some DLM races by using new hash table for lock handles (419)
4201 - permit the client to specify desired inodes, at replay
4202 - duplicate requests when we queue them for replay reintegration
4203 - fix last_rcvd offset calculation
4204 - sync after each recovered transaction, so we always make progress
4205 - never, not always, ERESTART requests without transnos
4206 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4208 - skip replay if the MDS says that the client is already connected
4209 - don't check for a recovery-enabled export to match lctl's UUID
4210 - don't INC_USE_COUNT for phantom exports
4211 - don't crash when cleaning up phantom exports (567)
4212 - don't double-finish or set replay data for errored mdc_open requests
4213 - abort requests when they time out, so we don't get old replies
4214 - send/receive replies for AST messages again
4215 - if the client says that it doesn't have the lock, cancel it on the
4217 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4218 mark it as destroyed, it will all work out when decref is called
4219 - fix module use counts (22, 581)
4221 - ASTs now expect a reply (server cancels lock on error reply)
4223 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4226 - fix many simultaneous client startup (392)
4227 - fix dentry->d_it clobbering
4228 - credentials weren't being shipped for readdir/getattr operations
4229 - remove invalid assertions triggered during some concurrent MD
4231 - proper Lustre versions added (336, 389)
4232 - fix memory leak for create error case (398)
4233 - fix LOV locking bug that would get cli/srv out of sync
4234 - fix echo client over LOV (409)
4235 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4236 - fix double-O_EXCL intent crash (424)
4237 - avoid sending multiple lock CANCELs (352)
4239 - MDS can do multi-client recovery (modulo bugs in new code)
4241 - many updates, edits, cleanups
4243 2002-11-18 Phil Schwan <phil@clusterfs.com>
4246 - fix null d_it dereference (346)
4247 - fix full OST/dbench hang (333)
4248 - fix permission problem with file removal (286)
4249 - fix removal of OSCs from LOV when they fail
4250 - fix NULL deref during bulk timeout (214)
4251 - fix problems related to multiple filesystems on one MDS (241)
4252 - fixed serious subtle metadata locking bugs
4253 - free locks on clients when inodes are removed due to memory
4255 - fix inode pointer in lock data (285)
4256 - partial support for multiple MDS on a single host (241)
4257 - data locks weren't cancelled at clear_inode time (290, 311)
4258 - intent locks could lead to unbounded lock growth (205)
4259 - added a maximum lock count, an LRU list, and a flusher
4260 - fix multiple rename (365)
4261 - properly abstracted the echo client
4262 - OSC locked 1 byte too many; fixed
4263 - rewrote brw callback code:
4264 - fixed recovery bugs related to LOVs (306)
4265 - fixed too-many-pages-in-one-write crash (191)
4266 - fixed (again) crash in sync_io_timeout (214)
4267 - probably fixed callback-related race (385)
4269 - Add capability to MDS protocol
4270 - LDLM cancellations and callbacks on different portals
4272 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4275 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4276 - defer open object destruction to close time (601981, 138)
4277 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4278 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4279 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4280 - don't call obd_setattr in ll_file_release for destroyed objects
4281 * protocol change to lustre_msg: move |version| and add |flags|
4282 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4283 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4284 * added replay of create, unlink, link and rename operations during
4285 MDS failover; recovery should be much more robust now
4286 * remove failed OSCs from LOVs (only lov_create uses this so far)
4287 * the lustre-HOWTO was brought (more) up to date (582544)
4289 2002-10-23 Phil Schwan <phil@clusterfs.com>
4292 - in-use dentries weren't being reused properly (617851)
4293 - prevent multiple LDLM setup (599178)
4294 - fix LOV size calculations for truncate (617853)
4295 - fix client handling of MDS intent errors (POSIX)
4296 - fix permission bug in lovstripe.c test (624321)
4297 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4298 - truncate past end of file could corrupt data
4299 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4300 - a race in recovery could return ETIMEDOUT to apps (623947)
4301 - building outside the source directory was fixed
4302 * the lustre-HOWTO was brought (more) up to date (582544)
4303 * major progress was made on recovery functionality
4305 2002-10-10 Phil Schwan <phil@clusterfs.com>
4308 - recovery deadlock fix
4309 - rm -rf causes LBUG fix (617817)
4310 - file open by multiple tasks fix (618962)
4311 - directory permissions bugs (602707 and 620007)
4312 - journal_stop fixed with locking (611313)
4313 - O_APPEND failures resolved (618273, perhaps 614459)
4314 - lconf PATH fix (619770)
4315 - IA64 build fix (621450)
4316 - RPC buffer sizes scale with amount of memory
4318 2002-10-01 Phil Schwan <phil@clusterfs.com>
4321 - locks would be cancelled without throwing away data pages,
4322 resulting in inconsistent data (605627)
4323 - inode attributes were not always being refreshed (605627, 612449)
4324 - lconf now continues to cleanup after lctl reports an error
4325 - MDS now enforces user permissions (602707)
4326 - lprocfs cleanup fixed, but not yet enabled (614157)
4327 - fixed infinite server hang, should a client not respond to an AST
4328 - avoid going into recovery if user calls readlink() with a buffer
4329 that's too small (613941)
4330 - AST RPCs no longer require replies (614867) -- this may be changed
4331 - don't crash server if client sends an IOV that's too big (611336)
4332 - fixed lock conversion deadlock (611892)
4333 - fixed the following of symlinks (614622)
4334 * recovery: the server can remove locks from a client that dies, other
4335 clients can make progress
4336 * more extN patch fixes
4337 * compile-time configurable ptlrpc buffer allocations
4339 - collaborative read cache document
4340 - Lustre Lite Performance CDR document-in-progress
4342 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4345 - fix typo in patch-2.4.18
4347 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4350 - clear ptlrpc request each time in handle_incoming_request()
4351 - unlink of files now destroys the object on the OST
4353 2002-09-19 Peter Braam <braam@clusterfs.com>
4355 * add hard link support
4356 * change obdfile creation method
4357 * kernel patch changed
4359 2002-09-19 Peter Braam <braam@clusterfs.com>
4362 - stack overflow bug in extN fixed
4364 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4366 * documentation updates
4367 - add man pages for config tools
4368 - update tests/README to describe testing with new config tools
4369 - finish metadata API descriptions
4370 * bug fixes and cleanups
4371 - statfs workaround for 16TB limit
4372 - LOV stripe allocation improved, can stripe on subset of OSTs
4373 - LOV file size/IO offset was wrong for files > 4GB in size
4374 - object EA data was being dropped, caused files to be unreadable
4375 - memory overflow with non-LOV OST caused memory corruption
4376 - fixed regression tests to work with new config tools, obdfilter
4377 - fixed bug when directory size became larger than 1 block
4378 - fixed bug (for single client case) when PWD was deleted
4379 - invalidate local directory pages when doing intent-based ops
4380 - avoid LDLM oops when lock callback contained bad data
4382 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4384 * documentation updates
4385 * bug fixes and cleanups
4386 - configuration tools
4389 - 64-bit compile warnings
4390 - 64-bit internal statfs data
4392 * test_brw on persistent OST devices
4394 * lprocfs (disabled)
4396 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4398 * documentation updates
4399 * bug fixes and cleanups
4400 * configuration tools
4402 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4405 * many small fixes to 0_5_4
4406 * io/network handling
4407 * thinkos in MDS operations
4409 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4412 * crucial basic fixes to 0.5.3
4413 * IOR, Iozone work over Elan
4416 2002-08-07 Phil Schwan <phil@clusterfs.com>
4417 * version 0_5_3, our first alpha
4418 * we use the new Portals iovs
4419 * documentation updates
4420 * bug fixes and cleanups
4421 * small changes in the DLM wire protocol
4423 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4424 * version 0_5_1 with some initial stability,
4425 * locking on MD and file I/O.
4426 * documentation updates
4427 * several bug fixes since 0.5.0
4428 * small changes in wire protocol
4430 2002-07-18 Phil Schwan <phil@clusterfs.com>
4432 * delivered as Lustre Light Alpha
4433 * fixed a crash after handling invalid MDS requests
4434 * fixed directory pages for architectures with non-4k pages sizes
4436 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4437 * release version v0_4_4
4438 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4440 * Fixes a number of interruption problems with OST operations.
4441 * Update documentation for portals header changes
4442 * Move all wire protocol structs/defines to lustre_idl.h
4443 * Fixes symlink length bug.
4444 * Add tcpdump to repository.
4446 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4447 * release version v0_4_3
4448 * Fixes statfs for inodes on extN.
4449 * Fixes bug in runtests which would delete /etc/hosts.
4450 * Use 64-bit object IDs wherever possible (not into VFS though)
4451 Remove ost_get_info, which is unused by lustre, and out of date.
4453 2002-07-03 Peter Braam <braam@clusterfs.com>
4454 * release version v0_4_2 Fixes a lookup error (type not passed)
4455 * move forward to head of Portals
4456 * move forward to latest Lustre kernel
4458 2002-06-25 Peter Braam <braam@clusterfs.com>
4459 * release version v0_4_1. Hopefully stable on single node use.