1 2007-08-27 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: 1.39.cfs8
15 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
19 Description: block reactivating mgc import until all deactivates complete
20 Details : Fix race when failing back MDT/MGS to itself (testing)
22 Severity : enhancement
24 Description: lfs setstripe enhancement
25 Details : Make lfs setstripe understand 'k', 'm' and 'g' for stripe size.
28 Frequency : mds/oss recovery
30 Description: llog ctxt is refrenced after it has been freed.
31 Details : llog ctxt refcount was added to avoide the race between ctxt free
32 and llog recovery process. Each llog user must hold ctxt refcount
33 before it access the llog. And the llog ctxt can only be freed
34 when its refcount is zero.
36 Severity : enhancement
38 Description: randomly memory allocation failure util
39 Details : Make lustre randomly failed allocating memory for testing purpose.
42 Frequency : only for SLES10
44 Description: Update kernel patch for SLES10 SP1
45 Details : Add patch blkdev_tunables-2.6-sles10.patch to 2.6-sles10.series.
47 Severity : enhancement
49 Description: omit set fsid for export NFS
50 Details : fix set/restore device id for avoid EMFILE error and mark lustre fs
51 as FS_REQUIRES_DEV for avoid problems with generate fsid.
55 Description: Add sync option to mount_lustre.c
56 Details : Client loses data written to lustre after a network interruption.
60 Description: Error message improvement.
61 Details : Merging of two LCONSOLE_ERROR_MSG into one.
65 Description: don't use GFP_* in generic Lustre code.
66 Details : Use cfs_alloc_* functions and CFS_* flags for code portability.
70 Description: obdclass is limited by single OBD_ALLOC(idarray)
71 Details : replace OBD_ALLOC/OBD_FREE with OBD_VMALLOC/OBD_VFREE
75 Description: updated patchess for new RHEL4 kernel
76 Details : Fixed ext3-unlink-race.patch per Kalpak's comment.
80 Description: warnings with build patchless client with vanila 2.6.19 and up
81 Details : change old ctl_table style and replace ctl_table/ctl_table_header
82 with cfs_sysctl_table_t/cfs_sysctl_table_header_t
86 Description: O_DIRECT bypasses client statistics.
87 Details : When running with O_DIRECT I/O, neither the client rpc_stats nor
88 read_ahead_stats were updated. Copied stats section from
89 osc_send_oap_rpc() into async_internal().
93 Frequency : only for Cray XT3
94 Description: peer credits not enough on many OST per OSS systems.
95 Details : Use new lnet way to add credits as we need those for pings and ASTs
99 Frequency : only with liblustre
100 Description: Liblustre is not releasing flock locks on file close.
101 Details : Release flock locks on file close.
105 Frequency : only for RHEL4
106 Description: Update kernel patches for kernel-2.6.9-55.0.2.EL
107 Details : Remove inode-nr_unused-2.6.9-rhel4.patch from 2.6-rhel4.series
108 Update target file and kernel config.
112 Description: lustre support for RHEL5
113 Details : Add support for RHEL5.
118 Description: ASSERTION(export != NULL) failed in target_handle_connect
119 Details : Assetion hit is result of rare race between disconnect and connet
120 to same nid. target_handle_connect found old connect cockie and
121 tried to reconnect, but can't find export for this cockie.
126 Description: umount blocks forever on error
127 Details : In result of wrong using obd_no_recov and obd_force flags client can
128 hand if cancel or some other requests is lost.
132 Frequency : Only for SLES
133 Description: sanity_quota fail test_1
134 Details : There are multiple occurences of $TSTUSR in SLES's /etc/group
135 file, which makes TSTID[2] inunique.
139 Frequency : Only for SLES9
140 Description: Kernel patches for SLES9 2.6.5-7.286 kernel
141 Details : Update target/ChangeLog/which_patch .
145 Frequency : Only for test_10 in sanity-quota.sh
146 Description: a bug in quota test code
147 Details : checking a test flag in a wrong place
151 Frequency : Only for test_13 in sanity-quota.sh
152 Description: a bug in quota test script
153 Details : shouldn't check overfull allocated quota
157 Description: jbd statistics
158 Details : Port older jbd statistics patch for sles10
163 Description: lvbo_init failed for resource with missing objects.
164 Details : Fix returning error if we do stat for file with missing/corrupted
165 objects and i_size set to all sum of size of all avaible objects.
166 if we to truncate/write to missing object - it is recreated.
171 Description: Oops in read and write path when failing to allocate lock.
172 Details : Check if lock allocation failed and return error back.
176 Frequency : When flocks are used.
177 Description: assertion failure in ldlm_cli_enquque_fini for non NULL lock.
178 Details : Flock locks might destroy just granted lock if it could be merged
179 with another existing flock, this is done in completion handler,
180 so teach ldlm_cli_enquque_fini that this is a valid case for
186 Description: reply_lock_interpret crash due to race with it and lock cancel.
187 Details : Do not replay locks that are being cancelled. Do not reference
188 locks by their address during replay, just by their handle.
191 --------------------------------------------------------------------------------
193 2007-07-30 Cluster File Systems, Inc. <info@clusterfs.com>
195 * Support for kernels:
198 2.6.16.46-0.14 (SLES 10)
199 2.6.18.8 vanilla (kernel.org)
200 * Client support for unpatched kernels:
201 (see http://wiki.lustre.org/index.php?title=Patchless_Client)
202 2.6.16 - 2.6.22 vanilla (kernel.org)
203 * Due to recently discovered recovery problems, we do not recommend
204 using patchless RHEL 4 clients with this or any earlier release.
205 * Recommended e2fsprogs version: 1.39.cfs8
206 * Note that reiserfs quotas are disabled on SLES 10 in this kernel.
207 * Starting with this release, the ldiskfs backing filesystem required
208 by Lustre is now in its own package, lustre-ldiskfs. This package
209 should be installed. It is versioned separately from Lustre and
210 may be released separately in future.
216 Description: server LBUG when shutting down
217 Details : Block umount forever until the mount refcount is zero rather
218 than giving up after an arbitrary timeout.
220 Severity : enhancement
222 Description: add optional extra BUILD_VERSION info
223 Details : add a new environment variable (namely LUSTRE_VERS) which allows
224 to override the lustre version.
227 Frequency : 2.6.18 servers only
229 Description: ll_kern_mount() doesn't release the module reference
230 Details : The ldiskfs module reference count never drops down to 0
231 because ll_kern_mount() doesn't release the module reference.
236 Description: server LBUG when using old ost_num_threads parameter
237 Details : Accept the old ost_num_threads parameter but warn that it
238 is deprecated, and fix an off-by-one error that caused an LBUG.
243 Description: Transient SCSI error results in persistent IO issue
244 Details : iobuf->dr_error is not reinitialized to 0 between two
248 Frequency : sometimes when underlying device returns I/O errors
250 Description: OSTs not going read-only during write failures
251 Details : OSTs are not remounted read-only when the journal commit threads
252 get I/O errors because fsfilt_ext3 calls journal_start/stop()
253 instead of the ext3 wrappers.
257 Description: poor connect scaling with increasing client count
258 Details : Don't run filter_grant_sanity_check for more than 100 exports
259 to improve scaling for large numbers of clients.
262 Frequency : SLES10 only
264 Description: sanity-quota.sh quotacheck failed: rc = -22
265 Details : Quotas cannot be enabled on SLES10.
268 Frequency : liblustre clients only
270 Description: getdirentries does not give error when run on compute nodes
271 Details : getdirentries does not fail when the size specified as an argument
272 is too small to contain at least one entry
274 Severity : enhancement
276 Description: Add LNET router traceability for debug purposes
277 Details : If a checksum failure occurs with a router as part of the
278 IO path, the NID of the last router that forwarded the bulk data
279 is printed so it can be identified.
284 Description: OST "spontaneously" evicts client; client has imp_pingable == 0
285 Details : Due to a race condition, liblustre clients were occasionally
288 Severity : enhancement
290 Description: lfs setstripe use optional parameters instead of postional
293 Severity : enhancement
295 Description: Nanosecond timestamp support for ldiskfs
296 Details : The on-disk ldiskfs filesystem has added support for nanosecond
297 resolution timestamps. There is not yet support for this at
298 the Lustre filesystem level.
301 Frequency : during server recovery
303 Description: MDS failing to send precreate requests due to OSCC_FLAG_RECOVERING
304 Details : request with rq_no_resend flag not awake l_wait_event if they get a
308 Frequency : nfs export on patchless client
310 Description: connectathon hang when test nfs export over patchless client
311 Details : Disconnected dentry cannot be found with lookup, so we do not need
312 to unhash it or make it invalid
315 Description: fix llapi_lov_get_uuids() to allow many OSTs to be returned
316 Details: : Change llapi_lov_get_uuids() to read the UUIDs from /proc instead
317 of using an ioctl. This allows lfsck for > 160 OSTs to succeed.
322 Description: open req refcounting wrong on reconnect
323 Details : If reconnect happened between getting open reply from server and
324 call to mdc_set_replay_data in ll_file_open, we will schedule
325 replay for unreferenced request that we are about to free.
326 Subsequent close will crash in variety of ways.
327 Check that request is still eligible for replay in
328 mdc_set_replay_data().
333 Description: disable writes to filesystem when reading health_check file
334 Details : the default for reading the health_check proc file has changed
335 to NOT do a journal transaction and write to disk, because this
336 can cause reads of the /proc file to hang and block HA state
337 checking on a healthy but otherwise heavily loaded system. It
338 is possible to return to the previous behaviour during configure
339 with --enable-health-write.
341 Severity : enhancement
343 Description: 64-bit inode version
344 Details: : Add a on-disk 64-bit inode version for ext3 to track changes made
345 to the inode. This will be required for version-based recovery.
350 Description: MDS fails to start if a duplicate client export is detected
351 Details : in some rare cases it was possible for a client to connect to
352 an MDS multiple times. Upon recovery the MDS would detect this
353 and fail during startup. Handle this more gracefully.
355 Severity : enhancement
357 Description: Add -o localflock option to simulate old noflock
359 Details : This will achieve local-only flock/fcntl locks
365 Description: log_commit_thread vs filter_destroy race leads to crash
366 Details : Take import reference before releasing llog record semaphore
371 Description: Wrong request locking in request set processing
372 Details : ptlrpc_check_set wrongly uses req->rq_lock for proctect add to
373 imp_delayed_list, in this place should be used imp_lock.
376 Frequency : when reconnection
378 Description: Grant Leak when osc reconnect to OST
379 Details : When osc reconnect ost, OST(filter) should check whether it
380 should grant more space to client by comparing fed_grant and
381 cl_avail_grant, and return the granted space to client instead
382 of "new granted" space, because client will call osc_init_grant
383 to update the client grant space info.
386 Frequency : when client reconnect to OST
388 Description: Grant Leak when osc do resend and replay bulk write
389 Details : When osc reconnect to OST, OST(filter)should clear grant info of
390 bulk write request, because the grant info while be sync between
391 OSC and OST when reconnect, and we should ignore the grant info
392 these of resend/replay write req.
397 Description: Grant space more than avaiable left space sometimes.
398 Details : When then OST is about to be full, if two bulk writing from
399 different clients came to OST. Accord the avaliable space of the
400 OST, the first req should be permitted, and the second one
401 should be denied by ENOSPC. But if the seconde arrived before
402 the first one is commited. The OST might wrongly permit second
403 writing, which will cause grant space > avaiable space.
406 Frequency : when client is evicted
408 Description: Grant might be wrongly erased when osc is evicted by OST
409 Details : when the import is evicted by server, it will fork another
410 thread ptlrpc_invalidate_import_thread to invalidate the
411 import, where the grant will be set to 0. While the original
412 thread will update the grant it got when connecting. So if
413 the former happened latter, the grant will be wrongly errased
414 because of this race.
419 Description: Checking Stale with correct fid
420 Details : ll_revalidate_it should uses de_inode instead of op_data.fid2
421 to check whether it is stale, because sometimes, we want the
422 enqueue happened anyway, and op_data.fid2 will not be initialized.
424 Severity : enhancement
426 Description: update patchless client
427 Details : Add support for patchless client with 2.6.20, 2.6.21 and RHEL 5
430 Frequency : only with 2.4 kernel
432 Description: random memory corruption
433 Details : size of struct ll_inode_info is to big for union inode.u and this
434 can be cause of random memory corruption.
439 Description: Memory leak in recovery
440 Details : Lov_mds_md was not free in an error handler in mds_create_object.
441 It should also check obd_fail before fsfilt_start, otherwise if
442 fsfilt_start return -EROFS,(failover mds during mds recovery).
443 then the req will return with repmsg->transno = 0 and rc = EROFS.
444 and we met hit the assert LASSERT(req->rq_reqmsg->transno ==
445 req->rq_repmsg->transno) in ptlrpc_replay_interpret. Fcc should
446 be freed no matter whether fsfilt_commit success or not.
449 Frequency : only with huge count clients
451 Description: Prevents from taking the superblock lock in llap_from_page for
453 Details : using LL_ORIGIN_REMOVEPAGE origin flag instead of LL_ORIGIN_UNKNOW
454 for llap_from_page call in ll_removepage prevents from taking the
455 superblock lock for a soon died page.
460 Description: Not check open intent error before release open handle
461 Details : in some rare cases, the open intent error is not checked before
462 release open handle, which may cause
463 ASSERTION(open_req->rq_transno != 0), because it tries to release
464 the failed open handle.
469 Description: Set cat log bitmap only after create log success.
470 Details : in some rare cases, the cat log bitmap is set too early. and it
471 should be set only after create log success.
475 Description: Accessing a block bevice can re-enable I/O when Lustre is
476 tearing down a device.
477 Details : dev_clear_rdonly(bdev) must be called in kill_bdev() instead of
482 Description: service threads may hog cpus when there are a lot of requests
483 Details : Insert cond_resched to give other threads a chance to use some CPU
488 Description: the cat log was not initialized in recovery
489 Details : When mds(mgs) do recovery, the tgt_count might be zero, so the
490 unlink log on mds will not be initialized until mds post
491 recovery. And also in mds post recovery, the unlink log will
492 initialization will be done asynchronausly, so there will be race
493 between add unlink log and unlink log initialization.
497 Description: brw_stats were being printed incorrectly
498 Details : brw_stats were being printed as log2 but all of them were not
499 recorded as log2. Also remove some code duplication arising from
500 filter_tally_{read,write}.
504 Frequency : rare, only in recovery.
505 Description: ASSERTION(req->rq_type != LI_POISON) failed
506 Details : imp_lock should be held while iterating over imp_sending_list for
507 prevent destroy request after get timeout in ptlrpc_queue_wait.
511 Description: replay-single.sh test 52 fails
512 Details : A lock's skiplist need to be cleanup when it being unlinked
513 from its resource list.
518 Description: Short directio read returns full requested size rather than
520 Details : Direct I/O operations should return actual amount of bytes
521 transferred rather than requested size.
523 Severity : enhancement
525 Description: metadata RPC reduction (e.g. for rm performance)
526 Details : decrease the amount of synchronous RPC between clients and servers
527 by canceling conflicing lock before the operation on the client
528 and packing thier handles into the main operation RPC to server.
530 Severity : enhancement
532 Description: add #ifdef HAVE_KERNEL_CONFIG_H
533 Details : kernels from 2.6.19 not need include linux/config.h, but add
534 include linux/autoconf.h in commpiler command line.
536 Severity : enhancement
538 Description: patchless client support for 2.6.22 kernel
539 Details : 2.6.22 has only one visble change, SLAB_CTOR_* constants is
540 removed. In this case we need drop using os depended interface to
541 kmem_cache and use cfs_mem_cache API.
546 Description: fix mal-formatted messages
547 Details : fix some mal-formatted DEBUG_REQ and LCONSOLE_ERROR_MSG messages
551 Frequency : always in liblustre
552 Description: wrong IS_ERR implementation in liblustre.h
553 Details : fix IS_ERR implementation in liblustre.h for right detect errors.
558 Description: Correct condition for output debug message.
559 Details : inode i_nlink equal zero is not enough for output message about
560 disk corruption, i_ctime and i_mode should be also checked.
564 Frequency : always in patchless client
565 Description: add configure check for truncate_complete_page
566 Details : improve checks for exported symbols. This allow run check without
567 sources, but with Module.symvers shipped with kernel distribution.
568 add check for truncate_complete_page used by patchless client.
573 Description: improve handling recoverable errors
574 Details : if request processig with error which can be recoverable on server
575 request should be resend, otherwise page released from cache and
580 Description: sanity.sh test_77h fails with "test_77h file compare failed"
581 Details : test_77h uses a file which was messed by other test case.
585 Description: Not Check whether lov_tgts is NULL in some lov functions
586 Details : Checking whether lov_tgts is NULL in some functions.
590 Description: replace obdo_alloc() with OBDO_ALLOC macro
591 Details : nothing special is done in obdo_alloc() function, and for
592 debugging purpose, it needs to be replaced with macros.
596 Description: bad return value and errno from fcntl call
597 Details : In liblustre API, errno should be a negative value if error
602 Description: ptlrpc_check_set() LBUG
603 Details : In case of positive reply from server and failed client bulk
604 callback after bulk transfer shouldn't LBUG, but process this
605 request as erroneous.
607 Severity : enhancement
609 Description: VFS operations stats tool.
610 Details : Tool which collects stats by tracking value written in pid,
611 ppid, gid and uses llstat to generate output to plot graph using
613 Updated lustre/utils/Makefile.am
614 Added lustre/utils/ltrack_stats.c
616 Severity : enhancement
618 Description: 2.6.18 server support (lustre 1.6.1)
619 Details : Support for 2.6.18 kernels on the server side.
624 Description: ASSERTION(imp->imp_conn_current) failed
625 Details : an assertion failure is hit if a client node boots and attempts to
626 mount a lustre filesystem faster than RECONNECT_INTERVAL seconds.
629 Frequency : only for i686
631 Description: 1.4.11 RC1 build fails for RHEL 4, i686
632 Details : Fixed config variable for build.
637 Description: Updated patchess for new RHEL4 kernel
638 Details : Updated patch inode-nr_unused-2.6.9-rhel4.patch
639 Updated patch jbd-stats-2.6.9.patch
640 Updated patch qsnet-rhel4-2.6.patch
641 Updated patch quota-deadlock-on-pagelock-core.patch
642 Updated patch vfs_intent-2.6-rhel4.patch
643 Updated patch vfs_races-2.6-rhel4.patch
644 Updated series file 2.6-rhel4-titech.series
645 Updated series file 2.6-rhel4.series
646 Updated kernel config files
651 Description: lquota slave complains LBUG when reconnecting with mds
653 Details : quota slave depends qctxt->lqc_import to send its quota request.
654 This pointer will be invalid if mds did failover or broke its
655 connect to osts, which leads to LBUG.
658 Frequency : when qunit size is too small(less than 20M)
660 Description: write is stopped by improper -EDQUOT
661 Details : If the master is busy and qunit size is small enough(let's say 1M),
662 the slave can not get quota from master on time, which will lead
663 slave to trigger a -EQUOTA to client.
668 Description: Deadlock during metadata tests
669 Details : in prune_dir_dentries(), shrink_dcache_parent() should not be
670 called with the per-dentry lock held.
673 Frequency : SLES9 only
675 Description: Lustre patched kernel for SLES9 SP3 has NR_CPUS set to 8
676 Details : set CONFIG_NR_CPUS to 128 instead of 8.
678 Severity : enhancement
680 Description: remove fs_prep_san_write operation and related patches
681 Details : remove the ext3-san-jdike patches which are no longer useful.
686 Description: LDISKFS-fs error (device sdc): ldiskfs_free_blocks
687 Details : a disk corruption can cause the mballoc code to assert on a
688 double free or other extent corruptions. Handle these with
689 ext3_error() instead of assertions.
692 Frequency : only with mballoc3 code and deep extent trees
694 Description: ldiskfs_ext_search_right: bad header in inode: unexpected eh_depth
695 Details : a wrong check of extent headers in ldiskfs_ext_search_right()
696 can cause the filesystem to be remounted read-only.
701 Description: lfsck built against 1.4.x cannot run against 1.6.0 lustre
702 Details : the definition for OBD_IOC_GETNAME changed in 1.6.0. One of the
703 few external users of this ioctl number is lfsck's call to
704 llapi_lov_get_uuids() and this caused lfsck to fail at startup.
705 Add the old ioctl number to the handler so both old and new
710 Description: parallel lock callbacks
711 Details : Instead of sending blocking and completion callbacks as separated
712 requests, adding them to a set and sending in parallel.
716 Description: Disable most debugging by default
717 Details : To improve performance, disable most logging (for debug purposes)
718 by default. VFSTRACE, RPCTRACE, and DLMTRACE are now off by
719 default, and HA includes fewer messages.
723 Frequency : only run on patchless client.
724 Description: use do_facet on sanity.sh for test handling recoverables errors
725 Details : use do_facet instead of direct use sysctl for set fail_loc on OST
729 Description: hash tables for lists of nids, connections and uuids
730 Details : Hash tables noticeably help when a lot of clients connect to a
731 server, to faster identify duplicate connections or reconnects,
732 also to faster find export to evict in manual eviction case.
736 Description: Sometimes, when the server evict a client, and the client will
737 not be evicted as soon as possible.
738 Details : In enqueue req, the error was returned by intent, instead of
739 rq_status which make ptlrpc layer not detect this error, and
740 does not evict the client. So enqueue error should be returned
744 Frequency : only at startup
746 Description: Delay client connections to MDT until fist MDT->OST connect
747 Details : If a client tried to create a new file before the MDT had
748 connected to any OSTs, the create would return EIO. Now
749 the client will simply block until the MDT connects to the
750 first OST and the create can succeed.
753 Frequency : only for SLES9
755 Description: Routinely utilize latest Quadrics drivers in CFS releases
756 Details : Update patch qsnet-suse-2.6.patch.
759 Frequency : only for sles10
761 Description: Update patches for SLES 10 SP1 kernel.
762 Details : Update the patch vfs_intent-2.6-sles10.patch.
767 Description: Routinely utilize latest Quadrics drivers in CFS releases
768 Details : Update patch qsnet-rhel4-2.6.patch.
771 Frequency : at statup only
773 Description: mds_lov_synchronize race leads to various problems
774 Details : simultaneous MDT->OST connections at startup can cause the
775 sync to abort, leaving the OSC in a bad state.
779 Description: Using wrong pointer in osc_brw_prep_request
780 Details : Access to array[-1] can produce panic if kernel compiled with
781 CONFIG_PAGE_ALLOC enabled
783 Severity : enhancement
785 Description: Async OSC create to avoid the blocking unnecessarily.
786 Details : If a OST has no remain object, system will block on the creating
787 when need to create a new object on this OST. Now, ways use
788 pre-created objects when available, instead of blocking on an
789 empty osc while others are not empty. If we must block, we block
790 for the shortest possible period of time.
794 Frequency : only in recovery
795 Description: Mark OST as early accessible if his start SYNC.
796 Details : osc_precreate return flag early accessible if oscc marked as
797 OSCC_FLAG_SYNC_IN_PROGRESS.
802 Description: Sometimes precreate code can triger create object on wrong ost
803 Details : Wrong protected or not not restored variables aftre precreate loop
804 can produce creation object on wrong ost.
807 Frequency : oss recovery
809 Description: llog_commit_thread cleanup should sync with llog_commit_thread
811 Details : llog_commit_thread_count should be synced between llog_commit
812 start and cleanup, so new llog_commit thread should not be started
813 when llog_commit threads being stopped to avoid accessing some
816 Severity : enhancement
818 Description: Add printing inode info into message about error in writepage.
820 --------------------------------------------------------------------------------
822 2007-05-03 Cluster File Systems, Inc. <info@clusterfs.com>
827 Frequency : on some architectures
829 Description: 1.6 client sometimes fails to mount from a 1.4 MDT
830 Details : Uninitialized flags sometimes cause configuration commands to
834 Frequency : patchless clients only
836 Description: missing __iget() symbol export
837 Details : The __iget() symbol export is missing. To avoid the need for
838 this on patchless clients the deathrow inode reaper is turned
839 off, and we depend on the VM to clean up old inodes. This
840 dependency was during via the fix for bug 12181.
845 Description: sanity.sh fail: test_52b
846 Details : The ll_inode_to_ext_flags() has a glitch which makes MDS return
847 incorrect inode's flags to client.
849 --------------------------------------------------------------------------------
851 2007-04-19 Cluster File Systems, Inc. <info@clusterfs.com>
853 * CONFIGURATION CHANGE. This version of Lustre WILL NOT
854 INTEROPERATE with older versions automatically. In many cases a
855 special upgrade step is needed. Please read the
856 user documentation before upgrading any part of a 1.4.x system.
857 * WARNING: Lustre configuration and startup changes are required with
858 this release. See https://mail.clusterfs.com/wikis/lustre/MountConf
860 * Support for kernels:
861 2.4.21-47.0.1.EL (RHEL 3)
863 2.6.9-42.0.10.EL (RHEL 4)
864 2.6.12.6 vanilla (kernel.org)
865 2.6.16.27-0.9 (SLES10)
866 * Client support for unpatched kernels:
867 (see https://mail.clusterfs.com/wikis/lustre/PatchlessClient)
868 2.6.16 - 2.6.19 vanilla (kernel.org)
869 2.6.9-42.0.8EL (RHEL 4)
870 * Recommended e2fsprogs version: 1.39.cfs6
871 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
874 Severity : enhancement
876 Description: Async OSC create to avoid the blocking unnecessarily.
877 Details : If a OST has no remain object, system will block on the creating
878 when need to create a new object on this OST. Now, ways use
879 pre-created objects when available, instead of blocking on an
880 empty osc while others are not empty. If we must block, we block
881 for the shortest possible period of time.
883 Severity : enhancement
885 Description: MountConf
886 Details : Lustre configuration is now managed via mkfs and mount
887 commands instead of lmc and lconf. New obd types (MGS, MGC)
888 are added for dynamic configuration management. See
889 https://mail.clusterfs.com/wikis/lustre/MountConf for
892 Severity : enhancement
894 Description: dynamic OST addition
895 Details : OSTs can now be added to a live filesystem
897 Severity : enhancement
899 Description: startup order invariance
900 Details : MDTs and OSTs can be started in any order. Clients only
901 require the MDT to complete startup.
903 Severity : enhancement
905 Description: parallel, asynchronous orphan cleanup
906 Details : orphan cleanup is now performed in separate threads for each
907 OST, allowing parallel non-blocking operation.
909 Severity : enhancement
911 Description: optimized stripe assignment
912 Details : stripe assignments are now made based on ost space available,
913 ost previous usage, and OSS previous usage, in order to try
914 to optimize storage space and networking resources.
916 Severity : enhancement
918 Description: Permanently set tunables
919 Details : All writable /proc/fs/lustre tunables can now be permanently
920 set on a per-server basis, at mkfs time or on a live
923 Severity : enhancement
925 Description: Lustre message v2
926 Details : Add lustre message format v2.
928 Severity : enhancement
930 Description: client OST exclusion list
931 Details : Clients can be started with a list of OSTs that should be
932 declared "inactive" for known non-responsive OSTs.
936 Description: ENOENT returned for valid filehandle during dbench.
937 Details : Check if a directory has children when invalidating dentries
938 associated with an inode during lock cancellation. This fixes
939 an incorrect ENOENT sometimes seen for valid filehandles during
943 Frequency : SFS test only (otherwise harmless)
945 Description: SPEC SFS validation failure on NFS v2 over lustre.
946 Details : Changes the blocksize for regular files to be 2x RPC size,
947 and not depend on stripe size.
949 Severity : enhancement
951 Description: fine-grained SMP locking inside DLM
952 Details : Improve DLM performance on SMP systems by removing the single
953 per-namespace lock and replace it with per-resource locks.
955 Severity : enhancement
957 Description: don't hold multiple extent locks at one time
958 Details : To avoid client eviction during large writes, locks are not
959 held on multiple stripes at one time or for very large writes.
960 Otherwise, clients can block waiting for a lock on a failed OST
961 while holding locks on other OSTs and be evicted.
963 Severity : enhancement
965 Description: Multiple MD RPCs in flight.
966 Details : Further unserialise some read-only MDT RPCs - learn about intents.
967 To avoid overly-overloading MDT, introduce a limit on number of
968 MDT RPCs in flight for a single client and add /proc controls
969 to adjust this limit.
971 Severity : enhancement
973 Description: client read/write statistics
974 Details : Add client read/write call usage stats for performance
975 analysis of user processes.
976 /proc/fs/lustre/llite/*/offset_stats shows non-sequential
977 file access. extents_stats shows chunk size distribution.
978 extents_stats_per_process show chunk size distribution per
981 Severity : enhancement
983 Description: per-client statistics on server
984 Details : Add ldlm and operations statistics for each client in
985 /proc/fs/lustre/mds|obdfilter/*/exports/
987 Severity : enhancement
989 Description: improved MDT statistics
990 Details : Add detailed MDT operations statistics in
991 /proc/fs/lustre/mds/*/stats
993 Severity : enhancement
995 Description: VFS operations stats
996 Details : Add client VFS call stats, trackable by pid, ppid, or gid
997 /proc/fs/lustre/llite/*/stats_track_[pid|ppid|gid]
1002 Description: Fix client-side osc byte counters
1003 Details : The osc read/write byte counters in
1004 /proc/fs/lustre/osc/*/stats are now working
1007 Frequency : always as root on SLES
1009 Description: Failure of copying files with lustre special EAs.
1010 Details : Client side always return success for setxattr call for lustre
1011 special xattr (currently only "trusted.lov").
1016 Description: Refcount LNET uuids
1017 Details : The global LNET uuid list grew linearly with every startup;
1018 refcount repeated list entries instead of always adding to
1021 Severity : enhancement
1023 Description: Dynamic service threads
1024 Details : Within a small range, start extra service threads
1025 automatically when the request queue builds up.
1028 Frequency : mixed-endian client/server environments
1030 Description: mixed-endian crashes
1031 Details : The new msg_v2 system had some failures in mixed-endian
1034 Severity : enhancement
1036 Description: Easy OST removal
1037 Details : OSTs can be permanently deactivated with e.g. 'lctl
1038 conf_param lustre-OST0001.osc.active=0'
1040 Severity : enhancement
1042 Description: MGS proc entries
1043 Details : Added basic proc entries for the MGS showing what filesystems
1046 Severity : enhancement
1048 Description: provide MGS failover
1049 Details : Added config lock reacquisition after MGS server failover.
1051 Severity : enhancement
1053 Description: add Linux 2.4 support
1054 Details : Added support for RHEL 2.4.21 kernel for 1.6 servers and clients
1058 Description: a large application tries to do I/O to the same resource and dies
1059 in the middle of it.
1060 Details : Check the req->rq_arrival time after the call to
1061 ost_brw_lock_get(), but before we do anything about
1062 processing it & sending the BULK transfer request. This
1063 should help move old stale pending locks off the queue as
1064 quickly as obd_timeout.
1067 Frequency : when an incorrect nid is specified during startup
1069 Description: ptlrpc connect to non-existant node causes kernel crash
1070 Details : LNET can't be re-entered from an event callback, which
1071 happened when we expire a message after the export has been
1072 cleaned up. Instead, hand the zombie cleanup off to another
1075 Severity : enhancement
1077 Description: plain/inodebits lock performance improvement
1078 Details : Grouping plain/inodebits in granted list by their request modes
1079 and bits policy, thus improving the performance of search through
1083 Frequency : only if OST filesystem is corrupted
1085 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1086 Details : for a short time RPCs with bulk IO are in the replay list,
1087 but replay of bulk IOs is unimplemented. If the OST filesystem
1088 is corrupted due to disk cache incoherency and then replay is
1089 started it is possible to trip an assertion. Avoid putting
1090 committed RPCs into the replay list at all to avoid this issue.
1093 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1095 Description: System hang on startup
1096 Details : This bug allowed the liblustre (e.g. catamount) client to
1097 return to the app before handling all startup RPCs. This
1098 could leave the node unresponsive to lustre network traffic
1099 and manifested as a server ptllnd timeout.
1101 Severity : enhancement
1103 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1104 Details : liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT
1105 boolean to control whether to print peer debug info when a
1106 client's RPC times out.
1109 Frequency : only for kernels with patches from Lustre below 1.4.3
1111 Description: Remove old rdonly API
1112 Details : Remove old rdonly API which unused from at least lustre 1.4.3
1115 Frequency : only for devices with external journals
1117 Description: Set external device read-only also
1118 Details : During a commanded failover stop, we set the disk device
1119 read-only while the server shuts down. We now also set any
1120 external journal device read-only at the same time.
1123 Frequency : when upgrading from 1.4 while trying to change parameters
1125 Description: The wrong (new) MDC name was used when setting parameters for
1126 upgraded MDT's. Also allows changing of OSC (and MDC)
1127 parameters if --writeconf is specified at tunefs upgrade time.
1130 Frequency : when setting specific ost indicies
1132 Description: QOS code breaks on skipped indicies
1133 Details : Add checks for missing OST indicies in the QOS code, so OSTs
1134 created with --index need not be sequential.
1136 Severity : enhancement
1138 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1139 Details : The uninit_groups feature works in conjunction with the kernel
1140 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1141 up the pass1 processing of e2fsck. This is a read-only feature
1142 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1143 mount filesystems that have had this feature enabled.
1145 Severity : enhancement
1147 Description: Improve multi-block allocation algorithm to avoid fragmentation
1148 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1149 allocation locality and avoid filesystem fragmentation.
1151 ------------------------------------------------------------------------------
1153 2007-04-01 Cluster File Systems, Inc. <info@clusterfs.com>
1155 * Support for kernels:
1156 2.4.21-47.0.1.EL (RHEL 3)
1157 2.6.5-7.283 (SLES 9)
1158 2.6.9-42.0.10.EL (RHEL 4)
1159 2.6.12.6 vanilla (kernel.org)
1160 2.6.16.27-0.9 (SLES 10)
1161 * Recommended e2fsprogs version: 1.39.cfs5
1163 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1167 Frequency : occasional, depends on client load and configuration
1168 Bugzilla : 12181, 12203
1169 Description: data loss for recently-modified files
1171 Details : In some cases it is possible that recently written or created
1172 files may not be written to disk in a timely manner (this should
1173 normally be within 30s unless client IO load is very high).
1174 The problem appears as zero-length files or files that are a
1175 multiple of 1MB in size after a client crash or client eviction
1176 that are missing data at the end of the file.
1178 This problem is more likely to be hit on clients where files are
1179 repeatedly created and unlinked in the same directory, clients
1180 have a large amount of RAM, have many CPUs, the filesystem has
1181 many OSTs, the clients are rebooted frequently, and/or the files
1182 are not accessed by other nodes after being written.
1184 The presence of the problem can be detected by looking at
1185 /proc/sys/fs/inode-state. If the first number (nr_inodes) is
1186 smaller than the second (nr_unused) then dirty files will not
1187 be flushed automatically to disk. "sync; sleep 10" should be
1188 run several times on the node before unmounting it to update
1189 Lustre (this is also safe to run on nodes without this problem).
1191 There is also a related kernel bug in the RHEL4 4 2.6.9 kernel
1192 that can cause this same problem, so customers using that kernel
1193 also need to update the kernel in addition to Lustre. In order
1194 to properly fix this bug, the RHEL3 2.4.21 kernel is also updated.
1196 It is normal that files written just before a client crash (less
1197 than 30s) may not yet have been flushed to disk, even for local
1201 Frequency : frequent on thin XT3 nodes
1203 Description: UUID collision on thin XT3 Linux nodes
1204 Details : UUIDs on Compute Node Linux XT3 nodes were not generated
1205 randomly, since we relied on an insufficiently-seeded PRNG.
1210 Description: OSS hangs after "All ost request buffers busy"
1211 Details : A deadlock between quota and journal operations caused OSS
1212 hangs after printing "All ost request buffers busy."
1215 Frequency : always on liblustre builds
1217 Description: Cleanup compiler warnings on liblustre
1220 Frequency : always on liblustre builds on XT3
1222 Description: LC_CONFIG_CDEBUG don't run while build liblustre on XT3.
1226 Description: Addition of EXT3_FEATURE_RO_COMPAT_DIR_NLINKS flag for
1227 > 32000 subdirectories
1228 Details : Add EXT3_FEATURE_RO_COMPAT_DIR_NLINK flag to
1229 EXT3_FEATURE_RO_COMPAT_SUPP. This flag will be set whenever
1230 subdirectory count crosses 32000. This will aid e2fsck to
1231 correctly handle more than 32000 subdirectories.
1234 Frequency : liblustre (e.g. catamount) on a large cluster with >= 8 OSTs/OSS
1236 Description: System hang on startup
1237 Details : This bug allowed the liblustre (e.g. catamount) client to
1238 return to the app before handling all startup RPCs. This
1239 could leave the node unresponsive to lustre network traffic
1240 and manifested as a server ptllnd timeout.
1242 Severity : enhancement
1244 Description: Add "/proc/sys/lustre/debug_peer_on_timeout"
1245 (liblustre envirable: LIBLUSTRE_DEBUG_PEER_ON_TIMEOUT)
1246 boolean to control whether to print peer debug info when a
1247 client's RPC times out.
1252 Description: make O_SYNC working on 2.6 kernels
1253 Details : 2.6 kernels use different method for mark pages for write,
1254 so need add a code to lustre for O_SYNC work.
1259 Description: Failure to close file and release space on NFS
1260 Details : Put inode details into lock acquired in ll_intent_file_open.
1261 Use mdc_intent_lock in ll_intent_open to properly
1262 detect all kind of errors unhandled by mdc_enqueue.
1267 Description: proc file read during shutdown sometimes raced obd removal,
1269 Details : Add lock to prevent obd access after proc file removal.
1272 Frequency : Only for files larger than 4GB on 32-bit clients.
1274 Description: improperly doing page alignment of locks
1275 Details : Modify lustre core code to use CFS_PAGE_* defines instead of
1276 PAGE_*. Make CFS_PAGE_MASK a 64-bit mask.
1281 Description: RPCs being resent when they shouldn't be
1282 Details : Some RPCs that should not be resent are being resent. This
1283 can cause inconsistencies in the RPC state machine. Do not
1284 resend such requests.
1287 Frequency : rare, only with NFS export
1289 Description: Crash on NFS re-export node
1290 Details : under very unusual load conditions an assertion is hit in
1291 ll_intent_file_open()
1294 Frequency : only if OST filesystem is corrupted
1296 Description: client incorrectly hits assertion in ptlrpc_replay_req()
1297 Details : for a short time RPCs with bulk IO are in the replay list,
1298 but replay of bulk IOs is unimplemented. If the OST filesystem
1299 is corrupted due to disk cache incoherency and then replay is
1300 started it is possible to trip an assertion. Avoid putting
1301 committed RPCs into the replay list at all to avoid this issue.
1306 Description: large O_DIRECT requests fail under memory pressure/fragmentation
1307 Details : Large single O_DIRECT read and write calls can fail to allocate
1308 a sufficiently large buffer to process the request. In case of
1309 allocation failure the allocation is retried with a smaller
1310 buffer and broken into smaller requests.
1312 Severity : enhancement
1314 Description: Add -o localflock option to simulate old noflock behaviour.
1315 Details : This will achieve local-only flock/fcntl locks coherentness.
1320 Description: versioning check is incomplete
1321 Details : Checking the version difference of client vs. server, report
1322 error if the gap is too big.
1327 Description: add support PG_writeback bit
1328 Details : add support for PG_writeback bit for Lustre, for more carefull
1329 work with page cache in 2.6 kernel. This also fix some deadlocks
1330 and remove hack for work O_SYNC with 2.6 kernel.
1332 Severity : enhancement
1334 Description: Add uninit_groups feature to ldiskfs2 to speed up e2fsck
1335 Details : The uninit_groups feature works in conjunction with the kernel
1336 filesystem code (ldiskfs2 only) and e2fsprogs-1.39-cfs6 to speed
1337 up the pass1 processing of e2fsck. This is a read-only feature
1338 in ldiskfs2 only, so older kernels and current ldiskfs cannot
1339 mount filesystems that have had this feature enabled.
1341 Severity : enhancement
1343 Description: Improve multi-block allocation algorithm to avoid fragmentation
1344 Details : The mballoc3 code (ldiskfs2 only) adds new mechanisms to improve
1345 allocation locality and avoid filesystem fragmentation.
1347 ------------------------------------------------------------------------------
1349 2007-02-09 Cluster File Systems, Inc. <info@clusterfs.com>
1351 * Support for kernels:
1352 2.6.9-42.0.3.EL (RHEL 4)
1353 2.6.5-7.276 (SLES 9)
1354 2.4.21-47.0.1.EL (RHEL 3)
1355 2.6.12.6 vanilla (kernel.org)
1356 2.6.16.21-0.8 (SLES10)
1357 * Recommended e2fsprogs version: 1.39.cfs2-0
1359 * The backwards-compatible /proc/sys/portals symlink has been removed
1360 in this release. Before upgrading, please ensure that you change
1361 any configuration scripts or /etc/sysctl.conf files that access
1362 /proc/sys/portals/* or sysctl portals.* to use the corresponding
1363 entry in /proc/sys/lnet or sysctl lnet.*. This change can be made
1364 in advance of the upgrade on any system running Lustre 1.4.6 or
1365 newer, since /proc/sys/lnet was added in that version.
1366 * Note that reiserfs quotas are disabled on SLES 10 in this kernel
1370 Frequency : only when quota is used
1372 Description: avoid scanning export list for quota master
1373 Details : Change the algorithms to avoid scanning export list in order
1374 to improve the efficiency.
1377 Frequency : MDS failover only, very rarely
1379 Description: "went back in time" messages on mds failover
1380 Details : The greatest transno may be lost when the current operation
1381 finishes with an error (transno==0) and the client's last_rcvd
1382 record is over-written. Save the greatest transno in the
1383 mds_last_transno for this case.
1386 Frequency : always for specific kernels and striping counts
1388 Description: client may get "Matching packet too big" without ACL support
1389 Details : Clients compiled without CONFIG_FS_POSIX_ACL get an error message
1390 when trying to access files in certain configurations. The
1391 clients should in fact be denied when mounting because they do
1392 not understand ACLs.
1395 Frequency : Cray XT3 with more than 4000 clients and multiple jobs
1397 Description: many clients connecting with IO in progress causes connect timeouts
1398 Details : Avoid synchronous journal commits to avoid delays caused by many
1399 clients connecting/disconnecting when bulk IO is in progress.
1400 Queue liblustre connect requests on OST_REQUEST_PORTAL instead of
1401 OST_IO_PORTAL to avoid delays behind potentially many pending
1405 Frequency : occasionally with multiple writers to a single file
1407 Description: shared writes to file may result in wrong size reported by stat()
1408 Details : Allow growing of kms when extent lock is cancelled
1411 Frequency : always with random mmap IO to multi-striped file
1413 Description: mmap write might be lost if we are writing to a 'hole' in stripe
1414 Details : Only if the hole is at the end of OST object so that kms is too
1415 small. Fix is to increase kms accordingly in ll_nopage.
1418 Frequency : rare, only if OST filesystem is inconsistent with MDS filesystem
1420 Description: writes to a missing object would leak memory on the OST
1421 Details : If there is an inconsistency between the MDS and OST filesystems,
1422 such that the MDS references an object that doesn't exist, writes
1423 to that object will leak memory due to incorrect cleanup in the
1424 error handling path, eventually running out of memory on the OST.
1429 Description: Creating too long symlink causes lustre errors
1430 Details : Check symlink and name lengths before sending requests to MDS.
1433 Frequency : only if flock is enabled (not on by default)
1435 Description: posix locks not released on fd closure on 2.6.9+
1436 Details : We failed to add posix locks to list of inode locks on 2.6.9+
1437 kernels, this caused such locks not to be released on fd close and
1438 then assertions on fs unmount about still used locks.
1441 Frequency : MDS failover only, very rarely
1443 Description: clients may get ASSERTION(granted_lock != NULL)
1444 Details : When request was taking a long time, and a client was resending
1445 a getattr by name lock request. The were multiple lock requests
1446 with the same client lock handle and
1447 mds_getattr_name->fixup_handle_for_resent_request found one of the
1448 lock handles but later failed with ASSERTION(granted_lock != NULL).
1453 Description: handle->h_buffer_credits > 0, assertion failure
1454 Details : h_buffer_credits is zero after truncate, causing assertion
1455 failure. This patch extends the transaction or creates a new
1459 Frequency : NFS re-export or patchless client
1460 Bugzilla : 11179, 10796
1461 Description: Crash on NFS re-export node (__d_move)
1462 Details : We do not want to hash the dentry if we don't have a lock.
1463 But if this dentry is later used in d_move, we'd hit uninitialised
1464 list head d_hash, so we just do this to init d_hash field but
1465 leave dentry unhashed.
1468 Frequency : NFS re-export or patchless client
1470 Description: NFS exports has problem with symbolic link
1471 Details : lustre client didn't properly install dentry when re-exported
1472 to NFS or running patchless client.
1475 Frequency : NFS re-export or patchless client
1477 Description: Various nfs/patchless fixes.
1478 Details : fixes reuse disconected alias for lookup process - this fixes
1479 warning "find_exported_dentry: npd != pd",
1480 fix permission error with open files at nfs.
1481 fix apply umask when do revalidate.
1484 Frequency : occasional
1486 Description: Crash on NFS re-export node
1487 Details : calling clear_page() on the wrong pointer triggered oops in
1488 generic_mapping_read().
1491 Frequency : rarely, using O_DIRECT IO
1493 Description: unaligned directio crashes client with LASSERT
1494 Details : check for unaligned buffers before trying any requests.
1497 Frequency : rarely, using CFS RAID5 patches in non-standard kernel series
1499 Description: stale data returned from RAID cache
1500 Details : If only a small amount of IO is done to the RAID device before
1501 reading it again it is possible to get stale data from the RAID
1502 cache instead of reading it from disk.
1505 Frequency : always for sles10 kernel
1507 Description: sles10 support
1508 Details : ll_follow_link: compile fixes and using of nd_set_link
1509 under newer kernels.
1512 Frequency : depends on arch, kernel and compiler version, always on sles10
1515 Description: recursive or deep enough symlinks cause stack overflow
1516 Details : getting rid of large stack-allocated variable in
1520 Frequency : depends on hardware
1522 Description: lustre write performance loss in the SLES10 kernel
1523 Details : the performance loss is caused by using of write barriers in the
1524 ext3 code. The SLES10 kernel turns barrier support on by
1525 default. The fix is to undo that change for ldiskfs.
1527 ------------------------------------------------------------------------------
1529 2006-12-09 Cluster File Systems, Inc. <info@clusterfs.com>
1531 * Support for kernels:
1532 2.6.9-42.0.3EL (RHEL 4)
1533 2.6.5-7.276 (SLES 9)
1534 2.4.21-47.0.1.EL (RHEL 3)
1535 2.6.12.6 vanilla (kernel.org)
1539 Frequency : quota enabled and large files being deleted
1541 Description: releasing more than 4GB of quota at once hangs OST
1542 Details : If a user deletes more than 4GB of files on a single OST it
1543 will cause the OST to spin in an infinite loop. Release
1544 quota in < 4GB chunks, or use a 64-bit value for 1.4.7.1+.
1549 Description: statfs data retrieved from /proc may be stale or zero
1550 Details : When reading per-device statfs data from /proc, in the
1551 {kbytes,files}_{total,free,avail} files, it may appear
1552 as zero or be out of date.
1555 Frequency : systems with MD RAID1 external journal devices
1557 Description: lconf's call to blkid is confused by RAID1 journal devices
1558 Details : Use the "blkid -l" flag to locate the MD RAID device instead
1559 of returning all block devices that match the journal UUID.
1562 Frequency : always, for aggregate stripe size over 4GB
1564 Description: "lfs setstripe" fails assertion when setting 4GB+ stripe width
1565 Details : Using "lfs setstripe" to set stripe size * stripe count over 4GB
1566 will fail the kernel with "ASSERTION(lsm->lsm_xfersize != 0)"
1569 Frequency : always if "lfs find" used on a local file/directory
1571 Description: "lfs find" segfaults if used on a local file/directory
1572 Details : The case where a directory component was not specified wasn't
1573 handled correctly. Handle this properly.
1576 Frequency : always on ppc64
1578 Description: the write to an ext3 filesystem mounted with mballoc got stuck
1579 Details : ext3_mb_generate_buddy() uses find_next_bit() which does not
1580 perform endianness conversion.
1583 Frequency : rarely (truncate to non-zero file size after write under load)
1584 Bugzilla : 10730, 10687
1585 Description: Files padded with zeros to next 4K multiple
1586 Details : With filesystems mounted using the "extents" option (2.6 kernels)
1587 it is possible that files that are truncated to a non-zero size
1588 immediately after being written are filled with zero bytes beyond
1589 the truncated size. No file data is lost.
1591 Severity : enhancement
1593 Description: Allow recovery/failover for liblustre clients.
1594 Details : liblustre clients were unaware of failover configurations until
1597 Severity : enhancement
1599 Description: user file locks should fail when not mounting with flock option
1600 Details : Set up an error-returning stub in ll_file_operations.lock field
1601 to prevent incorrect behaviour when client is mounted without
1602 flock option. Also, set up properly f_op->flock field for
1606 Frequency : always on ia64
1608 Description: "lfs df" loops on printing out MDS statfs information
1609 Details : The obd_ioctl_data was not initialized and in some systems
1610 this caused a failure during the ioctl that did not return
1611 an error. Initialize the struct and return an error on failure.
1614 Frequency : SLES 9 only
1616 Description: Error of copying files with lustre special EAs as root
1617 Details : Client side always return success for setxattr call for lustre
1618 special xattr (currently only "trusted.lov").
1621 Frequency : rarely on clusters with both ia64+i386 clients
1623 Description: ia64+i686 clients doing shared IO on the same file may LBUG
1624 Details : In rare cases when both ia64+i686 (or other mixed-PAGE_SIZE)
1625 clients are doing concurrent writes to the same file it is
1626 possible that the ia64 clients may LASSERT because the OST
1627 extent locks are not PAGE_SIZE aligned. Ensure that grown
1628 locks are always aligned on the request boundary.
1631 Frequency : specific use, occasional
1633 Description: Overwriting in use executable truncates on-disk binary image
1634 Details : If one node attempts to overwrite an executable in use by
1635 another node, we now correctly return ETXTBSY instead of
1636 truncating the file.
1638 Severity : enhancement
1640 Description: Async OSC create to avoid the blocking unnecessarily.
1641 Details : If a OST has no remain object, system will block on the creating
1642 when need to create a new object on this OST. Now, ways use
1643 pre-created objects when available, instead of blocking on an
1644 empty osc while others are not empty. If we must block, we block
1645 for the shortest possible period of time.
1650 Description: chmod on Lustre root is propagated to other clients
1651 Details : Re-validate root's dentry in ll_lookup_it to avoid having it
1652 invalid by the follow_mount time.
1655 Frequency : liblustre clients only
1657 Description: Race in 'instant cancel' lock handling could lead to such locks
1658 never to be granted in case of SMP MDS
1659 Details : Do not destroy not yet granted but cbpending locks in
1663 Frequency : replay/resend of open
1665 Description: non null lock assetion failure in mds_intent_policy
1666 Details : Trying to replay/resend lockless open requests resulted in
1667 mds_open() returning 0 with no lock. Now it sets a flag if
1668 a lock is going to be returned.
1670 Severity : enhancement
1672 Description: Checksum enhancements
1673 Details : New checksum enhancements allow for resending RPCs that failed
1676 Severity : enhancement
1678 Description: Tunables on number of dirty pages in cacche
1679 Details : Allow to set limit on number of dirty pages cached.
1684 Description: client crash on unmount - lock still has references
1685 Details : In some error handling cases it was possible to leak a lock
1686 reference on a client while accessing a file. This was not
1687 harmful to the client during operation, but would cause the
1688 client to crash when the filesystem is unmounted.
1691 Frequency : specific case, rare
1693 Description: ETXTBSY on mds though file not in use by client
1694 Details : ETXTBSY is no longer incorrectly returned when attempting to
1695 chmod or chown a directory that the user previously tried to
1696 execute or a currently-executing binary.
1699 Frequency : extremely rare except on liblustre-based clients
1701 Description: Lustre space not freed when files are deleted
1702 Details : Clean up open-unlinked files after client eviction. Previously
1703 the unlink was skipped and the files remained as orphans.
1708 Description: OST failure "would be an LBUG" in waiting_locks_callback()
1709 Details : In some cases it was possible to send a blocking callback to
1710 a client doing a glimpse, even though that client didn't get
1711 a lock granted. When the glimpse lock is cancelled on the OST
1712 the freed lock is left on the waiting list and corrupted the list.
1715 Frequency : all core dumps
1717 Description: Broke core dumps to lustre
1718 Details : Negative dentry may be unhashed if parent does not have UPDATE
1719 lock, but some callers, e.g. do_coredump, expect dentry to be
1720 hashed after successful create, hash it in ll_create_it.
1722 ------------------------------------------------------------------------------
1724 2006-09-13 Cluster File Systems, Inc. <info@clusterfs.com>
1726 * Support for kernels:
1727 2.6.9-42.0.2.EL (RHEL 4)
1728 2.6.5-7.276 (SLES 9)
1729 2.4.21-40.EL (RHEL 3)
1730 2.6.12.6 vanilla (kernel.org)
1734 Frequency : always on RHEL 3
1736 Description: Number of open files grows over time
1737 Details : The number of open files grows over time, whether or not
1738 Lustre is started. This was due to a filp leak introduced
1739 by one of our kernel patches.
1741 ------------------------------------------------------------------------------
1743 2006-08-20 Cluster File Systems, Inc. <info@clusterfs.com>
1745 * Support for kernels:
1746 2.6.9-42.EL (RHEL 4)
1747 2.6.5-7.267 (SLES 9)
1748 2.4.21-40.EL (RHEL 3)
1749 2.6.12.6 vanilla (kernel.org)
1754 Bugzilla : 5719, 9635, 9792, 9684
1755 Description: OST (or MDS) trips assertions in (re)connection under heavy load
1756 Details : If a server is under heavy load and cannot reply to new
1757 connection requests before the client resends the (re)connect,
1758 the connection handling code can behave badly if two service
1759 threads are concurrently handing separate (re)connections from
1760 the same client. Add better locking to the connection handling
1761 code, and ensure that only a single connection will be processed
1762 for a given client UUID, even if the lock is dropped.
1764 Severity : enhancement
1766 Description: add TCP zero-copy support to kernel
1767 Details : Add support to the kernel TCP stack to allow zero-copy bulk
1768 sends if the hardware supports scatter-gather and checksumming.
1769 This allows socklnd to do client-write and server-read more
1770 efficiently and reduce CPU utilization from skbuf copying.
1773 Frequency : only if NFS exporting from client
1775 Description: NULL pointer deref in ll_iocontrol() if chattr mknod file
1776 Details : If setting attributes on a file created under NFS that had
1777 never been opened it would be possible to oops the client
1778 if the file had no objects.
1782 Bugzilla : 9326, 10402, 10897
1783 Description: client crash in ptlrpcd_wake() thread when sending async RPC
1784 Details : It is possible that ptlrpcd_wake() dereferences a freed async
1785 RPC. In rare cases the ptlrpcd thread alread processed the RPC
1786 before ptlrpcd_wake() was called and the request was freed.
1789 Frequency : always for liblustre
1791 Description: liblustre client does MDS+OSTs setattr RPC for each write
1792 Details : When doing a write from a liblustre client, the client
1793 incorrectly issued an RPC to the MDS and each OST the file was
1794 striped over in order to update the timestamps. When writing
1795 with small chunks and many clients this could overwhelm the MDS
1796 with RPCs. In all cases it would slow down the write because
1797 these RPCs are unnecessary.
1799 Severity : enhancement
1801 Description: allow number of MDS service threads to be changed at module load
1802 Details : It is now possible to change the number of MDS service threads
1803 running. Adding "options mds mds_num_threads={N}" to the MDS's
1804 /etc/modprobe.conf will set the number of threads for the next
1805 time Lustre is restarted (assuming the "mds" module is also
1806 reloaded at that time). The default number of threads will
1807 stay the same, 32 for most systems.
1812 Description: OST crash if filesystem is unformatted or corrupt
1813 Details : If an OST is started on a device that has never been formatted
1814 or if the filesystem is corrupt and cannot even mount then the
1815 error handling cleanup routines would dereference a NULL pointer.
1820 Description: NULL pointer deref in llap_from_page.
1821 Details : get_cache_page_nowait can return a page with NULL (or otherwise
1822 incorrect) mapping if the page was truncated/reclaimed while it was
1823 searched for. Check for this condition and skip such pages when
1824 doing readahead. Introduce extra check to llap_from_page() to
1825 verify page->mapping->host is non-NULL (so page is not anonymous).
1828 Frequency : Sometimes when using sys_sendfile
1830 Description: "page not covered by a lock" warnings from ll_readpage
1831 Details : sendfile called ll_readpage without right page locks present.
1832 Now we introduced ll_file_sendfile that does necessary locking
1833 around call to generic_file_sendfile() much like we do in
1837 Frequency : with certain MDS communication failures at client mount time
1839 Description: NULL pointer deref after failed client mount
1840 Details : a client connection request may delayed by the network layer
1841 and not be sent until after the PTLRPC layer has timed out the
1842 request. If the client fails the mount immediately it will try
1843 to clean up before the network times out the request. Add a
1844 reference from the request import to the obd device and delay
1845 the cleanup until the network drops the request.
1848 Frequency : occasionally during client (re)connect
1850 Description: assertion failure during client (re)connect
1851 Details : processing a client connection request may be delayed by the
1852 client or server longer than the client connect timeout. This
1853 causes the client to resend the connection request. If the
1854 original connection request is replied in this interval, the
1855 client may trip an assertion failure in ptlrpc_connect_interpret()
1856 which thought it would be the only running connect process.
1859 Frequency : only with obd_echo servers and clients that are rebooted
1861 Description: kernel BUG accessing uninitialized data structure
1862 Details : When running an obd_echo server it did not start the ping_evictor
1863 thread, and when a client was evicted an uninitialized data
1864 structure was accessed. Start the ping_evictor in the RPC
1865 service startup instead of the OBD startup.
1867 Severity : enhancement
1868 Bugzilla : 10193 (patchless)
1869 Description: Remove dependency on various unexported kernel interfaces.
1870 Details : No longer need reparent_to_init, exit_mm, exit_files,
1871 sock_getsockopt, filemap_populate, FMODE_EXEC, put_filp.
1874 Frequency : rare (only users of deprecated and unsupported LDAP config)
1876 Description: write_conf for zeroconf mount queried LDAP incorrectly for client
1877 Details : LDAP apparently contains 'lustreName' attributes instead of
1878 'name'. A simple remapping of the name is sufficient.
1881 Frequency : rare (only with non-default dump_on_timeout debug enabled)
1883 Description: waiting_locks_callback trips kernel BUG if client is evicted
1884 Details : Running with the dump_on_timeout debug flag turned on makes
1885 it possible that the waiting_locks_callback() can try to dump
1886 the Lustre kernel debug logs from an interrupt handler. Defer
1887 this log dumping to the expired_lock_main() thread.
1889 Severity : enhancement
1891 Description: Support NFS exporting on 2.6 kernels.
1892 Details : Implement non-rawops metadata methods for NFS server to use without
1893 changing NFS server code.
1896 Frequency : very rare (synthetic metadata workload only)
1898 Description: two racing renames might cause an MDS thread to deadlock
1899 Details : Running the "racer" program may cause one MDS thread to rename
1900 a file from being the source of a rename to being the target of
1901 a rename at exactly the same time that another thread is doing
1902 so, and the second thread has already enqueued these locks after
1903 doing a lookup of the target and is trying to relock them in
1904 order. Ensure that we don't try to re-lock the same resource.
1907 Frequency : only very large systems with liblustre clients
1909 Description: slow eviction of liblustre clients with the "evict_by_nid" RPC
1910 Details : Use asynchronous set_info RPCs to send the "evict_by_nid" to
1911 all OSTs in parallel. This allows the eviction of stale liblustre
1912 clients to proceed much faster than if they were done in series,
1913 and also offers similar improvements for other set_info RPCs.
1918 Description: excessive CPU usage during initial read phase on client
1919 Details : During the initial read phase on a client, it would agressively
1920 retry readahead on the file, consuming too much CPU and impacting
1921 performance (since 1.4.5.8). Improve the readahead algorithm
1922 to avoid this, and also improve some other common cases (read
1923 of small files in particular, where "small" is files smaller than
1924 /proc/fs/lustre/llite/*/max_read_ahead_whole_mb, 2MB by default).
1929 Description: MDS crash when receiving packet with unknown intent.
1930 Details : Do not LBUG in unknown intent case, just return -EFAULT
1932 Severity : enhancement
1933 Bugzilla : 9293, 9385
1934 Description: MDS RPCs are serialised on client. This is unnecessary for some.
1935 Details : Do not serialize getattr (non-intent version) and statfs.
1938 Frequency : occasional, when OST network is overloaded/intermittent
1940 Description: client evicted by OST after bulk IO timeout
1941 Details : If a client sends a bulk IO request (read or write) the OST
1942 may evict the client if it is unresposive to its data GET/PUT
1943 request. This is incorrect if the network is overloaded (takes
1944 too long to transfer the RPC data) or dropped the OST GET/PUT
1945 request. There is no need to evict the client at all, since
1946 the pinger and/or lock callbacks will handle this, and the
1947 client can restart the bulk request.
1950 Frequency : Always when mmapping file with no objects
1952 Description: client crashes when mmapping file with no objects
1953 Details : Check that we actually have objects in a file before doing any
1954 operations on objects in ll_vm_open, ll_vm_close and
1960 Description: Request leak when working with deleted CWD
1961 Details : Introduce advanced request refcount tracking for requests
1962 referenced from lustre intent.
1964 Severity : Enhancement
1966 Description: Cache open file handles on client.
1967 Details : MDS now will return special lock along with openhandle, if
1968 requested and client is allowed to hold openhandle, even if unused,
1969 until such a lock is revoked. Helps NFS a lot, since NFS is opening
1970 closing files for every read/write openration.
1972 Severity : Enhancement
1974 Description: Cache open negative dentries on client when possible.
1975 Details : Guard negative dentries with UPDATE lock on parent dir, drop
1976 negative dentries on lock revocation.
1981 Description: Remounting a client read-only wasn't possible with a zconf mount
1982 Details : It wasn't possible to remount a client read-only with llmount.
1984 Severity : enhancement
1985 Description: Include MPICH 1.2.6 Lustre ADIO interface patch
1986 Details : In lustre/contrib/ or /usr/share/lustre in RPM a patch for
1987 MPICH is included to add Lustre-specific ADIO interfaces.
1988 This is based closely on the UFS ADIO layer and only differs
1989 in file creation, in order to allow the OST striping to be set.
1990 This is user-contributed code and not supported by CFS.
1995 Description: extended inode attributes (immutable, append-only) work improperly
1996 when 2.4 and 2.6 kernels are used on client/server or vice versa
1997 Details : Introduce kernel-independent values for these flags.
1999 Severity : enhancement
2002 Description: Allow fractional MB tunings for lustre in /proc/ filesystem.
2003 Details : Many of the /proc/ tunables can only be tuned at a megabyte
2004 granularity. Now, Fractional MB granularity is be supported,
2005 this is very useful for low memory system.
2007 Severity : enhancement
2009 Description: Getattr by fid
2010 Details : Getting a file attributes by its fid, obtaining UPDATE|LOOKUP
2011 locks, avoids extra getattr rpc requests to MDS, allows '/' to
2012 have locks and avoids getattr rpc requests for it on every stat.
2015 Frequency : Always, for filesystems larger than 2TB
2017 Description: ldiskfs crash at mount for filesystem larger than 2TB with mballoc
2018 Details : Kenrel kmalloc limits allocations to 128kB and this prevents
2019 filesystems larger than 2TB to be mounted with mballoc enabled.
2022 Frequency : Always, for 32-bit kernel without CONFIG_LBD and filesystem > 2TB
2024 Description: filesystem corruption for non-standard kernels and very large OSTs
2025 Details : If a 32-bit kernel is compiled without CONFIG_LBD enabled and a
2026 filesystems larger than 2TB is mounted then the kernel will
2027 silently corrupt the start of the filesystem. CONFIG_LBD is
2028 enabled for all CFS-supported kernels, but the possibility of
2029 this happening with a modified kernel config exists.
2031 Severity : enhancement
2033 Description: add client O_DIRECT support for 2.6 kernels
2034 Details : It is now possible to do O_DIRECT reads and writes to files
2035 in the Lustre client mountpoint on 2.6 kernel clients.
2037 Severity : enhancement
2039 Description: parallel glimpse, setattr, statfs, punch, destroy requests
2040 Details : Sends glimpse, setattr, statfs, punch, destroy requests to OSTs in
2041 parallel, not waiting for response from every OST before sending
2042 a rpc to the next OST.
2047 Description: setattr vs write race when updating file timestamps
2048 Details : Client processes that update a file timestamp into the past
2049 right after writing to the file (e.g. tar) it is possible that
2050 the updated file modification time can be reset to the current
2051 time due to a race between processing the setattr and write RPC.
2053 Severity : enhancement
2055 Description: Bring 'lfs find' closer in line with regular Linux find.
2056 Details : lfs find util supports -atime, -mtime, -ctime, -maxdepth, -print,
2057 -print0 options and obtains all the needed info through the lustre
2060 Severity : enhancement
2062 Description: support up to 1024 configured devices on one node
2063 Details : change obd_dev array from statically allocated to dynamically
2064 allocated structs as they are first used to reduce memory usage
2069 Description: Flush dirty partially truncated pages during truncate
2070 Details : Immediatelly flush partially truncated pages in filter_setattr,
2071 this way we completely avoid having any pages in page cache on OST
2072 and can retire ugly workarounds during writes to flush such pages.
2077 Description: i_sem vs transaction deadlock in mds_obd_destroy during unlink.
2078 Details : protect inode from truncation within vfs_unlink() context
2079 just take a reference before calling vfs_unlink() and release it
2080 when parent's i_sem is free.
2083 Frequency : always, if extents are used on OSTs
2085 Description: index ei_leaf_hi (48-bit extension) is not zeroed in extent index
2086 Details : OSTs using the extents format would not zero the high 16 bits of
2087 the index physical block number. This is not a problem for any
2088 OST filesystems smaller than 16TB, and no kernels support ext3
2089 filesystems larger than 16TB yet. This is fixed in 1.4.7 (all
2090 new/modified files) and can be fixed for existing filesystems
2091 with e2fsprogs-1.39-cfs1.
2096 Description: import connection selection may be incorrect if timer wraps
2097 Details : Using a 32-bit jiffies timer with HZ=1000 may cause backup
2098 import connections to be ignored if the 32-bit jiffies counter
2099 wraps. Use a 64-bit jiffies counter.
2102 Frequency : during server recovery
2104 Description: crash after server is denying duplicate export
2105 Details : If clients are resending connect requests to the server, the
2106 server refuses to allow a client to connect multiple times.
2107 Fixed a bug in the handling of this case.
2110 Frequency : very large clusters immediately after boot
2112 Description: LNET request buffers exhausted under heavy short-term load
2113 Details : If a large number of client requests are generated on a service
2114 that has previously never seen so many requests it is possible
2115 that the request buffer growth cannot keep up with the spike in
2116 demand. Instead of dropping incoming requests, they are held in
2117 the LND until the RPC service can accept more requests.
2120 Frequency : Sometimes during replay
2122 Description: Assertion failure in ll_local_open after replay.
2123 Details : If replay happened on an open request reply before we were able
2124 to set replay handler, reply will become not swabbed tripping the
2125 assertion in ll_local_open. Now we set the handler right after
2126 recognising of open request
2129 Frequency : very rare
2131 Description: kernel reports "badness in vsnprintf"
2132 Details : Reading from the "recovery_status" /proc file in small chunks
2133 may cause a negative length in lprocfs_obd_rd_recovery_status()
2134 call to vsnprintf() (which is otherwise harmless). Exit early
2135 if there is no more space in the output buffer.
2137 Severity : enhancement
2139 Description: clear OBD RPC statistics by writing to them
2140 Details : It is now possible to clear the OBD RPC statistics by writing
2141 to the "stats" file.
2146 Description: Client mtime is not the same on different clients after utimes
2147 Details : In some cases, the client was using the utimes() syscall on
2148 a file cached on another node. The clients now validate the
2149 ctime from the MDS + OSTs to determine which one is right.
2154 Description: Inability to activate failout mode
2155 Details : lconf script incorrectly assumed that in python string's numeric
2156 value is used in comparisons.
2159 Frequency : always with multiple stripes per file
2161 Description: Inefficient object allocation for mutli-stripe files
2162 Details : When selecting which OSTs to stripe files over, for files with
2163 a stripe count that divides evenly into the number of OSTs,
2164 the MDS is always picking the same starting OST for each file.
2165 Return the OST selection heuristic to the original design.
2170 Description: mount failures may take full timeout to return an error
2171 Details : Under some heavy load conditions it is possible that a
2172 failed mount can wait for the full obd_timeout interval,
2173 possibly several minutes, before reporting an error.
2174 Instead return an error as soon as the status is known.
2176 ------------------------------------------------------------------------------
2178 2006-02-14 Cluster File Systems, Inc. <info@clusterfs.com>
2180 * WIRE PROTOCOL CHANGE. This version of Lustre networking WILL NOT
2181 INTEROPERATE with older versions automatically. Please read the
2182 user documentation before upgrading any part of a live system.
2183 * WARNING: Lustre networking configuration changes are required with
2184 this release. See https://bugzilla.clusterfs.com/show_bug.cgi?id=10052
2187 * Support for kernels:
2188 2.6.9-22.0.2.EL (RHEL 4)
2189 2.6.5-7.244 (SLES 9)
2190 2.6.12.6 vanilla (kernel.org)
2193 Severity : enhancement
2194 Bugzilla : 7981/8208
2195 Description: Introduced Lustre Networking (LNET)
2196 Details : LNET is new networking infrastructure for Lustre, it includes
2197 a reorganized network configuration mode (see the user
2198 documentation for full details) as well as support for routing
2199 between different network fabrics. Lustre Networking Devices
2200 (LNDs) for the supported network fabrics have also been
2201 created for this new infrastructure.
2203 Severity : enhancement
2204 Description: Introduced Access control lists
2205 Details : clients can set ACLs on files and directories in order to have
2206 more fine-grained permissions than the standard Unix UGO+RWX.
2207 The MDS must be started with the "-o acl" mount option.
2209 Severity : enhancement
2210 Description: Introduced filesystem quotas
2211 Details : Administrators may now establish per-user quotas on the
2214 Severity : enhancement
2216 Description: Configuration change for the XT3
2217 The PTLLND is now used to run Lustre over Portals on the XT3
2218 The configure option(s) --with-cray-portals are no longer used.
2219 Rather --with-portals=<path-to-portals-includes> is used to
2220 enable building on the XT3. In addition to enable XT3 specific
2221 features the option --enable-cray-xt3 must be used.
2226 Description: Running on many-way SMP OSTs can trigger oops in llcd_send()
2227 Details : A race between allocating a new llcd and re-getting the llcd_lock
2228 allowed another thread to grab newly-allocated llcd.
2230 Severity : enhancement
2232 Description: 2.6 OST async journal commit and locking fix to improve performance
2233 Details : The filter_direct_io()+filter_commitrw_write() journal commits for
2234 2.6 kernels are now async as they already were in 2.4 kernels so
2235 that they can commit concurrently with the network bulk transfer.
2236 For block-allocated files the filter allocation semaphore is held
2237 to avoid filesystem fragmentation during allocation. BKL lock
2238 removed for 2.6 xattr operations where it is no longer needed.
2243 Description: lconf incorrectly determined whether two IP networks could talk
2244 Details : In some more complicated routing and multiple-network
2245 configurations, lconf will avoid trying to make a network
2246 connection to a disjoint part of the IP space. It was doing the
2247 math incorrectly for one set of cases.
2252 Description: Fix for potential infinite loop processing records in an llog.
2253 Details : If an llog record is corrupted/zeroed, it is possible to loop
2254 forever in llog_process(). Validate the llog record length
2255 and skip the remainder of the block on error.
2258 Frequency : occasional (liblustre only)
2260 Description: liblustre could not open files whose last component is a symlink
2261 Details : sysio_path_walk() would incorrectly pass the open intent to
2262 intermediate path components.
2265 Frequency : rare (liblustre only with non-standard tuning)
2266 Bugzilla : 7201 (7350)
2267 Description: Tuning the MDC DLM LRU size to zero triggers client LASSERT
2268 Details : llu_lookup_finish_locks() tries to set lock data on a lock
2269 after it has been released, only do this for referenced locks
2271 Severity : enhancement
2273 Description: specifying an (invalid) directory default stripe_size of -1
2274 would reset the directory default striping
2275 Details : stripe_size -1 was used internally to signal directory stripe
2276 removal, now use "all default" to signal dir stripe removal
2277 as a directory striping of "all default" is not useful
2280 Frequency : common for large clusters running liblustre clients
2282 Description: doing an ls when liblustre clients are running is slow
2283 Details : sending a glimpse AST to a liblustre client waits for every AST
2284 to time out, as liblustre clients will not respond. Since they
2285 cannot cache data we refresh the OST lock LVB from disk instead.
2287 Severity : enhancement
2289 Description: doing an ls at the same time as file IO can be slow
2290 Details : enqueue and other "small" requests can be blocked behind many
2291 large IO requests. Create a new OST IO portal for non-IO
2292 requests so they can be processed faster.
2295 Frequency : rare (only HPUX clients mounting unsupported re-exported NFS vol)
2297 Description: an HPUX NFS client would get -EACCESS when ftruncate()ing a newly
2298 created file with mode 000
2299 Details : the Linux NFS server relies on an MDS_OPEN_OWNEROVERRIDE hack to
2300 allow an ftruncate() as a non-root user to a file with mode 000.
2301 Lustre now respects this flag to disable mode checks when
2302 truncating a file owned by the user
2305 Frequency : liblustre-only, when liblustre client dies unexpectedly or becomes
2308 Description: Revoking locks from clients that went dead or catatonic might take
2310 Details : New lock flags FL_CANCEL_ON_BLOCK used by liblustre makes
2311 cancellation of such locks instant on servers without waiting for
2312 any reply from clients. Clients drops these locks when cancel
2313 notification from server is received without replying.
2316 Frequency : liblustre-only, when liblustre client dies or becomes busy
2318 Description: Doing ls on Linux clients can take a long time with active
2320 Details : Liblustre client cannot handle ASTs in timely manner, so avoid
2321 granting such locks to it in the first place if possible. Locks
2322 are taken by proxy on the OST during the read or write and
2323 dropped immediately afterward. Add connect flags handling, do
2324 not grant locks to liblustre clients for glimpse ASTs.
2326 Severity : enhancement
2328 Description: Improve read-ahead algorithm to avoid excessive IO for random reads
2329 Details : Existing read-ahead algorithm is tuned for the case of streamlined
2330 sequential reads and behaves badly with applications doing random
2331 reads. Improve it by reading ahead at least read region, and
2332 avoiding excessive large RPC for small reads.
2334 Severity : enhancement
2336 Description: Creating more than 1000 files for a single job may cause a load
2337 imbalance on the OSTs if there are also a large number of OSTs.
2338 Details : qos_prep_create() uses an OST index reseed value that is an
2339 even multiple of the number of available OSTs so that if the
2340 reseed happens in the middle of the object allocation it will
2341 still utilize the OSTs as uniformly as possible.
2346 Description: OST or MDS may oops in ping_evictor_main()
2347 Details : ping_evictor_main() drops obd_dev_lock if deleting a stale export
2348 but doesn't restart at beginning of obd_exports_timed list
2351 Severity : enhancement
2353 Description: improve by-nid export eviction on the MDS and OST
2354 Details : allow multiple exports with the same NID to be evicted at one
2355 time without re-searching the exports list.
2358 Frequency : rare, only with supplementary groups enabled on SMP 2.6 kernels
2360 Description: MDS may oops in groups_free()
2361 Details : in rare race conditions a newly allocated group_info struct is
2362 freed again, and this can be NULL. The 2.4 compatibility code
2363 for groups_free() checked for a NULL pointer, but 2.6 did not.
2366 Frequency : common for liblustre clients doing little filesystem IO
2367 Bugzilla : 9352, 7313
2368 Description: server may evict liblustre clients accessing contended locks
2369 Details : if a client is granted a lock or receives a completion AST
2370 with a blocking AST already set it would not reply to the AST
2371 for LDLM_FL_CANCEL_ON_BLOCK locks. It now replies to such ASTs.
2374 Frequency : lfs setstripe, only systems with more than 160 OSTs
2376 Description: unable to set striping with a starting offset beyond OST 160
2377 Details : llapi_create_file() incorrectly limited the starting stripe
2378 index to the maximum single-file stripe count.
2381 Frequency : LDAP users only
2383 Description: lconf did not handle in-kernel recovery with LDAP properly
2384 Details : lconf/LustreDB get_refs() is searching the wrong namespace
2386 Severity : enhancement
2388 Description: bind OST threads to NUMA nodes to improve performance
2389 Details : all OST threads are uniformly bound to CPUs on a single NUMA
2390 node and do their allocations there to localize memory access
2392 Severity : enhancement
2394 Description: llmount can determine client NID directly from Myrinet (GM)
2395 Details : the client NID code from gmnalnid was moved directly into
2396 llmount, removing the need to use this or specifying the
2397 client NID explicitly when mounting GM clients with zeroconf
2400 Frequency : if client is started with down MDS
2402 Description: if client is started with down MDS mount hangs in ptlrpc_queue_wait
2403 Details : Having an LWI_INTR() wait event (interruptible, but no timeout)
2404 will wait indefinitely in ptlrpc_queue_wait->l_wait_event() after
2405 ptlrpc_import_delayed_req() because we didn't check if the
2406 request was interrupted, and we also didn't break out of the
2407 event loop if there was no timeout
2412 Description: data loss during non-page-aligned writes to a single file from
2413 both multiple nodes and multiple threads on one node at same time
2414 Details : updates to KMS and lsm weren't protected by common lock. Resulting
2415 inconsistency led to false short-reads, that were cached and later
2416 used by ->prepare_write() to fill in partially written page,
2417 leading to data loss.
2420 Frequency : always, if lconf --abort_recovery used
2422 Description: lconf --abort_recovery fails with 'Operation not supported'
2423 Details : lconf was attempting to abort recovery on the MDT device and not
2426 Severity : enhancement
2428 Description: remove cleanup logs
2429 Details : replace lconf-generated cleanup logs with lustre internal
2430 cleanup routines. Eliminates the need for client-cleanup and
2433 Severity : enhancement
2435 Description: add support for EAs (user and system) on lustre filesystems
2436 Details : it is now possible to store extended attributes in the Lustre
2437 client filesystem, and with the user_xattr mount option it
2438 is possible to allow users to store EAs on their files also
2440 Severity : enhancement
2442 Description: Add possibility (config option) to show minimal available OST free
2444 Details : When compiled with --enable-mindf configure option, statfs(2)
2445 (and so, df) will return least minimal free space available from
2446 all OSTs as amount of free space on FS, instead of summary of
2447 free spaces of all OSTs.
2449 Severity : enhancement
2451 Description: do not expand extent locks acquired on OST-side
2452 Details : Modify ldlm_extent_policy() to not expand local locks, acquired
2453 by server: they are not cached anyway.
2456 Frequency : when mmap is used/binaries executed from Lustre
2458 Description: Unmmap pages before throwing them away from read cache.
2459 Details : llap_shrink cache now attempts to unmap pages before discarding
2460 them (if unmapping failed - do not discard). SLES9 kernel has
2461 extra checks that trigger if this unmapping is not done first.
2466 Description: lconf didn't resolve symlinks before checking to see whether a
2467 given mountpoint was already in use
2470 Frequency : when migrating failover services
2471 Bugzilla : 6395, 9514
2472 Description: When migrating a subset of services from a node (e.g. failback
2473 from a failover service node) the remaining services would
2474 time out and evict clients.
2475 Details : lconf --force (implied by --failover) sets the global obd_timeout
2476 to 5 seconds in order to quickly disconnect, but this caused
2477 other RPCs to time out too quickly. Do not change the global
2478 obd_timeout for force cleanup, only set it for DISCONNECT RPCs.
2480 Severity : enhancement
2481 Frequency : if MDS is started with down OST
2482 Bugzilla : 9439,5706
2483 Description: Allow startup/shutdown of an MDS without depending on the
2484 availability of the OSTs.
2485 Details : Asynchronously call mds_lov_synchronize during MDS startup.
2486 Add appropriate locking and lov-osc refcounts for safe
2487 cleaning. Add osc abort_inflight calls in case the
2488 synchronize never started.
2491 Frequency : occasional (Cray XT3 only)
2493 Description: root not authorized to access files in CRAY_PORTALS environment
2494 Details : The client process capabilities were not honoured on the MDS in
2495 a CRAY_PORTALS/CRAY_XT3 environment. If the file had previously
2496 been accessed by an authorized user then root was able to access
2497 the file on the local client also. The root user capabilities
2498 are now allowed on the MDS, as this environment has secure UID.
2501 Frequency : occasional
2503 Description: ldiskfs "too long searching" message happens too often
2504 Details : A debugging message (otherwise harmless) prints too often on
2505 the OST console. This has been reduced to only happen when
2506 there are fragmentation problems on the filesystem.
2511 Description: Division by zero in statfs when all OSCs are inactive
2512 Details : lov_get_stripecnt() returns zero due to incorrect order of checks,
2513 lov_statfs divides by value returned by lov_get_stripecnt().
2517 Bugzilla : 9489, 3273
2518 Description: First write from each client to each OST was only 4kB in size,
2519 to initialize client writeback cache, which caused sub-optimal
2520 RPCs and poor layout on disk for the first writen file.
2521 Details : Clients now request an initial cache grant at (re)connect time
2522 and so that they can start streaming writes to the cache right
2523 away and always do full-sized RPCs if there is enough data.
2524 If the OST is rebooted the client also re-establishes its grant
2525 so that client cached writes will be honoured under the grant.
2530 Description: Slow ls (and stat(2) syscall) on files residing on IO-loaded OSTs
2531 Details : Now I/O RPCs go to different portal number and (presumably) fast
2532 lock requests (and glimses) and other RPCs get their own service
2533 threads pool that should be able to service those RPCs
2536 Severity : enhancement
2538 Description: Ability to exchange lustre version between client and servers and
2539 issue warnings at client side if client is too old. Also for
2540 liblustre clients there is ability to refuse connection of too old
2542 Details : New 'version' field is added to connect data structure that is
2543 filled with version info. That info is later checked by server and
2547 Frequency : rare, liblustre only.
2548 Bugzilla : 9296, 9581
2549 Description: Two simultaneous writes from liblustre at offset within same page
2550 might proceed at the same time overwriting eachother with stale
2552 Details : I/O lock withing llu_file_prwv was released too early, before data
2553 actually was hitting the wire. Extended lock-holding time until
2554 server acknowledges receiving data.
2557 Frequency : extremely rare. Never observed in practice.
2559 Description: avoid generating lustre_handle cookie of 0.
2560 Details : class_handle_hash() generates handle cookies by incrementing
2561 global counter, and can hit 0 occasionaly (this is unlikely, but
2562 not impossible, because initial value of cookie counter is
2563 selected randonly). Value of 0 is used as a sentinel meaning
2564 "unassigned handle" --- avoid it. Also coalesce two critical
2565 sections in this function into one.
2567 Severity : enhancement
2569 Description: allow liblustre clients to delegate truncate locking to OST
2570 Details : To avoid overhead of locking, liblustre client instructs OST to
2571 take extent lock in ost_punch() on client's behalf. New connection
2572 flag is added to handle backward compatibility.
2574 Severity : enhancement
2575 Bugzilla : 4928, 7341, 9758
2576 Description: allow number of OST service threads to be specified
2577 Details : a module parameter allows the number of OST service threads
2578 to be specified via "options ost ost_num_threads={N}" in the
2579 OSS's /etc/modules.conf or /etc/modprobe.conf.
2583 Bugzilla : 6146, 9635, 9895
2584 Description: servers crash with bad pointer in target_handle_connect()
2585 Details : In rare cases when a client is reconnecting it was possible that
2586 the connection request was the last reference for that export.
2587 We would temporarily drop the export reference and get a new
2588 one, but this may have been the last reference and the export
2589 was just destroyed. Get new reference before dropping old one.
2591 Severity : enhancement
2592 Frequency : if client is started with failover MDS
2594 Description: Allow multiple MDS hostnames in the mount command
2595 Details : Try to read the configuration from all specified MDS
2596 hostnames during a client mount in case the "primary"
2599 Severity : enhancement
2601 Description: Stop sending data to evicted clients as soon as possible.
2602 Details : Check if the client we are about to send or are sending data to
2603 was evicted already. (Check is done every second of waiting,
2604 for which l_wait_event interface was extended to allow checking
2605 of exit condition at specified intervals).
2608 Frequency : rare, normally only when NFS exporting is done from client
2610 Description: 'bad disk LOV MAGIC: 0x00000000' error when chown'ing files
2612 Details : Make mds_get_md() recognise empty md case and set lmm size to 0.
2615 Frequency : always, if srand() is called before liblustre initialization
2617 Description: Liblustre uses system PRNG disturbing its usage by user application
2618 Details : Introduce internal to lustre fast and high-quality PRNG for
2619 lustre usage and make liblustre and some other places in generic
2620 lustre code to use it.
2622 Severity : enhancement
2623 Bugzilla : 9477, 9557, 9870
2624 Description: Verify that the MDS configuration logs are updated when xml is
2625 Details : Check if the .xml configuration logs are newer than the config
2626 logs stored on the MDS and report an error if this is the case.
2627 Request --write-conf, or allow starting with --old_conf.
2629 Severity : enhancement
2631 Description: Handle symlinks in the path when checking if Lustre is mounted.
2632 Details : Resolve intermediate symlinks when checking if a client has
2633 mounted a filesystem to avoid duplicate client mounts.
2638 Description: lconf can hit an error exception but still return success.
2639 Details : The lconf command catches the Command error exception at the top
2640 level script context and will exit with the associated exit
2641 status, but doesn't ensure that this exit status is non-zero.
2646 Description: failure of ptlrpc thread startup can cause oops
2647 Details : Starting a ptlrpc service thread can fail if there are a large
2648 number of threads or the server memory is very fragmented.
2649 Handle this without oopsing.
2652 Frequency : always, only if liblustre and non-default acceptor port was used
2654 Description: liblustre cannot connect to servers with non-default acceptor port
2655 Details : tcpnal_set_default_params() was not called and was therefore
2656 ignoring the environment varaible TCPNAL_PORT, as well as other
2657 TCPNAL_ environment variables
2662 Description: two objects could be created on the same OST for a single file
2663 Details : If an OST is down, in some cases it was possible to create two
2664 objects on a single OST for a single file. No problems other
2665 than potential performance impact and spurious error messages.
2669 Bugzilla : 5681, 9562
2670 Description: Client may oops in ll_unhash_aliases
2671 Details : Client dcache may become inconsistent in race condition.
2672 In some cases "getcwd" can fail if the current directory is
2678 Description: Inode refcounting problems in NFS export code
2679 Details : link_raw functions used to call d_instantiate without obtaining
2680 extra inode reference first.
2684 Bugzilla : 9942, 9903
2685 Description: Referencing freed requests leading to crash, memleaks with NFS.
2686 Details : We used to require that call to ll_revalidate_it was always
2687 followed by ll_lookup_it. Also with revalidate_special() it is
2688 possible to call ll_revalidate_it() twice for the same dentry
2689 even if first occurence returned success. This fix changes semantic
2690 between DISP_ENQ_COMPLETE disposition flag to mean there is extra
2691 reference on a request referred from the intent.
2692 ll_intent_release() then releases such a request.
2695 Frequency : rare, normally benchmark loads only
2697 Description: unlinked inodes were kept in memory on the client
2698 Details : If a client is repeatedly creating and unlinking files it
2699 can accumulate a lot of stale inodes in the inode slab cache.
2700 If there is no other client load running this can cause the
2701 client node to run out of memory. Instead flush old inodes
2702 from client cache that have the same inode number as a new inode.
2705 Frequency : SLES9 2.6.5 kernel and long filenames only
2706 Bugzilla : 9969, 10379
2707 Description: utime reports stale NFS file handle
2708 Details : SLES9 uses out-of-dentry names in some cases, which confused
2709 the lustre dentry revalidation. Change it to always use the
2713 Frequency : rare, unless heavy write-truncate concurrency is continuous
2714 Bugzilla : 4180, 6984, 7171, 9963, 9331
2715 Description: OST becomes very slow and/or deadlocked during object unlink
2716 Details : filter_destroy() was holding onto the parent directory lock
2717 while truncating+unlinking objects. For very large objects this
2718 may block other threads for a long time and slow overall OST
2719 responsiveness. It may also be possible to get a lock ordering
2720 deadlock in this case, or run out of journal credits because of
2721 the combined truncate+unlink. Solution is to do object truncate
2722 first in one transaction without parent lock, and then do the
2723 final unlink in a new transaction with the parent lock. This
2724 reduces the lock hold time dramatically.
2727 Frequency : rare, 2.4 kernels only
2729 Description: MDS or OST cleanup may trip kernel BUG when dropping kernel lock
2730 Details : mds_cleanup() and filter_cleanup() need to drop the kernel lock
2731 before unmounting their filesystem in order to avoid deadlock.
2732 The kernel_locked() function in 2.4 kernels only checks whether
2733 the kernel lock is held, not whether it is this process that is
2734 holding it as 2.6 kernels do.
2739 Description: MDS or OST may oops/LBUG if a client is connecting multiple times
2740 Details : The client ptlrpc code may be trying to reconnect to a down
2741 server before a previous connection attempt has timed out.
2742 Increase the reconnect interval to be longer than the connection
2743 timeout interval to avoid sending duplicate connections to
2747 Frequency : echo_client brw_test command
2749 Description: fix echo_client to work with OST preallocated code
2750 Details : OST preallocation code (5137) didn't take echo_client IO path
2751 into account: echo_client calls filter methods outside of any
2752 OST thread and, hence, there is no per-thread preallocated
2753 pages and buffers to use. Solution: hijack pga pages for IO. As
2754 a byproduct, this avoids unnecessary data copying.
2758 Bugzilla : 3555, 5962, 6025, 6155, 6296, 9574
2759 Description: Client can oops in mdc_commit_close() after open replay
2760 Details : It was possible for the MDS to return an open request with no
2761 transaction number in mds_finish_transno() if the client was
2762 evicted, but without actually returning an error. Clients
2763 would later try to replay that open and may trip an assertion
2764 Simplify the client close codepath, and always return an error
2765 from the MDS in case the open is not successful.
2768 Frequency : rare, 2.6 OSTs only
2770 Description: OST may deadlock under high load on fragmented files
2771 Details : If there was a heavy load and highly-fragmented OST filesystems
2772 it was possible to have all the OST threads deadlock waiting on
2773 allocation of biovecs, because the biovecs were not released
2774 until the entire RPC IO was completed. Instead, release biovecs
2775 as soon as they are complete to ensure forward IO progress.
2777 Severity : enhancement
2779 Description: Support for specifying external journal device at mount
2780 Details : If an OST or MDS device is formatted with an external journal
2781 device, this device major/minor is stored in the ext3 superblock
2782 and may not be valid for failover. Allow detecting and
2783 specifying the external journal at mount time.
2788 Description: Mounting an MDS with pending unlinked files may cause oops
2789 Details : target_finish_recovery() calls mds_postrecov() which returned
2790 the number of orphans unlinked. mds_lov_connect->mds_postsetup()
2791 considers this an error and immediately begins cleaning up the
2792 lov, just after starting the mds_lov process
2794 Severity : enhancement
2796 Description: Implement 'lfs df' to report actual free space on per-OST basis
2797 Details : Add sub-command 'df' on 'lfs' to report the disk space usage of
2798 MDS/OSDs. Usage: lfs df [-i][-h]. Command Options: '-i' to report
2799 usage of objects; '-h' to report in human readable format.
2801 ------------------------------------------------------------------------------
2803 2005-08-26 Cluster File Systems, Inc. <info@clusterfs.com>
2810 Description: Mounting an ldiskfs file system with mballoc may crash OST node.
2811 Details : ldiskfs mballoc code may reference an uninitialized buddy struct
2812 at startup during orphan unlinking. Instead, skip buddy update
2813 before setup, as it will be regenerated after recovery is complete.
2818 Description: If an OST is inactive, its locks might reference stale inodes.
2819 Details : lov_change_cbdata() must iterate over all namespaces, even if
2820 they are inactive to clear inode references from the lock.
2822 Severity : enhancement
2823 Frequency : occasional, if non-standard max_dirty_mb used
2825 Description: Client will block write RPCs if not enough grant
2826 Details : If a client has max_dirty_mb smaller than max_rpcs_in_flight,
2827 then the client will block writes while waiting for another RPC
2828 to complete instead of consuming its dirty limit. With change
2829 we get improved performance when max_dirty_mb is small.
2831 Severity : enhancement
2832 Bugzilla : 3389, 6253
2833 Description: Add support for supplementary groups on the MDS.
2834 Details : The MDS has an upcall /proc/fs/lustre/mds/{mds}/group_upcall
2835 (set to /usr/sbin/l_getgroups if enabled) which will do MDS-side
2836 lookups for user supplementary groups into a cache.
2840 Description: O_CREAT|O_EXCL open flags in liblustre always return -EEXIST
2841 Details : Make libsysio to not enforce O_EXCL by clearing the flag,
2842 for liblustre O_EXCL is enforced by MDS.
2846 Description: readdir never returns NULL in liblustre.
2847 Details : Corrected llu_iop_getdirentries logic, to return offset of next
2848 dentry in struct dirent.
2852 Frequency : liblustre only, depends on application IO pattern
2853 Description: liblustre clients evicted if not contacting servers
2854 Details : Don't put liblustre clients into the ping_evictor list, so
2855 they will not be evicted by the pinger ever.
2857 Severity : enhancement
2859 Description: Add ability to evict clients by NID from MDS.
2860 Details : By echoing "nid:$NID" string into
2861 /proc/fs/lustre/mds/.../evict_client client with nid that equals to
2862 $NID would be instantly evicted from this MDS and from all active
2863 OSTs connected to it.
2867 Description: Do not query file size twice, somewhat slowing stat(2) calls.
2868 Details : lookup_it_finish() used to query file size from OSTs that was not
2873 Description: service threads change working directory to that of init
2874 Details : Starting lustre service threads may pin the working directory
2875 of the parent thread, making that filesystem busy. Threads
2876 now change to the working directory of init to avoid this.
2880 Frequency : during shutdown only
2881 Description: shutdown with a failed MDS or OST can cause unmount to hang
2882 Details : Don't resend DISCONNECT messages in ptlrpc_disconnect_import()
2887 Frequency : 2.6 only
2888 Description: chmod/chown may include an extra supplementary group
2889 Details : ll{,u}_mdc_pack_op_data() does not properly initialize the
2890 supplementary group and if none is specified this is used.
2893 Bugzilla : 5479 (6816)
2895 Description: Racing open + rm can assert client in mdc_set_open_replay_data()
2896 Details : If lookup is in progress on a file that is unlinked we might try
2897 to revalidate the inode and fail in revalidate after lookup is
2898 complete and ll_file_open() enqueues the open again but
2899 it_open_error() was not checking DISP_OPEN_OPEN errors correctly.
2902 Frequency : always, if lconf --abort_recovery used
2904 Description: lconf --abort_recovery fails with 'Operation not supported'
2905 Details : lconf was attempting to abort recovery on the MDT device and not
2908 ------------------------------------------------------------------------------
2910 2005-08-08 Cluster File Systems, Inc. <info@clusterfs.com>
2915 Frequency : rare (only unsupported configurations with a node running as an
2917 Bugzilla : 6514, 5137
2918 Description: Mounting a Lustre file system on a node running as an OST could
2920 Details : OSTs now preallocates memory needed to write out data at
2921 startup, instead of when needed, to avoid having to
2922 allocate memory in possibly low memory situations.
2923 Specifically, if the file system is mounted on on OST,
2924 memory pressure could force it to try to write out data,
2925 which it needed to allocate memory to do. Due to the low
2926 memory, it would be unable to do so and the node would
2927 become unresponsive.
2929 Severity : enhancement
2931 Description: Addition of lconf --service command line option
2932 Details : lconf now accepts a '--service <arg>' option, which is
2933 shorthand for 'lconf --group <arg> --select <arg>=<hostname>'
2935 Severity : enhancement
2937 Description: Failover mode is now the default for OSTs.
2938 Details : By default, OSTs will now run in failover mode. To return to
2939 the old behaviour, add '--failout' to the lmc line for OSTs.
2941 Severity : enhancement
2943 Description: Health checks are now provided for MDS and OSTs
2944 Details : Additional detailed health check information on MSD and OSTs
2945 is now provided through the procfs health_check value.
2948 Frequency : occasional, depends on IO load
2950 Description: Disk fragmentation on the OSTs could eventually cause slowdowns
2951 after numerous create/delete cycles
2952 Details : The ext3 inode allocation policy would not allocate new inodes
2953 very well on the OSTs because there are no new directories
2954 being created. Instead we look for groups with free space if
2955 the parent directories are nearly full.
2960 Description: Network or server problems during mount may cause partially
2961 mounted clients instead of returning an error.
2962 Details : The config llog parsing code may overwrite the error return
2963 code during mount error handling, returning success instead
2969 Description: MDS can fail to allocate large reply buffers
2970 Details : After long uptimes the MDS can fail to allocate large reply
2971 buffers (e.g. zconf client mount config records) due to memory
2972 fragmentation or consumption by the buffer cache. Preallocate
2973 some large reply buffers so that these replies can be sent even
2974 under memory pressure.
2978 Frequency : rare (liblustre)
2979 Description: fsx running with liblustre complained that using truncate() to
2980 extend the file doesn't work. This patch corrects that issue.
2981 Details : This is the liblustre equivalent of the fix for bug 6196. Fixes
2982 ATTR_SIZE and lsm use in llu_setattr_raw.
2986 Frequency : rare, only 2.6 kernels
2987 Description: Unusual file access patterns on the MDS may result in inode
2988 data being lost in very rare circumstances.
2989 Details : Bad interaction between the ea-in-inode patch and the "no-read"
2990 code in the 2.6 kernel caused the inode and/or EA data not to
2991 be read from disk, causing single-file corruption.
2995 Frequency : rare, only 2.6 filesystems using extents
2996 Description: Heavy concurrent write and delete load may cause data corruption.
2997 Details : It was possible under high-load situations to have an extent
2998 metadata block in the block device cache from a just-unlinked
2999 file overwrite a newly-allocated data block. We now unmap any
3000 metadata buffers that alias just-allocated data blocks.
3004 Frequency : filesystems with default stripe_count larger than 77
3005 Description: lconf+mke2fs fail when formatting filesystem with > 77 stripes
3006 Details : lconf specifies an inode size of 4096 bytes when the default
3007 stripe_count is larger than 77. This conflicts with the default
3008 inode density of 1 per 4096 bytes. Allocate smaller inodes in
3009 this case to avoid pinning too much memory for large EAs.
3011 ------------------------------------------------------------------------------
3013 2005-07-07 Cluster File Systems, Inc. <info@clusterfs.com>
3018 Frequency : rare (extremely heavy IO load with hundreds of clients)
3020 Description: Client is evicted, gets IO error writing to file
3021 Details : lock ordering changes for bug 5492 reintroduced bug 3267 and
3022 caused clients to be evicted for AST timeouts. The fixes in
3023 bug 5192 mean we no longer need to have such short AST timeouts
3024 so ldlm_timeout has been increased.
3027 Frequency : occasional during --force or --failover shutdown under load
3028 Bugzilla : 5949, 4834
3029 Description: Server oops/LBUG if stopped with --force or --failover under load
3030 Details : a collection of import/export refcount and cleanup ordering
3031 issues fixed for safer force cleanup
3034 Frequency : only filesystems larger than 120 OSTs
3035 Bugzilla : 5990, 6223
3036 Description: lfs getstripe would oops on a very large filesystem
3037 Details : lov_getconfig used kfree on vmalloc'd memory
3040 Frequency : only filesystems exporting via NFS to Solaris 10 clients
3041 Bugzilla : 6242, 6243
3042 Description: reading from files that had been truncated to a non-zero size
3043 but never opened returned no data
3044 Details : ll_file_read() reads zeros from no-object files to EOF
3049 Description: A bug in MDS/OSS recovery could cause the OSS to fail an assertion
3050 Details : There's little harm in aborting MDS/OSS recovery and letting it
3051 try again, so I removed the LASSERT and return an error instead.
3053 Severity : enhancement
3055 Description: New debugging infrastructure for tracking down data corruption
3056 Details : The I/O checksum code was replaced to: (a) control it at runtime,
3057 (b) cover more of the client-side code path, and (c) try to narrow
3058 down where problems occurred
3062 Bugzilla : 3819, 4364, 4397, 6313
3063 Description: Racing close and eviction MDS could cause assertion in mds_close
3064 Details : It was possible to get multiple mfd references during close and
3065 client eviction, leading to one thread referencing a freed mfd.
3067 Severity: : enhancement
3068 Bugzilla : 3262, 6359
3069 Description: Attempts to reconnect to servers are now more aggressive.
3070 Details : This builds on the enhanced upcall-less recovery that was added
3071 in 1.4.2. When trying to reconnect to servers, clients will
3072 now try each server in the failover group every 10 seconds. By
3073 default, clients would previously try one server every 25 seconds.
3078 Description: After recovery, certain operations trigger a failed
3079 assertion on a client.
3080 Details : Failing over an mds, using lconf -d --failover, while a
3081 client was doing a readdir() call would cause the client to
3082 LBUG after recovery completed and the readdir() was resent.
3084 Severity : enhancement
3086 Description: Default groups are now added by lconf
3087 Details : You can now run lconf --group <servicename> without having to
3088 manually add groups with lmc.
3091 Frequency : occasional
3093 Description: Nodes with an elan id of 0 trigger a failed assertion
3096 Frequency : always when accessing e.g. tty/console device nodes
3098 Description: tty and some other devices nodes cannot be used on lustre
3099 Details : file's private_data field is used by device data and lustre
3100 values in there got lost. New field was added to struct file to
3101 store fs-specific private data.
3104 Frequency : when exporting Lustre via NFS
3106 Description: NFSD failed occasionally when looking up a path component
3107 Details : NFSD is looking up ".." which was broken in ext3 directories
3108 that had grown large enough to become hashed.
3111 Frequency : Clusters with multiple interfaces not on the same subnet
3113 Description: Nodes will repeatedly try to reconnect to an interface which it
3114 cannot reach and report an error to the log.
3115 Details : Extra peer list entries will be created by lconf with some peers
3116 unreachable. lconf now validates the peer before adding it.
3119 Frequency : Only if a default stripe is set on the filesystem root.
3121 Description: Setting a default stripe on the filesystem root prevented the
3122 filesystem from being remounted.
3123 Details : The client was sending extra request flags in the root getattr
3124 request and did not allocate a reply buffer for the dir EA.
3127 Frequency : occasional, higher if lots of files are accessed by one client
3128 Bugzilla : 6159, 6097
3129 Description: Client trips assertion regarding lsm mismatch/magic
3130 Details : While revalidating inodes the VFS looks up inodes with ifind()
3131 and in rare cases can find an inode that is being freed.
3132 The ll_test_inode() code will free the lsm during ifind()
3133 when it finds an existing inode and then the VFS later attaches
3134 this free lsm to a new inode.
3138 Bugzilla : 6422, 7030
3139 Description: MDS deadlock between mkdir and client eviction
3140 Details : Creating a new file via mkdir or mknod (starting a transaction
3141 and getting the ns lock) can deadlock with client eviction
3142 (gets ns lock and trying to finish a synchronous transaction).
3145 Frequency : occasional
3146 Description: While starting a server, the fsfilt_ext3 module could not be
3148 Details : CFS's improved ext3 filesystem is named ldiskfs for 2.6
3149 kernels. Previously, lconf would still use the ext3 name
3150 when trying to load modules. Now, it will correctly use
3151 ext3 on 2.4 and ldiskfs on 2.6.
3153 Severity : enhancement
3154 Description: The default stripe count has been changed to 1
3155 Details : The interpretation of the default stripe count (0, to lfs
3156 or lmc) has been changed to mean striping across a single
3157 OST, rather than all available. For general usage we have
3158 found a stripe count of 1 or 2 works best.
3160 Severity : enhancement
3161 Description: Add support for compiling against Cray portals.
3162 Details : Conditional compiling for some areas that are different
3166 Frequency : occasional
3167 Bugzilla : 6409, 6834
3168 Description: Creating files with an explicit stripe count may lead to
3169 a failed assertion on the MDS
3170 Details : If some OSTs are full or unavailable, creating files may
3171 trigger a failed assertion on the MDS. Now, Lustre will
3172 try to use other servers or return an error to the
3176 Frequency : occasional
3178 Description: Multiple concurrent overlapping read+write on multiple SMP nodes
3179 caused lock timeout during readahead (since 1.4.2).
3180 Details : Processes doing readahead might match a lock that hasn't been
3181 granted yet if there are overlapping and conflicting lock
3182 requests. The readahead process waits on ungranted lock
3183 (original lock is CBPENDING), while OST waits for that process
3184 to cancel CBPENDING read lock and eventually evicts client.
3186 Severity : enhancement
3188 Description: Initial enabling of flock support for clients
3189 Details : Implements fcntl advisory locking and file status functions.
3190 This feature is provided as an optional mount flag (default
3191 off), and is NOT CURRENTLY SUPPORTED. Not all types of record
3192 locking are implemented yet, and those that are are not guaranteed
3193 to be completely correct in production environments.
3194 mount -t lustre -o [flock|noflock] ...
3197 Frequency : occasional
3199 Description: OSTs running 2.4 kernels but with extents enabled might trip an
3200 assertion in the ext3 JBD (journaling) layer.
3201 Details : The b_committed_data struct is protected by the big kernel lock
3202 in 2.4 kernels, serializing journal_commit_transaction() and
3203 ext3_get_block_handle->ext3_new_block->find_next_usable_block()
3204 access to this struct. In 2.6 kernels there is finer grained
3205 locking to improve SMP performance of the JBD layer.
3209 Description: Changes the "SCSI I/O Stats" kernel patch to default to "enabled"
3211 -----------------------------------------------------------------------------
3213 2005-05-05 Cluster File Systems, Inc. <info@clusterfs.com>
3215 NOTE: Lustre 1.4.2 uses an incompatible network protocol than previous
3216 versions of Lustre. Please update all servers and clients to
3217 version 1.4.2 or later at the same time. You must also run
3218 "lconf --write-conf {config}.xml" on the MDS while it is stopped
3219 to update the configuration logs.
3221 - fix for HPUX NFS client breakage when NFS exporting Lustre (5781)
3222 - mdc_enqueue does not need max_mds_easize request buffer on send (5707)
3223 - swab llog records of type '0' so we get proper header size/idx (5861)
3224 - send llog cancel req to DLM cancel portal instead of cb portal (5515)
3225 - fix rename of one directory over another leaking an inode (5953)
3226 - avoid SetPageDirty on 2.6 (5981)
3227 - don't re-add just-being-destroyed locks to the waiting list (5653)
3228 - when creating new directories, inherit the parent's custom
3229 striping settings if present parent (3048)
3230 - flush buffers from cache before direct IO in 2.6 obdfilter (4982)
3231 - don't hold i_size_sem in ll_nopage() and ll_ap_refresh_count (6077)
3232 - don't hold client locks on temporary worklist from l_lru (5666)
3233 - handle IO errors in 2.6 obdfilter bio completion routine (6046)
3234 - automatically evict dead clients (5921)
3235 - Update file size properly in create+truncate+fstat case (6196)
3236 - Do not unhash mountpoint dentries, do not allow removal of
3238 - Avoid lock ordering deadlock issue with write/truncate (6203,5654)
3239 - reserve enough journal credits in fsfilt_start_log for setattr (4554)
3240 - ldlm_enqueue freed-export error path would always LBUG (6149,6184)
3241 - don't reference lr_lvb_data until after we hold lr_lvb_sem (6170)
3242 - don't overwrite last_rcvd if there is a *_client_add() error (6086)
3243 - Correctly handle reads of files with no objects (6243)
3244 - lctl recover will also mark a device active if deactivate used (5933)
3246 - by default create 1 inode per 4kB space on MDS, per 16kB on OSTs
3247 - allow --write-conf on an MDS with different nettype than client (5619)
3248 - don't write config llogs to MDS for mounts not from that MDS (5617)
3249 - lconf should create multiple TCP connections from a client (5201)
3250 - init scripts are now turned off by default; run chkconfig --on
3251 lustre and chkconfig --on lustrefs to use them
3252 - upcalls are no longer needed for clients to recover to failover
3254 - add --abort-recovery option to lconf to abort recovery on device
3256 - add support for an arbitrary number of OSTs (3026)
3257 - Quota support protocol changes.
3258 - forward compatibility changes to wire structs (6007)
3259 - rmmod NALs that might be loaded because of /etc/modules.conf (6133)
3260 - support for mountfsoptions and clientoptions to the Lustre LDAP (5873)
3261 - improved "lustre status" script
3262 - initialize blocksize for non-regular files (6062)
3263 - added --disable-server and --disable-client configure options (5782)
3264 - introduce a lookup cache for lconf to avoid repeated DB scans (6204)
3265 - Vanilla 2.4.29 support
3266 - increase maximum number of obd devices to 520 (6242)
3267 - remove the tcp-zero-copy patch from the suse-2.4 series (5902)
3268 - Quadrics Elan drivers are now included for the RHEL 3 2.4.21 and
3269 SLES 9 2.6.5 kernels
3270 - limit stripes per file to 160 (the maximum EA size) (6093)
3272 2005-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3275 - don't LASSERT in ll_release on NULL lld with NFS export (4655, 5760)
3276 - hold NS lock when calling handle_ast_error->del_waiting_lock (5746)
3277 - fix setattr mtime regression from lovcleanup merge (4829, 5669)
3278 - workaround for 2.6 crash in ll_unhash_aliases (5687, 5210)
3279 - small ext3 extents cleanups and fixes (5733)
3280 - improved mballoc code, several small races and bugs fixed (5733, 5638)
3281 - kernel version 43 - fix remove_suid bugs in both 2.4 and 2.6 (5695)
3282 - avoid needless client->OST connect, fix handle mismatch (5317)
3283 - fix DLM error path that led to out-of-sync client, long delays (5779)
3284 - support common vfs-enforced mount options (nodev,nosuid,noexec) (5637)
3285 - fix several locking issues related to i_size (5492,5624,5654,5672)
3286 - don't move pending lock onto export if it is already evicted (5683)
3287 - fix kernel oops when creating .foo in unlinked directory (5548)
3288 - fix deadlock in obdfilter statistics vs. object create (5811)
3289 - use time_{before,after} to avoid timer jiffies wrap (5882)
3290 - shutdown --force/--failover stability (3607,3651,4797,5203,4834)
3291 - Do not leak request if server was not able to process it (5154)
3292 - If mds_open unable to find parent dir, make that negative lookup(5154)
3293 - don't create new directories with extent-mapping (5909, 5936)
3295 - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
3296 - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
3297 - change obd_increase_kms to obd_adjust_kms (up or down) (5654)
3298 - lconf, lmc search both /usr/lib and /usr/lib64 for Python libs (5800)
3299 - support for RHEL4 kernel on i686 (5773)
3300 - provide error messages when incompatible logs are encountered (5898)
3302 2005-02-18 Cluster File Systems, Inc. <info@clusterfs.com>
3303 * version 1.4.0.10 (1.4.1 release candidate 1)
3305 - don't keep a lock reference when lock is not granted (4238)
3306 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3307 - add per-fs limit of Lustre pages in page cache, avoid OOM (4699)
3308 - drop import inflight refcount on signal_completed_replay error (5255)
3309 - unlock page after async write error during send (3677)
3310 - handle missing objects in filter_preprw_read properly (5265)
3311 - no transno return for symlink open, don't save no-trasno open (3440)
3312 - don't try to complete elan receive that already failed (4012)
3313 - free RPC server reply state on error (5406)
3314 - clean up thread from ptlrpc_start_thread() on error (5160)
3315 - readahead could read extra page into cache that wasn't ejected (5388)
3316 - prevent races in class_attach/setup/cleanup/detach (5260)
3317 - don't dereference de->d_inode after l_dput of de (5458)
3318 - use "int" for stripe value returned from lock_to_stripe (5544)
3319 - mballoc allocation and error-checking fixes in 2.6 (5504)
3320 - block device patches to fix I/O request sizes in 2.6 (5482)
3321 - look up hostnames for IB nals (5602)
3322 - 2.6 changed lock ordering of 2 semaphores, caused deadlock (5654)
3323 - don't start multiple acceptors for the same port (5277)
3324 - fix incorrect LASSERT in mds_getattr_name (5635)
3325 - export a proc file for general "ping" checking (5628)
3326 - fix "lfs check" to not block when the MDS is down (5628)
3328 - service request history (4965)
3329 - put {ll,lov,osc}_async_page structs in a single slab (4699)
3330 - create an "evict_client" /proc entry on OSTs, like the MDS has
3331 - fix mount usage message, return errors per mount(8) (5168)
3332 - change grep [] to grep "[]" in tests so they work in more UMLs
3333 - fix ppc64/x86_64 spec to use %{_libdir} instead of /usr/lib (5389)
3334 - remove ancient LOV_MAGIC_V0 EA support (5047)
3335 - add "disk I/Os in flight" and "I/O req time" stats in obdfilter
3336 - align r/w RPCs to PTLRPC_MAX_BRW_SIZE boundary for performance (3451)
3337 - allow readahead allocations to fail when low on memory (5383)
3338 - mmap locking landed again, after considerable improvement (2828)
3339 - add get_hostaddr() to lustreDB.py for LDAP support (5459)
3341 2004-11-23 Cluster File Systems, Inc. <info@clusterfs.com>
3344 - send OST transaction number in read/write reply to free req (4966)
3345 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3346 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3347 - the watchdog thread now runs as interruptible (5246)
3348 - flock/lockf fixes (but it's still disabled, pending 5135)
3349 - don't use EXT3 constants in llite code (5094)
3350 - memory shortage at startup could cause assertion (5176)
3352 - reorganization of lov code
3353 - single portals codebase
3355 - add extents/mballoc support (5025)
3356 - direct I/O reads in the obdfilter (4048)
3357 - kernel patches from LNXI for 2.6 (bluesmoke, perfctr, mtd, kexec)
3359 tbd Cluster File Systems, Inc. <info@clusterfs.com>
3362 - send OST transaction number in read/write reply to free req (4966)
3363 - don't ASSERT in ptl_send_rpc() if we run out of memory (5119)
3364 - lock /proc/sys/portals/routes internal state, avoiding oops (4827)
3365 - the watchdog thread now runs as interruptible (5246)
3366 - handle missing objects in filter_preprw_read properly (5265)
3367 - unsafe list practices (rarely) led to infinite eviction loop (4908)
3368 - drop import inflight refcount on signal_completed_replay error (5255)
3369 - unlock page after async write error during send (3677)
3370 - return original error code on reconstructed replies (3761)
3371 - no transno return for symlink open, don't save no-trasno open (3440)
3373 - add pid to ldlm debugging output (4922)
3374 - bump the watchdog timeouts -- we can't handle 30sec yet
3375 - extra debugging for orphan dentry/inode bug (5259)
3377 2004-11-16 Cluster File Systems, Inc. <info@clusterfs.com>
3380 - fix TCP_NODELAY bug, which caused extreme perf regression (5134)
3381 - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
3382 - don't LBUG on obdo_alloc() failure, use OBD_SLAB_ALLOC() (4800)
3383 - fix NULL dereference in /proc/sys/portals/routes (4827)
3384 - allow failed mdc_close() operations to be interrupted (4561)
3385 - stop precreate on OST before MDS would time out on it (4778)
3386 - don't send partial-page writes before EOF from client (4410)
3387 - discard client grant for sub-page writes on large-page clients (4520)
3388 - don't free dentries not owned by NFS code, check generation (4806)
3389 - fix lsm leak if mds_create_objects() fails (4801)
3390 - limit debug_daemon file size, always print CERROR messages (4789)
3391 - use transno after validating reply (3892)
3392 - process timed out requests if import state changes (3754)
3393 - update mtime on OST during writes, return in glimpse (4829)
3394 - add mkfsoptions to LDAP (4679)
3395 - use ->max_readahead method instead of zapping global ra (5039)
3396 - don't interrupt __l_wait_event() during strace
3398 - add software watchdogs to catch hung threads quickly (4941)
3399 - make lustrefs init script start after nfs is mounted
3400 - fix CWARN/ERROR duplication (4930)
3401 - return async write errors to application if possible (2248)
3402 - add /proc/sys/portal/memused (bytes allocated by PORTALS_ALLOC)
3403 - print NAL number in %x format (4645)
3404 - update barely-supported suse-2.4.21-171 series (4842)
3405 - support for sles 9 %post scripts
3406 - support for building 2.6 kernel-source packages
3407 - support for sles km_* packages
3409 2004-10-07 Cluster File Systems, Inc. <info@clusterfs.com>
3412 - ignore -ENOENT errors in osc_destroy (3639)
3413 - notify osc create thread that OSC is being cleaned up (4600)
3414 - add nettype argument for llmount in #5d in conf-sanity.sh (3936)
3415 - reconstruct ost_handle() like mds_handle() (4657)
3416 - create a new thread to do import eviction to avoid deadlock (3969)
3417 - let lconf resolve symlinked-to devices (4629)
3418 - don't unlink "objects" from directory with default EA (4554)
3419 - hold socknal file ref over connect in case target is down (4394)
3420 - allow more than 32000 subdirectories in a single directory (3244)
3421 - fix blocks count for O_DIRECT writes (3751)
3422 - OST returns ENOSPC from object create when no space left (4539)
3423 - don't send truncate RPC if file size isn't changing (4410)
3424 - limit OSC precreate to 1/2 of value OST considers bogus (4778)
3425 - bind to privileged port in socknal and tcpnal (3689)
3427 - rate limit CERROR/CWARN console message to avoid overload (4519)
3428 - GETFILEINFO dir ioctl returns LOV EA + MDS stat in 1 call (3327)
3429 - basic mmap support (3918)
3430 - kernel patch series update from b1_4 (4711)
3432 2004-09-16 Cluster File Systems, Inc. <info@clusterfs.com>
3435 - avoid crash during MDS cleanup with OST shut down (2775)
3436 - fix loi_list_lock/oig_lock inversion on interrupted IO (4136)
3437 - don't use bad inodes on the MDS (3744)
3438 - dynamic object preallocation to improve recovery speed (4236)
3439 - don't hold spinlock over lock dumping or change debug flags (4401)
3440 - don't zero obd_dev when it is force cleaned (3651)
3441 - print grants to console if they go negative (4431)
3442 - "lctl deactivate" will stop automatic recovery attempts (3406)
3443 - look for existing locks in ldlm_handle_enqueue() (3764)
3444 - don't resolve lock handle twice in recovery avoiding race (4401)
3445 - revalidate should check working dir is a directory (4134)
3447 - don't always mark "slow" obdfilter messages as errors (4418)
3449 2004-08-24 Cluster File Systems, Inc. <info@clusterfs.com>
3452 - don't close LustreDB during write_conf until it is done (3860)
3453 - fix typo in lconf for_each_profile (3821)
3454 - allow dumping logs from multiple threads at one time (3820)
3455 - don't allow multiple threads in OSC recovery (3812)
3456 - fix debug_size parameters (3864)
3457 - fix mds_postrecov to initialize import for llog ctxt (3121)
3458 - replace config semaphore with spinlock (3306)
3459 - be sure to send a reply for a CANCEL rpc with bad export (3863)
3460 - don't allow enqueue to complete on a destroyed export (3822)
3461 - down write_lock before checking llog header bitmap (3825)
3462 - recover from lock replay timeout (3764)
3463 - up llog sem before sending rpc (3652)
3464 - reduce ns lock hold times when setting kms (3267)
3465 - change a dlm LBUG to LASSERTF, to maybe learn something (4228)
3466 - fix NULL deref and obd_dev leak on setup error (3312)
3467 - replace some LBUG about llog ops with error handling (3841)
3468 - don't match INVALID dentries from d_lookup and spin (3784)
3469 - hold dcache_lock while marking dentries INVALID and hashing (4255)
3470 - fix invalid assertion in ptlrpc_set_wait (3880)
3472 - add libwrap support for the TCP acceptor (3996)
3473 - add /proc/sys/portals/routes for non-root route listing (3994)
3474 - allow setting MDS UUID in .xml (2580)
3475 - print the stack of a process that LBUGs (4228)
3477 2004-07-14 Cluster File Systems, Inc. <info@clusterfs.com>
3480 - don't cleanup request in ll_file_open() on failed MDS open (3430)
3481 - make sure to unset replay flag from failed open requests (3440)
3482 - if default stripe count is 0, use OST count for inode size (3636)
3483 - update parent mtime/ctime on client for create/unlink (2611)
3484 - drop dentry ref in ext3_add_link from open_connect_dentry (3266)
3485 - free recovery state on server during a forced cleanup (3571)
3486 - unregister_reply for resent reqs (3063)
3487 - loop back devices mounting and status check on 2.6 (3563)
3488 - fix resource-creation race that can provoke i_size == 0 (3513)
3489 - don't try to use bad inodes returned from MDS/OST fs lookup (3688)
3490 - more debugging for page-accounting assertion (3746)
3491 - return -ENOENT instead of asserting if ost getattr+unlink race (3558)
3492 - avoid deadlock after precreation failure (3758)
3493 - fix race and lock order deadlock in orphan handling (3450, 3750)
3494 - add validity checks when grabbing inodes from l_ast_data (3599)
3496 - add /proc/.../recovery_status to obdfilter (3428)
3497 - lightweight CDEBUG infrastructure, debug daemon (3668)
3498 - change default OSC RPC parameters to be better on small clusters
3499 - turn off OST read cache for files smaller than 32MB
3500 - install man pages and include them in rpms (3100)
3501 - add new init script for (un)mounting lustre filesystems (2593)
3502 - run chkconfig in %post for init scripts (3701)
3503 - drop scimac NAL (unmaintained)
3505 2004-06-17 Cluster File Systems, Inc. <info@clusterfs.com>
3508 - clean kiobufs before and after use (3485)
3509 - strip trailing '/'s before comparing paths with /proc/mounts (3486)
3510 - remove assertions to work around "in-flight rpcs" recovery bug (3063)
3511 - change init script to fail more clearly if not run as root (1528)
3512 - allow clients to reconnect during replay (1742)
3513 - fix ns_lock/i_sem lock ordering deadlock for kms update (3477)
3514 - don't do DNS lookups on NIDs too small for IP addresses (3442)
3515 - re-awaken ptlrpcd if new requests arrive during check_set (3554)
3516 - fix cond_resched (3554)
3517 - only evict unfinished clients after recovery (3515)
3518 - allow bulk resend, prevent data loss (3570)
3519 - dynamic ptlrpc request buffer allocation (2102)
3520 - don't allow unlinking open directory if it isn't empty (2904)
3521 - set MDS/OST threads to umask 0 to not clobber client modes (3359)
3522 - remove extraneous obd dereference causing LASSERT failure (3334)
3523 - don't use get_cycles() when creating temp. files on the mds (3156)
3524 - hold i_sem when setting i_size in ll_extent_lock() (3564)
3525 - handle EEXIST for set-stripe, set proper directory name (3336)
3527 - servers can dump a log evicting a client - lustre.dump_on_timeout=1
3528 - fix ksocknal_fmb_callback() error messages (2918)
3530 2004-05-27 Cluster File Systems, Inc. <info@clusterfs.com>
3533 - don't copy lvb into (possibly NULL) reply on error (2983)
3534 - don't deref dentry after dput, don't free lvb on error (2922)
3535 - use the kms to determine writeback rpc length (2947)
3536 - increment oti_logcookies when osc is inactive (2948)
3537 - update client's i_blocks count via lvb messages (2543)
3538 - handle intent open/close of special files properly (1557)
3539 - mount MDS with errors=remount-ro, like obdfilter (2009)
3540 - initialize lock handle to avoid ASSERT on error cleanup (3057)
3541 - don't use cancelling-locks' kms values (2947)
3542 - use highest lock extent for kms, not last one (2925)
3543 - don't dereference ERR_PTR() dentry in error handling path (3107)
3544 - fix thread race in portals_debug_dumplog() (3122)
3545 - create lprocfs device entries at setup instead of at attach (1519)
3546 - common AST error handler, don't evict client on completion race (3145)
3547 - zero nameidata in detach_mnt in 2.6 (3118)
3548 - verify d_inode after revalidate_special is valid in 2.6 (3116)
3549 - use lustre_put_super() to handle zconf unmounts in 2.6 (3064)
3550 - initialize RPC timeout timer earlier for 2.6 (3219)
3551 - don't dereference NULL reply buffer if mdc_close was never sent (2410)
3552 - print nal/nid for unknown nid (3258)
3553 - additional checks for oscc recovery before doing precreate (3284)
3554 - fix ll_extent_lock() error return code for 64-bit systems (3043)
3555 - don't crash in mdc_close for bad permissions on open (3285)
3556 - zero i_rdev for non-device files (3147)
3557 - clear page->private before handing to FS, better assertion (3119)
3558 - tune the read pipeline (3236)
3559 - fix incorrect decref of invalidated dentry (2350)
3560 - provide read-ahead stats and refine rpc in flight stats (3328)
3561 - don't hold journal transaction open across create RPC (3313)
3562 - update atime on MDS at close time (3265)
3563 - close LDAP connection when recovering to avoid server load (3315)
3564 - update iopen-2.6 patch with fixes from 2399,2517,2904 (3301)
3565 - don't leak open file on MDS after open resend (3325)
3566 - serialize filter_precreate and filter_destroy_precreated (3329)
3567 - loop device shouldn't call sync_dev() for nul device (3092)
3568 - clear page cache after eviction (2766)
3569 - resynchronize MDS->OST in background (2824)
3570 - refuse to mount the same filesystem twice on same mountpoint (3394)
3571 - allow llmount to create routes for mounting behind routers (3320)
3572 - push lock cancellation to blocking thread for glimpse ASTs (3409)
3573 - don't call osc_set_data_with_check() for TEST_LOCK matches (3159)
3574 - fix rare problem with rename on htree directories (3417)
3576 - allow default OST striping configuration per directory (1414)
3577 - fix compilation for qswnal for 2.6 kernels (3125)
3578 - increase maximum number of MDS request buffers for large systems
3579 - change liblustreapi to be useful for external progs like lfsck (3098)
3580 - increase local configuration timeout for slow disks (3353)
3581 - allow configuring ldlm AST timeout - lustre.ldlm_timeout=<seconds>
3583 2004-03-22 Cluster File Systems, Inc. <info@clusterfs.com>
3586 - fixes for glimpse AST timeouts / incorrectly 0-sized files (2818)
3587 - don't overwrite extent policy data in reply if lock was blocked (2901)
3588 - drop filter export grants atomically with removal from device (2663)
3589 - del obd_self_export from work_list in class_disconnect_exports (2908)
3590 - don't LBUG if MDS recovery times out during orphan cleanup (2530)
3591 - swab reply message in mdc_close, other PPC fixes (2464)
3592 - fix destroying of named logs (2325)
3593 - overwrite old logs when running lconf --write_conf (2264)
3594 - bump LLOG_CHUNKSIZE to 8k to allow for larger clusters (2306)
3595 - fix race in target_handle_connect (2898)
3596 - mds_reint_create() should take same inode create lock (2926)
3597 - correct journal credits calculated for CANCEL_UNLINK_LOG (2931)
3598 - don't close files for self_export to avoid uninitialized obd (2936)
3599 - allow MDS with the same name as client node (2939)
3600 - hold dentry reference for closed log files for unlink (2325)
3601 - reserve space for all logs during transactions (2059)
3602 - don't evict page beyond end of stripe extent (2925)
3603 - don't oops on a deleted current working directory (2399)
3604 - handle hard links to targets without a parent properly (2517)
3605 - don't dereference NULL lock when racing during eviction (2867)
3606 - don't grow lock extents when lots of conflicting locks (2919)
3608 2004-03-04 Cluster File Systems, Inc. <info@clusterfs.com>
3611 - account for cache space usage on clients to avoid data loss (974)
3612 - lfsck support in lustre kernel code (2349)
3613 - reduce journal credits needed for BRW writes (2370)
3614 - orphan handling to avoid losing space on client/server crashes
3615 - ptlrpcd can be blocked, stopping ALL progress (2477)
3616 - use lock value blocks to assist in proper KMS, faster stat (1021)
3617 - takes i_sem instead of DLM locks internally on obdfilter (2720)
3618 - recovery for initial connections (2355)
3619 - fixes for mds_cleanup_orphans (1934)
3620 - abort_recovery crashes MDS in b_eq (mds_unlink_orphan) (2584)
3621 - block all file creations until orphan recovery completes (1901)
3622 - client remove rq_connection from request struct (2423)
3623 - conf-sanity test_5, proper cleanup in umount log not availale (2640)
3624 - recovery timer race (2670)
3625 - mdc_close recovey bug (2532)
3626 - ptlrpc cleanup bug (2710)
3627 - mds timeout on local locks (2588)
3628 - namespace lock held during RPCs (2431)
3629 - handle interrupted sync write properly (2503)
3630 - don't try to handle a message that hasn't been replied to (2699)
3631 - client assert failure during cleanup after abort recovery (2701)
3632 - leak mdc device after failed mount (2712)
3633 - ptlrpc_check_set allows timedout requests to complete (2714)
3634 - wait for inflight reqs when ptlrpcd finishes (2710)
3635 - make sure unregistered services are removed from the srv_list
3636 - reset bulk XID's when resending them (caught by 1138 test)
3637 - unregister_bulk after timeout
3638 - fix lconf error (2694)
3639 - handle write after unfinished setstripe, stripe-only getstripe (2388)
3640 - readahead locks pages, leaves pending causing memory pressure (2673)
3641 - increase OST request buffers to 4096 on large machines (2729)
3642 - fix up permission of existing directories in simple_mkdir (2661)
3643 - init deleted item, add assertions ptlrpc_abort_inflight() (2725)
3644 - don't assign transno to errored transactions (2742)
3645 - don't delete objects on OST if given a bogus objid from MDS (2751)
3646 - handle large client PAGE_SIZE readdir on small PAGE_SIZE MDS (2777)
3647 - if rq_no_resend, then timeout request after recovery (2432)
3648 - fix MDS llog_logid record size, 64-bit array alignment (2733)
3649 - don't call usermode_helper from ptlrpcd, DEFAULT upcall (2773)
3650 - put magic in mount.lustre data, check for bad/NULL mount data (2529)
3651 - MDS recovery shouldn't delete objects that it has given out (2730)
3652 - if enqueue arrives after completion, don't clobber LVB (2819)
3653 - don't unlock pages twice when trigger_group_io returns error (2814)
3654 - don't deref NULL rq_repmsg if ldlm_handle_enqueue failed (2822)
3655 - don't write pages to disk if there was an error (1450)
3656 - don't ping imports that have recovery disabled (2676)
3657 - take buffered bytes into account when balancing socknal conn (2817)
3658 - hold a DLM lock over readdir always, use truncate_inode_pages (2706)
3659 - reconnect unlink llog connection after MDS reconnects to OST (2816)
3660 - remove little-endian swabbing of llog records (1987)
3661 - set/limit i_blksize to LL_MAX_BLKSIZE on client (2884)
3662 - retry reposting request buffers if they fail (1191)
3663 - grow extent at grant time to avoid granting a revoked lock (2809)
3664 - lock revoke doesn't evict page if covered by a second lock (2765)
3665 - disable VM readahead to avoid reading outside lock extents (2805)
3667 - return LL_SUPER_MAGIC from statfs for the filesystem type (1972)
3668 - updated kernel patches for hp-2.4.20 kernel (2681)
3670 2004-02-07 Cluster File Systems, Inc. <info@clusterfs.com>
3673 - fix truncated write corruption (2366)
3674 - fix for failed assertion in iopen_connect_dentry (1792,2517)
3676 - don't flag the ptlrpcd thread with PF_MEMALLOC (2636)
3677 - ensure len(uuid) < 37 in lmc (1171)
3678 - fix ia64 OOPS in llog_test (2255)
3679 - zero end of page at obdfilter for partial page writes (2648)
3680 - don't leave stale dentries around after renames (bug 2428)
3681 - fix timeouts when evicting a client with a single lock held (2642)
3682 - set deadline for the initial HELLO message to drain (2634)
3683 - print out dotted-quad IP addresses in the socknal (2302)
3685 - additional debugging for MDS client eviction problem (2443)
3686 - fix mkfsoptions support for osts (2603, 2604)
3688 2004-01-27 Cluster File Systems, Inc. <info@clusterfs.com>
3691 - add series for the vanilla 2.6.0 kernel
3692 - add series for the vanilla 2.4.24 kernel
3693 - add series for a cray x86/64 UL kernel drop
3694 - fix xattr patches for the vanilla 2.4.19 series
3696 - generate true UUIDs in lmc (1171)
3697 - have portals stack dumping break in UML (2466)
3698 - avoid bad dchild deref; avoid inum lock w/o creation (2362)
3699 - allocate with _NOFS in ldlm to avoid deadlock (1933)
3700 - wake callback waiting threads on client eviction (2460)
3701 - Add --ptldebug and --subsystem to lmc (1719)
3702 - update assertion to allow safe interrupt allocation
3703 - set rq_no_resend for cancel requests (2432)
3704 - recalculate ptlrpcd timeout after resend (2494)
3705 - call vfs_rmdir when removing pending directories (2368)
3706 - fix renaming a file to itself (2429)
3707 - lmc creates a default one-stripe lov (2454)
3708 - expand procfs space to handle large clusters (2326)
3709 - increase UML stack to avoid overflow
3710 - update lconf's list of debug and subsystem masks
3711 - fix lfs find --obd (2510)
3712 - /proc tunable for disabling filter read caching (2591)
3713 - stop rpm packages from altering slapd.conf (2301)
3714 - disable nagle in the socknal under 0conf (2578)
3715 - choose mds inode size based on stripe count (2572)
3716 - fix kernel-source rpm problems (2516)
3718 - add --disable-doc to avoid pdf generation (2421)
3719 - update documentation, tests, type-os, comments
3720 - avoid format warnings on ia64
3721 - remove the TOE NAL
3722 - tiny code cleanups by removing unused fields
3724 2004-01-07 Cluster File Systems, Inc. <info@clusterfs.com>
3727 - fix obvious semaphore misuse in as-yet-unused setattr path (2348)
3728 - remove the most blatant lies from BUILDING file (2371)
3729 - change default debug level to reasonable production setting
3730 - reduce client side cache size to reduce cache flush time
3731 - reduce max RPCs in flight to avoid unnecessary file fragmentation
3732 - make TCP zerocopy and pinger support enabled by default (2476)
3733 - sync writes completed after process exits caused crashes (2319)
3734 - maintain correct mount count on the MDS (2356)
3735 - backout 1557, because 2316 wasn't really fixed
3736 - better file I/O statistics gathering in /proc
3737 - don't take unnecessary, deadlock-inducing bug in readpage (2383)
3738 - another kernel patch to fix zero-copy TCP function export
3739 - don't take duplicate lock when processing re-sent getattr (2420)
3740 - lctl uses obd_self_export instead of creating new conn (2353)
3741 - MDS/OST recovery case which requires object creation asserted (2425)
3742 - move lfs from /usr/sbin to /usr/bin in packages
3743 - fix race between mds_client_add and mds_client_free (2417)
3744 - use kmalloc instead of slabs in portals (2430)
3745 - don't create duplicate records when a failover MDS is present (2442)
3746 - remove unnecessary mount age check (2332)
3747 - don't remove directory inodes from locks prematurely (2451)
3748 - don't break if MDS service name is the same as hostname (2103)
3749 - fix races in client write RPC generation when cache full (2482)
3751 2003-12-13 Cluster File Systems, Inc. <info@clusterfs.com>
3754 - remove now-unused request->rq_obd (278)
3755 - if an allocation fails, print out how much memory we've used (1933)
3756 - use PORTAL_SLAB_ALLOC for structures, to get GFP_MEMALLOC (1933)
3757 - add the "configurable stack size" patch to most series files (1256)
3758 - ability to write large log records, for 100+ OST configs (2306)
3759 - fix NULL deref when filter_prep fails (2314)
3760 - fix operator precedence error in filter_sync
3761 - dynamic allocation of socknal TX descriptors (2315)
3762 - fix a missed case in the GFP_MEMALLOC patch, can cause deadlock (2310)
3763 - fix gcc 2.96 compilation problem in xattr kernel patch (2294)
3764 - ensure that CWARN messages in Portals always get to the syslog
3765 - __init/__exit are not for prototype decls (ldlm_init/exit)
3766 - x86-64 compile warning fixes
3767 - fix gateway LMC keyword conflict (2318)
3768 - fix MDS lock inversions in getattr/reint paths (1844)
3769 - fix a rare lock re-ordering bug, which caused deadlock (2322)
3770 - fix i_sem/journal inversion in fsfilt_ext3_write_record (2306)
3771 - DLM race condition prevented some lock evictions (2328)
3772 - ENOMEM detection and retry on socknal sends (2230)
3773 - use GFP_NOFS throughout Lustre, to combat ENOMEM (2230)
3774 - move osc_rpcd into ptlrpc, for use in MDC and others (2329)
3775 - protect MDS inode fsdata with stronger locking; fixes assertion (2313)
3776 - better error messages when a client is rejected during recovery (1505)
3777 - avoid cancelling locks which were never granted, after failure (2330)
3778 - fix i_sem/journal inversion in mds_client_add (2333)
3779 - fix truncate/getattr lock cycle deadlock (2334)
3780 - use rpcd to send close; allows resend after timeout, avoid leak (1897)
3781 - fix two rare exit paths which could leak an l_lock() ref (2321)
3782 - fencepost error in MDS/OST orphan recovery (2226)
3783 - make log record alignment 8 bytes (1988)
3784 - lstripe now fails when requested offset > ost_count (2237)
3785 - ensure that all kernel series have a complete list.h (1607)
3786 - fix crashes in special-file operations (2316)
3787 - lctl create/brw OID mismatch, caused by obsolete filter loop (2339)
3789 - allow configurable automake binary, for testing new versions
3790 - small update to the lfs documentation
3792 2003-12-03 Cluster File Systems, Inc. <info@clusterfs.com>
3794 * fix negative export reference count in fsfilt_sync (2312)
3796 2003-12-01 Cluster File Systems, Inc. <info@clusterfs.com>
3797 * release candidate 0.9.1
3799 - orphans are moved into the PENDING directory for possible recovery
3800 - replayed opens now open by fid for orphan/rename safety (1042)
3801 - last close of an orphan inode generates a transno (683)
3802 - chdir() and mount() now pin the directory entry (1020)
3803 - avoid CERROR in normal ll_setattr_raw() error case (1500)
3804 - discard very old requests without processing them (1502)
3805 - remove some common, well-understood CERRORs (1505)
3806 - require O_DIRECT I/O to be page-sized to workaround IA64 crash (1609)
3807 - clear "grant" flags in OST replies until OST grant code lands (1644)
3808 - fix read performance by not clobbering i_blksize on client (1598)
3809 - fix __ldlm_handle2lock oops by not dereferencing lock after PUT (1625)
3810 - make LRU size a /proc tunable, clears locks when reduced (707)
3811 - fix some lprocfs rot that prevented ptlbd from loading (1732)
3812 - server locks take references on exports now (1558)
3813 - build fixes for 2.4.20-rh trees (1663)
3814 - return an error from lov_create if all OSCs are inactive (1751)
3815 - fix import levels when a reconnect happens without a timeout (1597)
3816 - exit early from mds_open if we get a lookup error (1749)
3817 - partial page read at EOF wouldn't wait for disk before sending (1642)
3818 - avoid NULL deref in obdfilter when reading page past EOF (1592)
3819 - avoid LASSERT in ll_intent_lock if server failed very early (1090)
3820 - fix LBUG in ll_it_open_error with rc = -2 (1861)
3821 - write/truncate lock inversion (1639)
3822 - Don't auto-load obdclass, portals modules during cleanup (1495)
3823 - fix timestamps from jumping to "now" (1763)
3824 - extra journal assertions (1648)
3825 - add an extra multiunlink test (1771)
3826 - fix read_record/write_record API (1776)
3827 - fix leak of offset_extent, possible incorrect i_size later (1772)
3828 - fix lasserts in mis-matched transnos during open-unlink testing (1541)
3829 - Debugging for the kqswnal_get_idle_tx problems (1820)
3830 - Allow recovery to be attempted multiple times (1536)
3831 - Write out MDS last_rcvd file after it is first created (1600)
3832 - Fix tx_descriptor leak in failed transmit situations (1827)
3833 - ext3 journaling fixes for assertion failure after IO error (1871)
3834 - class_export_put() on freed export after completion AST error (1896)
3835 - Fix revalidate looping in VFS (1322)
3836 - Don't access a freed export during MDS_REINT timeout (1521)
3837 - Add open-unlink recovery support on the MDS (1673,1764)
3838 - Return an error if no MDS data was read from last_rcvd (1946)
3839 - Fix for lookup "." or ".." crash on error (1932,1931,1935)
3840 - Don't setup a disk device that doesn't match exported UUID (317)
3841 - Reduce bulk RPC timeout to avoid cascading client/OST failures (1845)
3842 - avoid committing NULL handle in force close
3843 - local.sh is now a one-stripe LOV configuration
3844 - POSIX utime.4 -EPERM on FIFO not owned by user (56)
3845 - fix ext3 htree duplicate directory entry corruption (1516)
3846 - POSIX creat.13, fstat.1, open.18, stat.3 new file atime/mtime (2020)
3847 - update to new LOV EA format (2097)
3848 - interoperability for different PAGE_SIZE/wordsize (686,1821,1343,2042)
3850 2003-06-15 Phil Schwan <phil@clusterfs.com>
3853 - imports and exports cleanup too early, need refcounts (349, 879, 1045)
3854 - per-import/export recovery handling (958, 931, 959)
3855 - multiple last-rcvd slots, for serving multiple FSes (949)
3856 - connections are again shared between multiple imp/exports (963, 964)
3857 - "umount -f" would hang if any requests needed to be sent (393, 978)
3858 - avoid pinning large req buffer by copying for queued messages (989)
3859 - add "uuid" to "lctl device" command to help upcalls (991)
3860 - "open" RPCs with transnos would confuse recovery counters (1037)
3861 - do proper endian conversion of all wire messages (288, 340, 891)
3862 - remove OST bulk get LBUGs, fix ost_brw_write cleanup (1126)
3863 - call waiting locks callback from LDLM recovery thread (1127, 1151)
3864 - fix ptlrpc_connection leak in target_handle_connect (1174)
3865 - fix import refcounting bug in OST and MDS cleanup (1134)
3866 - if an invalid-at-open-time OSC returned before close(), LBUG (1150)
3867 - fix very unlikely obd_types race condition (501)
3868 - remove osc_open hack for echo_client (1187)
3869 - we leaked exports/dlmimps for forcibly disconnected clients (1143)
3870 - a failure in read_inode2 leads to deadlock (1139)
3871 - cancel ack-locks as soon as transaction is committed (1072)
3872 - fix major leaks and crashes in the bulk I/O path (937, 1057)
3873 - make sure to commitrw after any preprw to avoid deadlock (1162)
3874 - failing to execute a file in a lustre FS would lock inode (1203)
3875 - small DEBUG_REQ fix to avoid dereferencing a NULL (1227)
3876 - don't ASSERT while cleaning up an incompletely-setup obd (1248)
3877 - obd_uuid2tgt would walk off the end of the list (1255)
3878 - on IA64 the osc would give portals incorrect bulk size (1258)
3879 - fix debug daemon ioctl interface; allows daemon on ia64 (1274)
3880 - fix lock inversion caused by new llite matching code (1282)
3881 - limit the number of dirty pages on a client to 10MB (1286)
3882 - timed out locks were not being corrected cancelled (1289)
3883 - fix O_DIRECT above 4GB on IA-32 (1292)
3884 * major user-visible changes
3885 - fail out/fail over policy now controlled by the upcall (993)
3887 - add OBD_PING to check server availability and failure (954)
3888 - lustre messages are now sent in sending host order (288, 340, 891)
3889 - add eadatalen to MDS getattr reply (340)
3890 - OST read replies may contain second buffer, with per-page status (593)
3892 2003-03-11 Phil Schwan <phil@clusterfs.com>
3895 - LDLM_DEBUG macro fix, for gcc 3.2 (850)
3896 - failed open()s could cause deadlock; fixed (867, 869)
3897 - stop cancelling OST locks when files are closed (481)
3898 - overlapping XID spaces caused network corruption (851, 853)
3899 - fix unsafe fsfilt counter arithmetic; change to atomic_t
3900 - setattr_raw added, to do single-RPC, server-side setattrs
3901 - lmc/lconf syntax change for OST UUIDs
3902 - fix crashy race condition between ptlrpc_free_req and osc_close
3903 - don't use request in mdc_enqueue if we hit a timeout (889)
3904 - don't set the inode i_size for regular files from the MDS (896)
3905 - handle out of order completion AST (842)
3906 - don't LBUG if a lock request times out after receiving AST (913)
3907 - avoid d_rehash race in ll_find_alias by rehashing inside dcache_lock
3908 - if a bad lock AST arrives, send an error instead of dropping entirely
3909 - return 0 from revalidate2 if ll_intent_lock returns -EINTR (912)
3910 - fix leak in bulk IO when only partially completed (899, 900, 926)
3911 - fix O_DIRECT for ia64 (55)
3912 - (almost) eliminate Lustre-kernel-thread effects on load average (722)
3913 - C-z after timeout could hang a process forever; fixed (977)
3915 - client-side I/O cache (678, 924, 929, 941, 970)
3917 - READPAGE and SETATTRs which don't take server-side locks get
3920 2003-02-11 Phil Schwan <phil@clusterfs.com>
3923 - Fix ldlm_lock_match on the MDS to avoid matching remote locks (592)
3924 - Fix fsfilt_extN_readpage() to read a full page of directory
3925 entries, or fake the remainder if PAGE_SIZE != blocksize (500)
3926 - Avoid extra mdc_getattr() in ll_intent_lock when possible (534, 604)
3927 - Fix imbalanced LOV object allocation and out-of-bound access (469)
3928 - Most intent operations were removed, in favour of a new RPC mode
3929 that does a single RPC to the server and bypasses most of the VFS
3930 - All LDLM resource ID arrays were removed in favour of ldlm_res_id
3931 - Aggressively cancel local locks on DLM servers
3932 - mds_reint_unlink sends EA to the client if it's the last nlink.
3933 client uses that EA to unlink OST objects.
3934 - mds_reint_{rename,unlink,link} were rewritten to take ordered locks
3935 - recursive symlinks were fixed (439)
3936 - fixed NULL deref in DEBUG_REQ
3937 - filter_update_lastobjid no longer calls sync, which annoyed extN
3938 - fixed multi-client small-writes to a single file problem (445)
3939 - fixed mtime updates during file writes (607)
3940 - fixed vector writes on obdfilter causing problems when ENOSPC (670)
3941 - fixed bug in obd_brw_read/write() (under guise of testing 367)
3942 - fixed Linux OST size reporting problem (444, 656)
3943 - OST now updates object mtime with writes or setattr (607, 619)
3944 - client verifies file size before zeroing page past EOF (445)
3945 - OST now writes last allocated objid to disk with allocation (108)
3946 - LOV on echo now works (409)
3948 - mds_reint_unlink sends a new buffer, with the EA included. this
3949 buffer is only valid if body->valid & OBD_MD_FLEASIZE, which is only
3950 set if a regular file was being unlinked, and it was the last link
3951 - use PtlGet from the target for bulk writes (315)
3952 - OST now updates object mtime with writes or setattr (607, 619)
3953 - LDLM now has a grant-time callback to revalidate locked items, if
3955 - Many MDS operations were reorganized to combat race conditions
3957 - Merge b_intel branch (updated lprocfs code) - now at /proc/fs/lustre
3958 - configure check to avoid gcc version 2.96 20000731-2.96-98) (606)
3960 2003-01-06 Andreas Dilger <adilger@clusterfs.com>
3963 - Fully reactivate OST imports after reconnection (512, others)
3964 - Make sure client sees our -ENOTCONN from mds_handle (513 - partial)
3965 - More graceful error handling for truncating on dead OST (515)
3966 - Don't error out unless we're actually accessing dead stripes (474)
3967 - Fix garbage sizes when stripes are missing (410)
3968 - LRU counters were broken, causing constant lock purge (433, 432)
3969 - garbage on read from stripes with failed OSTs (441)
3970 - mark OSCs as active before reconnecting during recovery (438)
3971 - lov_enqueue and lov_cancel need to handle inactive OSTs (403)
3972 - lfind did not preserve OST order in output (443)
3973 - symlinks cause hung clients, incorrect data (439)
3974 - stop dereferencing request after dropping refcount (457)
3975 - don't LASSERT(spin_is_locked) on non-SMP (455)
3976 - fixes for many rename() bugs
3977 - fstat didn't correctly synchronize attributes (399)
3978 - server must handle lock cancellation during blocking AST prep (487)
3979 - bulk descriptors were free()d too soon (511)
3980 - fix paths in lconf, which would load incorrect modules (451, 507)
3981 - fix confusing lconf 'host not found' error message (386)
3982 - fix lock order deadlock on OST (O/R i_sem before journal ops, 478)
3983 - fix race condition in mdc_blocking_ast() for inode access (526)
3984 - fix lov_unpackmd() unpacking wrong number of stripes (537)
3985 - fix lov_set_osc_active() marking wrong OSC inactive (440)
3986 - fix bad lstripe lov_unpackmd() assertion (fix layering too) (527)
3987 - fix multiple writes of stripe MD to MDS (358, maybe 519)
3988 - fix lstripe in several ways (kernel side) (527)
3989 - fix request leak in ldlm_cli_enqueue (262)
3990 - incorrect OSC was marked inactive after OST failure
3991 - call mds_fs_cleanup before unmounting filesystem (524)
3992 - fix races between taking ns_lock and ldlm_lock_change_resource
3993 - fix races updating LOV export open file list
3994 - fix lov_enqueue error path, avoid decref-ing bad lock handle
3995 - fix recovery NULL deref in ldlm_cli_cancel_unused
3996 - fix some DLM races by using new hash table for lock handles (419)
3997 - permit the client to specify desired inodes, at replay
3998 - duplicate requests when we queue them for replay reintegration
3999 - fix last_rcvd offset calculation
4000 - sync after each recovered transaction, so we always make progress
4001 - never, not always, ERESTART requests without transnos
4002 - store the lov_desc in the MDS, so we don't depend on getlovinfo to
4004 - skip replay if the MDS says that the client is already connected
4005 - don't check for a recovery-enabled export to match lctl's UUID
4006 - don't INC_USE_COUNT for phantom exports
4007 - don't crash when cleaning up phantom exports (567)
4008 - don't double-finish or set replay data for errored mdc_open requests
4009 - abort requests when they time out, so we don't get old replies
4010 - send/receive replies for AST messages again
4011 - if the client says that it doesn't have the lock, cancel it on the
4013 - if we timeout during I/O, don't try to cancel an in-use lock; instead
4014 mark it as destroyed, it will all work out when decref is called
4015 - fix module use counts (22, 581)
4017 - ASTs now expect a reply (server cancels lock on error reply)
4019 2002-12-02 Andreas Dilger <adilger@clusterfs.com>
4022 - fix many simultaneous client startup (392)
4023 - fix dentry->d_it clobbering
4024 - credentials weren't being shipped for readdir/getattr operations
4025 - remove invalid assertions triggered during some concurrent MD
4027 - proper Lustre versions added (336, 389)
4028 - fix memory leak for create error case (398)
4029 - fix LOV locking bug that would get cli/srv out of sync
4030 - fix echo client over LOV (409)
4031 - fix dbench 2, extN refcount problem (170, 258, 356, 418)
4032 - fix double-O_EXCL intent crash (424)
4033 - avoid sending multiple lock CANCELs (352)
4035 - MDS can do multi-client recovery (modulo bugs in new code)
4037 - many updates, edits, cleanups
4039 2002-11-18 Phil Schwan <phil@clusterfs.com>
4042 - fix null d_it dereference (346)
4043 - fix full OST/dbench hang (333)
4044 - fix permission problem with file removal (286)
4045 - fix removal of OSCs from LOV when they fail
4046 - fix NULL deref during bulk timeout (214)
4047 - fix problems related to multiple filesystems on one MDS (241)
4048 - fixed serious subtle metadata locking bugs
4049 - free locks on clients when inodes are removed due to memory
4051 - fix inode pointer in lock data (285)
4052 - partial support for multiple MDS on a single host (241)
4053 - data locks weren't cancelled at clear_inode time (290, 311)
4054 - intent locks could lead to unbounded lock growth (205)
4055 - added a maximum lock count, an LRU list, and a flusher
4056 - fix multiple rename (365)
4057 - properly abstracted the echo client
4058 - OSC locked 1 byte too many; fixed
4059 - rewrote brw callback code:
4060 - fixed recovery bugs related to LOVs (306)
4061 - fixed too-many-pages-in-one-write crash (191)
4062 - fixed (again) crash in sync_io_timeout (214)
4063 - probably fixed callback-related race (385)
4065 - Add capability to MDS protocol
4066 - LDLM cancellations and callbacks on different portals
4068 2002-10-28 Andreas Dilger <adilger@clusterfs.com>
4071 - limit client IOV size to PTL_MD_MAX_IOV (611336, 191)
4072 - defer open object destruction to close time (601981, 138)
4073 - open/close OST file handle in obdo (OBD_MD_FLHANDLE) (601981, 138)
4074 - move LDLM_ENQUEUE/CONVERT back to MDS portal (625069)
4075 - abstract ll_lookup2, fix ll_revalidate2 to use abstraction (256)
4076 - don't call obd_setattr in ll_file_release for destroyed objects
4077 * protocol change to lustre_msg: move |version| and add |flags|
4078 * protocol change to osc_punch: "start" in "o_size", "end" in "o_blocks"
4079 * lock replay: for LDLM_FL_REPLAY trust client to do right thing
4080 * added replay of create, unlink, link and rename operations during
4081 MDS failover; recovery should be much more robust now
4082 * remove failed OSCs from LOVs (only lov_create uses this so far)
4083 * the lustre-HOWTO was brought (more) up to date (582544)
4085 2002-10-23 Phil Schwan <phil@clusterfs.com>
4088 - in-use dentries weren't being reused properly (617851)
4089 - prevent multiple LDLM setup (599178)
4090 - fix LOV size calculations for truncate (617853)
4091 - fix client handling of MDS intent errors (POSIX)
4092 - fix permission bug in lovstripe.c test (624321)
4093 - fix MDS thread deadlock - move LDLM handler to DLM portal (625069)
4094 - truncate past end of file could corrupt data
4095 - proper cleanup after timeouts, crashes, etc (592524, 550815)
4096 - a race in recovery could return ETIMEDOUT to apps (623947)
4097 - building outside the source directory was fixed
4098 * the lustre-HOWTO was brought (more) up to date (582544)
4099 * major progress was made on recovery functionality
4101 2002-10-10 Phil Schwan <phil@clusterfs.com>
4104 - recovery deadlock fix
4105 - rm -rf causes LBUG fix (617817)
4106 - file open by multiple tasks fix (618962)
4107 - directory permissions bugs (602707 and 620007)
4108 - journal_stop fixed with locking (611313)
4109 - O_APPEND failures resolved (618273, perhaps 614459)
4110 - lconf PATH fix (619770)
4111 - IA64 build fix (621450)
4112 - RPC buffer sizes scale with amount of memory
4114 2002-10-01 Phil Schwan <phil@clusterfs.com>
4117 - locks would be cancelled without throwing away data pages,
4118 resulting in inconsistent data (605627)
4119 - inode attributes were not always being refreshed (605627, 612449)
4120 - lconf now continues to cleanup after lctl reports an error
4121 - MDS now enforces user permissions (602707)
4122 - lprocfs cleanup fixed, but not yet enabled (614157)
4123 - fixed infinite server hang, should a client not respond to an AST
4124 - avoid going into recovery if user calls readlink() with a buffer
4125 that's too small (613941)
4126 - AST RPCs no longer require replies (614867) -- this may be changed
4127 - don't crash server if client sends an IOV that's too big (611336)
4128 - fixed lock conversion deadlock (611892)
4129 - fixed the following of symlinks (614622)
4130 * recovery: the server can remove locks from a client that dies, other
4131 clients can make progress
4132 * more extN patch fixes
4133 * compile-time configurable ptlrpc buffer allocations
4135 - collaborative read cache document
4136 - Lustre Lite Performance CDR document-in-progress
4138 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4141 - fix typo in patch-2.4.18
4143 2002-09-20 Andreas Dilger <adilger@clusterfs.com>
4146 - clear ptlrpc request each time in handle_incoming_request()
4147 - unlink of files now destroys the object on the OST
4149 2002-09-19 Peter Braam <braam@clusterfs.com>
4151 * add hard link support
4152 * change obdfile creation method
4153 * kernel patch changed
4155 2002-09-19 Peter Braam <braam@clusterfs.com>
4158 - stack overflow bug in extN fixed
4160 2002-09-18 Andreas Dilger <adilger@clusterfs.com>
4162 * documentation updates
4163 - add man pages for config tools
4164 - update tests/README to describe testing with new config tools
4165 - finish metadata API descriptions
4166 * bug fixes and cleanups
4167 - statfs workaround for 16TB limit
4168 - LOV stripe allocation improved, can stripe on subset of OSTs
4169 - LOV file size/IO offset was wrong for files > 4GB in size
4170 - object EA data was being dropped, caused files to be unreadable
4171 - memory overflow with non-LOV OST caused memory corruption
4172 - fixed regression tests to work with new config tools, obdfilter
4173 - fixed bug when directory size became larger than 1 block
4174 - fixed bug (for single client case) when PWD was deleted
4175 - invalidate local directory pages when doing intent-based ops
4176 - avoid LDLM oops when lock callback contained bad data
4178 2002-09-09 Andreas Dilger <adilger@clusterfs.com>
4180 * documentation updates
4181 * bug fixes and cleanups
4182 - configuration tools
4185 - 64-bit compile warnings
4186 - 64-bit internal statfs data
4188 * test_brw on persistent OST devices
4190 * lprocfs (disabled)
4192 2002-09-04 Andreas Dilger <adilger@clusterfs.com>
4194 * documentation updates
4195 * bug fixes and cleanups
4196 * configuration tools
4198 2002-08-30 Peter J. Braam <braam@clusterfs.com>
4201 * many small fixes to 0_5_4
4202 * io/network handling
4203 * thinkos in MDS operations
4205 2002-08-24 Peter J. Braam <braam@clusterfs.com>
4208 * crucial basic fixes to 0.5.3
4209 * IOR, Iozone work over Elan
4212 2002-08-07 Phil Schwan <phil@clusterfs.com>
4213 * version 0_5_3, our first alpha
4214 * we use the new Portals iovs
4215 * documentation updates
4216 * bug fixes and cleanups
4217 * small changes in the DLM wire protocol
4219 2002-07-25 Peter J. Braam <braam@clusterfs.com>
4220 * version 0_5_1 with some initial stability,
4221 * locking on MD and file I/O.
4222 * documentation updates
4223 * several bug fixes since 0.5.0
4224 * small changes in wire protocol
4226 2002-07-18 Phil Schwan <phil@clusterfs.com>
4228 * delivered as Lustre Light Alpha
4229 * fixed a crash after handling invalid MDS requests
4230 * fixed directory pages for architectures with non-4k pages sizes
4232 2002-07-11 Andreas Dilger <adilger@clusterfs.com>
4233 * release version v0_4_4
4234 * Moves TCP acceptor to be on port 2432 (unused Coda port) instead
4236 * Fixes a number of interruption problems with OST operations.
4237 * Update documentation for portals header changes
4238 * Move all wire protocol structs/defines to lustre_idl.h
4239 * Fixes symlink length bug.
4240 * Add tcpdump to repository.
4242 2002-07-05 Andreas Dilger <adilger@clusterfs.com>
4243 * release version v0_4_3
4244 * Fixes statfs for inodes on extN.
4245 * Fixes bug in runtests which would delete /etc/hosts.
4246 * Use 64-bit object IDs wherever possible (not into VFS though)
4247 Remove ost_get_info, which is unused by lustre, and out of date.
4249 2002-07-03 Peter Braam <braam@clusterfs.com>
4250 * release version v0_4_2 Fixes a lookup error (type not passed)
4251 * move forward to head of Portals
4252 * move forward to latest Lustre kernel
4254 2002-06-25 Peter Braam <braam@clusterfs.com>
4255 * release version v0_4_1. Hopefully stable on single node use.